From 308d22cbadc84e45102c23b97d95205dbe81c7b4 Mon Sep 17 00:00:00 2001 From: huppiflupp Date: Wed, 28 Oct 2026 09:56:22 +0100 Subject: [PATCH 01/53] M10: Enhance scanned file cleans scanned pages in place MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Scan & OCR ▸ Enhance scanned file (Acrobat's Enhance ▸ Scanned document) now cleans up scanned pages: background removal (the camera clean-up's level 0-100), edge shadow removal (book gutters, copier borders) and despeckle, each its own switch. The camera tool renders the page and replaces it with a picture, which would drop a scan's invisible OCR text, comments and links. Instead each scan image (pictures covering at least half the page, each at least a tenth) is decoded, cleaned and written back into its own image object at the same size and colour space: a JPEG stays a JPEG (quality 90), other grey/RGB images are stored losslessly, a 1-bit image stays 1-bit with its decode array, and every other dictionary entry (SMask, Interpolate, ...) is kept. An image several pages share is cleaned once. What can't be cleaned (CMYK, indexed, JPEG 2000, JBIG2, CCITT, colour-key masks, broken data, oversized images) is reported, never skipped silently; pages that are no scan are listed too. The engine works as a job like Optimize (scan_clean_job, run on a worker with progress and cancel, apply_scan_clean refuses a result for an older document state) and applies one undoable step, "Enhance scanned file". pdfcraft-optimize exposes decode_samples/encode_samples for in-place pixel edits. camera_clean gains clean_samples for grey and RGB samples with separate switches and an edge-shadow pass; the camera tool's output is unchanged. page_enhance_scan drives it headlessly. Not yet an option of Recognize text. Refs #4119. Co-Authored-By: Claude Opus 5.5 --- ROADMAP.md | 2 + crates/automation/src/dispatch.rs | 29 ++ crates/automation/src/tools.rs | 12 + crates/automation/tests/it/automation.rs | 56 +++ crates/engine/src/camera_clean.rs | 235 ++++++++++-- crates/engine/src/catalog.rs | 2 +- crates/engine/src/commands.rs | 1 + crates/engine/src/edit.rs | 7 + crates/engine/src/edit/labels.rs | 2 + crates/engine/src/edit/run.rs | 2 + crates/engine/src/history.rs | 1 + crates/engine/src/lib.rs | 1 + crates/engine/src/scan_clean.rs | 276 ++++++++++++++ crates/engine/src/scan_clean/tests.rs | 338 ++++++++++++++++++ crates/engine/src/scan_fields.rs | 23 +- crates/optimize/src/images.rs | 10 +- crates/optimize/src/lib.rs | 2 + crates/optimize/src/samples.rs | 273 ++++++++++++++ crates/ui-egui/src/app.rs | 3 + crates/ui-egui/src/app/control.rs | 1 + crates/ui-egui/src/app/frame.rs | 1 + crates/ui-egui/src/commands.rs | 1 + crates/ui-egui/src/dialogs.rs | 10 + crates/ui-egui/src/i18n/de.tsv | 19 + crates/ui-egui/src/lib.rs | 4 + crates/ui-egui/src/state.rs | 2 + crates/ui-egui/src/workflows.rs | 1 + crates/ui-egui/src/workflows/scan_clean_ui.rs | 253 +++++++++++++ crates/ui-egui/tests/it/main.rs | 1 + crates/ui-egui/tests/it/scan_clean.rs | 77 ++++ docs/gaps.md | 5 +- docs/roadmap.md | 3 +- docs/target-app-parity.md | 3 +- parity/acrobat-features.toml | 24 +- 34 files changed, 1613 insertions(+), 67 deletions(-) create mode 100644 crates/engine/src/scan_clean.rs create mode 100644 crates/engine/src/scan_clean/tests.rs create mode 100644 crates/optimize/src/samples.rs create mode 100644 crates/ui-egui/src/workflows/scan_clean_ui.rs create mode 100644 crates/ui-egui/tests/it/scan_clean.rs diff --git a/ROADMAP.md b/ROADMAP.md index f5de8044e1..91e8bf1ece 100644 --- a/ROADMAP.md +++ b/ROADMAP.md @@ -2,6 +2,7 @@ **Stage: alpha** · next: beta, ~25% (ready for real work 50% → 75%) and ~400–700 h away +> **Last reviewed:** 2026-10-28 | **Last updated:** 2026-10-28 | **Change:** minor (M10, #4119 partial: Enhance scanned file cleans scanned pages in place: background removal, edge shadow removal, despeckle; estimates unchanged) > **Last reviewed:** 2026-10-27 | **Last updated:** 2026-10-27 | **Change:** minor (M10: OCR model downloads use verified ONNX pairs with legacy discovery fallback; estimates unchanged) > **Last reviewed:** 2026-10-25 | **Last updated:** 2026-10-25 | **Change:** minor (M14 localization current state: Arabic logical catalog, explicit RTL label direction and mirrored stock chrome; native/browser RTL workflows unqualified. Native release Hans face via LXGW WenKai Lite; estimates unchanged) > **Last reviewed:** 2026-10-23 · **Last updated:** 2026-10-23 · **Change:** minor (M7, #2636 partial: pasting added text boxes and images in the engine and through `content_copy`; estimates unchanged) @@ -200,6 +201,7 @@ Ranked; detail and the full milestone table in [docs/roadmap.md](docs/roadmap.md ## Progress log +- **2026-10-28 (M10, #4119 partial):** Scan & OCR ▸ Enhance scanned file cleans scanned pages in place: background removal (level 0–100, as the camera clean-up), edge shadow removal (book gutters, copier borders) and despeckle. Each scan image is decoded, cleaned and written back into its own object at the same size and colour space (JPEG stays JPEG, others lossless, 1-bit stays 1-bit), so the OCR text layer, comments and links stay; one undo step, shared images once, unsupported images (CMYK, indexed, JPEG 2000, JBIG2, CCITT, broken data) reported. `page_enhance_scan` drives it headlessly. Not yet an option of Recognize text. Estimates unchanged. - **2026-10-27 (M10, OCR models):** Model staging downloads and verifies the ONNX pair after the upstream RTEN URL brownout; runtime discovery prefers complete ONNX pairs per directory, retains complete RTEN fallback and preserves directory precedence. - **2026-10-24 (M10, #4158 partial):** Word, HTML, RTF and PowerPoint export keep a block that continues another on its line (an italic case name, a link, a superscript) as a run in its own style inside that paragraph, with the source's word space, instead of a paragraph of its own. Style changes that wrap onto the next line still split paragraphs (the reading blocks keep style boundaries). Estimates unchanged. - **2026-10-24 (M3, window move on Wayland, #3970 follow-up):** After a move or resize by PdfCraft's own title bar or edges, the button release given back to egui goes before the pointer leaving the window, so hover no longer sticks at the old place until the pointer moves. Found by CodeRabbit in review of #3970; a unit test covers it. Estimates unchanged. diff --git a/crates/automation/src/dispatch.rs b/crates/automation/src/dispatch.rs index 87f47e8ebe..b8dfd36388 100644 --- a/crates/automation/src/dispatch.rs +++ b/crates/automation/src/dispatch.rs @@ -106,6 +106,35 @@ impl Automation { out["detected"] = json!(given.is_none()); out } + "page_enhance_scan" => { + let pages = match a.opt_ints("pages")? { + Some(_) => self.pages(&a, "pages")?, + None => Vec::new(), + }; + let level = match a.opt_int("level")? { + None => pdfcraft_engine::scan_clean::ScanCleanSettings::default().level, + Some(l @ 0..=100) => l as u8, + Some(l) => return Err(ToolError::InvalidArgs(format!("level must be between 0 and 100, not {l}"))), + }; + let mut settings = pdfcraft_engine::scan_clean::ScanCleanSettings::at(level); + if let Some(on) = a.opt_bool("edge_shadows")? { + settings.edge_shadows = on; + } + if let Some(on) = a.opt_bool("despeckle")? { + settings.despeckle = on; + } + let id = self.doc(&a)?.id; + let report = self.session.enhance_scanned_pages(id, &pages, settings).map_err(failed)?; + let numbered = |p: &[usize]| p.iter().map(|p| p + 1).collect::>(); + let skipped: Vec = + report.skipped.iter().map(|k| json!({ "pages": numbered(&k.pages), "name": k.name, "reason": k.reason })).collect(); + let mut out = summary(self.doc(&a)?); + out["pages_cleaned"] = json!(numbered(&report.pages)); + out["images"] = json!(report.images); + out["not_scanned"] = json!(numbered(&report.not_scanned)); + out["skipped"] = json!(skipped); + out + } "page_delete" => { let pages = self.pages(&a, "pages")?; self.apply(&a, Edit::DeletePages { pages })? diff --git a/crates/automation/src/tools.rs b/crates/automation/src/tools.rs index 4b4ac8873b..0d34911154 100644 --- a/crates/automation/src/tools.rs +++ b/crates/automation/src/tools.rs @@ -225,6 +225,18 @@ pub fn tools() -> Vec { }), &["doc", "page"], )), + t("page_enhance_scan", "Enhance scanned file", "Clean up scanned pages in place (Scan & OCR ▸ Enhance ▸ Scanned document). A page is a scan when its pictures cover at least half of it; the pictures covering at least a tenth of it are cleaned and written back into the same image objects at the same size and colour space (a JPEG stays a JPEG at quality 90, other grey/RGB images are stored losslessly, a 1-bit image stays 1-bit), so the page's invisible OCR text, comments, links and other content stay as they were, and an image drawn on several pages is cleaned once. level (0–100, default 50) is background removal: the paper's illumination is evened out so a grey, yellowed or unevenly lit background turns white, then the contrast is raised so faint strokes darken; colour ink stays colour. edge_shadows (default: on unless level is 0) lightens shadows along the scan's edges (book gutters, a copier's dark border). despeckle (default: from level 60) turns isolated dark specks white; it is the only change a 1-bit scan takes. Returns pages_cleaned (1-based), images (rewritten), not_scanned (pages that are no scan) and skipped ([{pages, name, reason}]: CMYK, indexed, JPEG 2000, JBIG2 or CCITT images, broken data, or nothing to change). One undoable step; none when nothing changed.") + .cmd("ocr.enhance") + .with(schema( + json!({ + "doc": doc(), + "pages": pages("to clean up (default: all)"), + "level": { "type": "integer", "minimum": 0, "maximum": 100 }, + "edge_shadows": { "type": "boolean" }, + "despeckle": { "type": "boolean" }, + }), + &["doc"], + )), t("page_delete", "Delete pages", "Delete pages. Undoable until saved.") .destructive() .cmd("page.delete") diff --git a/crates/automation/tests/it/automation.rs b/crates/automation/tests/it/automation.rs index 78038d8ebd..c6616f11f3 100644 --- a/crates/automation/tests/it/automation.rs +++ b/crates/automation/tests/it/automation.rs @@ -3376,6 +3376,62 @@ fn enhance_camera_image_through_tools() { )); } +#[test] +fn enhance_scanned_file_cleans_the_scan_in_place_through_tools() { + let dir = workdir("enhance-scan"); + // A 300 × 400 grey scan: dull paper, a dark gutter at the left, text bars and a speck. + let img = image::GrayImage::from_fn(300, 400, |x, y| { + let light = 0.55 + 0.45 * (x as f32 / 60.0).min(1.0); + let ink = if ((60..360).contains(&y) && (y - 60) % 40 < 6 && (60..240).contains(&x)) || (x, y) == (150, 25) { 0.12 } else { 0.78 }; + image::Luma([(ink * light * 255.0) as u8]) + }); + img.save(dir.join("scan.jpg")).unwrap(); + let mut a = auto(&dir); + let scan = ok(&mut a, "doc_create", json!({ "from": "images", "paths": ["scan.jpg"] }))["doc"].as_u64().unwrap(); + ok(&mut a, "image_save", json!({ "doc": scan, "page": 1, "image": 1, "path": "before.jpg" })); + + let r = ok(&mut a, "page_enhance_scan", json!({ "doc": scan, "level": 100 })); + assert_eq!(r["undo"], "Enhance scanned file"); + assert_eq!((r["pages_cleaned"].clone(), r["images"].clone(), r["not_scanned"].clone()), (json!([1]), json!(1), json!([]))); + assert_eq!(r["skipped"], json!([])); + // In place: still one picture of the same size, still a JPEG, now with white paper. + let images = ok(&mut a, "page_images", json!({ "doc": scan, "page": 1 })); + assert_eq!((images["count"].clone(), images["images"][0]["pixels"].clone()), (json!(1), json!([300, 400]))); + let saved = ok(&mut a, "image_save", json!({ "doc": scan, "page": 1, "image": 1, "path": "after" })); + assert_eq!(saved["format"], "jpg"); + let (before, after) = (image::open(dir.join("before.jpg")).unwrap().to_luma8(), image::open(dir.join("after.jpg")).unwrap().to_luma8()); + for (x, y) in [(150, 40), (3, 40), (3, 380), (150, 25)] { + assert!( + before.get_pixel(x, y)[0] < 200 && after.get_pixel(x, y)[0] >= 230, + "({x}, {y}): {:?} → {:?}", + before.get_pixel(x, y), + after.get_pixel(x, y) + ); + } + assert!(after.get_pixel(150, 62)[0] <= 60, "a stroke stays dark"); + // Undo puts the scan back as it was. + ok(&mut a, "edit_undo", json!({ "doc": scan })); + ok(&mut a, "image_save", json!({ "doc": scan, "page": 1, "image": 1, "path": "undone.jpg" })); + assert_eq!(std::fs::read(dir.join("undone.jpg")).unwrap(), std::fs::read(dir.join("before.jpg")).unwrap()); + + // Switches: despeckle alone leaves everything but the speck. + let r = ok(&mut a, "page_enhance_scan", json!({ "doc": scan, "pages": [1], "level": 0, "despeckle": true, "edge_shadows": false })); + assert_eq!(r["images"], 1); + // A document of text pages has no scans: nothing changes, and it says so. + let text = ok(&mut a, "doc_open", json!({ "path": "a.pdf" }))["doc"].as_u64().unwrap(); + let r = ok(&mut a, "page_enhance_scan", json!({ "doc": text })); + assert_eq!((r["images"].clone(), r["not_scanned"].clone(), r["undo"].clone()), (json!(0), json!([1, 2, 3]), Value::Null)); + // Bad arguments are refused before anything changes. + for bad in [ + json!({ "doc": scan, "level": 101 }), + json!({ "doc": scan, "pages": [9] }), + json!({ "doc": scan, "despeckle": "yes" }), + json!({ "doc": scan, "pages": [] }), + ] { + assert!(matches!(a.call("page_enhance_scan", &bad), Err(ToolError::InvalidArgs(_))), "{bad}"); + } +} + #[test] fn enhance_camera_image_finds_the_edges_by_itself() { let dir = workdir("enhance-camera-auto"); diff --git a/crates/engine/src/camera_clean.rs b/crates/engine/src/camera_clean.rs index 9c4120be56..d1768a2a83 100644 --- a/crates/engine/src/camera_clean.rs +++ b/crates/engine/src/camera_clean.rs @@ -1,6 +1,8 @@ -//! Clean up a straightened camera photo (Enhance ▸ Camera image ▸ enhancement level): even out -//! the lighting so the paper turns white (room-light colour casts, a hand's shadow, darker -//! edges go with it), then raise the contrast and drop isolated specks. Colour ink stays colour. +//! Clean up a straightened camera photo (Enhance ▸ Camera image ▸ enhancement level) or a +//! scanned page's picture (Enhance ▸ Scanned document, [`crate::scan_clean`]): even out the +//! lighting so the paper turns white (room-light colour casts, a hand's shadow, darker edges and +//! book gutters go with it), then raise the contrast and drop isolated specks. Colour ink stays +//! colour. //! //! The paper's illumination is estimated per channel on a coarse grid of blocks, from each //! block's bright end (the paper, not the ink), smoothed into a continuous map and divided out. @@ -8,6 +10,8 @@ /// The strongest a level can be. pub const MAX_LEVEL: u8 = 100; +/// The level from which [`clean`] also drops isolated specks. +pub const DESPECKLE_FROM: u8 = 60; /// How an enhanced camera page looks. #[derive(Clone, Copy, Debug, Default, PartialEq, Eq)] @@ -66,34 +70,158 @@ pub fn apply(rgb: &[u8], width: u32, height: u32, look: Look, level: u8) -> Opti let even = clean(rgb, width, height, 50)?; let t = 0.45 + 0.4 * level.min(MAX_LEVEL) as f32 / MAX_LEVEL as f32; let mut grey: Vec = even.as_chunks::<3>().0.iter().map(|p| if (lum(p) as f32 / 255.0) < t { 0 } else { 255 }).collect(); - despeckle_grey(&mut grey, w, h); + despeckle(&mut grey, 1, w, h); Pixels::Grey(grey) } }) } -/// Clean `rgb` (RGB8, `width` × `height`) at `level` (0 = unchanged … 100 = strongest). -/// Returns `None` when the buffer doesn't match the size. +/// Clean `rgb` (RGB8, `width` × `height`) at `level` (0 = unchanged … 100 = strongest), from +/// [`DESPECKLE_FROM`] without isolated specks. Returns `None` when the buffer doesn't match the +/// size. pub fn clean(rgb: &[u8], width: u32, height: u32, level: u8) -> Option> { + clean_samples(rgb, 3, width, height, CleanUp { level, edge_shadows: false, despeckle: level >= DESPECKLE_FROM }) +} + +/// What [`clean_samples`] does. +#[derive(Clone, Copy, Debug, PartialEq, Eq)] +pub struct CleanUp { + /// 0 (unchanged) … 100: even out the paper's illumination and raise the contrast, as + /// [`clean`]'s level. + pub level: u8, + /// Even out shadows along the picture's edges first (a book's gutter, a copier's dark edge): + /// lightened line by line to the paper beside them, within the outer fifth of each side. + pub edge_shadows: bool, + /// Drop isolated dark specks. + pub despeckle: bool, +} + +/// [`clean`] for samples of `n` components (1: grey, 3: RGB), with each step chosen apart (each +/// works at level 0 too). Returns `None` when `n` is neither or the buffer doesn't match the size. +pub fn clean_samples(px: &[u8], n: usize, width: u32, height: u32, how: CleanUp) -> Option> { let (w, h) = (width as usize, height as usize); - if w == 0 || h == 0 || w.checked_mul(h)?.checked_mul(3)? != rgb.len() { + if !(n == 1 || n == 3) || w == 0 || h == 0 || w.checked_mul(h)?.checked_mul(n)? != px.len() { return None; } - let level = level.min(MAX_LEVEL); - if level == 0 { - return Some(rgb.to_vec()); + let level = how.level.min(MAX_LEVEL); + let mut out = px.to_vec(); + if how.edge_shadows { + even_edges(&mut out, n, w, h); } + if level > 0 { + out = even_out(&out, n, w, h, level); + } + if how.despeckle { + despeckle(&mut out, n, w, h); + } + Some(out) +} + +/// The share of the width (and of the height) at each side where [`even_edges`] works. +const EDGE_BAND: f32 = 0.2; +/// The most [`even_edges`] lightens a line by; a darker line (a scanner's black border, a +/// picture) is not taken for shadowed paper and stays as it is. +const MAX_EDGE_GAIN: f32 = 3.0; + +/// Even out shadows that run along the picture's edges, in place: a book's gutter, a copier's +/// dark border, a page that curved away from the glass. Within the outer [`EDGE_BAND`] of each +/// side, each column (then each row) gets the paper's level along it (per channel, its 90th +/// percentile, so the text across it doesn't count), and is lightened to the brightest paper of +/// that band. The block grid of [`background`] is too coarse for such a narrow shadow: its +/// blocks borrow the brighter paper next to them. +fn even_edges(px: &mut [u8], n: usize, w: usize, h: usize) { + for vertical in [true, false] { + // Columns: lines across the width, each running down the height; rows the other way. + let (lines, along) = if vertical { (w, h) } else { (h, w) }; + let band = ((lines as f32 * EDGE_BAND) as usize).min(lines / 2); + if band == 0 { + continue; + } + let at = |line: usize, i: usize| if vertical { i * w + line } else { line * w + i }; + // The paper's level of every line in the bands and a few beyond them. + let reach = band.saturating_add(3).min(lines); + let measured = |line: usize| line < reach || line >= lines.saturating_sub(reach); + let mut level: Vec> = vec![None; lines]; + let mut hist = [[0u32; 256]; 3]; + for (line, slot) in level.iter_mut().enumerate().filter(|(line, _)| measured(*line)) { + hist.iter_mut().for_each(|hh| hh.fill(0)); + for i in 0..along { + let p = at(line, i) * n; + for (c, hh) in hist.iter_mut().enumerate().take(n) { + if let Some(v) = px.get(p + c) + && let Some(count) = hh.get_mut(*v as usize) + { + *count += 1; + } + } + } + let want = (along as f32 * 0.9) as u32; + *slot = Some(std::array::from_fn(|c| { + let mut seen = 0; + let v = hist.get(c).and_then(|hh| { + hh.iter().position(|&count| { + seen += count; + seen > want + }) + }); + v.unwrap_or(255) as f32 / 255.0 + })); + } + // Smoothed: the median of five neighbouring lines (fewer at the very edge, centred all + // the same, so a gradient isn't shifted), so one odd line (a rule, a fold) doesn't stand + // out. + let smooth: Vec> = (0..lines) + .map(|line| { + level.get(line).copied().flatten()?; + let r = line.min(lines - 1 - line).min(2); + Some(std::array::from_fn(|c| { + let mut near: Vec = (line - r..=line + r).filter_map(|l| level.get(l).copied().flatten()).map(|v| v[c]).collect(); + near.sort_by(f32::total_cmp); + near.get(near.len() / 2).copied().unwrap_or(1.0) + })) + }) + .collect(); + // Each side's band, and the lines its paper is measured on (the band and a few beyond). + for (shadowed, measured) in [(0..band, 0..reach), (lines - band..lines, lines.saturating_sub(reach)..lines)] { + // The brightest paper there: what the shadow is lightened to. + let paper: [f32; 3] = + std::array::from_fn(|c| measured.clone().filter_map(|l| smooth.get(l).copied().flatten()).map(|v| v[c]).fold(0.0, f32::max)); + for line in shadowed { + let Some(own) = smooth.get(line).copied().flatten() else { continue }; + let gain: [f32; 3] = std::array::from_fn(|c| { + let g = paper[c] / own[c].max(1e-3); + if (1.0..=MAX_EDGE_GAIN).contains(&g) { g } else { 1.0 } + }); + if gain.iter().take(n).all(|&g| g == 1.0) { + continue; + } + for i in 0..along { + let p = at(line, i) * n; + for (c, g) in gain.iter().enumerate().take(n) { + if let Some(v) = px.get_mut(p + c) { + *v = (*v as f32 * g).round().clamp(0.0, 255.0) as u8; + } + } + } + } + } + } +} + +/// The lighting evened out and the contrast raised at `level` (1–100); `px` holds `w` × `h` +/// pixels of `n` components. +fn even_out(px: &[u8], n: usize, w: usize, h: usize, level: u8) -> Vec { let k = level as f32 / MAX_LEVEL as f32; - let background = background(rgb, w, h); + let background = background(px, n, w, h); // Lighting is evened out fully from half strength on; contrast keeps growing to the top. let flatten = (2.0 * k).min(1.0); let white = 1.0 - 0.12 * k; // what reaches white let black = 0.30 * k; // what reaches black - let mut out = Vec::with_capacity(rgb.len()); - for (i, px) in rgb.as_chunks::<3>().0.iter().enumerate() { + let mut out = Vec::with_capacity(px.len()); + for (i, p) in px.chunks_exact(n).enumerate() { let (x, y) = (i % w, i / w); let bg = background.at(x, y); - for (c, &v) in px.iter().enumerate() { + for (c, &v) in p.iter().enumerate() { let v = v as f32 / 255.0; let b = bg.get(c).copied().unwrap_or(1.0).max(0.05); let even = (v / b).min(1.0); @@ -102,10 +230,7 @@ pub fn clean(rgb: &[u8], width: u32, height: u32, level: u8) -> Option> out.push((stretched * 255.0).round() as u8); } } - if level >= 60 { - despeckle(&mut out, w, h); - } - Some(out) + out } /// The paper's illumination: a grid of per-channel values, read back with bilinear @@ -140,9 +265,10 @@ impl Background { } } -/// Estimate the paper's illumination: per block, the 90th percentile of each channel (bright -/// paper, not dark ink); blocks that are mostly ink borrow from their neighbours; then smoothed. -fn background(rgb: &[u8], w: usize, h: usize) -> Background { +/// Estimate the paper's illumination: per block, the 90th percentile of each of the `n` channels +/// (bright paper, not dark ink); blocks that are mostly ink borrow from their neighbours; then +/// smoothed. Channels past `n` stay white. +fn background(px: &[u8], n: usize, w: usize, h: usize) -> Background { let block = (w.min(h) / 16).clamp(8, 128); let (cols, rows) = (w.div_ceil(block), h.div_ceil(block)); let mut cells = Vec::with_capacity(cols * rows); @@ -150,20 +276,20 @@ fn background(rgb: &[u8], w: usize, h: usize) -> Background { for by in 0..rows { for bx in 0..cols { hist.iter_mut().for_each(|hh| hh.fill(0)); - let mut n = 0u32; + let mut count = 0u32; for y in by * block..((by + 1) * block).min(h) { for x in bx * block..((bx + 1) * block).min(w) { - if let Some(p) = rgb.get((y * w + x) * 3..(y * w + x) * 3 + 3) { + if let Some(p) = px.get((y * w + x) * n..(y * w + x) * n + n) { for (c, &v) in p.iter().enumerate() { if let Some(slot) = hist.get_mut(c).and_then(|hh| hh.get_mut(v as usize)) { *slot += 1; } } - n += 1; + count += 1; } } } - let want = (n as f32 * 0.9) as u32; + let want = (count as f32 * 0.9) as u32; cells.push(std::array::from_fn(|c| { let mut seen = 0; let v = hist.get(c).and_then(|hh| { @@ -219,19 +345,15 @@ fn background(rgb: &[u8], w: usize, h: usize) -> Background { Background { cols, rows, block: block as f32, cells } } -/// [`despeckle`] for one grey byte per pixel. -fn despeckle_grey(grey: &mut [u8], w: usize, h: usize) { - let mut rgb: Vec = grey.iter().flat_map(|&v| [v, v, v]).collect(); - despeckle(&mut rgb, w, h); - for (g, p) in grey.iter_mut().zip(rgb.as_chunks::<3>().0.iter()) { - *g = p[0]; - } -} - -/// Lighten dark pixels with at most one dark neighbour (specks), in place. -fn despeckle(rgb: &mut [u8], w: usize, h: usize) { - let lum = |p: &[u8; 3]| (p[0] as u32 * 299 + p[1] as u32 * 587 + p[2] as u32 * 114) / 1000; - let dark: Vec = rgb.as_chunks::<3>().0.iter().map(|p| lum(p) < 128).collect(); +/// Lighten dark pixels with at most one dark neighbour (specks), in place; `px` holds `w` × `h` +/// pixels of `n` components (1 or 3). +fn despeckle(px: &mut [u8], n: usize, w: usize, h: usize) { + let lum = |p: &[u8]| match p { + [r, g, b] => (*r as u32 * 299 + *g as u32 * 587 + *b as u32 * 114) / 1000, + [v, ..] => *v as u32, + [] => 255, + }; + let dark: Vec = px.chunks_exact(n.max(1)).map(|p| lum(p) < 128).collect(); let is_dark = |x: isize, y: isize| { x >= 0 && y >= 0 && (x as usize) < w && (y as usize) < h && dark.get(y as usize * w + x as usize).copied().unwrap_or(false) }; @@ -243,7 +365,7 @@ fn despeckle(rgb: &mut [u8], w: usize, h: usize) { let neighbours = (-1..=1).flat_map(|dy| (-1..=1).map(move |dx| (dx, dy))).filter(|&(dx, dy)| (dx, dy) != (0, 0) && is_dark(x + dx, y + dy)).count(); if neighbours <= 1 - && let Some(p) = rgb.get_mut((y as usize * w + x as usize) * 3..(y as usize * w + x as usize) * 3 + 3) + && let Some(p) = px.get_mut((y as usize * w + x as usize) * n..(y as usize * w + x as usize) * n + n) { p.fill(255); } @@ -388,5 +510,40 @@ mod tests { assert_eq!(clean(&[], 0, 0, 50), None); assert_eq!(clean(&[0; 12], 2, 3, 50), None); assert_eq!(clean(&[9; 3], 1, 1, 255).map(|v| v.len()), Some(3)); + let every = CleanUp { level: 100, edge_shadows: true, despeckle: true }; + assert_eq!(clean_samples(&[0; 12], 2, 2, 3, every), None, "two components"); + assert_eq!(clean_samples(&[0; 5], 1, 2, 3, every), None); + for (w, h) in [(1, 1), (1, 9), (9, 1), (2, 2), (5, 3)] { + assert_eq!(clean_samples(&vec![100; w * h], 1, w as u32, h as u32, every).map(|v| v.len()), Some(w * h)); + } + } + + #[test] + fn edge_shadows_lighten_shadowed_paper_but_not_a_black_border_or_the_middle() { + // Grey paper (200) with a shadow darkening the left 20 columns to 120, a black scanner + // border in the right 8 columns, and a dark block in the middle. + let (w, h) = (100usize, 60usize); + let mut grey = vec![200u8; w * h]; + for y in 0..h { + for x in 0..w { + let v = if x >= w - 8 { + 5 + } else if (40..60).contains(&x) && (20..40).contains(&y) { + 30 + } else if x < 20 { + 120 + (80 * x / 20) as u8 + } else { + 200 + }; + grey[y * w + x] = v; + } + } + let out = clean_samples(&grey, 1, w as u32, h as u32, CleanUp { level: 0, edge_shadows: true, despeckle: false }).unwrap(); + for x in [0, 5, 19] { + assert!(out[30 * w + x].abs_diff(200) <= 3, "column {x}: {} → {}", grey[30 * w + x], out[30 * w + x]); + } + assert_eq!(out[30 * w + w - 3], 5, "a black border is no shadow"); + assert_eq!(out[30 * w + 50], 30, "the middle is untouched"); + assert_eq!(out[30 * w + 30], 200); } } diff --git a/crates/engine/src/catalog.rs b/crates/engine/src/catalog.rs index e479232bbb..47394a7bb5 100644 --- a/crates/engine/src/catalog.rs +++ b/crates/engine/src/catalog.rs @@ -265,7 +265,7 @@ pub static TOOL_GROUPS: &[ToolGroup] = &[ items: &[ item("In this file", "scan-text", "ocr.recognize", Ready), item("In multiple files", "files", "ocr.recognize_batch", Ready), - item("Enhance scanned file", "sparkles", "ocr.enhance", Planned("M10")), + item("Enhance scanned file", "sparkles", "ocr.enhance", Ready), item("Enhance camera image", "camera", "ocr.enhance_camera", Ready), item("Correct recognized text", "text-select", "ocr.correct", Planned("M10")), ], diff --git a/crates/engine/src/commands.rs b/crates/engine/src/commands.rs index 1f25c23c98..8c83996c6d 100644 --- a/crates/engine/src/commands.rs +++ b/crates/engine/src/commands.rs @@ -352,6 +352,7 @@ pub const COMMANDS: &[CommandSpec] = &[ c("tools.document_js", "Document JavaScripts…", None, None, Modification, "file-code"), c("ocr.recognize", "Recognize text…", None, None, Modification, "scan-text"), c("ocr.enhance_camera", "Enhance camera image", None, None, Modification, "camera"), + c("ocr.enhance", "Enhance scanned file…", None, None, Modification, "sparkles"), c("ocr.recognize_batch", "Recognize text in multiple files…", None, None, Nothing, "files"), c("a11y.check", "Check for accessibility…", None, None, Document, "accessibility"), c("a11y.find_untagged", "Find untagged content", None, None, Document, "search"), diff --git a/crates/engine/src/edit.rs b/crates/engine/src/edit.rs index c31984ca19..655a1ce558 100644 --- a/crates/engine/src/edit.rs +++ b/crates/engine/src/edit.rs @@ -375,6 +375,13 @@ pub enum Edit { page: usize, words: Vec, }, + /// Scan & OCR ▸ Enhance ▸ Scanned document: put the scan images + /// [`ScanCleanJob::run`](crate::scan_clean::ScanCleanJob::run) cleaned in place of the old + /// ones (each object must still be an image). Built by + /// [`Session::apply_scan_clean`](crate::Session::apply_scan_clean). + EnhanceScan { + images: Vec<(pdfcraft_cos::ObjRef, pdfcraft_cos::Stream)>, + }, /// Edit a PDF ▸ Edit text: replace the text of line `line` (from `Document::text_lines`) on /// `page`, in its own font when it can show it, else in Helvetica. EditTextLine { diff --git a/crates/engine/src/edit/labels.rs b/crates/engine/src/edit/labels.rs index ea25898068..ff86f21a08 100644 --- a/crates/engine/src/edit/labels.rs +++ b/crates/engine/src/edit/labels.rs @@ -98,6 +98,7 @@ pub const EDIT_LABELS: &[&str] = &[ "Set alternate text", "Mark figure as decorative", "Recognize text", + "Enhance scanned file", "Run JavaScript", "Run form script", "Save as {level}", @@ -239,6 +240,7 @@ impl Edit { Edit::MarkDecorative { .. } => Label::of("Mark figure as decorative"), Edit::MarkTagArtifact { .. } => Label::of("Mark content as artifact"), Edit::AddOcrText { .. } => Label::of("Recognize text"), + Edit::EnhanceScan { .. } => Label::of("Enhance scanned file"), Edit::ApplyScriptChanges { .. } => Label::of("Run JavaScript"), Edit::XfaEvent { .. } => Label::of("Run form script"), // The level is the standard's own name ("PDF/A-2b"), so it goes in as it stands. diff --git a/crates/engine/src/edit/run.rs b/crates/engine/src/edit/run.rs index 49f19fa637..e2a99665aa 100644 --- a/crates/engine/src/edit/run.rs +++ b/crates/engine/src/edit/run.rs @@ -121,6 +121,7 @@ pub(crate) fn check_permission(edit: &Edit, p: &pdfcraft_cos::Permissions) -> Re | Edit::MarkDecorative { .. } | Edit::MarkTagArtifact { .. } | Edit::AddOcrText { .. } + | Edit::EnhanceScan { .. } | Edit::SetDocumentScript { .. } | Edit::SetFieldScript { .. } | Edit::RemoveStandardId { .. } @@ -491,6 +492,7 @@ pub(crate) fn run_edit(doc: &mut pdfcraft_cos::Document, edit: &Edit, cx: &mut E Edit::AddOcrText { page, words } => { pdfcraft_edit::stamp(doc, *page, "OCR", pdfcraft_ocr::text_layer(words))?; } + Edit::EnhanceScan { images } => crate::scan_clean::write_back(doc, images)?, Edit::EditPageImage { page, index, change } => { let img = pdfcraft_edit::page_images(doc, *page)? .into_iter() diff --git a/crates/engine/src/history.rs b/crates/engine/src/history.rs index fafa65b4c2..0fe874e711 100644 --- a/crates/engine/src/history.rs +++ b/crates/engine/src/history.rs @@ -77,6 +77,7 @@ pub(crate) fn scope_of(edit: &Edit) -> Scope { | Edit::EditTextLine { .. } | Edit::EditTextBlock { .. } | Edit::EditPageImage { .. } + | Edit::EnhanceScan { .. } | Edit::AddText { .. } | Edit::AddImage { .. } | Edit::UpdateContent { .. } diff --git a/crates/engine/src/lib.rs b/crates/engine/src/lib.rs index 889b839368..9e2eb54389 100644 --- a/crates/engine/src/lib.rs +++ b/crates/engine/src/lib.rs @@ -43,6 +43,7 @@ mod pictures; pub mod print_raster; pub mod printcheck; mod protection; +pub mod scan_clean; mod scan_colours; mod scan_fields; mod session; diff --git a/crates/engine/src/scan_clean.rs b/crates/engine/src/scan_clean.rs new file mode 100644 index 0000000000..bfc8188b29 --- /dev/null +++ b/crates/engine/src/scan_clean.rs @@ -0,0 +1,276 @@ +//! Scan & OCR ▸ Enhance ▸ Scanned document: clean up the pictures scanned pages are made of. +//! +//! Unlike Enhance ▸ Camera image, the page is not rendered and replaced: each scan image is +//! decoded, cleaned ([`crate::camera_clean::clean_samples`]) and written back **into the same +//! image object**, at the same size and in the same colour space. So everything else on the +//! page stays as it was: an invisible OCR text layer, comments, links, form fields, other +//! content, and the image's own soft mask, `/Interpolate` and other entries. A JPEG stays a JPEG +//! (at quality 90), any other grey or RGB image is stored losslessly (Flate), and a black and +//! white (1-bit) image stays 1-bit. An image several pages draw is cleaned once. +//! +//! A page is a scan when its pictures cover at least half of it (as for Recognize text); the +//! pictures each covering at least a tenth of the page are its scan. What is not cleaned is +//! reported, never skipped silently: pages that are no scan, and scan images in a form PdfCraft +//! can't rewrite yet (CMYK, indexed, JPEG 2000, JBIG2, CCITT fax, a stencil mask, a remapping +//! decode array) or whose data is broken. +//! +//! Like Optimize, the work is a job: [`Session::scan_clean_job`] captures the document, +//! [`ScanCleanJob::run`] cleans anywhere (a worker thread), and [`Session::apply_scan_clean`] +//! makes the result one undoable step, "Enhance scanned file" (the name the Scan & OCR tool has). + +use pdfcraft_cos::{ObjRef, Object, Stream}; + +use crate::camera_clean::{CleanUp, MAX_LEVEL, clean_samples}; +use crate::edit::check_permission; +use crate::{DocId, Edit, EditError, Session, guard, optimize}; + +pub use crate::camera_clean::DESPECKLE_FROM; + +/// The JPEG quality a cleaned JPEG scan is written back at (as Enhance camera image). +const QUALITY: u8 = 90; +/// On a scanned page, an image covering at least this share of the page is part of the scan; +/// smaller ones (a logo or a signature pasted on) are left alone. +const PART_OF_SCAN: f64 = 0.1; + +/// How scanned pages are cleaned up. +#[derive(Clone, Copy, Debug, PartialEq, Eq)] +pub struct ScanCleanSettings { + /// Background removal, 0 (off) … 100 (strongest), as Enhance camera image's clean-up level: + /// the paper's illumination is evened out (fully from 50), so a grey, yellowed or unevenly lit + /// background turns white; then the contrast is raised so faint strokes darken. Colour ink + /// stays colour. + pub level: u8, + /// Edge shadow removal: shadows along the scan's edges (a book's gutter, a copier's dark + /// border) are lightened to the paper next to them, line by line, before the level applies. + pub edge_shadows: bool, + /// Despeckle: isolated dark specks (dust, scanner noise), dark pixels with at most one dark + /// neighbour, turn white. The only change a black and white (1-bit) scan takes. + pub despeckle: bool, +} + +impl ScanCleanSettings { + /// The level `level` (capped at 100) with edge shadow removal (unless the level is 0), and + /// despeckling from [`DESPECKLE_FROM`] as the camera tool does. + pub fn at(level: u8) -> Self { + let level = level.min(MAX_LEVEL); + Self { level, edge_shadows: level > 0, despeckle: level >= DESPECKLE_FROM } + } + + /// Whether these settings change anything at all. + pub fn changes_something(self) -> bool { + self.level > 0 || self.edge_shadows || self.despeckle + } +} + +impl Default for ScanCleanSettings { + fn default() -> Self { + Self::at(50) + } +} + +/// What a clean-up did. +#[derive(Clone, Debug, Default, PartialEq)] +pub struct ScanCleanReport { + /// Pages (0-based) whose scan was cleaned, in order. + pub pages: Vec, + /// Images rewritten. One drawn on several pages counts once. + pub images: usize, + /// Pages left alone because no picture covers most of them. + pub not_scanned: Vec, + /// Scan images left as they were, and why. + pub skipped: Vec, +} + +/// A scan image that was not cleaned. +#[derive(Clone, Debug, PartialEq)] +pub struct SkippedImage { + /// The pages (0-based) that draw it, among those asked for. + pub pages: Vec, + /// Its resource name on the first of them. + pub name: String, + /// Why, in a sentence that follows "skipped because". + pub reason: String, +} + +/// One image to clean: its object, its name and the pages that draw it. +struct Target { + object: ObjRef, + name: String, + pages: Vec, +} + +/// What cleaning scanned pages needs, independent of the session. +pub struct ScanCleanJob { + cos: pdfcraft_cos::Document, + targets: Vec, + settings: ScanCleanSettings, + /// The document's edit generation it was taken at: a result is for that state only. + generation: u64, + report: ScanCleanReport, +} + +/// The cleaned images, ready for [`Session::apply_scan_clean`], and the report. +pub struct ScanCleaned { + images: Vec<(ObjRef, Stream)>, + generation: u64, + pub report: ScanCleanReport, +} + +impl Session { + /// Capture what cleaning up the scans on `pages` (0-based; empty = all) of document `id` + /// needs. Errors when the document can't be changed (read-only, or its permissions forbid + /// changes) or a page doesn't exist. + pub fn scan_clean_job(&self, id: DocId, pages: &[usize], settings: ScanCleanSettings) -> Result { + let doc = self.get(id).ok_or(EditError::NoDocument)?; + if let Some(why) = &doc.read_only_reason { + return Err(EditError::ReadOnly(why.clone())); + } + let editor = doc.editor.as_ref().ok_or_else(|| EditError::ReadOnly(String::new()))?; + if let Some(p) = editor.cos.permissions() { + check_permission(&Edit::EnhanceScan { images: Vec::new() }, &p)?; + } + let all = doc.info.pages.len(); + let mut wanted: Vec = if pages.is_empty() { (0..all).collect() } else { pages.to_vec() }; + if let Some(p) = wanted.iter().find(|p| **p >= all) { + return Err(EditError::Invalid(format!("there is no page {}: the document has {all}", p.saturating_add(1)))); + } + wanted.sort_unstable(); + wanted.dedup(); + let mut report = ScanCleanReport::default(); + let mut targets: Vec = Vec::new(); + for page in wanted { + let Some(info) = doc.info.pages.get(page) else { continue }; + let Some(images) = crate::scan_fields::scan_images(doc, page, info) else { + report.not_scanned.push(page); + continue; + }; + for (image, _) in images.into_iter().filter(|(_, share)| *share >= PART_OF_SCAN) { + let Some(object) = image.object else { + report.skipped.push(SkippedImage { + pages: vec![page], + name: image.name, + reason: "it is stored inside the page's resources rather than as an object of its own".into(), + }); + continue; + }; + match targets.iter_mut().find(|t| t.object == object) { + Some(t) if !t.pages.contains(&page) => t.pages.push(page), + Some(_) => {} + None => targets.push(Target { object, name: image.name, pages: vec![page] }), + } + } + } + Ok(ScanCleanJob { cos: editor.cos.clone(), targets, settings, generation: doc.edit_generation(), report }) + } + + /// Write the images [`ScanCleanJob::run`] cleaned into document `id` as one undoable step, + /// "Enhance scanned file" (none when nothing was cleaned), and return the report. Refused + /// when the document changed since the job was taken. + pub fn apply_scan_clean(&mut self, id: DocId, cleaned: ScanCleaned) -> Result { + let doc = self.get(id).ok_or(EditError::NoDocument)?; + if doc.edit_generation() != cleaned.generation { + return Err(EditError::Invalid("the document changed while its scans were being cleaned up; nothing was changed, try again".into())); + } + if !cleaned.images.is_empty() { + self.apply(id, Edit::EnhanceScan { images: cleaned.images })?; + } + Ok(cleaned.report) + } + + /// Enhance ▸ Scanned document in one call: [`Self::scan_clean_job`], run, then + /// [`Self::apply_scan_clean`] (the CLI and agents use this). + pub fn enhance_scanned_pages(&mut self, id: DocId, pages: &[usize], settings: ScanCleanSettings) -> Result { + let job = self.scan_clean_job(id, pages, settings)?; + let cleaned = job.run(|_, _| true)?; + self.apply_scan_clean(id, cleaned) + } +} + +impl ScanCleanJob { + /// The images to clean (one drawn on several pages counts once). + pub fn total(&self) -> usize { + self.targets.len() + } + + /// Clean every scan image. `progress(done, total)` is called before each image; returning + /// `false` stops with [`EditError::Cancelled`]. A panic is returned as an error, so a worker + /// thread always reports back. + pub fn run(self, mut progress: impl FnMut(usize, usize) -> bool) -> Result { + guard(move || self.work(&mut progress)).map_err(|m| EditError::Invalid(format!("cleaning up the scans failed unexpectedly ({m})")))? + } + + fn work(self, progress: &mut dyn FnMut(usize, usize) -> bool) -> Result { + let ScanCleanJob { cos, targets, settings, generation, mut report } = self; + let total = targets.len(); + let mut images = Vec::new(); + let mut changed_pages: Vec = Vec::new(); + for (done, target) in targets.into_iter().enumerate() { + if !progress(done, total) { + return Err(EditError::Cancelled); + } + let mut skip = |reason: String| report.skipped.push(SkippedImage { pages: target.pages.clone(), name: target.name.clone(), reason }); + match clean_image(&cos, target.object, settings) { + Ok(Some(stream)) => { + images.push((target.object, stream)); + changed_pages.extend(target.pages.iter().copied()); + } + Ok(None) => skip("it is already clean: nothing changes at these settings".into()), + Err(reason) => skip(reason), + } + } + changed_pages.sort_unstable(); + changed_pages.dedup(); + report.images = images.len(); + report.pages = changed_pages; + report.skipped.sort_by_key(|s| s.pages.first().copied()); + Ok(ScanCleaned { images, generation, report }) + } +} + +/// Image `object` cleaned with `settings`: the new stream, `None` when nothing changes, or why it +/// can't be cleaned. +fn clean_image(cos: &pdfcraft_cos::Document, object: ObjRef, settings: ScanCleanSettings) -> Result, String> { + let obj = cos.get(object); + let Object::Stream(s) = &*obj else { return Err("it is not an image".into()) }; + if s.dict.name(b"Subtype") != Some(b"Image") { + return Err("it is not an image".into()); + } + // A colour-key mask names sample values to leave unpainted: cleaned samples would match others. + if matches!(s.dict.get(b"Mask").map(|m| cos.resolve(m)).as_deref(), Some(Object::Array(_))) { + return Err("it has a colour-key mask, which cleaning would change".into()); + } + let samples = optimize::decode_samples(cos, s).map_err(|e| e.to_string())?; + let cleaned = if samples.bilevel { + if !settings.despeckle { + return Err("it is black and white, which only Despeckle changes, and Despeckle is off".into()); + } + clean_samples(&samples.data, 1, samples.width, samples.height, CleanUp { level: 0, edge_shadows: false, despeckle: true }) + } else { + if !settings.changes_something() { + return Ok(None); + } + let how = CleanUp { level: settings.level, edge_shadows: settings.edge_shadows, despeckle: settings.despeckle }; + clean_samples(&samples.data, samples.components, samples.width, samples.height, how) + } + .ok_or("its samples don't match its size")?; + if cleaned == samples.data { + return Ok(None); + } + let cleaned = optimize::Samples { data: cleaned, ..samples }; + optimize::encode_samples(s, &cleaned, QUALITY).map(Some).ok_or_else(|| "it could not be written back".into()) +} + +/// [`Edit::EnhanceScan`]: put the cleaned images in place. Each object must still be an image. +pub(crate) fn write_back(doc: &mut pdfcraft_cos::Document, images: &[(ObjRef, Stream)]) -> Result<(), EditError> { + for (object, stream) in images { + let is_image = matches!(&*doc.get(*object), Object::Stream(s) if s.dict.name(b"Subtype") == Some(b"Image")); + if !is_image { + return Err(EditError::Invalid(format!("object {} is no longer an image; the scan was not changed", object.num))); + } + doc.set(*object, Object::Stream(stream.clone())); + } + Ok(()) +} + +#[cfg(test)] +mod tests; diff --git a/crates/engine/src/scan_clean/tests.rs b/crates/engine/src/scan_clean/tests.rs new file mode 100644 index 0000000000..561ba9e2e4 --- /dev/null +++ b/crates/engine/src/scan_clean/tests.rs @@ -0,0 +1,338 @@ +use std::sync::Arc; + +use pdfcraft_testkit::Pdf; + +use super::*; + +const W: usize = 300; +const H: usize = 400; +/// Isolated specks on the paper, away from the text rows. +const SPECKS: [(usize, usize); 3] = [(150, 25), (200, 85), (120, 125)]; + +/// The light falling on the paper at `x`: full in the middle, darker towards a shadowed left +/// edge (as a book's gutter) and a little at the right edge. +fn light(x: usize) -> f32 { + let left = (x as f32 / 60.0).min(1.0); + let right = ((W - 1 - x) as f32 / 30.0).min(1.0); + (0.55 + 0.45 * left) * (0.8 + 0.2 * right) +} + +fn is_stroke(x: usize, y: usize) -> bool { + (60..360).contains(&y) && (y - 60) % 40 < 6 && (60..240).contains(&x) +} + +/// A synthetic scan, `n` components (1 grey, 3 RGB): greyish yellowed paper, a shadowed gutter, +/// dark text strokes and a few specks. +fn scan_pixels(n: usize) -> Vec { + let mut out = Vec::with_capacity(W * H * n); + for y in 0..H { + for x in 0..W { + let ink = if is_stroke(x, y) || SPECKS.contains(&(x, y)) { 0.12 } else { 0.8 }; + let v = ink * light(x); + let px = [v, v, v * 0.85]; + for c in px.iter().take(n) { + out.push((c * 255.0).round().clamp(0.0, 255.0) as u8); + } + } + } + out +} + +fn jpeg(px: &[u8], n: usize) -> Vec { + let mut out = Vec::new(); + let ty = if n == 1 { image::ExtendedColorType::L8 } else { image::ExtendedColorType::Rgb8 }; + image::ImageEncoder::write_image(image::codecs::jpeg::JpegEncoder::new_with_quality(&mut out, 95), px, W as u32, H as u32, ty).unwrap(); + out +} + +/// Three pages: 1 a scan (the image `image_dict`/`data` as object 4) with an invisible text +/// layer and a comment, 2 a page of real text, 3 the same scan image drawn again. +fn scan_pdf(image_dict: &str, data: &[u8]) -> Vec { + let resources = "/Resources << /XObject << /Im0 4 0 R >> /Font << /F1 5 0 R >> >>"; + Pdf::new() + .obj("<< /Type /Catalog /Pages 2 0 R >>") + .obj("<< /Type /Pages /Kids [3 0 R 8 0 R 9 0 R] /Count 3 >>") + .obj(format!("<< /Type /Page /Parent 2 0 R /MediaBox [0 0 {W} {H}] {resources} /Contents 6 0 R /Annots [7 0 R] >>")) + .stream(&format!("/Type /XObject /Subtype /Image {image_dict}"), data) + .obj("<< /Type /Font /Subtype /Type1 /BaseFont /Helvetica >>") + .stream("", format!("q {W} 0 0 {H} 0 0 cm /Im0 Do Q BT 3 Tr /F1 12 Tf 20 380 Td (Invisible OCR words) Tj ET")) + .obj("<< /Type /Annot /Subtype /Square /Rect [10 10 50 50] /C [1 0 0] /Contents (A comment) >>") + .obj(format!("<< /Type /Page /Parent 2 0 R /MediaBox [0 0 {W} {H}] {resources} /Contents 10 0 R >>")) + .obj(format!("<< /Type /Page /Parent 2 0 R /MediaBox [0 0 {W} {H}] {resources} /Contents 11 0 R >>")) + .stream("", "BT /F1 12 Tf 20 380 Td (Real text) Tj ET") + .stream("", format!("q {W} 0 0 {H} 0 0 cm /Im0 Do Q")) + .build() +} + +fn grey_scan() -> Vec { + scan_pdf(&format!("/Width {W} /Height {H} /ColorSpace /DeviceGray /BitsPerComponent 8 /Interpolate true"), &scan_pixels(1)) +} + +fn open(bytes: Vec) -> (Session, DocId) { + let mut s = Session::new().with_clock(|| 1_700_000_000); + let id = s.open("scan.pdf", None, Arc::new(bytes), None).unwrap(); + (s, id) +} + +fn object(s: &Session, id: DocId, num: u32) -> Arc { + s.get(id).unwrap().editor.as_ref().unwrap().cos.get(ObjRef::new(num, 0)) +} + +/// The samples of the image page 1 draws, now. +fn samples(s: &Session, id: DocId) -> optimize::Samples { + let doc = s.get(id).unwrap(); + let object = doc.page_images(0)[0].object.unwrap(); + let cos = &doc.editor.as_ref().unwrap().cos; + let Object::Stream(img) = &*cos.get(object) else { panic!("the image") }; + optimize::decode_samples(cos, img).unwrap() +} + +fn grey_at(px: &optimize::Samples, x: usize, y: usize) -> u8 { + px.data[(y * W + x) * px.components] +} + +#[test] +fn a_scan_gets_white_paper_even_edges_dark_strokes_and_no_specks_in_place() { + let (mut s, id) = open(grey_scan()); + let before = samples(&s, id); + let kept: Vec> = [3, 5, 6, 7, 8, 10, 11].iter().map(|&n| object(&s, id, n)).collect(); + let text_before = s.get(id).unwrap().text_lines(0); + let original = s.get(id).unwrap().bytes.clone(); + + let report = s.enhance_scanned_pages(id, &[], ScanCleanSettings::at(100)).unwrap(); + assert_eq!(report.pages, vec![0, 2], "the shared image counts for both its pages"); + assert_eq!(report.images, 1); + assert_eq!(report.not_scanned, vec![1], "the text page is no scan"); + assert!(report.skipped.is_empty(), "{:?}", report.skipped); + assert_eq!(s.get(id).unwrap().can_undo(), Some("Enhance scanned file")); + + let after = samples(&s, id); + assert_eq!((after.width, after.height, after.components, after.jpeg), (W as u32, H as u32, 1, false)); + // Paper, also in the shadowed gutter and at the right edge, turns (nearly) white. + for (x, y) in [(150, 40), (150, 380), (3, 40), (3, 380), (30, 200), (W - 2, 40)] { + assert!(grey_at(&before, x, y) < 210, "the fixture's paper is grey at ({x}, {y})"); + assert!(grey_at(&after, x, y) >= 235, "paper at ({x}, {y}): {} → {}", grey_at(&before, x, y), grey_at(&after, x, y)); + } + // The edge is evened out: the gutter is as white as the middle. + assert!(grey_at(&after, 150, 40).abs_diff(grey_at(&after, 3, 40)) <= 12); + // Strokes stay dark (darker, with the contrast), also in the gutter's light. + for (x, y) in [(150, 62), (61, 102), (238, 342)] { + assert!(grey_at(&after, x, y) <= 40, "stroke at ({x}, {y}): {}", grey_at(&after, x, y)); + } + for (x, y) in SPECKS { + assert!(grey_at(&before, x, y) < 60); + assert!(grey_at(&after, x, y) >= 200, "speck at ({x}, {y}): {}", grey_at(&after, x, y)); + } + // Everything but the image is as it was: the page, its invisible text layer, the comment, the + // font, the other pages. The image keeps its entries. + let now: Vec> = [3, 5, 6, 7, 8, 10, 11].iter().map(|&n| object(&s, id, n)).collect(); + assert_eq!(kept, now); + assert_eq!(s.get(id).unwrap().info.pages.len(), 3); + let text_after = s.get(id).unwrap().text_lines(0); + assert!(text_after.iter().any(|l| l.text.contains("Invisible OCR words")), "{text_after:?}"); + assert_eq!(text_before.len(), text_after.len()); + let Object::Stream(img) = &*object(&s, id, 4) else { panic!("the image") }; + assert_eq!(img.dict.get(b"Interpolate"), Some(&Object::Bool(true))); + assert_eq!(img.dict.name(b"ColorSpace"), Some(&b"DeviceGray"[..])); + + // Undo brings the original image, and the original file, back. + s.undo(id).unwrap(); + assert_eq!(samples(&s, id), before); + assert_eq!(s.get(id).unwrap().bytes, original); + assert_eq!(s.get(id).unwrap().can_undo(), None); +} + +#[test] +fn a_jpeg_scan_stays_a_jpeg_in_colour() { + let px = scan_pixels(3); + let (mut s, id) = open(scan_pdf(&format!("/Width {W} /Height {H} /ColorSpace /DeviceRGB /BitsPerComponent 8 /Filter /DCTDecode"), &jpeg(&px, 3))); + let report = s.enhance_scanned_pages(id, &[0], ScanCleanSettings::at(80)).unwrap(); + // Only page 1 was asked for; the image it shares with page 3 is changed for both. + assert_eq!((report.pages.as_slice(), report.images), (&[0][..], 1)); + let after = samples(&s, id); + assert!(after.jpeg && after.components == 3); + let Object::Stream(img) = &*object(&s, id, 4) else { panic!("the image") }; + assert_eq!(img.dict.name(b"Filter"), Some(&b"DCTDecode"[..])); + // Yellowed paper becomes white; the strokes stay dark. + let at = |x: usize, y: usize| after.data[(y * W + x) * 3..][..3].to_vec(); + for (x, y) in [(150, 40), (5, 300)] { + assert!(at(x, y).iter().all(|&v| v >= 230), "paper at ({x}, {y}): {:?}", at(x, y)); + } + assert!(at(150, 63).iter().all(|&v| v <= 60), "stroke: {:?}", at(150, 63)); + // The file still renders. + let doc = s.get(id).unwrap(); + let shot = pdfcraft_render::PageRenderer::new(doc.bytes.clone(), doc.config.clone()).render(pdfcraft_render::RenderRequest { + page: 0, + scale: 1.0, + ..Default::default() + }); + assert!(shot.error.is_none(), "{:?}", shot.error); +} + +#[test] +fn despeckle_is_its_own_switch() { + let speck = |settings: ScanCleanSettings| { + let (mut s, id) = open(grey_scan()); + s.enhance_scanned_pages(id, &[0], settings).unwrap(); + let after = samples(&s, id); + SPECKS.iter().map(|&(x, y)| grey_at(&after, x, y)).max().unwrap_or(0) + }; + assert!(speck(ScanCleanSettings { despeckle: false, ..ScanCleanSettings::at(80) }) < 100, "kept without despeckle"); + assert!(speck(ScanCleanSettings::at(80)) >= 200); + // Despeckle alone, at level 0, changes nothing but the specks. + let (mut s, id) = open(grey_scan()); + let before = samples(&s, id); + s.enhance_scanned_pages(id, &[], ScanCleanSettings { level: 0, edge_shadows: false, despeckle: true }).unwrap(); + let after = samples(&s, id); + let changed: Vec = before.data.iter().zip(&after.data).enumerate().filter(|(_, (a, b))| a != b).map(|(i, _)| i).collect(); + let mut specks: Vec = SPECKS.iter().map(|&(x, y)| y * W + x).collect(); + specks.sort_unstable(); + assert_eq!(changed, specks); + // Nothing to do: no undo step, and the report says why. + let (mut s, id) = open(grey_scan()); + let report = s.enhance_scanned_pages(id, &[], ScanCleanSettings { level: 0, edge_shadows: false, despeckle: false }).unwrap(); + assert_eq!(report.images, 0); + assert!(report.pages.is_empty()); + assert_eq!(report.skipped.len(), 1); + assert!(report.skipped[0].reason.contains("already clean"), "{:?}", report.skipped); + assert_eq!(s.get(id).unwrap().can_undo(), None); + assert!(!s.get(id).unwrap().dirty); +} + +#[test] +fn a_black_and_white_scan_stays_one_bit_and_loses_its_specks() { + // 64 × 64, one bit per pixel, 1 = white: a black bar and two specks. + let (w, h) = (64usize, 64usize); + let mut bits = vec![0xFFu8; w / 8 * h]; + let mut black = |x: usize, y: usize| bits[y * w / 8 + x / 8] &= !(0x80 >> (x % 8)); + for y in 20..30 { + for x in 10..50 { + black(x, y); + } + } + black(5, 5); + black(58, 58); + let bytes = Pdf::new() + .obj("<< /Type /Catalog /Pages 2 0 R >>") + .obj("<< /Type /Pages /Kids [3 0 R] /Count 1 >>") + .obj("<< /Type /Page /Parent 2 0 R /MediaBox [0 0 64 64] /Resources << /XObject << /Im0 4 0 R >> >> /Contents 5 0 R >>") + .stream("/Type /XObject /Subtype /Image /Width 64 /Height 64 /ColorSpace /DeviceGray /BitsPerComponent 1", &bits) + .stream("", "q 64 0 0 64 0 0 cm /Im0 Do Q") + .build(); + let (mut s, id) = open(bytes); + // Only Despeckle changes a black and white scan: without it the image is reported. + let report = s.enhance_scanned_pages(id, &[], ScanCleanSettings { despeckle: false, ..ScanCleanSettings::at(100) }).unwrap(); + assert_eq!(report.images, 0); + assert!(report.skipped.first().is_some_and(|k| k.reason.contains("black and white")), "{:?}", report.skipped); + let report = s.enhance_scanned_pages(id, &[], ScanCleanSettings::at(100)).unwrap(); + assert_eq!(report.images, 1); + let Object::Stream(img) = &*object(&s, id, 4) else { panic!("the image") }; + assert_eq!(img.dict.int(b"BitsPerComponent"), Some(1)); + let px = samples(&s, id); + assert!(px.bilevel); + let at = |x: usize, y: usize| px.data[y * w + x]; + assert_eq!((at(5, 5), at(58, 58)), (255, 255), "specks gone"); + assert_eq!((at(10, 20), at(30, 25), at(49, 29)), (0, 0, 0), "the bar stays"); + assert_eq!(px.data.iter().filter(|&&v| v == 0).count(), 40 * 10); +} + +#[test] +fn hostile_images_are_reported_not_a_panic() { + // Each page draws one image over the whole page. + let images: [(&str, Vec); 6] = [ + ("/Width 100000 /Height 100000 /ColorSpace /DeviceGray /BitsPerComponent 8", vec![0; 64]), + ("/Width 300 /Height 400 /ColorSpace /DeviceGray /BitsPerComponent 8 /Filter /DCTDecode", vec![0xFF, 0xD8, 0xFF, 0xE0, 0, 3, 9]), + ("/Width 300 /Height 400 /ColorSpace /DeviceGray /BitsPerComponent 8", vec![200; 1000]), + ("/Width 1 /Height 1 /ColorSpace /DeviceGray /BitsPerComponent 8", vec![128]), + ("/Width 8 /Height 8 /ColorSpace /DeviceCMYK /BitsPerComponent 8", vec![0; 256]), + ("/Width 300 /Height 400 /ColorSpace /DeviceGray /BitsPerComponent 8 /Mask [250 255]", vec![200; 300 * 400]), + ]; + let n = images.len(); + let kids: String = (0..n).map(|i| format!("{} 0 R ", 3 + 3 * i)).collect(); + let mut pdf = Pdf::new().obj("<< /Type /Catalog /Pages 2 0 R >>").obj(format!("<< /Type /Pages /Kids [{kids}] /Count {n} >>")); + for (i, (dict, data)) in images.iter().enumerate() { + let (img, contents) = (4 + 3 * i, 5 + 3 * i); + pdf = pdf + .obj(format!( + "<< /Type /Page /Parent 2 0 R /MediaBox [0 0 300 400] /Resources << /XObject << /Im0 {img} 0 R >> >> /Contents {contents} 0 R >>" + )) + .stream(&format!("/Type /XObject /Subtype /Image {dict}"), data) + .stream("", "q 300 0 0 400 0 0 cm /Im0 Do Q"); + } + let (mut s, id) = open(pdf.build()); + let report = s.enhance_scanned_pages(id, &[], ScanCleanSettings::at(100)).unwrap(); + assert_eq!(report.images, 0); + assert_eq!(report.not_scanned, vec![3], "a 1 × 1 image is no scan"); + let why: Vec<(usize, &str)> = report.skipped.iter().map(|k| (k.pages[0], k.reason.as_str())).collect(); + assert_eq!(why.len(), 5, "{why:?}"); + assert!(why[0].0 == 0 && why[0].1.contains("80 million"), "{why:?}"); + assert!(why[1].0 == 1 && why[1].1.contains("could not be decoded"), "{why:?}"); + assert!(why[2].0 == 2 && why[2].1.contains("shorter"), "{why:?}"); + assert!(why[3].0 == 4 && why[3].1.contains("colour space"), "{why:?}"); + assert!(why[4].0 == 5 && why[4].1.contains("colour-key mask"), "{why:?}"); + assert_eq!(s.get(id).unwrap().can_undo(), None); + // Pages that don't exist are an error, not a panic. + assert!(matches!(s.enhance_scanned_pages(id, &[99], ScanCleanSettings::default()), Err(EditError::Invalid(_)))); +} + +#[test] +fn protected_documents_keep_their_encryption_and_restrictions() { + let protect = |user: &str, permissions: i32| { + let mut doc = pdfcraft_cos::Document::open(Arc::new(grey_scan())).unwrap(); + doc.set_encryption(&pdfcraft_cos::NewEncryption { + algorithm: pdfcraft_cos::Algorithm::Aes256, + user_password: user, + owner_password: "owner", + permissions, + encrypt_metadata: true, + seed: [7; 32], + }) + .unwrap(); + Arc::new(pdfcraft_cos::write_full(&doc, &Default::default()).unwrap()) + }; + // Opened with the password: cleaned, saved still encrypted, and the clean scan reads back. + let mut s = Session::new(); + let id = s.open("x.pdf", None, protect("pw", -1), Some("pw")).unwrap(); + assert_eq!(s.enhance_scanned_pages(id, &[], ScanCleanSettings::at(100)).unwrap().images, 1); + let saved = s.save_bytes(id).unwrap(); + assert!(pdfcraft_cos::Document::open(saved.clone()).is_err(), "still protected"); + let id2 = s.open("y.pdf", None, saved, Some("pw")).unwrap(); + assert!(grey_at(&samples(&s, id2), 3, 40) >= 235); + // Changes withheld: refused before any work. + let id = s.open("r.pdf", None, protect("", 0b0100), None).unwrap(); + assert_eq!(s.enhance_scanned_pages(id, &[], ScanCleanSettings::default()).unwrap_err(), EditError::NotPermitted("changes to the document")); + assert!(!s.get(id).unwrap().dirty); +} + +#[test] +fn a_result_for_an_older_state_is_refused() { + let (mut s, id) = open(grey_scan()); + let job = s.scan_clean_job(id, &[], ScanCleanSettings::at(70)).unwrap(); + assert_eq!(job.total(), 1); + let cleaned = job.run(|_, _| true).unwrap(); + s.apply(id, Edit::RotatePages { pages: vec![1], degrees: 90 }).unwrap(); + assert!(matches!(s.apply_scan_clean(id, cleaned), Err(EditError::Invalid(_)))); + assert_eq!(s.get(id).unwrap().can_undo(), Some("Rotate page")); + // Cancelled from the progress callback. + let job = s.scan_clean_job(id, &[], ScanCleanSettings::at(70)).unwrap(); + assert!(matches!(job.run(|_, _| false), Err(EditError::Cancelled))); +} + +#[test] +fn edge_shadow_removal_whitens_the_gutter_background_removal_alone_leaves() { + let gutter = |edge_shadows: bool| { + let (mut s, id) = open(grey_scan()); + s.enhance_scanned_pages(id, &[0], ScanCleanSettings { edge_shadows, ..ScanCleanSettings::at(100) }).unwrap(); + let after = samples(&s, id); + [(3, 40), (3, 380), (10, 200)].map(|(x, y)| grey_at(&after, x, y)) + }; + assert!(gutter(false).iter().any(|&v| v < 200), "{:?}", gutter(false)); + assert!(gutter(true).iter().all(|&v| v >= 235), "{:?}", gutter(true)); + // On its own, at level 0, it lightens the gutter's paper to the paper beside it. + let (mut s, id) = open(grey_scan()); + let before = samples(&s, id); + s.enhance_scanned_pages(id, &[0], ScanCleanSettings { level: 0, edge_shadows: true, despeckle: false }).unwrap(); + let after = samples(&s, id); + assert!(grey_at(&after, 3, 40).abs_diff(grey_at(&before, 150, 40)) <= 4, "{} vs {}", grey_at(&after, 3, 40), grey_at(&before, 150, 40)); + assert_eq!(grey_at(&after, 150, 40), grey_at(&before, 150, 40), "the middle is left alone"); +} diff --git a/crates/engine/src/scan_fields.rs b/crates/engine/src/scan_fields.rs index 8287743fc5..769beaefca 100644 --- a/crates/engine/src/scan_fields.rs +++ b/crates/engine/src/scan_fields.rs @@ -25,24 +25,33 @@ const MIN_WORD_HEIGHT: f64 = 4.0; /// Whether `page` is a picture of a page: images cover most of it. pub(crate) fn is_scan(doc: &Document, page: usize, info: &PageInfo) -> bool { + scan_images(doc, page, info).is_some() +} + +/// The raster images (larger than one pixel) `page` draws, each with the share of the page (its +/// crop box) it covers, when together they cover most of it ([`is_scan`]); `None` otherwise. +/// Enhance ▸ Scanned document cleans these up ([`crate::scan_clean`]). +pub(crate) fn scan_images(doc: &Document, page: usize, info: &PageInfo) -> Option> { let [cx0, cy0, cx1, cy1] = info.crop.map(f64::from); let page_area = (cx1 - cx0) * (cy1 - cy0); if !page_area.is_finite() || page_area <= 0.0 { - return false; + return None; } - let covered: f64 = doc + let images: Vec<(pdfcraft_edit::PageImage, f64)> = doc .page_images(page) - .iter() + .into_iter() .filter(|i| !i.is_form && i.width > 1 && i.height > 1) .map(|i| { let r = i.visible_rect.unwrap_or(i.rect); let w = (r[2].min(cx1) - r[0].max(cx0)).max(0.0); let h = (r[3].min(cy1) - r[1].max(cy0)).max(0.0); - w * h + let share = w * h / page_area; + (i, share) }) - .filter(|a| a.is_finite()) - .sum(); - covered >= COVERED * page_area + .filter(|(_, share)| share.is_finite()) + .collect(); + let covered: f64 = images.iter().map(|(_, share)| share).sum(); + (covered >= COVERED).then_some(images) } /// The page as displayed, in points with y up: where detection on a scan works. diff --git a/crates/optimize/src/images.rs b/crates/optimize/src/images.rs index b9761b85fc..f19d7d8fc7 100644 --- a/crates/optimize/src/images.rs +++ b/crates/optimize/src/images.rs @@ -8,7 +8,7 @@ use pdfcraft_cos::{DecodeBudget, Dict, Document, ObjRef, Object, Stream}; use crate::{Compression, ImageSettings, OptimizeError, Report, Settings, Stage}; /// Images with more pixels than this are left alone (memory). -const MAX_PIXELS: u64 = 80_000_000; +pub(crate) const MAX_PIXELS: u64 = 80_000_000; /// The resources of a page, inherited from the page tree if the page has none. pub(crate) fn page_resources(doc: &Document, page: ObjRef) -> Option { @@ -203,7 +203,7 @@ pub(crate) fn components(doc: &Document, cs: &Object) -> Option { } } -fn is_dct(s: &Stream) -> Option { +pub(crate) fn is_dct(s: &Stream) -> Option { match s.dict.get(b"Filter") { None => Some(false), Some(Object::Name(n)) => match n.as_slice() { @@ -288,7 +288,7 @@ pub(crate) fn pixels(doc: &Document, s: &Stream, n: usize) -> Option> { /// /ColorTransform contradict each other (PdfCraft's renderer lets /ColorTransform 0 win where /// the standard lets the marker win), either holds a value other than 0 or 1, or the parameters /// or the JPEG's segments can't be read. -fn ycc_encoded(doc: &Document, s: &Stream) -> Option { +pub(crate) fn ycc_encoded(doc: &Document, s: &Stream) -> Option { let entry = color_transform(doc, s)?; match adobe_transform(&s.raw)? { Some(t @ (0 | 1)) => entry.is_none_or(|c| c == i64::from(t)).then_some(t == 1), @@ -412,7 +412,7 @@ pub(crate) fn resample(px: &[u8], n: usize, w: u32, h: u32, nw: u32, nh: u32) -> }) } -fn jpeg(px: &[u8], n: usize, w: u32, h: u32, quality: u8) -> Option> { +pub(crate) fn jpeg(px: &[u8], n: usize, w: u32, h: u32, quality: u8) -> Option> { let mut out = Vec::new(); let mut enc = image::codecs::jpeg::JpegEncoder::new_with_quality(&mut out, quality.clamp(1, 100)); let ty = if n == 1 { image::ExtendedColorType::L8 } else { image::ExtendedColorType::Rgb8 }; @@ -435,7 +435,7 @@ pub(crate) fn decodable(doc: &Document, s: &Stream) -> Option { /// Whether `decode` is the default decode array for `n` components ([0 1] each), which changes /// nothing. Scanners write it on page images (#490). -fn is_identity_decode(doc: &Document, decode: &Object, n: usize) -> bool { +pub(crate) fn is_identity_decode(doc: &Document, decode: &Object, n: usize) -> bool { let decode = doc.resolve(decode); let Some(a) = decode.as_array() else { return false }; a.len() == 2 * n && a.chunks(2).all(|p| matches!(p, [lo, hi] if doc.resolve(lo).as_f64() == Some(0.0) && doc.resolve(hi).as_f64() == Some(1.0))) diff --git a/crates/optimize/src/lib.rs b/crates/optimize/src/lib.rs index 9f9fcdf2be..cb6b537115 100644 --- a/crates/optimize/src/lib.rs +++ b/crates/optimize/src/lib.rs @@ -27,6 +27,7 @@ mod audit; pub mod fragments; mod images; mod links; +mod samples; mod unused; pub use audit::{SpaceCategory, SpaceUse, audit_space}; @@ -35,6 +36,7 @@ use pdfcraft_cos::{Dict, Document, MAX_PAGE_DECODED, ObjRef, Object, Stream, ser use pdfcraft_filters::{Filter, Params}; pub use images::effective_resolutions; +pub use samples::{NotDecoded, Samples, decode_samples, encode_samples}; #[derive(Debug, thiserror::Error)] pub enum OptimizeError { diff --git a/crates/optimize/src/samples.rs b/crates/optimize/src/samples.rs new file mode 100644 index 0000000000..7f402b51b8 --- /dev/null +++ b/crates/optimize/src/samples.rs @@ -0,0 +1,273 @@ +//! An image's samples read out and written back at the same size, for tools that change the +//! pixels of an image in place (Scan & OCR ▸ Enhance ▸ Scanned document). The image dictionary +//! stays as it was apart from the encoding: its size, colour space, masks, `/Interpolate`, +//! `/Intent`, `/Metadata` and every other entry are kept. + +use pdfcraft_cos::{Document, Object, Stream}; + +use crate::images::{MAX_PIXELS, components, is_dct, is_identity_decode, jpeg, pixels, ycc_encoded}; + +/// An image's samples, one byte each, row-major. +#[derive(Clone, Debug, PartialEq)] +pub struct Samples { + pub width: u32, + pub height: u32, + /// 1 (grey) or 3 (RGB). + pub components: usize, + /// The samples. A bi-level image's are 0 (black) or 255 (white) as drawn, whatever its + /// decode array. + pub data: Vec, + /// Stored as a JPEG (written back as one); otherwise written back losslessly (Flate). + pub jpeg: bool, + /// One bit per pixel (written back as one). + pub bilevel: bool, + /// A bi-level image whose 1 bits are black (decode array `[1 0]`): written back the same way, + /// so its decode array stays true. + pub inverted: bool, +} + +/// Why an image's samples were not read. Each says what the image is, so a report can name it. +#[derive(Clone, Copy, Debug, PartialEq, Eq, thiserror::Error)] +pub enum NotDecoded { + #[error("it is a stencil mask, not a picture")] + StencilMask, + #[error("its colour space is not grey or RGB (CMYK, indexed and special colours are not supported yet)")] + ColourSpace, + #[error("it has {0} bits per sample (8, and 1 for black and white, are supported)")] + BitDepth(i64), + #[error("its decode array remaps the samples")] + Decode, + #[error("it is compressed in a format PdfCraft cannot read back yet (JPEG 2000, JBIG2 or CCITT fax)")] + Filter, + #[error("its JPEG colour transform is ambiguous, so readers may disagree about its colours")] + Ambiguous, + #[error("it is larger than 80 million pixels")] + TooLarge, + #[error("its data could not be decoded, or is shorter than its size says")] + Corrupt, +} + +/// Read the samples of image stream `s` (a dictionary with `/Subtype /Image`). +pub fn decode_samples(doc: &Document, s: &Stream) -> Result { + let d = &s.dict; + if matches!(d.get(b"ImageMask").map(|v| doc.resolve(v)).as_deref(), Some(Object::Bool(true))) { + return Err(NotDecoded::StencilMask); + } + let n = d.get(b"ColorSpace").and_then(|cs| components(doc, cs)).ok_or(NotDecoded::ColourSpace)?; + let bpc = d.get(b"BitsPerComponent").map(|v| doc.resolve(v)).and_then(|v| v.as_int()).unwrap_or(0); + let bilevel = match bpc { + 8 => false, + 1 if n == 1 => true, + other => return Err(NotDecoded::BitDepth(other)), + }; + let jpeg = is_dct(s).ok_or(NotDecoded::Filter)?; + let width = d.get(b"Width").map(|v| doc.resolve(v)).and_then(|v| v.as_int()).and_then(|v| u32::try_from(v).ok()).unwrap_or(0); + let height = d.get(b"Height").map(|v| doc.resolve(v)).and_then(|v| v.as_int()).and_then(|v| u32::try_from(v).ok()).unwrap_or(0); + if width == 0 || height == 0 { + return Err(NotDecoded::Corrupt); + } + if u64::from(width) * u64::from(height) > MAX_PIXELS { + return Err(NotDecoded::TooLarge); + } + if bilevel { + if jpeg { + // A JPEG has 8 bits per sample: the dictionary contradicts the data. + return Err(NotDecoded::Corrupt); + } + let inverted = match d.get(b"Decode") { + None => false, + Some(a) if is_identity_decode(doc, a, 1) => false, + Some(a) if is_inverted_decode(doc, a) => true, + Some(_) => return Err(NotDecoded::Decode), + }; + let data = unpack_bits(s, width, height, inverted).ok_or(NotDecoded::Corrupt)?; + return Ok(Samples { width, height, components: 1, data, jpeg: false, bilevel: true, inverted }); + } + if d.get(b"Decode").is_some_and(|a| !is_identity_decode(doc, a, n)) { + return Err(NotDecoded::Decode); + } + if jpeg && n == 3 && ycc_encoded(doc, s).is_none() { + return Err(NotDecoded::Ambiguous); + } + let data = pixels(doc, s, n).ok_or(NotDecoded::Corrupt)?; + Ok(Samples { width, height, components: n, data, jpeg, bilevel: false, inverted: false }) +} + +/// `[1 0]`: a bi-level image whose 1 bits are black. +fn is_inverted_decode(doc: &Document, decode: &Object) -> bool { + let decode = doc.resolve(decode); + matches!(decode.as_array().map(Vec::as_slice), Some([a, b]) if doc.resolve(a).as_f64() == Some(1.0) && doc.resolve(b).as_f64() == Some(0.0)) +} + +/// The bytes a `w` × `h` bi-level image takes: each row padded to whole bytes. +fn packed_len(w: u32, h: u32) -> Option { + usize::try_from(w).ok()?.div_ceil(8).checked_mul(usize::try_from(h).ok()?) +} + +/// A bi-level image's pixels, 0 (black) or 255 (white) as drawn. +fn unpack_bits(s: &Stream, w: u32, h: u32, inverted: bool) -> Option> { + let stride = usize::try_from(w).ok()?.div_ceil(8); + let need = packed_len(w, h)?; + // Some writers pad the data; anything far longer than the image needs is not this image. + let bits = s.decoded_within(need.saturating_add(need / 2).saturating_add(4096)).ok()?; + let bits = bits.get(..need)?; + let (w, h) = (usize::try_from(w).ok()?, usize::try_from(h).ok()?); + let mut out = Vec::with_capacity(w.checked_mul(h)?); + for row in bits.chunks(stride).take(h) { + for x in 0..w { + let byte = row.get(x / 8).copied().unwrap_or(0); + let one = byte & (0x80 >> (x % 8)) != 0; + // Without a decode array 0 is black and 1 white; [1 0] swaps them. + out.push(if one != inverted { 255 } else { 0 }); + } + } + Some(out) +} + +/// The pixels (0 black … 255 white, threshold at half) packed back to bits. +fn pack_bits(px: &[u8], w: u32, h: u32, inverted: bool) -> Option> { + let stride = usize::try_from(w).ok()?.div_ceil(8); + let mut out = vec![0u8; packed_len(w, h)?]; + let (w, h) = (usize::try_from(w).ok()?, usize::try_from(h).ok()?); + if px.len() != w.checked_mul(h)? { + return None; + } + for (y, row) in px.chunks(w.max(1)).enumerate() { + for (x, &v) in row.iter().enumerate() { + let white = v >= 128; + if white != inverted + && let Some(b) = out.get_mut(y * stride + x / 8) + { + *b |= 0x80 >> (x % 8); + } + } + } + Some(out) +} + +/// Image stream `old` with its samples replaced by `samples` (same size and components): a JPEG +/// stays a JPEG (at `quality`, 1–100), anything else becomes Flate, and a bi-level image stays +/// one bit per pixel with its decode array. Filter parameters go (the new data needs none); every +/// other entry stays. `None` if the samples don't fit the image. +pub fn encode_samples(old: &Stream, samples: &Samples, quality: u8) -> Option { + let Samples { width, height, components: n, data, jpeg: as_jpeg, bilevel, inverted } = samples; + let (w, h, n) = (*width, *height, *n); + if !(n == 1 || n == 3) || data.len() != usize::try_from(w).ok()?.checked_mul(usize::try_from(h).ok()?)?.checked_mul(n)? { + return None; + } + let mut d = old.dict.clone(); + for k in [&b"Filter"[..], b"DecodeParms", b"Length", b"DL"] { + d.remove(k); + } + if *bilevel { + if n != 1 { + return None; + } + d.set(b"BitsPerComponent".to_vec(), Object::Int(1)); + return Some(Stream::flate(d, &pack_bits(data, w, h, *inverted)?)); + } + d.set(b"BitsPerComponent".to_vec(), Object::Int(8)); + if *as_jpeg { + let bytes = jpeg(data, n, w, h, quality)?; + d.set(b"Filter".to_vec(), Object::name("DCTDecode")); + Some(Stream::from_raw(d, bytes)) + } else { + Some(Stream::flate(d, data)) + } +} + +#[cfg(test)] +mod tests { + use super::*; + use pdfcraft_cos::Dict; + + fn image(dict: &[(&str, Object)], raw: Vec) -> Stream { + let mut d = Dict::new(); + d.set(b"Type".to_vec(), Object::name("XObject")); + d.set(b"Subtype".to_vec(), Object::name("Image")); + for (k, v) in dict { + d.set(k.as_bytes().to_vec(), v.clone()); + } + Stream::from_raw(d, raw) + } + + fn doc() -> Document { + let pdf = b"%PDF-1.7\n1 0 obj\n<< /Type /Catalog /Pages 2 0 R >>\nendobj\n2 0 obj\n<< /Type /Pages /Kids [] /Count 0 >>\nendobj\ntrailer\n<< /Root 1 0 R >>\n%%EOF\n"; + Document::open(std::sync::Arc::new(pdf.to_vec())).unwrap() + } + + #[test] + fn grey_samples_round_trip_losslessly_and_keep_the_other_entries() { + let px: Vec = (0..12 * 5).map(|i| (i * 4) as u8).collect(); + let s = image( + &[ + ("Width", Object::Int(12)), + ("Height", Object::Int(5)), + ("ColorSpace", Object::name("DeviceGray")), + ("BitsPerComponent", Object::Int(8)), + ("Interpolate", Object::Bool(true)), + ], + px.clone(), + ); + let got = decode_samples(&doc(), &s).unwrap(); + assert_eq!((got.width, got.height, got.components, got.jpeg, got.bilevel), (12, 5, 1, false, false)); + assert_eq!(got.data, px); + let back = encode_samples(&s, &got, 90).unwrap(); + assert_eq!(back.dict.name(b"Filter"), Some(&b"FlateDecode"[..])); + assert_eq!(back.dict.get(b"Interpolate"), Some(&Object::Bool(true))); + assert_eq!(decode_samples(&doc(), &back).unwrap().data, px); + } + + #[test] + fn bilevel_samples_keep_their_decode_array_and_bits() { + // 10 × 2: first row black on the left half, second row all white; [1 0] makes 1 black. + let raw = vec![0b1111_1000, 0b0000_0000, 0, 0]; + let s = image( + &[ + ("Width", Object::Int(10)), + ("Height", Object::Int(2)), + ("ColorSpace", Object::name("DeviceGray")), + ("BitsPerComponent", Object::Int(1)), + ("Decode", Object::Array(vec![Object::Int(1), Object::Int(0)])), + ], + raw.clone(), + ); + let got = decode_samples(&doc(), &s).unwrap(); + assert!(got.bilevel); + assert_eq!(&got.data[..10], &[0, 0, 0, 0, 0, 255, 255, 255, 255, 255]); + assert!(got.data[10..].iter().all(|&v| v == 255)); + assert!(got.inverted); + let back = encode_samples(&s, &got, 90).unwrap(); + assert_eq!(back.dict.int(b"BitsPerComponent"), Some(1)); + assert!(back.dict.get(b"Decode").is_some()); + assert_eq!(back.decoded().unwrap(), raw); + } + + #[test] + fn what_is_not_supported_says_why_and_never_panics() { + let base = |extra: &[(&str, Object)], raw: Vec| { + let mut d: Vec<(&str, Object)> = vec![ + ("Width", Object::Int(4)), + ("Height", Object::Int(4)), + ("ColorSpace", Object::name("DeviceGray")), + ("BitsPerComponent", Object::Int(8)), + ]; + d.extend(extra.iter().cloned()); + decode_samples(&doc(), &image(&d, raw)) + }; + assert_eq!(base(&[("ColorSpace", Object::name("DeviceCMYK"))], vec![0; 64]), Err(NotDecoded::ColourSpace)); + assert_eq!(base(&[("BitsPerComponent", Object::Int(4))], vec![0; 8]), Err(NotDecoded::BitDepth(4))); + assert_eq!(base(&[("Filter", Object::name("JBIG2Decode"))], vec![0; 8]), Err(NotDecoded::Filter)); + assert_eq!(base(&[("ImageMask", Object::Bool(true))], vec![0; 8]), Err(NotDecoded::StencilMask)); + assert_eq!(base(&[("Decode", Object::Array(vec![Object::Int(1), Object::Int(0)]))], vec![0; 16]), Err(NotDecoded::Decode)); + // Too short, a huge size, a broken JPEG, no size at all. + assert_eq!(base(&[], vec![0; 5]), Err(NotDecoded::Corrupt)); + assert_eq!(base(&[("Width", Object::Int(1 << 30)), ("Height", Object::Int(1 << 30))], vec![0; 16]), Err(NotDecoded::TooLarge)); + assert_eq!(base(&[("Filter", Object::name("DCTDecode"))], vec![0xFF, 0xD8, 1, 2, 3]), Err(NotDecoded::Corrupt)); + assert_eq!(base(&[("Width", Object::Int(-3))], vec![0; 16]), Err(NotDecoded::Corrupt)); + // Samples that don't fit the image are refused when written back. + let s = image(&[("Width", Object::Int(4)), ("Height", Object::Int(4))], vec![0; 16]); + let wrong = Samples { width: 4, height: 4, components: 1, data: vec![0; 15], jpeg: false, bilevel: false, inverted: false }; + assert!(encode_samples(&s, &wrong, 90).is_none()); + } +} diff --git a/crates/ui-egui/src/app.rs b/crates/ui-egui/src/app.rs index 8925a0dca1..1fd46a956f 100644 --- a/crates/ui-egui/src/app.rs +++ b/crates/ui-egui/src/app.rs @@ -179,6 +179,8 @@ impl PdfCraftApp { stamp_draft: Default::default(), optimize_draft: OptimizeDraft::default(), optimize_run: None, + scan_clean_draft: scan_clean_ui::ScanCleanDraft::default(), + scan_clean_run: None, progress_notice: None, page_clipboard: None, grid_drop: None, @@ -508,6 +510,7 @@ impl eframe::App for PdfCraftApp { self.auto_ocr_scanned_page(); self.poll_ocr(); self.poll_optimize(); + self.poll_scan_clean(); self.poll_action(); self.process_file_requests(); #[cfg(not(target_arch = "wasm32"))] diff --git a/crates/ui-egui/src/app/control.rs b/crates/ui-egui/src/app/control.rs index bb5a65cb34..3fe737b40b 100644 --- a/crates/ui-egui/src/app/control.rs +++ b/crates/ui-egui/src/app/control.rs @@ -101,6 +101,7 @@ impl PdfCraftApp { "export-all-images" => Some(Dialog::Export(export_ui::ExportKind::AllImages)), "accessibility-options" => Some(Dialog::AccessibilityOptions), "recognize-text" => Some(Dialog::RecognizeText), + "enhance-scan" => Some(Dialog::EnhanceScan), "js-console" => Some(Dialog::JsConsole), "document-js" => Some(Dialog::DocumentJs), "preferences" => Some(Dialog::Preferences), diff --git a/crates/ui-egui/src/app/frame.rs b/crates/ui-egui/src/app/frame.rs index e0a8370c60..3d6f493e10 100644 --- a/crates/ui-egui/src/app/frame.rs +++ b/crates/ui-egui/src/app/frame.rs @@ -5,6 +5,7 @@ impl PdfCraftApp { pub(super) fn show_progress(&mut self, ctx: &egui::Context) { if widgets::progress_notice(self, ctx) { self.cancel_optimize(); + self.cancel_scan_clean(); } } diff --git a/crates/ui-egui/src/commands.rs b/crates/ui-egui/src/commands.rs index acccbcc55a..dec99e8b58 100644 --- a/crates/ui-egui/src/commands.rs +++ b/crates/ui-egui/src/commands.rs @@ -500,6 +500,7 @@ impl PdfCraftApp { "export.image" => self.dialog = Some(Dialog::Export(crate::export_ui::ExportKind::Image)), "export.text" => self.dialog = Some(Dialog::Export(crate::export_ui::ExportKind::Text)), "ocr.recognize" => self.dialog = Some(Dialog::RecognizeText), + "ocr.enhance" => self.dialog = Some(Dialog::EnhanceScan), "ocr.recognize_batch" => self.ocr_files_dialog(), "edit.edit_text" => { self.quick_tool = crate::QuickTool::EditText; diff --git a/crates/ui-egui/src/dialogs.rs b/crates/ui-egui/src/dialogs.rs index c21604a4da..a77b0e4d19 100644 --- a/crates/ui-egui/src/dialogs.rs +++ b/crates/ui-egui/src/dialogs.rs @@ -56,6 +56,7 @@ struct Pending { open_revision: Option, a11y_now: bool, ocr_now: bool, + scan_clean_now: bool, compare_now: bool, images_now: bool, stamp_now: bool, @@ -372,6 +373,12 @@ fn dialog_body(ui: &mut egui::Ui, app: &mut PdfCraftApp, dialog: Dialog, req: &m req.close = go || cancel; return false; } + Dialog::EnhanceScan => { + let (go, cancel) = crate::scan_clean_ui::body(ui, app, &t); + req.scan_clean_now = go; + req.close = go || cancel; + return false; + } Dialog::AccessibilityOptions => { let (start, cancel) = crate::a11y_ui::options_body(ui, app, &t); req.a11y_now = start; @@ -1818,6 +1825,9 @@ fn finish_close(app: &mut PdfCraftApp, dialog: Dialog, modal_closed: bool, req: if req.ocr_now { app.start_ocr(); } + if req.scan_clean_now { + app.start_scan_clean(); + } if req.compare_now { app.run_compare(); } diff --git a/crates/ui-egui/src/i18n/de.tsv b/crates/ui-egui/src/i18n/de.tsv index 8586a21d43..076fb6bf66 100644 --- a/crates/ui-egui/src/i18n/de.tsv +++ b/crates/ui-egui/src/i18n/de.tsv @@ -2558,3 +2558,22 @@ animation effects Off Aus Rotate the selected pages clockwise Ausgewählte Seiten im Uhrzeigersinn drehen Extract the selected pages… Ausgewählte Seiten extrahieren… Delete the selected pages Ausgewählte Seiten löschen + Enhance scanned file… Gescannte Datei verbessern… + Enhance Scanned File Gescannte Datei verbessern + Cleans up the scanned pictures in place: recognized text, comments and links on the pages stay as they are. Bereinigt die gescannten Bilder an Ort und Stelle: erkannter Text, Kommentare und Links auf den Seiten bleiben erhalten. + Filters Filter + Background removal Hintergrund entfernen + Evens out the paper so a grey or yellowed background turns white, and darkens faint strokes. 0 is off. Gleicht das Papier aus, sodass ein grauer oder vergilbter Hintergrund weiß wird, und dunkelt blasse Striche nach. 0 ist aus. + Edge shadow removal Randschatten entfernen + Lightens shadows along the edges of the scan, such as a book's gutter or a copier's dark border Hellt Schatten an den Rändern des Scans auf, etwa den Bundsteg eines Buchs oder den dunklen Rand eines Kopierers + Despeckle Flecken entfernen + Removes isolated dark specks such as dust and scanner noise. The only filter black and white scans take. Entfernt einzelne dunkle Flecken wie Staub und Scannerrauschen. Der einzige Filter für Schwarzweiß-Scans. + Enhance Verbessern + No scanned pages were cleaned up. Es wurden keine gescannten Seiten bereinigt. + Scanned pages are already being enhanced Gescannte Seiten werden bereits verbessert + Couldn't enhance the scanned pages Die gescannten Seiten konnten nicht verbessert werden + Enhancing scans… image {d} of {t} Scans werden verbessert… Bild {d} von {t} + Enhancing the scanned pages was cancelled Das Verbessern der gescannten Seiten wurde abgebrochen +@plural Cleaned up the scan on {n} page.|Cleaned up the scans on {n} pages. Scan auf {n} Seite bereinigt.|Scans auf {n} Seiten bereinigt. +@plural {n} page is not a scan.|{n} pages are not scans. {n} Seite ist kein Scan.|{n} Seiten sind keine Scans. +@plural {n} image was left as it was (page {p}: {why}).|{n} images were left as they were (page {p}: {why}, …). {n} Bild blieb unverändert (Seite {p}: {why}).|{n} Bilder blieben unverändert (Seite {p}: {why}, …). diff --git a/crates/ui-egui/src/lib.rs b/crates/ui-egui/src/lib.rs index 65bc7e4b1b..e61e3827a5 100644 --- a/crates/ui-egui/src/lib.rs +++ b/crates/ui-egui/src/lib.rs @@ -162,6 +162,7 @@ use workflows::js_ui; use workflows::ocr_ui; use workflows::optimize_ui; use workflows::print_ui; +use workflows::scan_clean_ui; use workflows::web_print; /// Files dropped on a document's page grid or its Pages panel. @@ -494,6 +495,9 @@ pub struct PdfCraftApp { pub optimize_draft: OptimizeDraft, /// The running optimization (Optimize PDF ▸ Advanced optimization). pub optimize_run: Option, + /// Scan & OCR ▸ Enhance scanned file: the dialog's choices and the running clean-up. + pub scan_clean_draft: scan_clean_ui::ScanCleanDraft, + pub scan_clean_run: Option, /// A background job's progress card (see [`widgets::progress_notice`]). pub progress_notice: Option, /// Pages copied or cut in Organize Pages, ready to paste (into any document). diff --git a/crates/ui-egui/src/state.rs b/crates/ui-egui/src/state.rs index dd3b3e90aa..72d3b6ddf2 100644 --- a/crates/ui-egui/src/state.rs +++ b/crates/ui-egui/src/state.rs @@ -162,6 +162,8 @@ pub enum Dialog { AccessibilityOptions, /// Scan & OCR ▸ Recognize text. RecognizeText, + /// Scan & OCR ▸ Enhance scanned file. + EnhanceScan, /// The JavaScript console (⌘J). JsConsole, /// Document JavaScripts. diff --git a/crates/ui-egui/src/workflows.rs b/crates/ui-egui/src/workflows.rs index 63686a70cc..9314914ec4 100644 --- a/crates/ui-egui/src/workflows.rs +++ b/crates/ui-egui/src/workflows.rs @@ -11,4 +11,5 @@ pub(super) mod js_ui; pub(super) mod ocr_ui; pub(super) mod optimize_ui; pub(super) mod print_ui; +pub(super) mod scan_clean_ui; pub(super) mod web_print; diff --git a/crates/ui-egui/src/workflows/scan_clean_ui.rs b/crates/ui-egui/src/workflows/scan_clean_ui.rs new file mode 100644 index 0000000000..a275b5af20 --- /dev/null +++ b/crates/ui-egui/src/workflows/scan_clean_ui.rs @@ -0,0 +1,253 @@ +//! Scan & OCR ▸ Enhance scanned file: the dialog (pages, background removal level, edge +//! shadows, despeckle) and the background run with its progress card. The scans are cleaned +//! on a worker thread ([`pdfcraft_engine::scan_clean`]); the result is applied as one undoable +//! step when it is done, unless the document changed meanwhile. + +use std::sync::{Arc, Mutex}; + +use egui::{Align, Layout}; +use pdfcraft_engine::scan_clean::{ScanCleanReport, ScanCleanSettings, ScanCleaned}; +use pdfcraft_engine::{DocId, EditError}; + +use crate::ocr_ui::OcrPages; +use crate::theme::{self, Tokens}; +use crate::{PdfCraftApp, widgets}; + +/// The dialog's choices (kept between runs). +#[derive(Clone, Debug, PartialEq)] +pub struct ScanCleanDraft { + pub pages: OcrPages, + pub from: usize, + pub to: usize, + pub settings: ScanCleanSettings, +} + +impl Default for ScanCleanDraft { + fn default() -> Self { + ScanCleanDraft { pages: OcrPages::All, from: 1, to: 1, settings: ScanCleanSettings::default() } + } +} + +/// Progress of a run: images done, total, the result once finished, and a cancel request. +#[derive(Default)] +pub struct ScanCleanProgress { + pub done: usize, + pub total: usize, + pub result: Option>, + pub cancel: bool, +} + +pub struct ScanCleanRun { + pub doc: DocId, + pub progress: Arc>, +} + +pub(crate) fn body(ui: &mut egui::Ui, app: &mut PdfCraftApp, t: &Tokens) -> (bool, bool) { + let pages = app.active_ids().and_then(|(_, id)| app.session.get(id)).map_or(1, |d| d.info.pages.len().max(1)); + let d = &mut app.scan_clean_draft; + d.to = d.to.clamp(1, pages); + d.from = d.from.clamp(1, d.to); + ui.label(egui::RichText::new(tl!("Enhance Scanned File")).font(theme::semibold(18.0))); + ui.add_space(4.0); + ui.label( + egui::RichText::new(tl!("Cleans up the scanned pictures in place: recognized text, comments and links on the pages stay as they are.")) + .small() + .color(t.text_muted), + ); + ui.add_space(8.0); + let group = |ui: &mut egui::Ui, title: &str, body: &mut dyn FnMut(&mut egui::Ui)| { + ui.label(egui::RichText::new(tl!(title)).font(theme::semibold(13.0))); + egui::Frame::new().fill(t.hover).corner_radius(egui::CornerRadius::same(6)).inner_margin(egui::Margin::same(10)).show(ui, |ui| { + ui.set_width(ui.available_width()); + body(ui); + }); + ui.add_space(8.0); + }; + group(ui, "Pages", &mut |ui| { + widgets::radio_value(ui, &mut d.pages, OcrPages::All, tl!("All pages")); + widgets::radio_value(ui, &mut d.pages, OcrPages::Current, tl!("Current page")); + ui.horizontal(|ui| { + let from = widgets::radio_value(ui, &mut d.pages, OcrPages::Range, tl!("From")); + let on = d.pages == OcrPages::Range; + ui.add_enabled(on, egui::DragValue::new(&mut d.from).range(1..=pages)).labelled_by(from.id); + let to = ui.label(tl!("to")); + ui.add_enabled(on, egui::DragValue::new(&mut d.to).range(1..=pages)).labelled_by(to.id); + }); + }); + group(ui, "Filters", &mut |ui| { + ui.horizontal(|ui| { + let label = ui.label(tl!("Background removal")); + ui.spacing_mut().slider_width = 220.0; + ui.add(egui::Slider::new(&mut d.settings.level, 0..=pdfcraft_engine::camera_clean::MAX_LEVEL).trailing_fill(true)) + .labelled_by(label.id) + .on_hover_text(tl!("Evens out the paper so a grey or yellowed background turns white, and darkens faint strokes. 0 is off.")); + }); + ui.add(widgets::checkbox(&mut d.settings.edge_shadows, tl!("Edge shadow removal"))) + .on_hover_text(tl!("Lightens shadows along the edges of the scan, such as a book's gutter or a copier's dark border")); + ui.add(widgets::checkbox(&mut d.settings.despeckle, tl!("Despeckle"))) + .on_hover_text(tl!("Removes isolated dark specks such as dust and scanner noise. The only filter black and white scans take.")); + }); + ui.add_space(6.0); + let (mut go, mut cancel) = (false, false); + ui.horizontal(|ui| { + ui.with_layout(Layout::right_to_left(Align::Center), |ui| { + let ready = d.settings.changes_something(); + if ui.add_enabled_ui(ready, |ui| widgets::pill_button(ui, tl!("Enhance"), true)).inner.clicked() { + go = true; + } + if widgets::pill_button(ui, tl!("Cancel"), false).clicked() { + cancel = true; + } + }) + }); + (go, cancel) +} + +/// What a finished run says: what was cleaned, what was not a scan, and what was left and why. +fn summary(r: &ScanCleanReport) -> String { + let lang = crate::i18n::current(); + let mut out = if r.pages.is_empty() { + tl!("No scanned pages were cleaned up.").to_string() + } else { + crate::i18n::trn(lang, r.pages.len() as u64, "Cleaned up the scan on {n} page.", "Cleaned up the scans on {n} pages.") + }; + if !r.not_scanned.is_empty() { + out.push(' '); + out.push_str(&crate::i18n::trn(lang, r.not_scanned.len() as u64, "{n} page is not a scan.", "{n} pages are not scans.")); + } + if let Some(first) = r.skipped.first() { + out.push(' '); + let page = first.pages.first().map_or_else(String::new, |p| (p + 1).to_string()); + // The reason is the engine's own sentence, in English. + out.push_str(&crate::i18n::fmt( + &crate::i18n::trn( + lang, + r.skipped.len() as u64, + "{n} image was left as it was (page {p}: {why}).", + "{n} images were left as they were (page {p}: {why}, …).", + ), + &[("p", &page), ("why", &first.reason)], + )); + } + out +} + +impl PdfCraftApp { + /// Clean up the scans on the pages chosen in the dialog, in the background. + pub fn start_scan_clean(&mut self) { + let Some((vi, id)) = self.active_ids() else { return }; + if self.scan_clean_run.is_some() { + self.notify_tr("Scanned pages are already being enhanced"); + return; + } + // What's typed in a form field is part of the document (#166). + if !self.commit_form_typing() { + return; + } + let d = &self.scan_clean_draft; + let pages: Vec = match d.pages { + OcrPages::All => Vec::new(), + OcrPages::Current => self.views.get(vi).map(|v| vec![v.current]).unwrap_or_default(), + OcrPages::Range => (d.from.saturating_sub(1)..d.to).collect(), + }; + let job = match self.session.scan_clean_job(id, &pages, d.settings) { + Ok(job) => job, + Err(e) => return self.notify(e.to_string()), + }; + let progress = Arc::new(Mutex::new(ScanCleanProgress { total: job.total(), ..Default::default() })); + let p = progress.clone(); + let work = move || { + let result = job.run(|done, total| { + let Ok(mut s) = p.lock() else { return false }; + s.done = done; + s.total = total; + !s.cancel + }); + if let Ok(mut s) = p.lock() { + s.result = Some(result); + } + }; + #[cfg(not(target_arch = "wasm32"))] + if self.run_inline { + work(); + } else if let Err(e) = std::thread::Builder::new().name("pdfcraft-scan-clean".into()).spawn(work) { + // No worker: report the failure instead of waiting for a result that never comes. + if let Ok(mut s) = progress.lock() { + s.result = Some(Err(EditError::Invalid(e.to_string()))); + } + } + #[cfg(target_arch = "wasm32")] + work(); + self.scan_clean_run = Some(ScanCleanRun { doc: id, progress }); + self.poll_scan_clean(); + } + + /// Stop a running clean-up (nothing changes). + pub fn cancel_scan_clean(&mut self) { + if let Some(r) = &self.scan_clean_run + && let Ok(mut s) = r.progress.lock() + { + s.cancel = true; + } + } + + /// Show progress; apply the result once the worker is done. + pub(crate) fn poll_scan_clean(&mut self) { + let Some(run) = self.scan_clean_run.as_ref() else { return }; + let (doc, progress) = (run.doc, run.progress.clone()); + let Ok(mut s) = progress.lock() else { + // The worker panicked while holding the lock: it will never report back. + self.scan_clean_run = None; + self.progress_notice = None; + self.notify_tr("Couldn't enhance the scanned pages"); + return; + }; + let Some(result) = s.result.take() else { + let (cancelling, done, total) = (s.cancel, s.done, s.total); + drop(s); + let label = if cancelling { + tl!("Cancelling…").to_string() + } else { + crate::i18n::fmt( + tl!("Enhancing scans… image {d} of {t}"), + &[("d", &(done + 1).min(total.max(1)).to_string()), ("t", &total.max(1).to_string())], + ) + }; + let fraction = if total == 0 { 1.0 } else { done.min(total) as f32 / total as f32 }; + self.progress_notice = Some(crate::widgets::ProgressNotice { label, fraction, cancellable: !cancelling }); + if let Some(ctx) = &self.ctx { + ctx.request_repaint_after(std::time::Duration::from_millis(50)); + } + return; + }; + drop(s); + self.scan_clean_run = None; + self.progress_notice = None; + match result.and_then(|cleaned| self.session.apply_scan_clean(doc, cleaned)) { + Ok(report) => { + let msg = summary(&report); + self.notify(msg); + } + Err(EditError::Cancelled) => self.notify_tr("Enhancing the scanned pages was cancelled"), + Err(e) => self.notify(e.to_string()), + } + } +} + +#[cfg(test)] +mod tests { + use super::*; + use pdfcraft_engine::scan_clean::SkippedImage; + + #[test] + fn the_summary_says_what_was_done_and_what_was_left() { + let r = ScanCleanReport { + pages: vec![0, 2], + images: 1, + not_scanned: vec![1], + skipped: vec![SkippedImage { pages: vec![3], name: "Im0".into(), reason: "it is CMYK".into() }], + }; + assert_eq!(summary(&r), "Cleaned up the scans on 2 pages. 1 page is not a scan. 1 image was left as it was (page 4: it is CMYK)."); + assert_eq!(summary(&ScanCleanReport::default()), "No scanned pages were cleaned up."); + } +} diff --git a/crates/ui-egui/tests/it/main.rs b/crates/ui-egui/tests/it/main.rs index a5c3192358..2df736c14c 100644 --- a/crates/ui-egui/tests/it/main.rs +++ b/crates/ui-egui/tests/it/main.rs @@ -68,6 +68,7 @@ mod reading_colors; mod recovery; mod redact; mod reload; +mod scan_clean; mod side_panels; mod signing; mod stamps; diff --git a/crates/ui-egui/tests/it/scan_clean.rs b/crates/ui-egui/tests/it/scan_clean.rs new file mode 100644 index 0000000000..afb4a4f75c --- /dev/null +++ b/crates/ui-egui/tests/it/scan_clean.rs @@ -0,0 +1,77 @@ +//! Scan & OCR ▸ Enhance scanned file in the real shell (egui_kittest): the dialog, the run and +//! its report, and that the page keeps its comment while its scan is cleaned in place. + +use egui_kittest::Harness; +use egui_kittest::kittest::Queryable; +use pdfcraft_engine::{Edit, NewAnnotation, NoteIcon, Session, Shape, Style}; +use pdfcraft_ui_egui::PdfCraftApp; + +/// A two-page PDF: page 1 a grey scan with a darker gutter and text bars (and a sticky note), +/// page 2 the same scan again. +fn scan() -> Vec { + let img = image::GrayImage::from_fn(300, 400, |x, y| { + let light = 0.55 + 0.45 * (x as f32 / 60.0).min(1.0); + let ink = if (60..360).contains(&y) && (y - 60) % 40 < 6 && (60..240).contains(&x) { 0.12 } else { 0.78 }; + image::Luma([(ink * light * 255.0) as u8]) + }); + let mut png = Vec::new(); + img.write_to(&mut std::io::Cursor::new(&mut png), image::ImageFormat::Png).unwrap(); + let mut s = Session::new(); + let bytes = s.create_from_images(&[("a.png".into(), png.clone()), ("b.png".into(), png)]).unwrap(); + let id = s.open("scan.pdf", None, bytes, None).unwrap(); + let shape = Shape::Note { at: [20.0, 20.0], icon: NoteIcon::Comment }; + let note = NewAnnotation { page: 0, style: Style::default_for(&shape), shape, contents: "Keep me".into(), author: "Tester".into() }; + s.apply(id, Edit::AddAnnotation(note)).unwrap(); + s.save_bytes(id).unwrap().to_vec() +} + +fn harness() -> Harness<'static, PdfCraftApp> { + let mut h = Harness::builder().with_size(egui::vec2(1400.0, 900.0)).build_eframe(|_cc| { + let mut app = PdfCraftApp::new(); + app.set_option("language", "en").unwrap(); + app.open_bytes("scan.pdf", None, scan()).unwrap(); + app.run_inline = true; + app + }); + h.run_steps(4); + h +} + +#[test] +fn enhance_scanned_file_cleans_the_chosen_pages_and_says_what_it_did() { + let mut h = harness(); + let id = h.state().active_ids().unwrap().1; + assert!(h.state().session.get(id).unwrap().comment_props(0, 0).is_some()); + assert!(h.state_mut().execute("ocr.enhance")); + h.run_steps(2); + assert_eq!(h.query_all_by_label("Background removal").count(), 2, "the label and the slider it names"); + h.get_by_label("Edge shadow removal"); + h.get_by_label("Despeckle"); + h.get_by_role_and_label(egui::accesskit::Role::RadioButton, "Current page").click(); + h.run_steps(1); + h.get_by_label("Enhance").click(); + h.run_steps(3); + let app = h.state(); + assert!(app.scan_clean_run.is_none() && app.dialog.is_none(), "finished and closed"); + let toast = app.toast.clone().unwrap().0; + assert_eq!(toast, "Cleaned up the scan on 1 page."); + let doc = app.session.get(id).unwrap(); + assert_eq!(doc.can_undo(), Some("Enhance scanned file")); + assert_eq!(doc.info.pages.len(), 2); + assert_eq!(doc.comment_props(0, 0).map(|p| p.author), Some("Tester".to_string()), "the sticky note stays"); +} + +#[test] +fn nothing_to_enhance_is_said_not_done() { + let mut h = harness(); + let id = h.state().active_ids().unwrap().1; + h.state_mut().scan_clean_draft.settings = pdfcraft_engine::scan_clean::ScanCleanSettings { level: 0, edge_shadows: false, despeckle: false }; + assert!(h.state_mut().execute("ocr.enhance")); + h.run_steps(2); + // With every filter off the button is disabled; starting anyway changes nothing. + h.state_mut().start_scan_clean(); + h.run_steps(2); + let app = h.state(); + assert!(app.toast.clone().unwrap().0.starts_with("No scanned pages were cleaned up."), "{:?}", app.toast); + assert_eq!(app.session.get(id).unwrap().can_undo(), None); +} diff --git a/docs/gaps.md b/docs/gaps.md index 56ec3ae748..ab6f083ab1 100644 --- a/docs/gaps.md +++ b/docs/gaps.md @@ -1,5 +1,6 @@ # Where PdfCraft falls short of Acrobat Pro +> **Last reviewed:** 2026-10-28 | **Last updated:** 2026-10-28 | **Change:** minor (M10, #4119 partial: Enhance scanned file cleans scanned pages in place: background removal, edge shadow removal, despeckle; estimates unchanged) > **Last reviewed:** 2026-10-25 · **Last updated:** 2026-10-25 · **Change:** minor (gaps 5 and 14 current state: native release Hans face via LXGW WenKai Lite; Arabic logical catalog, explicit RTL direction and mirrored stock chrome, with native/browser/IME workflows unqualified; no readiness re-estimate) > **Last reviewed:** 2026-10-15 · **Last updated:** 2026-10-19 · **Change:** minor (gap 2: composite fonts with `/ToUnicode /Identity-H` read as Unicode, hayro-interpret patch 44; estimates unchanged) · **Target:** Adobe Acrobat Pro (Acrobat DC, continuous track 26.002.21931, macOS) @@ -164,7 +165,9 @@ update the numbers. ### 7. OCR reads only unaccented Latin - **Missing:** every other script and accented Latin; editable-text output; deskew, rotation, - despeckle, background and camera clean-up; MRC compression; pluggable engines; OCR in the + despeckle and background clean-up as options of Recognize text (Enhance scanned file does them + on grey, RGB and 1-bit scans, not CMYK, indexed, JBIG2 or CCITT ones); MRC compression; + pluggable engines; OCR in the browser; scanner input. - **Evidence:** `ocr.ocr-languages` partial ("the models read only the Latin alphabet without accents"); 19 of 26 OCR features planned; [#744](https://github.com/storytold/pdfcraft/issues/744) diff --git a/docs/roadmap.md b/docs/roadmap.md index 25205e92c1..6d910fc795 100644 --- a/docs/roadmap.md +++ b/docs/roadmap.md @@ -1,5 +1,6 @@ # PdfCraft roadmap: milestones and what's next +> **Last reviewed:** 2026-10-28 | **Last updated:** 2026-10-28 | **Change:** minor (M10, #4119 partial: Enhance scanned file cleans scanned pages in place: background removal, edge shadow removal, despeckle; estimates unchanged) > **Last reviewed:** 2026-10-27 | **Last updated:** 2026-10-27 | **Change:** minor (M10: OCR model downloads use verified ONNX pairs with legacy discovery fallback; estimates unchanged) > **Last reviewed:** 2026-10-25 | **Last updated:** 2026-10-25 | **Change:** minor (M14 localization current state: Arabic logical catalog, explicit RTL label direction and mirrored stock chrome; native/browser RTL workflows unqualified. Native release Hans face via LXGW WenKai Lite; estimates unchanged) > **Last reviewed:** 2026-10-23 · **Last updated:** 2026-10-23 · **Change:** minor (M7, #2636 partial: pasting added text boxes and images in the engine and through `content_copy`; estimates unchanged) @@ -194,7 +195,7 @@ total matches the parity estimate. | M7 | Content editing (text, images, header/footer, watermark) | 250–500 | 20% | 100–180 | Done: header & footer, watermarks, backgrounds, Bates; added text and images that stay editable (move, resize, format, rotate, flip, crop, replace); links (Link tool, Link Properties, create from URLs, remove all). Missing: editing existing text and images in place (the longest pole), image/PDF watermarks | | M8 | Security + redaction | 100–180 | 68% | 15–30 | Done: opening protected documents, permissions, Protect Using Password, Remove security; redaction (mark text/areas/pages, Search & Redact with patterns, apply removing glyphs, image pixels, vectors, XObject content, annotations and fields, verification, full rewrite on save); Remove hidden information and Sanitize.
Done since: redaction codes (U.S. FOIA and Privacy Act sets as overlay text, in Redaction properties and `redact_mark`). Redaction cleans indirect and form structure references, preserving surviving form tags. Redaction retires inherited resource dictionaries, so a parent `/Pages` can't keep a removed XObject reachable; unredacted pages are left untouched. Image redaction removes alternate images and unsupported masks only on affected images. Changed marked-content spans lose replacement descriptions; untouched spans keep theirs. Hidden-text cleanup retires unused named descriptions across inherited/shared page and form resources (64 focused tests and scoped Rust 1.96 Clippy/wasm checks pass; publication CI pending). The redaction audit counts a filled cover that hides half of any one word of a long line (#2662), applied or not. Phone-pattern redaction also matches labelled Italian geographic numbers with `+39` and a slash after the area code (#3205 follow-up); unlabelled international slash numbers remain unsupported. Missing: certificate security, custom redaction code sets and pattern locales, DCT re-encoding | | M9 | Signatures (PAdES, validation) | 160–280 | 60% | 40–70 | Done: new `sign` crate (DER, X.509, CMS, PKCS #12 on RustCrypto; aws-lc-rs for RSA private keys); PAdES B-B signing (visible, invisible, existing fields, certification with DocMDP); validation with trust store and changes-after-signing classification; self-signed digital IDs; Signatures panel, message bar, sign dialogs; agent tools. Checked with pdfsig and OpenSSL.
Done since: macOS Keychain and Windows Current User Personal store signing (software-backed CNG keys), certificate viewer, signed documents protected from rewrites, page validation overlays separate from original signature artwork (baked Not Verified captions remain; #79 open), snapshot tabs that reuse instead of multiplying. Missing: timestamps (B-T), LTV (DSS, OCSP, CRL), FieldMDP, smart cards, PKCS #11 | -| M10 | OCR, create, export, print | 200–350 | 33% | 120–215 | Done: create from blank/text/PNG/JPEG/TIFF (multi-page)/GIF/BMP; export PNG/JPEG/TIFF and text; Print (Acrobat's sizing, n-up, booklet, poster, comments & forms, preview, the CUPS and Windows spoolers, print-ready PDF).
Done since: embedded/72/custom DPI choices for image imports; export all images; OCR (searchable image for pages, ranges and multiple files); single-sided cut-and-stack imposition with cut marks, through Print and doc_print; architectural/ANSI/large ISO sheet presets, Match page at 100% and PDF-scale preview (#599 follow-up); Create PDF from multiple files (PDFs, images and text, edited as pages in the grid); OCR can also read pictures on pages that have text, without adding that text twice (#1028). Model staging fetches SHA-256-verified ONNX detection/recognition files after the upstream RTEN URL brownout; runtime discovery prefers complete ONNX pairs per directory, then complete legacy RTEN pairs, without mixing formats or changing directory precedence. Done since: Word invoice export uses borderless inferred tables, floating positioned text/artwork only where text overlaps an image or is rotated, and per-page sizes, covered by synthetic invoice automation regressions (#773); printing on the web through the browser's print dialog, sheets drawn by PdfCraft (gap 4). Lossless image export applies per-component Decode, including Indexed mapping (#1201 partial). Existing 16-bit precision and color-management limits remain. #3532 partial: vector printing retains existing appearances with absent Subtype through a print-owned Form copy; source artwork stays unchanged. #3120: Office export reads form XObjects 50 levels deep, as the renderer does, and up to 50 000 a page; pages past that are listed in `truncated_pages`. Word table cells use the grid columns they span, including cells after a merged cell. Done since: text in a table slot covered by another cell's span remains as paragraph text in Word, HTML, RTF and PowerPoint. Missing: OCR languages beyond Latin, editable-text OCR output, full Office layout fidelity/import. Done since: dotted-leader contents entries stay in separate complete paragraphs (#1757) #2156 candidate: last selected print paper persists across sessions; other print preferences unchanged. Image export applies soft-mask Decode opacity intervals before resizing, including indirect entries and inverted ranges. Vector printing gives annotation appearances unoccupied XObject names so existing page artwork survives. | +| M10 | OCR, create, export, print | 200–350 | 33% | 120–215 | Done: create from blank/text/PNG/JPEG/TIFF (multi-page)/GIF/BMP; export PNG/JPEG/TIFF and text; Print (Acrobat's sizing, n-up, booklet, poster, comments & forms, preview, the CUPS and Windows spoolers, print-ready PDF).
Done since: embedded/72/custom DPI choices for image imports; export all images; OCR (searchable image for pages, ranges and multiple files); single-sided cut-and-stack imposition with cut marks, through Print and doc_print; architectural/ANSI/large ISO sheet presets, Match page at 100% and PDF-scale preview (#599 follow-up); Create PDF from multiple files (PDFs, images and text, edited as pages in the grid); OCR can also read pictures on pages that have text, without adding that text twice (#1028). Model staging fetches SHA-256-verified ONNX detection/recognition files after the upstream RTEN URL brownout; runtime discovery prefers complete ONNX pairs per directory, then complete legacy RTEN pairs, without mixing formats or changing directory precedence. Done since: Word invoice export uses borderless inferred tables, floating positioned text/artwork only where text overlaps an image or is rotated, and per-page sizes, covered by synthetic invoice automation regressions (#773); printing on the web through the browser's print dialog, sheets drawn by PdfCraft (gap 4). Lossless image export applies per-component Decode, including Indexed mapping (#1201 partial). Existing 16-bit precision and color-management limits remain. #3532 partial: vector printing retains existing appearances with absent Subtype through a print-owned Form copy; source artwork stays unchanged. #3120: Office export reads form XObjects 50 levels deep, as the renderer does, and up to 50 000 a page; pages past that are listed in `truncated_pages`. Word table cells use the grid columns they span, including cells after a merged cell. Done since: text in a table slot covered by another cell's span remains as paragraph text in Word, HTML, RTF and PowerPoint. Missing: OCR languages beyond Latin, editable-text OCR output, full Office layout fidelity/import. Done since: dotted-leader contents entries stay in separate complete paragraphs (#1757) #2156 candidate: last selected print paper persists across sessions; other print preferences unchanged. Image export applies soft-mask Decode opacity intervals before resizing, including indirect entries and inverted ranges. Vector printing gives annotation appearances unoccupied XObject names so existing page artwork survives. Done since: Enhance scanned file cleans scanned pages in place (background removal, edge shadow removal, despeckle; the OCR text layer and comments stay; #4119 partial). | | M11 | Optimize, preflight, PDF/A/X/UA, print production | 200–350 | 18% | 80–140 | Done: new `optimize` crate: Reduce File Size and the PDF Optimizer (images measured where drawn, bicubic downsampling, JPEG/ZIP recompression only when smaller, discard objects and user data, Flate clean-up, resource merging, object streams). Target-version compatibility is partial: newer content is not detected or converted. Content stream boundaries preserve comments, split operands and graphics state when image fragments merge. Image downsampling includes page-local `/UserUnit` scaling when measuring physical resolution. Done since: reject invalid image and soft-mask dimensions with checked sizing (#2178). Missing: fonts and transparency panels, space audit, preflight, PDF/A/X/UA | | M12 | Accessibility, compare, measure, search, XFA | 200–380 | 30% | 90–160 | Done: new `a11y` crate with the Accessibility Checker (all 32 rules, report, Fix/Skip/Explain, options dialog and results panel, agent tools). Locatable figure, annotation and form findings highlight their page bounds. Character encoding accepts Identity-H/V with Adobe CJK collections and predefined CJK CMaps independently (#3673); per-glyph validation and embedded-font extraction remain separate.
Done since: 2D distance, perimeter and area measurements with persistent viewport calibration, snapping, live information and CSV export; a bounded, text-only auto-tag proposal tool for one page. Marking a figure decorative rewrites only that stream (#1324). An XFA radio item whose value contains a space is written back unchanged (#1319). Done since: #2569: Growable multiline XFA text reflows after edits, with undo/redo, save/reopen and print coverage; oversized fields still become read-only page text. Structured-data synchronization warnings retain accepted whitespace without relayout from stale data. Done since: #2572: Every repeated page-master button resolves its own field to the shared XFA click script, including print buttons. Done since: #2570: Dynamic XFA rich-text fields carry exData/maxLength into AcroForm MaxLen for reported and enforced Unicode character limits, retained after save/reopen. Done since: #2589: Active XFA skips PDF-container fallback scripts while retaining requested messages and saved scripts; ordinary/static/unlaid-out documents and explicit console/field calls keep their existing behavior. Missing: writing proposed tags and parent tree, field autotag, visual Tags/Order/Content panels, Reading Order tool, compare, geospatial/3D measurement, search index, XFA
Reading colours: pending page backgrounds follow the configured colour (#2059). | | M13 | Automation (MCP, Action Wizard, CLI) + AI providers | 60–120 | 45% | 25–45 | Done: MCP resources (document info, text, page images); headless tool table (123 tools incl. signing, optimizing, initial view, links, stamps, data exchange, comment review, forms authoring and scripts, redaction, sanitize, print, add content), opt-in MCP server over stdio, CLI `run`/`tools` (closed stdout pipes exit cleanly), UI control channel with drag. Missing: Action Wizard, AI providers
Done since: CLI run and MCP reject missing, empty, repeated and option-valued --root arguments before any tool/script execution or MCP initialization. A single-hyphen directory name remains valid; synthetic CLI regressions preserve a would-be overwritten file. `text_find` also caps the per-page match ranges it materializes at the remaining result limit plus one continuation match. CLI robustness scans visit each directory once through overlapping roots and symlink aliases. | diff --git a/docs/target-app-parity.md b/docs/target-app-parity.md index 6c7920e33c..3c77c491cb 100644 --- a/docs/target-app-parity.md +++ b/docs/target-app-parity.md @@ -1,5 +1,6 @@ # PdfCraft parity with Adobe Acrobat Pro +> **Last reviewed:** 2026-10-28 | **Last updated:** 2026-10-28 | **Change:** minor (M10, #4119 partial: Enhance scanned file cleans scanned pages in place: background removal, edge shadow removal, despeckle; estimates unchanged) > **Last reviewed:** 2026-10-25 · **Last updated:** 2026-10-25 · **Change:** minor (localization current state: Arabic logical catalog, explicit RTL direction and mirrored stock chrome; native/browser RTL workflows unqualified. Native release Hans face via LXGW WenKai Lite. Scores and estimates unchanged) > **Last reviewed:** 2026-10-10 · **Last updated:** 2026-10-13 · **Change:** minor (localization ≈59% → ≈68%; weighted readiness 49.25 → 49.7 ≈ 50%) · **Target:** Adobe Acrobat Pro (Acrobat DC, continuous track 26.002.21931, macOS) @@ -206,7 +207,7 @@ full parity in that area. | F Forms and JavaScript | 79 | 58 | 8 | 78.5% | 65% | 40–75 | Fill and author every AcroForm field type, Acrobat's AF functions in Acrobat's event order, sandboxed JavaScript (boa), data exchange, Fill & Sign; XFA static/dynamic layout, FormCalc and XFA data (partial). CLI info, text and render use the laid-out dynamic pages (#1645). Missing: wider JavaScript object model, document actions, submit, debugger, barcode fields, XFA flattening and full dynamic fidelity | | G Protect, redact, sanitize | 41 | 23 | 2 | 58.5% | 55% | 15–30 | Passwords and permissions, Remove security, redaction that removes glyphs, pixels, vectors, annotations and tags, with verification; word lists, patterns, FOIA/Privacy Act codes; sanitize. Missing: certificate security, pattern locales, custom code sets, partial-word redaction, folder redaction, protected view | | H Digital signatures | 45 | 23 | 10 | 62.2% | 45% | 40–70 | PAdES B-B signing and validation (PKCS #12, macOS Keychain, Windows store with PIN prompts), certification, encrypted documents, certificate usage checks; B-T, LTV, OCSP/CRL with caller-supplied evidence. Missing: PKCS #11 and smart cards outside Windows, timestamp-server preferences, online revocation fetching, FieldMDP and field locking, OS trust store, EU Trusted Lists, compare signed version | -| I Scan and OCR | 26 | 5 | 2 | 23.1% | 15% | 40–75 | Searchable image (ocrs, Latin script without accents) on pages, ranges and many files. Missing: every other language (#744 Dutch; accents, CJK, Cyrillic, Arabic), editable-text output, deskew, rotation, despeckle, camera images beyond straightening (found or placed corners): orientation, clean-up, MRC compression, scanners | +| I Scan and OCR | 26 | 5 | 2 | 23.1% | 15% | 40–75 | Searchable image (ocrs, Latin script without accents) on pages, ranges and many files. Missing: every other language (#744 Dutch; accents, CJK, Cyrillic, Arabic), editable-text output, deskew, rotation, camera images beyond straightening (found or placed corners): orientation, clean-up, MRC compression, scanners. Background removal, edge shadow removal and despeckle work on scanned pages (Enhance scanned file) but not yet as options of Recognize text, nor on CMYK, indexed, JBIG2 or CCITT scans | | J Create and export | 34 | 12 | 3 | 39.7% | 25% | 55–100 | Create from images, multipage TIFF, text, clipboard, blank, many files at once; export PNG/JPEG/TIFF, text, RTF; Word and HTML partial. Missing: Office/HTML/web/PostScript to PDF, Excel and PowerPoint export, Word layout fidelity (#773), SVG/XML/EPS/PS/JPEG 2000 export, accessible text | | K Optimize and standards | 58 | 15 | 10 | 34.5% | 20% | 65–115 | Reduce File Size and the Optimizer (images, discards, clean-up, space audit); PDF/A-2b/3b verify and convert (partial). Missing: Preflight (profiles, fixups, reports), PDF/A-1/4 and a/u levels, PDF/X, PDF/UA, PDF/E, PDF/VT, transparency flattening, font unembedding, monochrome JBIG2/CCITT | | L Print | 39 | 15 | 1 | 39.7% | 35% | 30–55 | Acrobat's sizing, n-up, booklet, poster, comments & forms, preview, print-ready PDF, CUPS on macOS/Linux, the Windows spooler (#756) and the browser's print dialog on the web. Missing: print as image, PostScript, all print-production tools (output preview, separations, ink manager, printer marks, colour conversion) | diff --git a/parity/acrobat-features.toml b/parity/acrobat-features.toml index 47ac06fdd4..1ea04254aa 100644 --- a/parity/acrobat-features.toml +++ b/parity/acrobat-features.toml @@ -6674,10 +6674,10 @@ area = "I" tier = "P1" milestone = "M10" status = "partial" -commands = ["ocr.enhance_camera"] -tools = ["page_enhance_camera"] -evidence = ["crates/engine/src/camera_clean.rs::a_shadow_and_a_yellow_cast_become_white_paper", "crates/engine/src/tests.rs::enhance_camera_page_at_a_level_whitens_a_shadowed_sheet", "crates/ui-egui/tests/it/camera.rs::the_clean_up_level_enhances_again_from_the_photo_and_keeps_the_turn"] -notes = "Camera images only: from clean-up level 60, isolated dark pixels are removed. Missing: for scanned pages and as an OCR option." +commands = ["ocr.enhance_camera", "ocr.enhance"] +tools = ["page_enhance_camera", "page_enhance_scan"] +evidence = ["crates/engine/src/camera_clean.rs::a_shadow_and_a_yellow_cast_become_white_paper", "crates/engine/src/tests.rs::enhance_camera_page_at_a_level_whitens_a_shadowed_sheet", "crates/ui-egui/tests/it/camera.rs::the_clean_up_level_enhances_again_from_the_photo_and_keeps_the_turn", "crates/engine/src/scan_clean/tests.rs::despeckle_is_its_own_switch", "crates/automation/tests/it/automation.rs::enhance_scanned_file_cleans_the_scan_in_place_through_tools", "crates/ui-egui/tests/it/scan_clean.rs::enhance_scanned_file_cleans_the_chosen_pages_and_says_what_it_did"] +notes = "Camera images, and scanned pages through Enhance scanned file (ocr.enhance, page_enhance_scan despeckle): isolated dark pixels (at most one dark neighbour) turn white, on grey, RGB and 1-bit scans, in place (the OCR text layer, comments and links stay). Missing: as an option of Recognize text; specks larger than two pixels; CMYK, indexed, JPEG 2000, JBIG2 and CCITT scans (reported, not changed)." [[feature]] id = "ocr.background-removal" @@ -6686,10 +6686,10 @@ area = "I" tier = "P1" milestone = "M10" status = "partial" -commands = ["ocr.enhance_camera"] -tools = ["page_enhance_camera"] -evidence = ["crates/engine/src/camera_clean.rs::a_shadow_and_a_yellow_cast_become_white_paper", "crates/engine/src/tests.rs::enhance_camera_page_at_a_level_whitens_a_shadowed_sheet", "crates/ui-egui/tests/it/camera.rs::the_clean_up_level_enhances_again_from_the_photo_and_keeps_the_turn"] -notes = "Camera images only: Enhance camera image's clean-up level (and page_enhance_camera level) evens out the paper's illumination per channel, so the background turns white and colour casts go; colour ink stays. Missing: the same for scanned pages and as an OCR option." +commands = ["ocr.enhance_camera", "ocr.enhance"] +tools = ["page_enhance_camera", "page_enhance_scan"] +evidence = ["crates/engine/src/camera_clean.rs::a_shadow_and_a_yellow_cast_become_white_paper", "crates/engine/src/tests.rs::enhance_camera_page_at_a_level_whitens_a_shadowed_sheet", "crates/ui-egui/tests/it/camera.rs::the_clean_up_level_enhances_again_from_the_photo_and_keeps_the_turn", "crates/engine/src/scan_clean/tests.rs::a_scan_gets_white_paper_even_edges_dark_strokes_and_no_specks_in_place", "crates/automation/tests/it/automation.rs::enhance_scanned_file_cleans_the_scan_in_place_through_tools", "crates/ui-egui/tests/it/scan_clean.rs::enhance_scanned_file_cleans_the_chosen_pages_and_says_what_it_did"] +notes = "Camera images (Enhance camera image level) and scanned pages (Enhance scanned file ▸ Background removal, page_enhance_scan level): the paper's illumination is evened out per channel so a grey, yellowed or unevenly lit background turns white, then the contrast is raised; colour ink stays. Scans are cleaned in place in their own image objects (JPEG stays JPEG at quality 90, others lossless), so the OCR text layer, comments and links stay; one undo step. Missing: as an option of Recognize text; CMYK, indexed, JPEG 2000, JBIG2 and CCITT scans and scans inside form XObjects (reported, not changed); a preview." [[feature]] id = "ocr.descreen" @@ -6722,10 +6722,10 @@ area = "I" tier = "P2" milestone = "M10" status = "partial" -commands = ["ocr.enhance_camera"] -tools = ["page_enhance_camera"] -evidence = ["crates/engine/src/camera_clean.rs::a_shadow_and_a_yellow_cast_become_white_paper", "crates/engine/src/tests.rs::enhance_camera_page_at_a_level_whitens_a_shadowed_sheet", "crates/ui-egui/tests/it/camera.rs::the_clean_up_level_enhances_again_from_the_photo_and_keeps_the_turn"] -notes = "Camera images only: the clean-up level divides out a smooth illumination estimate, which removes shadows and darker edges on the straightened photo. Missing: for scanned pages (book gutters) and as an OCR option." +commands = ["ocr.enhance_camera", "ocr.enhance"] +tools = ["page_enhance_camera", "page_enhance_scan"] +evidence = ["crates/engine/src/camera_clean.rs::a_shadow_and_a_yellow_cast_become_white_paper", "crates/engine/src/tests.rs::enhance_camera_page_at_a_level_whitens_a_shadowed_sheet", "crates/ui-egui/tests/it/camera.rs::the_clean_up_level_enhances_again_from_the_photo_and_keeps_the_turn", "crates/engine/src/scan_clean/tests.rs::edge_shadow_removal_whitens_the_gutter_background_removal_alone_leaves", "crates/automation/tests/it/automation.rs::enhance_scanned_file_cleans_the_scan_in_place_through_tools", "crates/ui-egui/tests/it/scan_clean.rs::enhance_scanned_file_cleans_the_chosen_pages_and_says_what_it_did"] +notes = "Camera images (the clean-up level divides out a smooth illumination estimate) and scanned pages (Enhance scanned file ▸ Edge shadow removal, page_enhance_scan edge_shadows): within the outer fifth of each side, every column and row is lightened to the brightest paper of that band, which removes book gutters and copier borders the coarse estimate misses; a black scanner border is left alone. Missing: as an option of Recognize text; curved gutters that also bend the text (no dewarping); the scan formats background removal also skips." [[feature]] id = "ocr.camera-perspective" From 622625a178b9bec514b586f12b53fe1c28ab8f33 Mon Sep 17 00:00:00 2001 From: huppiflupp Date: Wed, 28 Oct 2026 13:37:00 +0100 Subject: [PATCH 02/53] M10: Enhance scanned file reports scans inside Form XObjects and names JPEG 2000 - A scan drawn only through grouped Form artwork is reported in skipped (once per form, with every page that draws it) instead of being counted as no scan; a Form of vector artwork holding only a small picture (under 36 dpi over its area) still counts as no scan. - decode_samples checks the filter before the colour space and bit depth, so a JPEG 2000 image without /ColorSpace or /BitsPerComponent is reported as JPEG 2000, not as a colour-space or bit-depth problem. - Optimize and Enhance scanned file no longer start while the other runs: they share the progress card and capture the same document state. - docs/gaps.md: background and edge-shadow removal apply to grey and RGB scans, despeckle also to 1-bit ones. Co-Authored-By: Claude Opus 5.5 --- crates/automation/src/tools.rs | 2 +- crates/engine/src/copy_image.rs | 2 +- crates/engine/src/scan_clean.rs | 25 ++++++- crates/engine/src/scan_clean/tests.rs | 53 ++++++++++++++ crates/engine/src/scan_fields.rs | 70 +++++++++++++++---- crates/optimize/src/samples.rs | 9 ++- crates/ui-egui/src/workflows/optimize_ui.rs | 23 ++++++ crates/ui-egui/src/workflows/scan_clean_ui.rs | 5 ++ docs/gaps.md | 6 +- parity/acrobat-features.toml | 2 +- 10 files changed, 176 insertions(+), 21 deletions(-) diff --git a/crates/automation/src/tools.rs b/crates/automation/src/tools.rs index 0d34911154..c718e4a47a 100644 --- a/crates/automation/src/tools.rs +++ b/crates/automation/src/tools.rs @@ -225,7 +225,7 @@ pub fn tools() -> Vec { }), &["doc", "page"], )), - t("page_enhance_scan", "Enhance scanned file", "Clean up scanned pages in place (Scan & OCR ▸ Enhance ▸ Scanned document). A page is a scan when its pictures cover at least half of it; the pictures covering at least a tenth of it are cleaned and written back into the same image objects at the same size and colour space (a JPEG stays a JPEG at quality 90, other grey/RGB images are stored losslessly, a 1-bit image stays 1-bit), so the page's invisible OCR text, comments, links and other content stay as they were, and an image drawn on several pages is cleaned once. level (0–100, default 50) is background removal: the paper's illumination is evened out so a grey, yellowed or unevenly lit background turns white, then the contrast is raised so faint strokes darken; colour ink stays colour. edge_shadows (default: on unless level is 0) lightens shadows along the scan's edges (book gutters, a copier's dark border). despeckle (default: from level 60) turns isolated dark specks white; it is the only change a 1-bit scan takes. Returns pages_cleaned (1-based), images (rewritten), not_scanned (pages that are no scan) and skipped ([{pages, name, reason}]: CMYK, indexed, JPEG 2000, JBIG2 or CCITT images, broken data, or nothing to change). One undoable step; none when nothing changed.") + t("page_enhance_scan", "Enhance scanned file", "Clean up scanned pages in place (Scan & OCR ▸ Enhance ▸ Scanned document). A page is a scan when its pictures cover at least half of it; the pictures covering at least a tenth of it are cleaned and written back into the same image objects at the same size and colour space (a JPEG stays a JPEG at quality 90, other grey/RGB images are stored losslessly, a 1-bit image stays 1-bit), so the page's invisible OCR text, comments, links and other content stay as they were, and an image drawn on several pages is cleaned once. level (0–100, default 50) is background removal: the paper's illumination is evened out so a grey, yellowed or unevenly lit background turns white, then the contrast is raised so faint strokes darken; colour ink stays colour. edge_shadows (default: on unless level is 0) lightens shadows along the scan's edges (book gutters, a copier's dark border). despeckle (default: from level 60) turns isolated dark specks white; it is the only change a 1-bit scan takes. Returns pages_cleaned (1-based), images (rewritten), not_scanned (pages that are no scan) and skipped ([{pages, name, reason}]: CMYK, indexed, JPEG 2000, JBIG2 or CCITT images, a scan drawn inside grouped Form artwork, broken data, or nothing to change). One undoable step; none when nothing changed.") .cmd("ocr.enhance") .with(schema( json!({ diff --git a/crates/engine/src/copy_image.rs b/crates/engine/src/copy_image.rs index b377903ee4..1407c9f1d2 100644 --- a/crates/engine/src/copy_image.rs +++ b/crates/engine/src/copy_image.rs @@ -113,7 +113,7 @@ pub(crate) fn added_picture(cos: &Cos, file: &[u8]) -> Option { } /// The largest raster image Form `form` draws through its resources (nested Forms included). -fn picture_in_form(cos: &Cos, form: ObjRef) -> Option { +pub(crate) fn picture_in_form(cos: &Cos, form: ObjRef) -> Option { let mut seen = HashSet::from([form]); let mut best: Option<(u64, ObjRef)> = None; let res = resources(cos, form)?; diff --git a/crates/engine/src/scan_clean.rs b/crates/engine/src/scan_clean.rs index bfc8188b29..48ff15093a 100644 --- a/crates/engine/src/scan_clean.rs +++ b/crates/engine/src/scan_clean.rs @@ -12,7 +12,7 @@ //! pictures each covering at least a tenth of the page are its scan. What is not cleaned is //! reported, never skipped silently: pages that are no scan, and scan images in a form PdfCraft //! can't rewrite yet (CMYK, indexed, JPEG 2000, JBIG2, CCITT fax, a stencil mask, a remapping -//! decode array) or whose data is broken. +//! decode array, a scan drawn inside grouped Form artwork) or whose data is broken. //! //! Like Optimize, the work is a job: [`Session::scan_clean_job`] captures the document, //! [`ScanCleanJob::run`] cleans anywhere (a worker thread), and [`Session::apply_scan_clean`] @@ -31,6 +31,8 @@ const QUALITY: u8 = 90; /// On a scanned page, an image covering at least this share of the page is part of the scan; /// smaller ones (a logo or a signature pasted on) are left alone. const PART_OF_SCAN: f64 = 0.1; +/// Why a scan drawn through grouped Form artwork stays as it is. +const IN_FORM: &str = "it is drawn inside grouped artwork (a form XObject), which can't be cleaned up yet"; /// How scanned pages are cleaned up. #[derive(Clone, Copy, Debug, PartialEq, Eq)] @@ -138,10 +140,29 @@ impl Session { wanted.dedup(); let mut report = ScanCleanReport::default(); let mut targets: Vec = Vec::new(); + // Form artwork already reported, and where in `report.skipped`. + let mut forms_reported: Vec<(ObjRef, usize)> = Vec::new(); for page in wanted { let Some(info) = doc.info.pages.get(page) else { continue }; let Some(images) = crate::scan_fields::scan_images(doc, page, info) else { - report.not_scanned.push(page); + // A scan drawn through a Form XObject is a scan all the same: say why it stays. + let forms: Vec<_> = crate::scan_fields::scan_forms(doc, page, info).into_iter().filter(|(_, share)| *share >= PART_OF_SCAN).collect(); + if forms.is_empty() { + report.not_scanned.push(page); + } + for (form, _) in forms { + let known = form.object.and_then(|o| forms_reported.iter().find(|(r, _)| *r == o)).and_then(|(_, k)| report.skipped.get_mut(*k)); + match known { + Some(s) if !s.pages.contains(&page) => s.pages.push(page), + Some(_) => {} + None => { + if let Some(o) = form.object { + forms_reported.push((o, report.skipped.len())); + } + report.skipped.push(SkippedImage { pages: vec![page], name: form.name, reason: IN_FORM.into() }); + } + } + } continue; }; for (image, _) in images.into_iter().filter(|(_, share)| *share >= PART_OF_SCAN) { diff --git a/crates/engine/src/scan_clean/tests.rs b/crates/engine/src/scan_clean/tests.rs index 561ba9e2e4..f39f8c0432 100644 --- a/crates/engine/src/scan_clean/tests.rs +++ b/crates/engine/src/scan_clean/tests.rs @@ -275,6 +275,59 @@ fn hostile_images_are_reported_not_a_panic() { assert!(matches!(s.enhance_scanned_pages(id, &[99], ScanCleanSettings::default()), Err(EditError::Invalid(_)))); } +#[test] +fn a_scan_drawn_inside_a_form_xobject_is_reported_not_called_no_scan() { + // Pages 1 and 3 draw the scan through Form Fm0 (as some scanners write it), page 2 is text. + let page = |contents: usize| { + format!( + "<< /Type /Page /Parent 2 0 R /MediaBox [0 0 {W} {H}] /Resources << /XObject << /Fm0 4 0 R >> /Font << /F1 9 0 R >> >> /Contents {contents} 0 R >>" + ) + }; + let bytes = Pdf::new() + .obj("<< /Type /Catalog /Pages 2 0 R >>") + .obj("<< /Type /Pages /Kids [3 0 R 7 0 R 10 0 R] /Count 3 >>") + .obj(page(6)) + .stream( + &format!("/Type /XObject /Subtype /Form /BBox [0 0 {W} {H}] /Resources << /XObject << /Im0 5 0 R >> >>"), + format!("q {W} 0 0 {H} 0 0 cm /Im0 Do Q"), + ) + .stream(&format!("/Type /XObject /Subtype /Image /Width {W} /Height {H} /ColorSpace /DeviceGray /BitsPerComponent 8"), scan_pixels(1)) + .stream("", "/Fm0 Do") + .obj(page(8)) + .stream("", "BT /F1 12 Tf 20 380 Td (Real text) Tj ET") + .obj("<< /Type /Font /Subtype /Type1 /BaseFont /Helvetica >>") + .obj(page(11)) + .stream("", "q /Fm0 Do Q") + .build(); + let (mut s, id) = open(bytes); + let before = object(&s, id, 5); + let report = s.enhance_scanned_pages(id, &[], ScanCleanSettings::at(100)).unwrap(); + assert!(report.pages.is_empty() && report.images == 0, "{report:?}"); + assert_eq!(report.not_scanned, vec![1], "only the text page is no scan"); + assert_eq!(report.skipped.len(), 1, "the form is reported once: {report:?}"); + let k = &report.skipped[0]; + assert_eq!((k.pages.as_slice(), k.name.as_str()), (&[0, 2][..], "Fm0")); + assert!(k.reason.contains("form XObject"), "{k:?}"); + assert_eq!(object(&s, id, 5), before, "the picture is left as it was"); + assert_eq!(s.get(id).unwrap().can_undo(), None); + + // A page of vector artwork in a Form, holding only a small logo, is no scan. + let logo = Pdf::new() + .obj("<< /Type /Catalog /Pages 2 0 R >>") + .obj("<< /Type /Pages /Kids [3 0 R] /Count 1 >>") + .obj(format!("<< /Type /Page /Parent 2 0 R /MediaBox [0 0 {W} {H}] /Resources << /XObject << /Fm0 4 0 R >> >> /Contents 6 0 R >>")) + .stream( + &format!("/Type /XObject /Subtype /Form /BBox [0 0 {W} {H}] /Resources << /XObject << /Im0 5 0 R >> >>"), + format!("0 0 {W} {H} re f q 40 0 0 20 10 370 cm /Im0 Do Q"), + ) + .stream("/Type /XObject /Subtype /Image /Width 40 /Height 20 /ColorSpace /DeviceGray /BitsPerComponent 8", [128; 800]) + .stream("", "/Fm0 Do") + .build(); + let (mut s, id) = open(logo); + let report = s.enhance_scanned_pages(id, &[], ScanCleanSettings::at(100)).unwrap(); + assert_eq!((report.not_scanned.as_slice(), report.skipped.len()), (&[0][..], 0), "{report:?}"); +} + #[test] fn protected_documents_keep_their_encryption_and_restrictions() { let protect = |user: &str, permissions: i32| { diff --git a/crates/engine/src/scan_fields.rs b/crates/engine/src/scan_fields.rs index 769beaefca..d2e663f9a5 100644 --- a/crates/engine/src/scan_fields.rs +++ b/crates/engine/src/scan_fields.rs @@ -19,6 +19,9 @@ const SCALE: f32 = 2.0; const MAX_SIDE: f32 = 3000.0; /// A page is a scan when its images cover at least this share of it. const COVERED: f64 = 0.5; +/// A picture inside Form artwork is taken for a scan when it has at least this resolution over +/// the area the artwork covers (a quarter of the pixels a 72 dpi scan would have). +const MIN_FORM_DPI: f64 = 36.0; /// Recognised "words" less tall than this (points) are specks read as letters, and would make /// the box they sit on look filled in. const MIN_WORD_HEIGHT: f64 = 4.0; @@ -32,28 +35,69 @@ pub(crate) fn is_scan(doc: &Document, page: usize, info: &PageInfo) -> bool { /// crop box) it covers, when together they cover most of it ([`is_scan`]); `None` otherwise. /// Enhance ▸ Scanned document cleans these up ([`crate::scan_clean`]). pub(crate) fn scan_images(doc: &Document, page: usize, info: &PageInfo) -> Option> { - let [cx0, cy0, cx1, cy1] = info.crop.map(f64::from); - let page_area = (cx1 - cx0) * (cy1 - cy0); - if !page_area.is_finite() || page_area <= 0.0 { - return None; - } + let area = page_area(info)?; let images: Vec<(pdfcraft_edit::PageImage, f64)> = doc .page_images(page) .into_iter() .filter(|i| !i.is_form && i.width > 1 && i.height > 1) - .map(|i| { - let r = i.visible_rect.unwrap_or(i.rect); - let w = (r[2].min(cx1) - r[0].max(cx0)).max(0.0); - let h = (r[3].min(cy1) - r[1].max(cy0)).max(0.0); - let share = w * h / page_area; - (i, share) - }) - .filter(|(_, share)| share.is_finite()) + .filter_map(|i| Some((share(info, area, &i)?, i))) + .map(|(share, i)| (i, share)) .collect(); let covered: f64 = images.iter().map(|(_, share)| share).sum(); (covered >= COVERED).then_some(images) } +/// The grouped Form artwork on `page` that holds a raster picture, each with the share of the +/// page it covers, when it and the page's own images together cover most of the page: a scan +/// drawn through a Form XObject, which [`scan_images`] does not count and Enhance ▸ Scanned +/// document can't clean yet, so it reports these. Empty otherwise. +pub(crate) fn scan_forms(doc: &Document, page: usize, info: &PageInfo) -> Vec<(pdfcraft_edit::PageImage, f64)> { + let (Some(area), Some(editor)) = (page_area(info), doc.editor.as_ref()) else { return Vec::new() }; + let mut covered = 0.0; + let mut forms = Vec::new(); + for i in doc.page_images(page) { + let Some(share) = share(info, area, &i) else { continue }; + if !i.is_form { + if i.width > 1 && i.height > 1 { + covered += share; + } + } else if i.object.is_some_and(|form| holds_scan(&editor.cos, form, share * area)) { + covered += share; + forms.push((i, share)); + } + } + if covered >= COVERED { forms } else { Vec::new() } +} + +/// Whether Form `form`, covering `points` square points of the page, draws a picture that could +/// be a scan of it: one with at least [`MIN_FORM_DPI`] over that area, not a logo inside a page +/// of vector artwork. +fn holds_scan(cos: &pdfcraft_cos::Document, form: pdfcraft_cos::ObjRef, points: f64) -> bool { + let Some(picture) = crate::copy_image::picture_in_form(cos, form) else { return false }; + let obj = cos.get(picture); + let pdfcraft_cos::Object::Stream(s) = &*obj else { return false }; + let px = |k: &[u8]| s.dict.int(k).unwrap_or(0).max(0) as f64; + let per_point = MIN_FORM_DPI / 72.0; + px(b"Width") * px(b"Height") >= points * per_point * per_point +} + +/// The area of `info`'s crop box, when it is a usable one. +fn page_area(info: &PageInfo) -> Option { + let [cx0, cy0, cx1, cy1] = info.crop.map(f64::from); + let area = (cx1 - cx0) * (cy1 - cy0); + (area.is_finite() && area > 0.0).then_some(area) +} + +/// The share of the crop box (of area `area`) that `image` covers, when it is a number. +fn share(info: &PageInfo, area: f64, image: &pdfcraft_edit::PageImage) -> Option { + let [cx0, cy0, cx1, cy1] = info.crop.map(f64::from); + let r = image.visible_rect.unwrap_or(image.rect); + let w = (r[2].min(cx1) - r[0].max(cx0)).max(0.0); + let h = (r[3].min(cy1) - r[1].max(cy0)).max(0.0); + let share = w * h / area; + share.is_finite().then_some(share) +} + /// The page as displayed, in points with y up: where detection on a scan works. struct Frame<'a>(&'a PageInfo); diff --git a/crates/optimize/src/samples.rs b/crates/optimize/src/samples.rs index 7f402b51b8..1ed9f16b1f 100644 --- a/crates/optimize/src/samples.rs +++ b/crates/optimize/src/samples.rs @@ -53,6 +53,9 @@ pub fn decode_samples(doc: &Document, s: &Stream) -> Result if matches!(d.get(b"ImageMask").map(|v| doc.resolve(v)).as_deref(), Some(Object::Bool(true))) { return Err(NotDecoded::StencilMask); } + // The filter first: a JPEG 2000 image may leave its colour space and bit depth to the data, + // and the report should name the format, not the missing entries. + let jpeg = is_dct(s).ok_or(NotDecoded::Filter)?; let n = d.get(b"ColorSpace").and_then(|cs| components(doc, cs)).ok_or(NotDecoded::ColourSpace)?; let bpc = d.get(b"BitsPerComponent").map(|v| doc.resolve(v)).and_then(|v| v.as_int()).unwrap_or(0); let bilevel = match bpc { @@ -60,7 +63,6 @@ pub fn decode_samples(doc: &Document, s: &Stream) -> Result 1 if n == 1 => true, other => return Err(NotDecoded::BitDepth(other)), }; - let jpeg = is_dct(s).ok_or(NotDecoded::Filter)?; let width = d.get(b"Width").map(|v| doc.resolve(v)).and_then(|v| v.as_int()).and_then(|v| u32::try_from(v).ok()).unwrap_or(0); let height = d.get(b"Height").map(|v| doc.resolve(v)).and_then(|v| v.as_int()).and_then(|v| u32::try_from(v).ok()).unwrap_or(0); if width == 0 || height == 0 { @@ -258,6 +260,11 @@ mod tests { assert_eq!(base(&[("ColorSpace", Object::name("DeviceCMYK"))], vec![0; 64]), Err(NotDecoded::ColourSpace)); assert_eq!(base(&[("BitsPerComponent", Object::Int(4))], vec![0; 8]), Err(NotDecoded::BitDepth(4))); assert_eq!(base(&[("Filter", Object::name("JBIG2Decode"))], vec![0; 8]), Err(NotDecoded::Filter)); + // JPEG 2000 may leave the colour space and bit depth to its data: the format is the reason. + let jpx = |dict: &[(&str, Object)]| decode_samples(&doc(), &image(dict, vec![0; 8])); + let size = [("Width", Object::Int(4)), ("Height", Object::Int(4)), ("Filter", Object::name("JPXDecode"))]; + assert_eq!(jpx(&size), Err(NotDecoded::Filter)); + assert_eq!(jpx(&[size[0].clone(), size[1].clone(), size[2].clone(), ("ColorSpace", Object::name("DeviceRGB"))]), Err(NotDecoded::Filter)); assert_eq!(base(&[("ImageMask", Object::Bool(true))], vec![0; 8]), Err(NotDecoded::StencilMask)); assert_eq!(base(&[("Decode", Object::Array(vec![Object::Int(1), Object::Int(0)]))], vec![0; 16]), Err(NotDecoded::Decode)); // Too short, a huge size, a broken JPEG, no size at all. diff --git a/crates/ui-egui/src/workflows/optimize_ui.rs b/crates/ui-egui/src/workflows/optimize_ui.rs index 3bb80ac439..4e1a7c20bf 100644 --- a/crates/ui-egui/src/workflows/optimize_ui.rs +++ b/crates/ui-egui/src/workflows/optimize_ui.rs @@ -360,6 +360,11 @@ impl PdfCraftApp { self.notify_tr("Optimization is already running"); return; } + // Both jobs share the progress card and capture the same document state. + if self.scan_clean_run.is_some() { + self.notify_tr("Scanned pages are already being enhanced"); + return; + } let job = match self.session.optimize_job(id, settings, discard) { Ok(job) => job, Err(e) => return self.save_optimized(id, kind, Err(e)), @@ -491,4 +496,22 @@ mod tests { app.optimize_with_draft(); assert_eq!(app.toast.as_ref().map(|t| t.0.as_str()), Some("Optimization is already running")); } + + #[test] + fn optimize_and_enhance_scanned_file_do_not_run_at_once() { + let mut app = PdfCraftApp::new(); + app.set_option("language", "en").unwrap(); + app.open_bytes("notes.txt", None, b"hello".to_vec()).unwrap(); + let (_, doc) = app.active_ids().unwrap(); + app.scan_clean_run = Some(crate::scan_clean_ui::ScanCleanRun { doc, progress: Arc::default() }); + app.optimize_with_draft(); + assert_eq!(app.toast.as_ref().map(|t| t.0.as_str()), Some("Scanned pages are already being enhanced")); + assert!(app.optimize_run.is_none()); + + app.scan_clean_run = None; + app.optimize_run = Some(OptimizeRun { doc, kind: OptimizeKind::Reduce, progress: Arc::default() }); + app.start_scan_clean(); + assert_eq!(app.toast.as_ref().map(|t| t.0.as_str()), Some("Optimization is already running")); + assert!(app.scan_clean_run.is_none()); + } } diff --git a/crates/ui-egui/src/workflows/scan_clean_ui.rs b/crates/ui-egui/src/workflows/scan_clean_ui.rs index a275b5af20..64f8ce48eb 100644 --- a/crates/ui-egui/src/workflows/scan_clean_ui.rs +++ b/crates/ui-egui/src/workflows/scan_clean_ui.rs @@ -140,6 +140,11 @@ impl PdfCraftApp { self.notify_tr("Scanned pages are already being enhanced"); return; } + // Both jobs share the progress card and capture the same document state. + if self.optimize_run.is_some() { + self.notify_tr("Optimization is already running"); + return; + } // What's typed in a form field is part of the document (#166). if !self.commit_form_typing() { return; diff --git a/docs/gaps.md b/docs/gaps.md index ab6f083ab1..dacca2515b 100644 --- a/docs/gaps.md +++ b/docs/gaps.md @@ -165,8 +165,10 @@ update the numbers. ### 7. OCR reads only unaccented Latin - **Missing:** every other script and accented Latin; editable-text output; deskew, rotation, - despeckle and background clean-up as options of Recognize text (Enhance scanned file does them - on grey, RGB and 1-bit scans, not CMYK, indexed, JBIG2 or CCITT ones); MRC compression; + despeckle and background clean-up as options of Recognize text (Enhance scanned file removes + the background and edge shadows on grey and RGB scans and despeckles those and 1-bit ones; it + reports CMYK, indexed, JPEG 2000, JBIG2 and CCITT scans and scans drawn inside Form XObjects + and leaves them as they are); MRC compression; pluggable engines; OCR in the browser; scanner input. - **Evidence:** `ocr.ocr-languages` partial ("the models read only the Latin alphabet without diff --git a/parity/acrobat-features.toml b/parity/acrobat-features.toml index 1ea04254aa..5f48c815a9 100644 --- a/parity/acrobat-features.toml +++ b/parity/acrobat-features.toml @@ -6677,7 +6677,7 @@ status = "partial" commands = ["ocr.enhance_camera", "ocr.enhance"] tools = ["page_enhance_camera", "page_enhance_scan"] evidence = ["crates/engine/src/camera_clean.rs::a_shadow_and_a_yellow_cast_become_white_paper", "crates/engine/src/tests.rs::enhance_camera_page_at_a_level_whitens_a_shadowed_sheet", "crates/ui-egui/tests/it/camera.rs::the_clean_up_level_enhances_again_from_the_photo_and_keeps_the_turn", "crates/engine/src/scan_clean/tests.rs::despeckle_is_its_own_switch", "crates/automation/tests/it/automation.rs::enhance_scanned_file_cleans_the_scan_in_place_through_tools", "crates/ui-egui/tests/it/scan_clean.rs::enhance_scanned_file_cleans_the_chosen_pages_and_says_what_it_did"] -notes = "Camera images, and scanned pages through Enhance scanned file (ocr.enhance, page_enhance_scan despeckle): isolated dark pixels (at most one dark neighbour) turn white, on grey, RGB and 1-bit scans, in place (the OCR text layer, comments and links stay). Missing: as an option of Recognize text; specks larger than two pixels; CMYK, indexed, JPEG 2000, JBIG2 and CCITT scans (reported, not changed)." +notes = "Camera images, and scanned pages through Enhance scanned file (ocr.enhance, page_enhance_scan despeckle): isolated dark pixels (at most one dark neighbour) turn white, on grey, RGB and 1-bit scans, in place (the OCR text layer, comments and links stay). Missing: as an option of Recognize text; specks larger than two pixels; CMYK, indexed, JPEG 2000, JBIG2 and CCITT scans and scans inside form XObjects (reported, not changed)." [[feature]] id = "ocr.background-removal" From 7878a4cb20604a57b7ae0cbc964bcf3c29b3a711 Mon Sep 17 00:00:00 2001 From: huppiflupp Date: Thu, 29 Oct 2026 16:37:23 +0100 Subject: [PATCH 03/53] M10: Enhance scanned file is tested to leave unchosen pages' scans alone A new engine test builds two pages that each draw their own scan image, cleans one page, and checks that page's image changed while the other image object, and its stream bytes, stayed identical; then the same with the other page chosen. Taking every page instead of the chosen ones fails it, on the report and, with that assertion removed, on the other image. Co-Authored-By: Claude Opus 5.5 --- crates/engine/src/scan_clean/tests.rs | 34 +++++++++++++++++++++++++++ 1 file changed, 34 insertions(+) diff --git a/crates/engine/src/scan_clean/tests.rs b/crates/engine/src/scan_clean/tests.rs index f39f8c0432..11a1bca9c9 100644 --- a/crates/engine/src/scan_clean/tests.rs +++ b/crates/engine/src/scan_clean/tests.rs @@ -389,3 +389,37 @@ fn edge_shadow_removal_whitens_the_gutter_background_removal_alone_leaves() { assert!(grey_at(&after, 3, 40).abs_diff(grey_at(&before, 150, 40)) <= 4, "{} vs {}", grey_at(&after, 3, 40), grey_at(&before, 150, 40)); assert_eq!(grey_at(&after, 150, 40), grey_at(&before, 150, 40), "the middle is left alone"); } + +/// Two pages, each with its own scan image (objects 4 and 7), so cleaning one page can be told +/// apart from cleaning the other. +fn two_scans_pdf() -> Vec { + let dict = format!("/Type /XObject /Subtype /Image /Width {W} /Height {H} /ColorSpace /DeviceGray /BitsPerComponent 8"); + let draw = format!("q {W} 0 0 {H} 0 0 cm /Im0 Do Q"); + Pdf::new() + .obj("<< /Type /Catalog /Pages 2 0 R >>") + .obj("<< /Type /Pages /Kids [3 0 R 6 0 R] /Count 2 >>") + .obj(format!("<< /Type /Page /Parent 2 0 R /MediaBox [0 0 {W} {H}] /Resources << /XObject << /Im0 4 0 R >> >> /Contents 5 0 R >>")) + .stream(&dict, scan_pixels(1)) + .stream("", &draw) + .obj(format!("<< /Type /Page /Parent 2 0 R /MediaBox [0 0 {W} {H}] /Resources << /XObject << /Im0 7 0 R >> >> /Contents 8 0 R >>")) + .stream(&dict, scan_pixels(1)) + .stream("", &draw) + .build() +} + +#[test] +fn only_the_chosen_pages_scan_is_cleaned() { + // Clean page 1, then page 2 of a fresh copy: the other page's image stays byte-identical. + for (chosen, image, other) in [(0usize, 4u32, 7u32), (1, 7, 4)] { + let (mut s, id) = open(two_scans_pdf()); + let (image_before, other_before) = (object(&s, id, image), object(&s, id, other)); + assert_eq!(s.get(id).unwrap().page_images(chosen)[0].object, Some(ObjRef::new(image, 0)), "the fixture"); + let report = s.enhance_scanned_pages(id, &[chosen], ScanCleanSettings::at(100)).unwrap(); + assert_eq!((report.pages.as_slice(), report.images), (&[chosen][..], 1), "page {}", chosen + 1); + assert_ne!(object(&s, id, image), image_before, "page {}'s image was not cleaned", chosen + 1); + assert_eq!(object(&s, id, other), other_before, "cleaning page {} changed the other page's image", chosen + 1); + let (Object::Stream(was), Object::Stream(now)) = (&*other_before, &*object(&s, id, other)) else { panic!("the images") }; + assert_eq!(was.raw, now.raw, "the other image's stream bytes"); + assert_eq!(s.get(id).unwrap().can_undo(), Some("Enhance scanned file")); + } +} From cc9d95e1fbd6f9bff0ef93e4a194e9ac1ebf3cbd Mon Sep 17 00:00:00 2001 From: KenM76 Date: Fri, 30 Oct 2026 00:29:23 -0400 Subject: [PATCH 04/53] fix(edit): an edited OCR phrase under its scan image is drawn on top (#3973) Some OCR tools write the invisible text layer first and the opaque page image after it. Edit text rewrote such a phrase in the text's place, so the cover and the new visible text were painted under the scan: the edit was saved (text extraction finds it) but nothing showed. When the rewritten paragraph was invisible and an XObject, inline image or shading is painted after its place, the cover and the new text now go on top of the page in their own q/Q (place_tagged, tagged Edited), drawn from the initial graphics state with the paragraph's CTM. Only the state the following text expects stays in place. Both rewrite paths (single style and styled rows) do this. An OCR layer drawn over its scan, the common case, is rewritten in place as before, so reading order is kept. The test builds a scan whose OCR word comes before the image and checks the cover and new text show, for one line and for a typed line break. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SqmHYx8FiWM5PcsoiErgp9 --- crates/edit/src/lib.rs | 2 +- crates/edit/src/place.rs | 36 ++++++++++++ crates/edit/src/text/blocks.rs | 3 +- crates/edit/src/text/interp.rs | 20 +++++++ crates/edit/src/text/rewrite.rs | 26 +++++++-- crates/edit/src/text_regions/styled.rs | 12 +++- crates/engine/tests/it/ocr_edit.rs | 81 ++++++++++++++++++++++++++ parity/acrobat-features.toml | 4 +- 8 files changed, 174 insertions(+), 10 deletions(-) diff --git a/crates/edit/src/lib.rs b/crates/edit/src/lib.rs index 7c8556332b..1ca43d20a8 100644 --- a/crates/edit/src/lib.rs +++ b/crates/edit/src/lib.rs @@ -68,7 +68,7 @@ use marks::{matrix_number, n}; pub(crate) use place::undrawable; #[cfg(test)] use place::{ISOLATION_MAX_DEPTH, OLD_WRAP_CLOSE, WRAP_CLOSE, WRAP_OPEN, is_wrap_close, page_fonts, stream_bytes, stream_tag, tagged, wrapper_role}; -use place::{check, check_on_top, contents, page_list, place_tagged}; +use place::{check, check_on_top, contents, draw_on_top, lift_on_top, page_list, place_tagged, revealed_under_paint}; pub use place::{stamp, undrawable_message}; #[cfg(test)] diff --git a/crates/edit/src/place.rs b/crates/edit/src/place.rs index 2f2c24c862..3056d3f71b 100644 --- a/crates/edit/src/place.rs +++ b/crates/edit/src/place.rs @@ -346,6 +346,42 @@ pub(crate) fn place_tagged(doc: &mut Document, page: &pdfcraft_model::Page, tag: Ok(mark_ref) } +/// Draw `ops`, written for the space whose CTM is `ctm`, on top of everything on `page`, tagged +/// `Edited`. A rewritten invisible paragraph goes here when [`revealed_under_paint`]: an OCR +/// layer written before the scan image it lies under would keep the revealed text hidden (#3973). +pub(crate) fn draw_on_top(doc: &mut Document, page: &pdfcraft_model::Page, ctm: [f64; 6], ops: &[pdfcraft_content::Op]) -> Result<(), EditError> { + if ops.is_empty() { + return Ok(()); + } + use pdfcraft_content::Op; + let mut all = vec![Op::new("q", vec![]), Op::new("cm", ctm.iter().map(|v| pdfcraft_content::num(*v)).collect())]; + all.extend_from_slice(ops); + all.push(Op::new("Q", vec![])); + place_tagged(doc, page, "Edited", pdfcraft_content::serialize_ops(&all), false).map(drop) +} + +/// Whether a rewritten paragraph that was `invisible` goes on top of the page: anything after +/// operator `at` can paint over it there (an XObject, an inline image or a shading), as the scan +/// image does over an OCR layer written before it (#3973). +pub(crate) fn revealed_under_paint(invisible: bool, ops: &[pdfcraft_content::Op], at: usize) -> bool { + invisible && ops.get(at..).unwrap_or_default().iter().any(|op| op.is("Do") || op.is("BI") || op.is("sh")) +} + +/// The operators to draw on top of `page` ([`draw_on_top`]): all of `paragraph` when `over`, +/// after checking the page takes them, so a refusal leaves the document as it was. +pub(crate) fn lift_on_top( + doc: &Document, + page: &pdfcraft_model::Page, + over: bool, + paragraph: &mut Vec, +) -> Result, EditError> { + if !over { + return Ok(Vec::new()); + } + check_on_top(doc, page)?; + Ok(std::mem::take(paragraph)) +} + fn too_deep_on_top() -> EditError { EditError::Invalid("the page's marked content is nested too deeply to add anything on top of it".into()) } diff --git a/crates/edit/src/text/blocks.rs b/crates/edit/src/text/blocks.rs index 615e7699a2..d1cb997c89 100644 --- a/crates/edit/src/text/blocks.rs +++ b/crates/edit/src/text/blocks.rs @@ -28,7 +28,8 @@ pub struct TextBlock { /// Byte ranges in `editor_text` and their source line indexes. pub spans: Vec<(std::ops::Range, usize)>, /// Every line is invisible text (rendering mode 3): an OCR layer over a scanned image, whose - /// pixels show the words. A rewrite draws the new text visibly (see [`BlockStyle::cover`]). + /// pixels show the words. A rewrite draws the new text visibly, on top of the page when + /// the scan is drawn after it (see [`BlockStyle::cover`]). pub invisible: bool, } diff --git a/crates/edit/src/text/interp.rs b/crates/edit/src/text/interp.rs index 16b04b406f..08e9b2c5d7 100644 --- a/crates/edit/src/text/interp.rs +++ b/crates/edit/src/text/interp.rs @@ -25,6 +25,26 @@ pub(super) struct TextState { } impl TextState { + /// The state a page's content starts in (ISO 32000-2 §8.4.1, §9.3.1): what is in force on + /// top of the page, after its content wrapped in `q`/`Q`. + pub(crate) fn initial() -> Self { + TextState { + font: None, + char_spacing: 0.0, + word_spacing: 0.0, + scale: 1.0, + leading: 0.0, + rise: 0.0, + fill: Vec::new(), + render_mode: 0, + stroke: Vec::new(), + line_width: None, + ext_gstate: Vec::new(), + dash: (Vec::new(), 0.0), + line_cap: 0, + } + } + pub(super) fn of(ts: &Ts) -> Self { TextState { font: ts.font.as_ref().map(|(n, _)| (String::from_utf8_lossy(n).into_owned(), ts.size)), diff --git a/crates/edit/src/text/rewrite.rs b/crates/edit/src/text/rewrite.rs index 513077a027..f03174885a 100644 --- a/crates/edit/src/text/rewrite.rs +++ b/crates/edit/src/text/rewrite.rs @@ -865,9 +865,26 @@ pub(crate) fn rewrite_block_from( char_spacing: o_state.char_spacing, lead, }; - let plan = - BlockPlan { block: &b, lines: &lines, style, emit, after, movement, state: o_state, tm, size, underline: style.underline == Some(true) }; + let at = if split { start } else { emit.bt_op }; + // Revealed OCR text that something is painted over goes on top of the page, where it starts + // from the initial state. + let over = crate::revealed_under_paint(b.invisible, &ops, at); + let initial = TextState::initial(); + let drawn_from = if over { &initial } else { after }; + let plan = BlockPlan { + block: &b, + lines: &lines, + style, + emit, + after: drawn_from, + movement, + state: o_state, + tm, + size, + underline: style.underline == Some(true), + }; let mut block_ops = block_ops(&plan, &writer, &wrapped, font.show_font.clone())?; + let on_top = crate::lift_on_top(doc, &p, over, &mut block_ops)?; let n = pdfcraft_content::num; block_ops.extend(after.ops().into_iter().filter(|op| !op.is("Tf") || after.font.is_some())); if split { @@ -875,7 +892,6 @@ pub(crate) fn rewrite_block_from( block_ops.push(Op::new("BT", vec![])); block_ops.push(Op::new("Tm", emit.tlm.iter().map(|v| n(*v)).collect())); } - let at = if split { start } else { emit.bt_op }; let new_data = splice(&pieces, &ops, |i| { let insert = if i == at { std::mem::take(&mut block_ops) } else { Vec::new() }; (insert, !dropped.contains(&i)) @@ -892,12 +908,14 @@ pub(crate) fn rewrite_block_from( d.set(b"Resources".to_vec(), Object::Dict(res)); } })?; + crate::draw_on_top(doc, &p, emit.ctm, &on_top)?; Ok(LineEdit { substituted: font.substituted }) } /// The cover (when [`BlockStyle::cover`] asks for one) and visible rendering that start a /// rewritten invisible paragraph, drawn in the space whose CTM is `ctm`. Nothing for visible text. -/// The paragraph's `q`/`Q` restores the rendering mode for the invisible text after it. +/// The paragraph's `q`/`Q` restores the rendering mode for the invisible text after it; one +/// painted over where it was is drawn on top of the page instead ([`crate::draw_on_top`]). pub(crate) fn reveal_ops(block: &TextBlock, lines: &[TextLine], style: &BlockStyle, ctm: [f64; 6]) -> Result, EditError> { if !block.invisible { return Ok(Vec::new()); diff --git a/crates/edit/src/text_regions/styled.rs b/crates/edit/src/text_regions/styled.rs index e8b3b683d0..c03b75d3f4 100644 --- a/crates/edit/src/text_regions/styled.rs +++ b/crates/edit/src/text_regions/styled.rs @@ -1,3 +1,4 @@ +use super::super::interp::TextState; use super::grouping::{across, is_centred, row_rect}; use super::{ BlockStyle, CraftFont, Dict, Document, EditError, HashMap, LineEdit, MAX_TYPE3_GLYPHS, Matrix, Metrics, Object, Op, PdfString, Pieces, @@ -783,6 +784,12 @@ pub(crate) fn rewrite_styled_block( let o = &lead.origin; // What is in force where the rewrite is drawn, and what the text after it expects. let after = if split { &o.state } else { &o.bt_state }; + let at = if split { start } else { o.bt_op }; + // Revealed OCR text that something is painted over goes on top of the page, where it starts + // from the initial state. + let over = crate::revealed_under_paint(block.invisible, &ops, at); + let initial = TextState::initial(); + let drawn_from = if over { &initial } else { after }; validate_contextual_rows(&rows, &kept)?; let measured: Vec<_> = rows.iter().map(|row| measure(row)).collect::>()?; let shaped = prepare_shaped_rows(doc, &fonts, &rows, &kept, advance)?; @@ -869,7 +876,7 @@ pub(crate) fn rewrite_styled_block( // object (`gs`, stroke, line width, `Tr`; #3508/#3509) from the state the rewrite is // drawn in: a `gs` has no inverse, so one run's ExtGState would stay for the next. block_ops.push(Op::new("q", vec![])); - block_ops.extend(state.paint_ops(after)); + block_ops.extend(state.paint_ops(drawn_from)); if let Some((_, shaped, _)) = shaped_runs.get(&(row_index, pos)) { let resource = glyph_resources.get(&(row_index, pos)).ok_or_else(|| EditError::Invalid("shaped font is missing".into()))?; let mut properties = Dict::new(); @@ -967,13 +974,13 @@ pub(crate) fn rewrite_styled_block( }; } block_ops.push(Op::new("Q", vec![])); + let on_top = crate::lift_on_top(doc, &p, over, &mut block_ops)?; block_ops.extend(after.ops()); if split { block_ops.insert(0, Op::new("ET", vec![])); block_ops.push(Op::new("BT", vec![])); block_ops.push(Op::new("Tm", o.tlm.iter().map(|v| n(*v)).collect())); } - let at = if split { start } else { o.bt_op }; let data = splice(&pieces, &ops, |i| (if i == at { std::mem::take(&mut block_ops) } else { Vec::new() }, !drop.contains(&i))); let contents = rewritten_contents(doc, &streams, data); res.set(b"Font".to_vec(), Object::Dict(fonts)); @@ -981,5 +988,6 @@ pub(crate) fn rewrite_styled_block( d.set(b"Contents".to_vec(), contents); d.set(b"Resources".to_vec(), Object::Dict(res)); })?; + crate::draw_on_top(doc, &p, o.ctm, &on_top)?; Ok(LineEdit { substituted }) } diff --git a/crates/engine/tests/it/ocr_edit.rs b/crates/engine/tests/it/ocr_edit.rs index 8009ef1bbf..86fc1c9df2 100644 --- a/crates/engine/tests/it/ocr_edit.rs +++ b/crates/engine/tests/it/ocr_edit.rs @@ -207,3 +207,84 @@ fn visible_stretched_text_rewrites_unchanged() { let i = ((y * out.width + x) * 4) as usize; assert_eq!(&out.rgba[i..i + 3], &[255, 255, 255]); } + +/// A scan whose OCR layer is drawn first and the opaque page image over it, as some OCR tools +/// write it (#3973): invisible Helvetica `text` on the scanned word, then the image. +fn ocr_under_scan(text: &str) -> std::sync::Arc> { + use pdfcraft_cos::{Dict, Document, Object, SaveOptions, Stream}; + let (w, h) = (200u32, 100u32); + let mut rgb = Vec::new(); + for y in 0..h { + for x in 0..w { + let ink = (WORD[0]..WORD[2]).contains(&x) && (WORD[1]..WORD[3]).contains(&y); + rgb.extend(if ink { INK } else { PAPER }); + } + } + let mut d = Document::new_empty(); + let pages = d.add(Object::Dict(Dict::new())); + let content = format!("BT /F1 20 Tf 3 Tr 42 44 Td ({text}) Tj ET q 200 0 0 100 0 0 cm /Im Do Q"); + let contents = d.add(Object::Stream(Stream::flate(Dict::new(), content.as_bytes()))); + let mut image = Dict::new(); + image.set(b"Type".to_vec(), Object::name("XObject")); + image.set(b"Subtype".to_vec(), Object::name("Image")); + image.set(b"Width".to_vec(), Object::Int(i64::from(w))); + image.set(b"Height".to_vec(), Object::Int(i64::from(h))); + image.set(b"ColorSpace".to_vec(), Object::name("DeviceRGB")); + image.set(b"BitsPerComponent".to_vec(), Object::Int(8)); + let image = d.add(Object::Stream(Stream::flate(image, &rgb))); + let mut font = Dict::new(); + font.set(b"Type".to_vec(), Object::name("Font")); + font.set(b"Subtype".to_vec(), Object::name("Type1")); + font.set(b"BaseFont".to_vec(), Object::name("Helvetica")); + let mut fonts = Dict::new(); + fonts.set(b"F1".to_vec(), Object::Dict(font)); + let mut xobjects = Dict::new(); + xobjects.set(b"Im".to_vec(), Object::Ref(image)); + let mut res = Dict::new(); + res.set(b"Font".to_vec(), Object::Dict(fonts)); + res.set(b"XObject".to_vec(), Object::Dict(xobjects)); + let mut p = Dict::new(); + p.set(b"Type".to_vec(), Object::name("Page")); + p.set(b"Parent".to_vec(), Object::Ref(pages)); + p.set(b"Contents".to_vec(), Object::Ref(contents)); + p.set(b"Resources".to_vec(), Object::Dict(res)); + p.set(b"MediaBox".to_vec(), Object::Array([0, 0, 200, 100].map(Object::Int).to_vec())); + let page = d.add(Object::Dict(p)); + d.update_dict(pages, |p| { + p.set(b"Type".to_vec(), Object::name("Pages")); + p.set(b"Kids".to_vec(), Object::Array(vec![Object::Ref(page)])); + p.set(b"Count".to_vec(), Object::Int(1)); + }) + .unwrap(); + let mut root = Dict::new(); + root.set(b"Type".to_vec(), Object::name("Catalog")); + root.set(b"Pages".to_vec(), Object::Ref(pages)); + let root = d.add(Object::Dict(root)); + d.trailer_mut().set(b"Root".to_vec(), Object::Ref(root)); + std::sync::Arc::new(pdfcraft_cos::write_full(&d, &SaveOptions::default()).unwrap()) +} + +/// The OCR layer is under the scan image: the rewritten paragraph is drawn on top of the image, +/// not in the OCR text's place where the image would hide it (#3973). One line takes the +/// single-style rewrite, a typed line break the styled one. +#[test] +fn an_ocr_layer_under_the_scan_image_shows_its_edit() { + for edit in ["Jane", "Jane\nDoe"] { + let mut s = Session::new(); + let id = s.open("ocr-under-scan.pdf", None, ocr_under_scan("John"), None).unwrap(); + let blocks = s.get(id).unwrap().text_blocks(0); + assert_eq!(blocks.len(), 1); + assert!(blocks[0].invisible, "the OCR layer is invisible text"); + assert!(word_pixels(&s, id).iter().all(|p| near(*p, INK)), "the scanned word fills its box"); + + s.apply(id, Edit::EditTextBlock { page: 0, block: 0, text: edit.into(), style: Default::default() }).unwrap(); + + let blocks = s.get(id).unwrap().text_blocks(0); + assert_eq!(blocks.iter().map(|b| b.text.replace('\n', " ")).collect::>(), [edit.replace('\n', " ")], "{edit:?}"); + let px = word_pixels(&s, id); + let paper = px.iter().filter(|p| near(**p, PAPER)).count(); + let ink = px.iter().filter(|p| near(**p, INK)).count(); + assert!(paper * 10 > px.len() * 4, "{edit:?}: the cover is drawn over the scan: {paper} of {} paper", px.len()); + assert!(ink * 100 > px.len() * 3, "{edit:?}: the new text shows: {ink} of {} ink", px.len()); + } +} diff --git a/parity/acrobat-features.toml b/parity/acrobat-features.toml index 3b8dba920c..aaaf789f39 100644 --- a/parity/acrobat-features.toml +++ b/parity/acrobat-features.toml @@ -6628,8 +6628,8 @@ milestone = "M10" status = "partial" commands = ["edit.edit_text"] tools = ["text_edit"] -evidence = ["crates/engine/tests/it/ocr_edit.rs::editing_an_ocr_word_covers_the_scan_and_shows_the_new_text", "crates/engine/tests/it/ocr_edit.rs::an_explicit_colour_is_kept_for_the_new_text", "crates/engine/tests/it/ocr_edit.rs::ocr_words_group_into_phrases_that_edit_in_place", "crates/engine/tests/it/ocr_edit.rs::a_longer_ocr_phrase_grows_rightwards", "crates/engine/tests/it/ocr_edit.rs::a_rewritten_ocr_phrase_keeps_its_word_widths", "crates/ui-egui/tests/it/ocr.rs::an_ocr_word_opens_for_editing_and_stays_open", "crates/ui-egui/tests/it/ocr.rs::editing_a_scanned_page_recognizes_its_text_first_and_only_once"] -notes = "On demand, per edit: in Edit text & images, recognised text (the invisible OCR layer over a scan) is grouped into phrases (words across a word space; a wider gap such as a table cell, and each scanned line, stay separate) and opens for editing on a click. Applying draws the new text visibly in the scan's sampled ink colour (or the chosen one) and first covers the old words with the paper colour sampled around them, so the scanned word no longer shows; the phrase keeps each word's position and width and grows rightwards instead of wrapping. Turning Edit text & images on for a scanned page (an image covering at least half of a page that has no text of its own, once per page per window; comments and text boxes over it do not count as text, a page with real text and a background picture is left alone; one undoable \"Recognize text\" step) now recognizes its words first, as Acrobat's \"Converting scanned page to Editable Text and Images\" does (English only until more OCR languages exist, gap 7). Still missing: separating the photographs on the page from the lettering, matching the scanned typeface (the layer's Helvetica is used), and covering on a textured or patterned background (one flat colour is used)." +evidence = ["crates/engine/tests/it/ocr_edit.rs::editing_an_ocr_word_covers_the_scan_and_shows_the_new_text", "crates/engine/tests/it/ocr_edit.rs::an_explicit_colour_is_kept_for_the_new_text", "crates/engine/tests/it/ocr_edit.rs::ocr_words_group_into_phrases_that_edit_in_place", "crates/engine/tests/it/ocr_edit.rs::a_longer_ocr_phrase_grows_rightwards", "crates/engine/tests/it/ocr_edit.rs::a_rewritten_ocr_phrase_keeps_its_word_widths", "crates/engine/tests/it/ocr_edit.rs::an_ocr_layer_under_the_scan_image_shows_its_edit", "crates/ui-egui/tests/it/ocr.rs::an_ocr_word_opens_for_editing_and_stays_open", "crates/ui-egui/tests/it/ocr.rs::editing_a_scanned_page_recognizes_its_text_first_and_only_once"] +notes = "On demand, per edit: in Edit text & images, recognised text (the invisible OCR layer over a scan) is grouped into phrases (words across a word space; a wider gap such as a table cell, and each scanned line, stay separate) and opens for editing on a click. Applying draws the new text visibly in the scan's sampled ink colour (or the chosen one) and first covers the old words with the paper colour sampled around them, so the scanned word no longer shows; the new text is drawn on top of the page, so it shows also when the OCR layer was written under the scan image (#3973); the phrase keeps each word's position and width and grows rightwards instead of wrapping. Turning Edit text & images on for a scanned page (an image covering at least half of a page that has no text of its own, once per page per window; comments and text boxes over it do not count as text, a page with real text and a background picture is left alone; one undoable \"Recognize text\" step) now recognizes its words first, as Acrobat's \"Converting scanned page to Editable Text and Images\" does (English only until more OCR languages exist, gap 7). Still missing: separating the photographs on the page from the lettering, matching the scanned typeface (the layer's Helvetica is used), and covering on a textured or patterned background (one flat colour is used)." [[feature]] id = "ocr.ocr-downsample" From 270a11e699f8f224b67a5e8d45cc85596797fd5f Mon Sep 17 00:00:00 2001 From: KenM76 Date: Fri, 30 Oct 2026 00:43:25 -0400 Subject: [PATCH 05/53] fix(edit): refuse before changing anything; cover undo, reopen, re-edit and the split case An OCR paragraph that may go on top of the page is checked (`check_on_top`) before either rewrite path changes the document, so a page whose marked content nests too deep is refused with nothing changed. `lift_on_top` no longer checks. Tests: undo restores the hidden layer and the scan, redo, save and reopen, and a second edit of the moved text all show the edit; a phrase split from visible text in the same text object shows its edit; a refused rewrite on a too-deep page leaves the saved bytes unchanged. Each new test was seen to fail with its part of the fix reverted. Parity: the note says when the edit is drawn on top and lists what is still missing. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SqmHYx8FiWM5PcsoiErgp9 --- crates/edit/src/place.rs | 17 ++---- crates/edit/src/tests.rs | 15 ++++++ crates/edit/src/text/rewrite.rs | 7 ++- crates/edit/src/text_regions/styled.rs | 2 +- crates/engine/tests/it/ocr_edit.rs | 73 ++++++++++++++++++++++---- parity/acrobat-features.toml | 4 +- 6 files changed, 92 insertions(+), 26 deletions(-) diff --git a/crates/edit/src/place.rs b/crates/edit/src/place.rs index 3056d3f71b..f1acb6bdca 100644 --- a/crates/edit/src/place.rs +++ b/crates/edit/src/place.rs @@ -367,19 +367,10 @@ pub(crate) fn revealed_under_paint(invisible: bool, ops: &[pdfcraft_content::Op] invisible && ops.get(at..).unwrap_or_default().iter().any(|op| op.is("Do") || op.is("BI") || op.is("sh")) } -/// The operators to draw on top of `page` ([`draw_on_top`]): all of `paragraph` when `over`, -/// after checking the page takes them, so a refusal leaves the document as it was. -pub(crate) fn lift_on_top( - doc: &Document, - page: &pdfcraft_model::Page, - over: bool, - paragraph: &mut Vec, -) -> Result, EditError> { - if !over { - return Ok(Vec::new()); - } - check_on_top(doc, page)?; - Ok(std::mem::take(paragraph)) +/// The operators to draw on top of the page ([`draw_on_top`]): all of `paragraph` when `over`, +/// none otherwise. The page was checked to take them ([`check_on_top`]) before the edit began. +pub(crate) fn lift_on_top(over: bool, paragraph: &mut Vec) -> Vec { + if over { std::mem::take(paragraph) } else { Vec::new() } } fn too_deep_on_top() -> EditError { diff --git a/crates/edit/src/tests.rs b/crates/edit/src/tests.rs index 359155015f..2d3ba5301f 100644 --- a/crates/edit/src/tests.rs +++ b/crates/edit/src/tests.rs @@ -4550,6 +4550,21 @@ fn group_selection_refuses_incomplete_content_and_resource_budget_overruns() { assert!(editable_objects(&doc, 0).is_err(), "an undecodable stream is refused"); } +/// Revealed OCR text may have to go on top of the page; a page whose marked content nests too deep +/// for that is refused before either rewrite path changes anything. +#[test] +fn an_ocr_rewrite_refused_on_top_leaves_the_document_as_it_was() { + let content = format!("{} BT /F1 20 Tf 3 Tr 42 44 Td (John) Tj ET /Im Do {}", "/X BMC ".repeat(10_001), "EMC ".repeat(10_001)); + let mut doc = text_page(&content); + assert!(text::text_blocks(&doc, 0).unwrap()[0].invisible); + let before = write_full(&doc, &SaveOptions::default()).unwrap(); + for edit in ["Jane", "Jane\nDoe"] { + let error = text::rewrite_block(&mut doc, 0, 0, Some(edit), &text::BlockStyle::default()).unwrap_err(); + assert!(error.to_string().contains("nested too deeply"), "{edit:?}: {error}"); + assert!(write_full(&doc, &SaveOptions::default()).unwrap() == before, "{edit:?}: a refused rewrite changes nothing"); + } +} + #[test] fn group_move_budget_refusals_keep_the_original_page_selectable() { let text = text_page(&format!("{} BT /F1 12 Tf 40 400 Td (Target) Tj ET", "n ".repeat(99_995))); diff --git a/crates/edit/src/text/rewrite.rs b/crates/edit/src/text/rewrite.rs index f03174885a..42db5c4af3 100644 --- a/crates/edit/src/text/rewrite.rs +++ b/crates/edit/src/text/rewrite.rs @@ -755,6 +755,11 @@ pub(crate) fn rewrite_block_from( let first = members.first().copied().ok_or_else(|| EditError::Invalid("the paragraph has no source runs".into()))?; members.iter().try_for_each(|line| line.ensure_editable())?; let p = page_dict(doc, page)?; + // Revealed OCR text may go on top of the page (see `revealed_under_paint`): a page that can't + // take that is refused here, before either rewrite path changes anything. + if b.invisible { + crate::check_on_top(doc, &p)?; + } let mut res = p.dict.get(b"Resources").map(|r| doc.resolve(r)).and_then(|r| r.as_dict().cloned()).unwrap_or_default(); let mut fonts_res = res.get(b"Font").map(|f| doc.resolve(f)).and_then(|f| f.as_dict().cloned()).unwrap_or_default(); let dispatch = choose_rewrite_path(doc, &fonts_res, &b, &members, first, text, style)?; @@ -884,7 +889,7 @@ pub(crate) fn rewrite_block_from( underline: style.underline == Some(true), }; let mut block_ops = block_ops(&plan, &writer, &wrapped, font.show_font.clone())?; - let on_top = crate::lift_on_top(doc, &p, over, &mut block_ops)?; + let on_top = crate::lift_on_top(over, &mut block_ops); let n = pdfcraft_content::num; block_ops.extend(after.ops().into_iter().filter(|op| !op.is("Tf") || after.font.is_some())); if split { diff --git a/crates/edit/src/text_regions/styled.rs b/crates/edit/src/text_regions/styled.rs index c03b75d3f4..8ff4b0b62e 100644 --- a/crates/edit/src/text_regions/styled.rs +++ b/crates/edit/src/text_regions/styled.rs @@ -974,7 +974,7 @@ pub(crate) fn rewrite_styled_block( }; } block_ops.push(Op::new("Q", vec![])); - let on_top = crate::lift_on_top(doc, &p, over, &mut block_ops)?; + let on_top = crate::lift_on_top(over, &mut block_ops); block_ops.extend(after.ops()); if split { block_ops.insert(0, Op::new("ET", vec![])); diff --git a/crates/engine/tests/it/ocr_edit.rs b/crates/engine/tests/it/ocr_edit.rs index 86fc1c9df2..a51ce8ddcb 100644 --- a/crates/engine/tests/it/ocr_edit.rs +++ b/crates/engine/tests/it/ocr_edit.rs @@ -211,6 +211,11 @@ fn visible_stretched_text_rewrites_unchanged() { /// A scan whose OCR layer is drawn first and the opaque page image over it, as some OCR tools /// write it (#3973): invisible Helvetica `text` on the scanned word, then the image. fn ocr_under_scan(text: &str) -> std::sync::Arc> { + under_scan(&format!("BT /F1 20 Tf 3 Tr 42 44 Td ({text}) Tj ET")) +} + +/// A page whose `text` content is followed by the opaque 200 × 100 scan image. +fn under_scan(text: &str) -> std::sync::Arc> { use pdfcraft_cos::{Dict, Document, Object, SaveOptions, Stream}; let (w, h) = (200u32, 100u32); let mut rgb = Vec::new(); @@ -222,7 +227,7 @@ fn ocr_under_scan(text: &str) -> std::sync::Arc> { } let mut d = Document::new_empty(); let pages = d.add(Object::Dict(Dict::new())); - let content = format!("BT /F1 20 Tf 3 Tr 42 44 Td ({text}) Tj ET q 200 0 0 100 0 0 cm /Im Do Q"); + let content = format!("{text} q 200 0 0 100 0 0 cm /Im Do Q"); let contents = d.add(Object::Stream(Stream::flate(Dict::new(), content.as_bytes()))); let mut image = Dict::new(); image.set(b"Type".to_vec(), Object::name("XObject")); @@ -264,9 +269,27 @@ fn ocr_under_scan(text: &str) -> std::sync::Arc> { std::sync::Arc::new(pdfcraft_cos::write_full(&d, &SaveOptions::default()).unwrap()) } +/// The scanned word's box shows the cover and new ink, not the scan. +fn assert_edit_shows(s: &Session, id: pdfcraft_engine::DocId, label: &str) { + let px = word_pixels(s, id); + let paper = px.iter().filter(|p| near(**p, PAPER)).count(); + let ink = px.iter().filter(|p| near(**p, INK)).count(); + assert!(paper * 10 > px.len() * 4, "{label}: the cover is drawn over the scan: {paper} of {} paper", px.len()); + assert!(ink * 100 > px.len() * 3, "{label}: the new text shows: {ink} of {} ink", px.len()); +} + +fn texts(s: &Session, id: pdfcraft_engine::DocId) -> Vec { + s.get(id).unwrap().text_blocks(0).iter().map(|b| b.text.replace('\n', " ")).collect() +} + +fn edit_block(s: &mut Session, id: pdfcraft_engine::DocId, block: usize, text: &str) { + s.apply(id, Edit::EditTextBlock { page: 0, block, text: text.into(), style: Default::default() }).unwrap(); +} + /// The OCR layer is under the scan image: the rewritten paragraph is drawn on top of the image, /// not in the OCR text's place where the image would hide it (#3973). One line takes the -/// single-style rewrite, a typed line break the styled one. +/// single-style rewrite, a typed line break the styled one. Undo restores the hidden layer, and +/// the edit survives redo, a save and reopen, and being edited again. #[test] fn an_ocr_layer_under_the_scan_image_shows_its_edit() { for edit in ["Jane", "Jane\nDoe"] { @@ -277,14 +300,46 @@ fn an_ocr_layer_under_the_scan_image_shows_its_edit() { assert!(blocks[0].invisible, "the OCR layer is invisible text"); assert!(word_pixels(&s, id).iter().all(|p| near(*p, INK)), "the scanned word fills its box"); - s.apply(id, Edit::EditTextBlock { page: 0, block: 0, text: edit.into(), style: Default::default() }).unwrap(); + edit_block(&mut s, id, 0, edit); + assert_eq!(texts(&s, id), [edit.replace('\n', " ")], "{edit:?}"); + assert_edit_shows(&s, id, edit); + s.undo(id).unwrap(); + assert_eq!(texts(&s, id), ["John"], "{edit:?}: undo"); + assert!(s.get(id).unwrap().text_blocks(0)[0].invisible, "{edit:?}: undo hides the layer again"); + assert!(word_pixels(&s, id).iter().all(|p| near(*p, INK)), "{edit:?}: undo shows the scan again"); + s.redo(id).unwrap(); + assert_edit_shows(&s, id, &format!("{edit:?} redo")); + + let saved = s.save_bytes(id).unwrap(); + let mut reopened = Session::new(); + let id = reopened.open("saved.pdf", None, saved, None).unwrap(); + assert_eq!(texts(&reopened, id), [edit.replace('\n', " ")], "{edit:?}: reopened"); + assert_edit_shows(&reopened, id, &format!("{edit:?} reopened")); + + edit_block(&mut reopened, id, 0, "Ann"); + assert_eq!(texts(&reopened, id), ["Ann"], "{edit:?}: edited again"); + assert_edit_shows(&reopened, id, &format!("{edit:?} edited again")); + } +} + +/// Visible text earlier in the same text object makes the rewrite split it; the OCR phrase still +/// goes on top of the scan, and the visible text stays. +#[test] +fn an_ocr_phrase_split_from_visible_text_under_the_scan_shows_its_edit() { + for edit in ["Jane", "Jane\nDoe"] { + let mut s = Session::new(); + let page = under_scan("BT /F1 20 Tf 10 80 Td (Head) Tj 3 Tr 32 -36 Td (John) Tj ET"); + let id = s.open("split.pdf", None, page, None).unwrap(); let blocks = s.get(id).unwrap().text_blocks(0); - assert_eq!(blocks.iter().map(|b| b.text.replace('\n', " ")).collect::>(), [edit.replace('\n', " ")], "{edit:?}"); - let px = word_pixels(&s, id); - let paper = px.iter().filter(|p| near(**p, PAPER)).count(); - let ink = px.iter().filter(|p| near(**p, INK)).count(); - assert!(paper * 10 > px.len() * 4, "{edit:?}: the cover is drawn over the scan: {paper} of {} paper", px.len()); - assert!(ink * 100 > px.len() * 3, "{edit:?}: the new text shows: {ink} of {} ink", px.len()); + let john = blocks.iter().position(|b| b.text == "John").unwrap(); + assert!(blocks[john].invisible && blocks.iter().any(|b| b.text == "Head" && !b.invisible), "{blocks:?}"); + edit_block(&mut s, id, john, edit); + let mut after = texts(&s, id); + after.sort(); + let mut want = vec!["Head".to_string(), edit.replace('\n', " ")]; + want.sort(); + assert_eq!(after, want, "{edit:?}"); + assert_edit_shows(&s, id, &format!("{edit:?} split")); } } diff --git a/parity/acrobat-features.toml b/parity/acrobat-features.toml index aaaf789f39..68716cf5fd 100644 --- a/parity/acrobat-features.toml +++ b/parity/acrobat-features.toml @@ -6628,8 +6628,8 @@ milestone = "M10" status = "partial" commands = ["edit.edit_text"] tools = ["text_edit"] -evidence = ["crates/engine/tests/it/ocr_edit.rs::editing_an_ocr_word_covers_the_scan_and_shows_the_new_text", "crates/engine/tests/it/ocr_edit.rs::an_explicit_colour_is_kept_for_the_new_text", "crates/engine/tests/it/ocr_edit.rs::ocr_words_group_into_phrases_that_edit_in_place", "crates/engine/tests/it/ocr_edit.rs::a_longer_ocr_phrase_grows_rightwards", "crates/engine/tests/it/ocr_edit.rs::a_rewritten_ocr_phrase_keeps_its_word_widths", "crates/engine/tests/it/ocr_edit.rs::an_ocr_layer_under_the_scan_image_shows_its_edit", "crates/ui-egui/tests/it/ocr.rs::an_ocr_word_opens_for_editing_and_stays_open", "crates/ui-egui/tests/it/ocr.rs::editing_a_scanned_page_recognizes_its_text_first_and_only_once"] -notes = "On demand, per edit: in Edit text & images, recognised text (the invisible OCR layer over a scan) is grouped into phrases (words across a word space; a wider gap such as a table cell, and each scanned line, stay separate) and opens for editing on a click. Applying draws the new text visibly in the scan's sampled ink colour (or the chosen one) and first covers the old words with the paper colour sampled around them, so the scanned word no longer shows; the new text is drawn on top of the page, so it shows also when the OCR layer was written under the scan image (#3973); the phrase keeps each word's position and width and grows rightwards instead of wrapping. Turning Edit text & images on for a scanned page (an image covering at least half of a page that has no text of its own, once per page per window; comments and text boxes over it do not count as text, a page with real text and a background picture is left alone; one undoable \"Recognize text\" step) now recognizes its words first, as Acrobat's \"Converting scanned page to Editable Text and Images\" does (English only until more OCR languages exist, gap 7). Still missing: separating the photographs on the page from the lettering, matching the scanned typeface (the layer's Helvetica is used), and covering on a textured or patterned background (one flat colour is used)." +evidence = ["crates/engine/tests/it/ocr_edit.rs::editing_an_ocr_word_covers_the_scan_and_shows_the_new_text", "crates/engine/tests/it/ocr_edit.rs::an_explicit_colour_is_kept_for_the_new_text", "crates/engine/tests/it/ocr_edit.rs::ocr_words_group_into_phrases_that_edit_in_place", "crates/engine/tests/it/ocr_edit.rs::a_longer_ocr_phrase_grows_rightwards", "crates/engine/tests/it/ocr_edit.rs::a_rewritten_ocr_phrase_keeps_its_word_widths", "crates/engine/tests/it/ocr_edit.rs::an_ocr_layer_under_the_scan_image_shows_its_edit", "crates/engine/tests/it/ocr_edit.rs::an_ocr_phrase_split_from_visible_text_under_the_scan_shows_its_edit", "crates/ui-egui/tests/it/ocr.rs::an_ocr_word_opens_for_editing_and_stays_open", "crates/ui-egui/tests/it/ocr.rs::editing_a_scanned_page_recognizes_its_text_first_and_only_once"] +notes = "On demand, per edit: in Edit text & images, recognised text (the invisible OCR layer over a scan) is grouped into phrases (words across a word space; a wider gap such as a table cell, and each scanned line, stay separate) and opens for editing on a click. Applying draws the new text visibly in the scan's sampled ink colour (or the chosen one) and first covers the old words with the paper colour sampled around them, so the scanned word no longer shows; when an image, inline image or shading is painted after the OCR layer (a layer written under its scan, #3973), the cover and new text are drawn on top of the page instead, so they show; the phrase keeps each word's position and width and grows rightwards instead of wrapping. Turning Edit text & images on for a scanned page (an image covering at least half of a page that has no text of its own, once per page per window; comments and text boxes over it do not count as text, a page with real text and a background picture is left alone; one undoable \"Recognize text\" step) now recognizes its words first, as Acrobat's \"Converting scanned page to Editable Text and Images\" does (English only until more OCR languages exist, gap 7). Still missing: separating the photographs on the page from the lettering, matching the scanned typeface (the layer's Helvetica is used), covering on a textured or patterned background (one flat colour is used), a scan painted by a pattern fill or a filled rectangle over a layer written under it, and, for text drawn on top, the clip, optional content and marked content (tags) that enclosed the OCR layer." [[feature]] id = "ocr.ocr-downsample" From 5a2f9273a39de3c529b9108fd41735478ff43f46 Mon Sep 17 00:00:00 2001 From: KenM76 Date: Fri, 30 Oct 2026 00:46:27 -0400 Subject: [PATCH 06/53] test(edit): pin when revealed OCR text moves on top A table test for `revealed_under_paint`: only invisible text with an XObject, inline image or shading after its place moves; a filled path, visible text or nothing after it does not. Seen to fail with the check reduced to `invisible`. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SqmHYx8FiWM5PcsoiErgp9 --- crates/edit/src/tests.rs | 16 ++++++++++++++++ 1 file changed, 16 insertions(+) diff --git a/crates/edit/src/tests.rs b/crates/edit/src/tests.rs index 2d3ba5301f..a51612106b 100644 --- a/crates/edit/src/tests.rs +++ b/crates/edit/src/tests.rs @@ -4550,6 +4550,22 @@ fn group_selection_refuses_incomplete_content_and_resource_budget_overruns() { assert!(editable_objects(&doc, 0).is_err(), "an undecodable stream is refused"); } +/// Only invisible text with an XObject, inline image or shading painted after its place moves on top. +#[test] +fn revealed_text_moves_on_top_only_when_painted_over() { + let ops = pdfcraft_content::parse(b"BT (a) Tj ET q /Im Do Q BT (b) Tj ET BI /W 1 ID x EI /Sh sh").ops; + let at = |name: &str, n: usize| ops.iter().enumerate().filter(|(_, o)| o.is(name)).nth(n).map(|(i, _)| i).unwrap(); + let after_bi = at("BI", 0) + 1; + assert!(revealed_under_paint(true, &ops, at("BT", 0))); + assert!(revealed_under_paint(true, &ops, at("BT", 1)), "inline image and shading after it"); + assert!(revealed_under_paint(true, &ops, after_bi), "a shading after it"); + assert!(!revealed_under_paint(true, &ops, ops.len()), "nothing after it"); + assert!(!revealed_under_paint(true, &ops, ops.len() + 5), "past the end"); + assert!(!revealed_under_paint(false, &ops, 0), "visible text stays where it is"); + let text_only = pdfcraft_content::parse(b"BT (a) Tj ET 0 0 10 10 re f").ops; + assert!(!revealed_under_paint(true, &text_only, 0), "a filled path does not move it"); +} + /// Revealed OCR text may have to go on top of the page; a page whose marked content nests too deep /// for that is refused before either rewrite path changes anything. #[test] From 3c11186dbd9d919aefedb3f74f6f808570dd353a Mon Sep 17 00:00:00 2001 From: KenM76 Date: Fri, 30 Oct 2026 01:22:41 -0400 Subject: [PATCH 07/53] fix(edit): move revealed OCR text on top only when an XObject covers it An unrelated image later on the page no longer moves the paragraph and its cover on top, where the cover could hide visible content inside the paragraph's box. An XObject now counts only when it is drawn at or after the paragraph and its visible box overlaps the paragraph's (`images::painted_over`, from `page_images`). An inline image or a shading after the paragraph still counts anywhere, and so does any XObject when the page's images cannot be listed. The trigger table now covers an image elsewhere, an image drawn first, visible text, a filled path, a shading and an inline image. It fails with the old any-`Do` trigger. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SqmHYx8FiWM5PcsoiErgp9 --- crates/edit/src/images.rs | 6 ++++++ crates/edit/src/place.rs | 16 +++++++++----- crates/edit/src/tests.rs | 29 +++++++++++++++----------- crates/edit/src/text/rewrite.rs | 2 +- crates/edit/src/text_regions/styled.rs | 2 +- parity/acrobat-features.toml | 2 +- 6 files changed, 37 insertions(+), 20 deletions(-) diff --git a/crates/edit/src/images.rs b/crates/edit/src/images.rs index 5180e76018..7322b8f9d6 100644 --- a/crates/edit/src/images.rs +++ b/crates/edit/src/images.rs @@ -536,6 +536,12 @@ pub(crate) fn selectable_images( .collect()) } +/// Whether an XObject drawn at or after operator `from` of the page's joined operators shows +/// inside `rect` (user space). +pub(crate) fn painted_over(doc: &Document, page: usize, rect: [f64; 4], from: usize) -> Result { + Ok(page_images(doc, page)?.iter().any(|i| i.op >= from && i.visible_rect.and_then(|v| intersection(v, rect)).is_some())) +} + /// Find the saved graphics-state scope containing just this image and its crop. Applying /// the transform before that scope moves the actual clipping path as well as the pixels. /// A shared clip must never be moved along with unrelated artwork or text. diff --git a/crates/edit/src/place.rs b/crates/edit/src/place.rs index f1acb6bdca..e040acf1f3 100644 --- a/crates/edit/src/place.rs +++ b/crates/edit/src/place.rs @@ -360,11 +360,17 @@ pub(crate) fn draw_on_top(doc: &mut Document, page: &pdfcraft_model::Page, ctm: place_tagged(doc, page, "Edited", pdfcraft_content::serialize_ops(&all), false).map(drop) } -/// Whether a rewritten paragraph that was `invisible` goes on top of the page: anything after -/// operator `at` can paint over it there (an XObject, an inline image or a shading), as the scan -/// image does over an OCR layer written before it (#3973). -pub(crate) fn revealed_under_paint(invisible: bool, ops: &[pdfcraft_content::Op], at: usize) -> bool { - invisible && ops.get(at..).unwrap_or_default().iter().any(|op| op.is("Do") || op.is("BI") || op.is("sh")) +/// Whether a rewritten paragraph that was `invisible` goes on top of page `page`: something after +/// operator `at` of the page's joined `ops` paints over its `rect` (user space), as the scan image +/// does over an OCR layer written before it (#3973). XObjects count where they are drawn; an inline +/// image or a shading after it counts anywhere, and so does any XObject when the page's images +/// cannot be listed. +pub(crate) fn revealed_under_paint(doc: &Document, page: usize, invisible: bool, rect: [f64; 4], ops: &[pdfcraft_content::Op], at: usize) -> bool { + let later = ops.get(at..).unwrap_or_default(); + if !invisible || later.iter().any(|op| op.is("BI") || op.is("sh")) { + return invisible; + } + later.iter().any(|op| op.is("Do")) && crate::images::painted_over(doc, page, rect, at).unwrap_or(true) } /// The operators to draw on top of the page ([`draw_on_top`]): all of `paragraph` when `over`, diff --git a/crates/edit/src/tests.rs b/crates/edit/src/tests.rs index a51612106b..4349ce7ad6 100644 --- a/crates/edit/src/tests.rs +++ b/crates/edit/src/tests.rs @@ -4550,20 +4550,25 @@ fn group_selection_refuses_incomplete_content_and_resource_budget_overruns() { assert!(editable_objects(&doc, 0).is_err(), "an undecodable stream is refused"); } -/// Only invisible text with an XObject, inline image or shading painted after its place moves on top. +/// Invisible text moves on top only when something painted after it covers it: an XObject drawn +/// over its box, or any inline image or shading. #[test] fn revealed_text_moves_on_top_only_when_painted_over() { - let ops = pdfcraft_content::parse(b"BT (a) Tj ET q /Im Do Q BT (b) Tj ET BI /W 1 ID x EI /Sh sh").ops; - let at = |name: &str, n: usize| ops.iter().enumerate().filter(|(_, o)| o.is(name)).nth(n).map(|(i, _)| i).unwrap(); - let after_bi = at("BI", 0) + 1; - assert!(revealed_under_paint(true, &ops, at("BT", 0))); - assert!(revealed_under_paint(true, &ops, at("BT", 1)), "inline image and shading after it"); - assert!(revealed_under_paint(true, &ops, after_bi), "a shading after it"); - assert!(!revealed_under_paint(true, &ops, ops.len()), "nothing after it"); - assert!(!revealed_under_paint(true, &ops, ops.len() + 5), "past the end"); - assert!(!revealed_under_paint(false, &ops, 0), "visible text stays where it is"); - let text_only = pdfcraft_content::parse(b"BT (a) Tj ET 0 0 10 10 re f").ops; - assert!(!revealed_under_paint(true, &text_only, 0), "a filled path does not move it"); + let on_top = |content: &str| { + let doc = image_page_with(content); + let block = text::text_blocks(&doc, 0).unwrap().remove(0); + let ops = pdfcraft_content::parse(&page_content_bytes(&doc, 0)).ops; + let at = ops.iter().position(|o| o.is("BT")).unwrap(); + revealed_under_paint(&doc, 0, block.invisible, block.rect, &ops, at) + }; + let scan = "q 200 0 0 100 100 100 cm /Im0 Do Q"; + assert!(on_top(&format!("BT /F1 12 Tf 3 Tr 110 150 Td (Under) Tj ET {scan}")), "the scan covers it"); + assert!(!on_top(&format!("BT /F1 12 Tf 3 Tr 400 600 Td (Away) Tj ET {scan}")), "an image elsewhere"); + assert!(!on_top(&format!("{scan} BT /F1 12 Tf 3 Tr 110 150 Td (Over) Tj ET")), "the scan is drawn first"); + assert!(!on_top(&format!("BT /F1 12 Tf 0 Tr 110 150 Td (Shown) Tj ET {scan}")), "visible text stays where it is"); + assert!(!on_top("BT /F1 12 Tf 3 Tr 110 150 Td (Path) Tj ET 0 0 600 800 re f"), "a filled path does not move it"); + assert!(on_top("BT /F1 12 Tf 3 Tr 400 600 Td (Shade) Tj ET /Sh sh"), "a shading after it"); + assert!(on_top("BT /F1 12 Tf 3 Tr 400 600 Td (Inline) Tj ET BI /W 1 /H 1 /CS /G /BPC 8 ID x EI"), "an inline image after it"); } /// Revealed OCR text may have to go on top of the page; a page whose marked content nests too deep diff --git a/crates/edit/src/text/rewrite.rs b/crates/edit/src/text/rewrite.rs index 42db5c4af3..768693e70a 100644 --- a/crates/edit/src/text/rewrite.rs +++ b/crates/edit/src/text/rewrite.rs @@ -873,7 +873,7 @@ pub(crate) fn rewrite_block_from( let at = if split { start } else { emit.bt_op }; // Revealed OCR text that something is painted over goes on top of the page, where it starts // from the initial state. - let over = crate::revealed_under_paint(b.invisible, &ops, at); + let over = crate::revealed_under_paint(doc, page, b.invisible, b.rect, &ops, at); let initial = TextState::initial(); let drawn_from = if over { &initial } else { after }; let plan = BlockPlan { diff --git a/crates/edit/src/text_regions/styled.rs b/crates/edit/src/text_regions/styled.rs index 8ff4b0b62e..6ea69b4dc8 100644 --- a/crates/edit/src/text_regions/styled.rs +++ b/crates/edit/src/text_regions/styled.rs @@ -787,7 +787,7 @@ pub(crate) fn rewrite_styled_block( let at = if split { start } else { o.bt_op }; // Revealed OCR text that something is painted over goes on top of the page, where it starts // from the initial state. - let over = crate::revealed_under_paint(block.invisible, &ops, at); + let over = crate::revealed_under_paint(doc, page, block.invisible, block.rect, &ops, at); let initial = TextState::initial(); let drawn_from = if over { &initial } else { after }; validate_contextual_rows(&rows, &kept)?; diff --git a/parity/acrobat-features.toml b/parity/acrobat-features.toml index 68716cf5fd..5b5927d91d 100644 --- a/parity/acrobat-features.toml +++ b/parity/acrobat-features.toml @@ -6629,7 +6629,7 @@ status = "partial" commands = ["edit.edit_text"] tools = ["text_edit"] evidence = ["crates/engine/tests/it/ocr_edit.rs::editing_an_ocr_word_covers_the_scan_and_shows_the_new_text", "crates/engine/tests/it/ocr_edit.rs::an_explicit_colour_is_kept_for_the_new_text", "crates/engine/tests/it/ocr_edit.rs::ocr_words_group_into_phrases_that_edit_in_place", "crates/engine/tests/it/ocr_edit.rs::a_longer_ocr_phrase_grows_rightwards", "crates/engine/tests/it/ocr_edit.rs::a_rewritten_ocr_phrase_keeps_its_word_widths", "crates/engine/tests/it/ocr_edit.rs::an_ocr_layer_under_the_scan_image_shows_its_edit", "crates/engine/tests/it/ocr_edit.rs::an_ocr_phrase_split_from_visible_text_under_the_scan_shows_its_edit", "crates/ui-egui/tests/it/ocr.rs::an_ocr_word_opens_for_editing_and_stays_open", "crates/ui-egui/tests/it/ocr.rs::editing_a_scanned_page_recognizes_its_text_first_and_only_once"] -notes = "On demand, per edit: in Edit text & images, recognised text (the invisible OCR layer over a scan) is grouped into phrases (words across a word space; a wider gap such as a table cell, and each scanned line, stay separate) and opens for editing on a click. Applying draws the new text visibly in the scan's sampled ink colour (or the chosen one) and first covers the old words with the paper colour sampled around them, so the scanned word no longer shows; when an image, inline image or shading is painted after the OCR layer (a layer written under its scan, #3973), the cover and new text are drawn on top of the page instead, so they show; the phrase keeps each word's position and width and grows rightwards instead of wrapping. Turning Edit text & images on for a scanned page (an image covering at least half of a page that has no text of its own, once per page per window; comments and text boxes over it do not count as text, a page with real text and a background picture is left alone; one undoable \"Recognize text\" step) now recognizes its words first, as Acrobat's \"Converting scanned page to Editable Text and Images\" does (English only until more OCR languages exist, gap 7). Still missing: separating the photographs on the page from the lettering, matching the scanned typeface (the layer's Helvetica is used), covering on a textured or patterned background (one flat colour is used), a scan painted by a pattern fill or a filled rectangle over a layer written under it, and, for text drawn on top, the clip, optional content and marked content (tags) that enclosed the OCR layer." +notes = "On demand, per edit: in Edit text & images, recognised text (the invisible OCR layer over a scan) is grouped into phrases (words across a word space; a wider gap such as a table cell, and each scanned line, stay separate) and opens for editing on a click. Applying draws the new text visibly in the scan's sampled ink colour (or the chosen one) and first covers the old words with the paper colour sampled around them, so the scanned word no longer shows; when an image or form is drawn over the OCR layer's box after it, or an inline image or shading after it (a layer written under its scan, #3973), the cover and new text are drawn on top of the page instead, so they show; the phrase keeps each word's position and width and grows rightwards instead of wrapping. Turning Edit text & images on for a scanned page (an image covering at least half of a page that has no text of its own, once per page per window; comments and text boxes over it do not count as text, a page with real text and a background picture is left alone; one undoable \"Recognize text\" step) now recognizes its words first, as Acrobat's \"Converting scanned page to Editable Text and Images\" does (English only until more OCR languages exist, gap 7). Still missing: separating the photographs on the page from the lettering, matching the scanned typeface (the layer's Helvetica is used), covering on a textured or patterned background (one flat colour is used), a scan painted by a pattern fill or a filled rectangle over a layer written under it, and, for text drawn on top, the clip, optional content and marked content (tags) that enclosed the OCR layer." [[feature]] id = "ocr.ocr-downsample" From 705d862328fb40e2aeda77874d3e41d420fed4b6 Mon Sep 17 00:00:00 2001 From: KenM76 Date: Fri, 30 Oct 2026 02:08:08 -0400 Subject: [PATCH 08/53] fix(edit): check revealed OCR text where the rewrite lands The on-top trigger compared later paint with the paragraph's old box. A rewrite that is moved (BlockStyle::offset), rewrapped wider or grown to more lines can land under an image the old box is clear of, and stayed underneath it. Both rewrite paths now test a Reach: the old box (where the cover goes) plus the moved box grown to the wrap width either way and down to the drawn line count (the widest measured line when wider), and up by one line pitch for a larger size; turned text grows by the lines' diagonal; any non-finite number makes it unbounded. Over-reaching is harmless: the cover still fills only the old box. The parity notes say so. Tests: an_ocr_rewrite_goes_on_top_of_paint_over_where_it_lands (move, width and line growth, both paths) and a_rewrites_reach_covers_its_move_ width_and_lines. Seen failing with each path reverted to the old box and with growth zeroed. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SqmHYx8FiWM5PcsoiErgp9 --- crates/edit/src/lib.rs | 2 +- crates/edit/src/place.rs | 48 ++++++++++++++++++++++++-- crates/edit/src/tests.rs | 40 +++++++++++++++++++++ crates/edit/src/text/rewrite.rs | 4 ++- crates/edit/src/text_regions/styled.rs | 15 +++++--- parity/acrobat-features.toml | 2 +- 6 files changed, 100 insertions(+), 11 deletions(-) diff --git a/crates/edit/src/lib.rs b/crates/edit/src/lib.rs index 1ca43d20a8..059e94d927 100644 --- a/crates/edit/src/lib.rs +++ b/crates/edit/src/lib.rs @@ -68,7 +68,7 @@ use marks::{matrix_number, n}; pub(crate) use place::undrawable; #[cfg(test)] use place::{ISOLATION_MAX_DEPTH, OLD_WRAP_CLOSE, WRAP_CLOSE, WRAP_OPEN, is_wrap_close, page_fonts, stream_bytes, stream_tag, tagged, wrapper_role}; -use place::{check, check_on_top, contents, draw_on_top, lift_on_top, page_list, place_tagged, revealed_under_paint}; +use place::{Reach, check, check_on_top, contents, draw_on_top, lift_on_top, page_list, place_tagged, revealed_under_paint}; pub use place::{stamp, undrawable_message}; #[cfg(test)] diff --git a/crates/edit/src/place.rs b/crates/edit/src/place.rs index e040acf1f3..0d2f138a41 100644 --- a/crates/edit/src/place.rs +++ b/crates/edit/src/place.rs @@ -360,17 +360,59 @@ pub(crate) fn draw_on_top(doc: &mut Document, page: &pdfcraft_model::Page, ctm: place_tagged(doc, page, "Edited", pdfcraft_content::serialize_ops(&all), false).map(drop) } +/// Where a rewritten paragraph can be drawn, in user space ([`revealed_under_paint`]). +pub(crate) struct Reach { + /// The paragraph's old box, where its cover goes. + pub rect: [f64; 4], + /// The move asked for (`BlockStyle::offset`). + pub offset: [f64; 2], + /// The widest line the rewrite may draw, how many lines it has, and how far apart they are. + pub width: f64, + pub rows: usize, + pub pitch: f64, + pub upright: bool, +} + +impl Reach { + /// The old box and the moved one, grown to `width` either way (alignment may grow a line left + /// or right), down to `rows` lines and up to one `pitch` above its bottom (a larger size); turned text grows by the lines' diagonal every way. + /// Unbounded when a number is not finite. + pub(crate) fn bounds(&self) -> [f64; 4] { + const UNBOUNDED: [f64; 4] = [f64::NEG_INFINITY, f64::NEG_INFINITY, f64::INFINITY, f64::INFINITY]; + let ([dx, dy], r) = (self.offset, self.rect); + if !r.iter().chain(&self.offset).chain([&self.width, &self.pitch]).all(|v| v.is_finite()) { + return UNBOUNDED; + } + let moved = [r[0] + dx, r[1] + dy, r[2] + dx, r[3] + dy]; + let rows = f64::from(u32::try_from(self.rows).unwrap_or(u32::MAX)); + let (width, height) = (self.width.abs(), self.pitch.abs() * rows); + let grown = if self.upright { + [ + moved[0].min(moved[2] - width), + moved[1].min(moved[3] - height), + moved[2].max(moved[0] + width), + moved[3].max(moved[1] + self.pitch.abs()), + ] + } else { + let d = width.hypot(height); + [moved[0] - d, moved[1] - d, moved[2] + d, moved[3] + d] + }; + let b = [r[0].min(grown[0]), r[1].min(grown[1]), r[2].max(grown[2]), r[3].max(grown[3])]; + if b.iter().all(|v| v.is_finite()) { b } else { UNBOUNDED } + } +} + /// Whether a rewritten paragraph that was `invisible` goes on top of page `page`: something after -/// operator `at` of the page's joined `ops` paints over its `rect` (user space), as the scan image +/// operator `at` of the page's joined `ops` paints over `reach` (user space, [`Reach::bounds`]), as the scan image /// does over an OCR layer written before it (#3973). XObjects count where they are drawn; an inline /// image or a shading after it counts anywhere, and so does any XObject when the page's images /// cannot be listed. -pub(crate) fn revealed_under_paint(doc: &Document, page: usize, invisible: bool, rect: [f64; 4], ops: &[pdfcraft_content::Op], at: usize) -> bool { +pub(crate) fn revealed_under_paint(doc: &Document, page: usize, invisible: bool, reach: [f64; 4], ops: &[pdfcraft_content::Op], at: usize) -> bool { let later = ops.get(at..).unwrap_or_default(); if !invisible || later.iter().any(|op| op.is("BI") || op.is("sh")) { return invisible; } - later.iter().any(|op| op.is("Do")) && crate::images::painted_over(doc, page, rect, at).unwrap_or(true) + later.iter().any(|op| op.is("Do")) && crate::images::painted_over(doc, page, reach, at).unwrap_or(true) } /// The operators to draw on top of the page ([`draw_on_top`]): all of `paragraph` when `over`, diff --git a/crates/edit/src/tests.rs b/crates/edit/src/tests.rs index 4349ce7ad6..95119c802e 100644 --- a/crates/edit/src/tests.rs +++ b/crates/edit/src/tests.rs @@ -4571,6 +4571,46 @@ fn revealed_text_moves_on_top_only_when_painted_over() { assert!(on_top("BT /F1 12 Tf 3 Tr 400 600 Td (Inline) Tj ET BI /W 1 /H 1 /CS /G /BPC 8 ID x EI"), "an inline image after it"); } +/// Revealed OCR text is checked where the rewrite lands, not only where it was: an image drawn +/// after it over just the moved or grown text still lifts the text on top, on both rewrite paths. +#[test] +fn an_ocr_rewrite_goes_on_top_of_paint_over_where_it_lands() { + let scan = "q 200 0 0 100 100 100 cm /Im0 Do Q"; + let moved = text::BlockStyle { offset: Some([-290.0, -450.0]), ..Default::default() }; + let cases = [ + ("BT /F1 12 Tf 3 Tr 400 600 Td (Away) Tj ET", "Moved", &moved), + ("BT /F1 12 Tf 3 Tr 400 600 Td (Away) Tj ET", "Moved\nTwice", &moved), + ("BT /F1 12 Tf 3 Tr 40 150 Td (Hi) Tj ET", "Hello there, this line now runs well past the scan's left edge", &text::BlockStyle::default()), + ("BT /F1 12 Tf 3 Tr 40 150 Td (Hi) Tj ET", "Hello there,\nthis line now runs well past the scan's left edge", &text::BlockStyle::default()), + ("BT /F1 12 Tf 3 Tr 110 230 Td (Above) Tj ET", "One\nTwo\nThree\nFour\nFive\nSix", &text::BlockStyle::default()), + ]; + for (content, edit, style) in cases { + let mut doc = image_page_with(&format!("{content} {scan}")); + let before = text::text_blocks(&doc, 0).unwrap().remove(0); + let ops = pdfcraft_content::parse(&page_content_bytes(&doc, 0)).ops; + let at = ops.iter().position(|o| o.is("BT")).unwrap(); + assert!(!revealed_under_paint(&doc, 0, true, before.rect, &ops, at), "{edit:?}: the old box alone is clear of the scan"); + text::rewrite_block(&mut doc, 0, 0, Some(edit), style).unwrap(); + let ops = pdfcraft_content::parse(&page_content_bytes(&doc, 0)).ops; + let scan_at = ops.iter().position(|o| o.is("Do")).unwrap(); + let shows = ["Tj", "TJ", "'", "\""]; + assert!(ops.iter().skip(scan_at).any(|o| shows.iter().any(|s| o.is(s))), "{edit:?}: the rewrite is drawn over the scan"); + } +} + +/// [`crate::Reach`] grows the old box to the move, the width either way and the lines below. +#[test] +fn a_rewrites_reach_covers_its_move_width_and_lines() { + let reach = |offset, width, rows, upright| crate::Reach { rect: [10.0, 90.0, 30.0, 100.0], offset, width, rows, pitch: 10.0, upright }.bounds(); + assert_eq!(reach([0.0, 0.0], 20.0, 1, true), [10.0, 90.0, 30.0, 100.0]); + assert_eq!(reach([5.0, -50.0], 20.0, 1, true), [10.0, 40.0, 35.0, 100.0]); + assert_eq!(reach([0.0, 0.0], 50.0, 3, true), [-20.0, 70.0, 60.0, 100.0]); + assert_eq!(reach([0.0, 0.0], 30.0, 4, false), [-40.0, 40.0, 80.0, 150.0]); + assert_eq!(reach([f64::NAN, 0.0], 20.0, 1, true)[0], f64::NEG_INFINITY); + let larger = crate::Reach { rect: [10.0, 90.0, 30.0, 100.0], offset: [0.0, 0.0], width: 20.0, rows: 1, pitch: 24.0, upright: true }; + assert_eq!(larger.bounds(), [10.0, 76.0, 30.0, 114.0], "a larger size grows the line up as well as down"); +} + /// Revealed OCR text may have to go on top of the page; a page whose marked content nests too deep /// for that is refused before either rewrite path changes anything. #[test] diff --git a/crates/edit/src/text/rewrite.rs b/crates/edit/src/text/rewrite.rs index 768693e70a..e08d54ca4e 100644 --- a/crates/edit/src/text/rewrite.rs +++ b/crates/edit/src/text/rewrite.rs @@ -873,7 +873,9 @@ pub(crate) fn rewrite_block_from( let at = if split { start } else { emit.bt_op }; // Revealed OCR text that something is painted over goes on top of the page, where it starts // from the initial state. - let over = crate::revealed_under_paint(doc, page, b.invisible, b.rect, &ops, at); + let widest = wrapped.iter().map(|l| advance.of(l)).fold(width, f64::max); + let reach = crate::Reach { rect: b.rect, offset: [dx, dy], width: widest, rows: wrapped.len(), pitch: lead * k, upright }; + let over = crate::revealed_under_paint(doc, page, b.invisible, reach.bounds(), &ops, at); let initial = TextState::initial(); let drawn_from = if over { &initial } else { after }; let plan = BlockPlan { diff --git a/crates/edit/src/text_regions/styled.rs b/crates/edit/src/text_regions/styled.rs index 6ea69b4dc8..2b19a00108 100644 --- a/crates/edit/src/text_regions/styled.rs +++ b/crates/edit/src/text_regions/styled.rs @@ -785,11 +785,6 @@ pub(crate) fn rewrite_styled_block( // What is in force where the rewrite is drawn, and what the text after it expects. let after = if split { &o.state } else { &o.bt_state }; let at = if split { start } else { o.bt_op }; - // Revealed OCR text that something is painted over goes on top of the page, where it starts - // from the initial state. - let over = crate::revealed_under_paint(doc, page, block.invisible, block.rect, &ops, at); - let initial = TextState::initial(); - let drawn_from = if over { &initial } else { after }; validate_contextual_rows(&rows, &kept)?; let measured: Vec<_> = rows.iter().map(|row| measure(row)).collect::>()?; let shaped = prepare_shaped_rows(doc, &fonts, &rows, &kept, advance)?; @@ -815,6 +810,16 @@ pub(crate) fn rewrite_styled_block( }) .rfind(|v| *v > 0.0) .unwrap_or(first.size * 1.5); + // Revealed OCR text that something is painted over goes on top of the page, where it starts + // from the initial state. + let largest = style.ranges.iter().filter_map(|r| r.size).chain(style.size).fold(first.size, f64::max); + let pitch = source_leading.max(largest * style.line_spacing.unwrap_or(1.2)); + let widest = measured.iter().copied().fold(width, f64::max); + // This path lays rows out along user-space baselines, so the paragraph is upright here. + let reach = crate::Reach { rect: block.rect, offset: [dx, dy], width: widest, rows: rows.len(), pitch, upright: true }; + let over = crate::revealed_under_paint(doc, page, block.invisible, reach.bounds(), &ops, at); + let initial = TextState::initial(); + let drawn_from = if over { &initial } else { after }; let mut baseline = first.origin.baseline; for (row_index, row) in rows.iter().enumerate() { let anchor = block.rows.get(row_index).and_then(|r| r.first()).and_then(|i| lines.get(*i)); diff --git a/parity/acrobat-features.toml b/parity/acrobat-features.toml index 5b5927d91d..ef7f0f53fb 100644 --- a/parity/acrobat-features.toml +++ b/parity/acrobat-features.toml @@ -6629,7 +6629,7 @@ status = "partial" commands = ["edit.edit_text"] tools = ["text_edit"] evidence = ["crates/engine/tests/it/ocr_edit.rs::editing_an_ocr_word_covers_the_scan_and_shows_the_new_text", "crates/engine/tests/it/ocr_edit.rs::an_explicit_colour_is_kept_for_the_new_text", "crates/engine/tests/it/ocr_edit.rs::ocr_words_group_into_phrases_that_edit_in_place", "crates/engine/tests/it/ocr_edit.rs::a_longer_ocr_phrase_grows_rightwards", "crates/engine/tests/it/ocr_edit.rs::a_rewritten_ocr_phrase_keeps_its_word_widths", "crates/engine/tests/it/ocr_edit.rs::an_ocr_layer_under_the_scan_image_shows_its_edit", "crates/engine/tests/it/ocr_edit.rs::an_ocr_phrase_split_from_visible_text_under_the_scan_shows_its_edit", "crates/ui-egui/tests/it/ocr.rs::an_ocr_word_opens_for_editing_and_stays_open", "crates/ui-egui/tests/it/ocr.rs::editing_a_scanned_page_recognizes_its_text_first_and_only_once"] -notes = "On demand, per edit: in Edit text & images, recognised text (the invisible OCR layer over a scan) is grouped into phrases (words across a word space; a wider gap such as a table cell, and each scanned line, stay separate) and opens for editing on a click. Applying draws the new text visibly in the scan's sampled ink colour (or the chosen one) and first covers the old words with the paper colour sampled around them, so the scanned word no longer shows; when an image or form is drawn over the OCR layer's box after it, or an inline image or shading after it (a layer written under its scan, #3973), the cover and new text are drawn on top of the page instead, so they show; the phrase keeps each word's position and width and grows rightwards instead of wrapping. Turning Edit text & images on for a scanned page (an image covering at least half of a page that has no text of its own, once per page per window; comments and text boxes over it do not count as text, a page with real text and a background picture is left alone; one undoable \"Recognize text\" step) now recognizes its words first, as Acrobat's \"Converting scanned page to Editable Text and Images\" does (English only until more OCR languages exist, gap 7). Still missing: separating the photographs on the page from the lettering, matching the scanned typeface (the layer's Helvetica is used), covering on a textured or patterned background (one flat colour is used), a scan painted by a pattern fill or a filled rectangle over a layer written under it, and, for text drawn on top, the clip, optional content and marked content (tags) that enclosed the OCR layer." +notes = "On demand, per edit: in Edit text & images, recognised text (the invisible OCR layer over a scan) is grouped into phrases (words across a word space; a wider gap such as a table cell, and each scanned line, stay separate) and opens for editing on a click. Applying draws the new text visibly in the scan's sampled ink colour (or the chosen one) and first covers the old words with the paper colour sampled around them, so the scanned word no longer shows; when an image or form is drawn after the OCR layer over where the edit lands (its old box, moved, rewrapped or grown to more lines), or an inline image or shading after it (a layer written under its scan, #3973), the cover and new text are drawn on top of the page instead, so they show; the phrase keeps each word's position and width and grows rightwards instead of wrapping. Turning Edit text & images on for a scanned page (an image covering at least half of a page that has no text of its own, once per page per window; comments and text boxes over it do not count as text, a page with real text and a background picture is left alone; one undoable \"Recognize text\" step) now recognizes its words first, as Acrobat's \"Converting scanned page to Editable Text and Images\" does (English only until more OCR languages exist, gap 7). Still missing: separating the photographs on the page from the lettering, matching the scanned typeface (the layer's Helvetica is used), covering on a textured or patterned background (one flat colour is used), a scan painted by a pattern fill or a filled rectangle over a layer written under it, and, for text drawn on top, the clip, optional content and marked content (tags) that enclosed the OCR layer." [[feature]] id = "ocr.ocr-downsample" From 8fd151ff79ec2f8b35449bf6f1c24dc32a644095 Mon Sep 17 00:00:00 2001 From: KenM76 Date: Fri, 30 Oct 2026 02:49:28 -0400 Subject: [PATCH 09/53] fix(edit): reach the styled rewrite's rows at the step they are written The styled path writes rows `line_spacing` (1.5 by default) times the size apart when a size or spacing is set, but its Reach assumed 1.2, so an image over only the lower rows of a larger rewrite was missed. The pitch now comes from the row writer's own step (row_step, shared by both), at the largest size among the paragraph's lines and ranges. The landing test gains a six-row rewrite at twice the size over an image covering only its last row; it fails with the 1.2 step. A case with only line_spacing set covers the other branch. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SqmHYx8FiWM5PcsoiErgp9 --- crates/edit/src/tests.rs | 9 ++++++++- crates/edit/src/text_regions/styled.rs | 21 ++++++++++++++------- 2 files changed, 22 insertions(+), 8 deletions(-) diff --git a/crates/edit/src/tests.rs b/crates/edit/src/tests.rs index 95119c802e..e728a90393 100644 --- a/crates/edit/src/tests.rs +++ b/crates/edit/src/tests.rs @@ -4577,6 +4577,8 @@ fn revealed_text_moves_on_top_only_when_painted_over() { fn an_ocr_rewrite_goes_on_top_of_paint_over_where_it_lands() { let scan = "q 200 0 0 100 100 100 cm /Im0 Do Q"; let moved = text::BlockStyle { offset: Some([-290.0, -450.0]), ..Default::default() }; + let sized = text::BlockStyle { size: Some(24.0), ..Default::default() }; + let spaced = text::BlockStyle { line_spacing: Some(3.0), ..Default::default() }; let cases = [ ("BT /F1 12 Tf 3 Tr 400 600 Td (Away) Tj ET", "Moved", &moved), ("BT /F1 12 Tf 3 Tr 400 600 Td (Away) Tj ET", "Moved\nTwice", &moved), @@ -4584,7 +4586,12 @@ fn an_ocr_rewrite_goes_on_top_of_paint_over_where_it_lands() { ("BT /F1 12 Tf 3 Tr 40 150 Td (Hi) Tj ET", "Hello there,\nthis line now runs well past the scan's left edge", &text::BlockStyle::default()), ("BT /F1 12 Tf 3 Tr 110 230 Td (Above) Tj ET", "One\nTwo\nThree\nFour\nFive\nSix", &text::BlockStyle::default()), ]; - for (content, edit, style) in cases { + let low = "q 200 0 0 30 100 300 cm /Im0 Do Q"; + let cases = cases.into_iter().map(|(c, e, s)| (c, e, s, scan)).chain([ + ("BT /F1 12 Tf 3 Tr 110 500 Td (Above) Tj ET", "One\nTwo\nThree\nFour\nFive\nSix", &sized, low), + ("BT /F1 12 Tf 3 Tr 110 500 Td (Above) Tj ET", "One\nTwo\nThree\nFour\nFive\nSix", &spaced, low), + ]); + for (content, edit, style, scan) in cases { let mut doc = image_page_with(&format!("{content} {scan}")); let before = text::text_blocks(&doc, 0).unwrap().remove(0); let ops = pdfcraft_content::parse(&page_content_bytes(&doc, 0)).ops; diff --git a/crates/edit/src/text_regions/styled.rs b/crates/edit/src/text_regions/styled.rs index 2b19a00108..a606952233 100644 --- a/crates/edit/src/text_regions/styled.rs +++ b/crates/edit/src/text_regions/styled.rs @@ -402,6 +402,16 @@ fn paragraph_geometry(doc: &Document, page: &pdfcraft_model::Page, block: &TextB Ok((dx, dy, style.width.unwrap_or(natural).clamp(1.0, 14_400.0))) } +/// How far below a row of `size` the next one is written: the source's leading unless a size +/// or line spacing is chosen. +fn row_step(style: &BlockStyle, size: f64, source_leading: f64) -> f64 { + if style.line_spacing.is_none() && style.size.is_none() { + source_leading + } else { + style.line_spacing.unwrap_or(1.5).clamp(0.5, 5.0) * style.size.unwrap_or(size) + } +} + fn wrap_styled_rows( chars: &[StyledChar], width: f64, @@ -812,8 +822,9 @@ pub(crate) fn rewrite_styled_block( .unwrap_or(first.size * 1.5); // Revealed OCR text that something is painted over goes on top of the page, where it starts // from the initial state. - let largest = style.ranges.iter().filter_map(|r| r.size).chain(style.size).fold(first.size, f64::max); - let pitch = source_leading.max(largest * style.line_spacing.unwrap_or(1.2)); + let largest = + block.lines.iter().filter_map(|i| lines.get(*i)).map(|l| l.size).chain(style.ranges.iter().filter_map(|r| r.size)).fold(first.size, f64::max); + let pitch = source_leading.max(row_step(style, largest, source_leading)); let widest = measured.iter().copied().fold(width, f64::max); // This path lays rows out along user-space baselines, so the paragraph is upright here. let reach = crate::Reach { rect: block.rect, offset: [dx, dy], width: widest, rows: rows.len(), pitch, upright: true }; @@ -972,11 +983,7 @@ pub(crate) fn rewrite_styled_block( x += segment_width; pos = end; } - baseline -= if style.line_spacing.is_none() && style.size.is_none() { - source_leading - } else { - style.line_spacing.unwrap_or(1.5).clamp(0.5, 5.0) * style.size.unwrap_or(anchor.size) - }; + baseline -= row_step(style, anchor.size, source_leading); } block_ops.push(Op::new("Q", vec![])); let on_top = crate::lift_on_top(over, &mut block_ops); From 8e5f423d6d777380db7807a4e1c9dafb098368c9 Mon Sep 17 00:00:00 2001 From: KenM76 Date: Fri, 30 Oct 2026 03:31:03 -0400 Subject: [PATCH 10/53] fix(edit): reach as high as a larger range's glyphs With only a range enlarged, row_step is the source leading, so the styled rewrite's reach rose one source line above its old box while a 120-point range draws far higher, and an image there was missed. The pitch is now at least 1.2 times the largest size. The landing test gains a 120-point range on a 12-point OCR line under an image above the old box; it fails without the floor. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SqmHYx8FiWM5PcsoiErgp9 --- crates/edit/src/tests.rs | 13 +++++++++++++ crates/edit/src/text_regions/styled.rs | 3 ++- 2 files changed, 15 insertions(+), 1 deletion(-) diff --git a/crates/edit/src/tests.rs b/crates/edit/src/tests.rs index e728a90393..dead2dcef0 100644 --- a/crates/edit/src/tests.rs +++ b/crates/edit/src/tests.rs @@ -4579,6 +4579,18 @@ fn an_ocr_rewrite_goes_on_top_of_paint_over_where_it_lands() { let moved = text::BlockStyle { offset: Some([-290.0, -450.0]), ..Default::default() }; let sized = text::BlockStyle { size: Some(24.0), ..Default::default() }; let spaced = text::BlockStyle { line_spacing: Some(3.0), ..Default::default() }; + let range = text::RangeStyle { + range: 0..2, + family: None, + resource: None, + size: Some(120.0), + color: None, + underline: None, + char_spacing: None, + scale: None, + }; + let enlarged = text::BlockStyle { ranges: vec![range], ..Default::default() }; + let above = "q 200 0 0 60 100 200 cm /Im0 Do Q"; let cases = [ ("BT /F1 12 Tf 3 Tr 400 600 Td (Away) Tj ET", "Moved", &moved), ("BT /F1 12 Tf 3 Tr 400 600 Td (Away) Tj ET", "Moved\nTwice", &moved), @@ -4590,6 +4602,7 @@ fn an_ocr_rewrite_goes_on_top_of_paint_over_where_it_lands() { let cases = cases.into_iter().map(|(c, e, s)| (c, e, s, scan)).chain([ ("BT /F1 12 Tf 3 Tr 110 500 Td (Above) Tj ET", "One\nTwo\nThree\nFour\nFive\nSix", &sized, low), ("BT /F1 12 Tf 3 Tr 110 500 Td (Above) Tj ET", "One\nTwo\nThree\nFour\nFive\nSix", &spaced, low), + ("BT /F1 12 Tf 3 Tr 110 150 Td (Hi there) Tj ET", "Hi there", &enlarged, above), ]); for (content, edit, style, scan) in cases { let mut doc = image_page_with(&format!("{content} {scan}")); diff --git a/crates/edit/src/text_regions/styled.rs b/crates/edit/src/text_regions/styled.rs index a606952233..7fe7a6a066 100644 --- a/crates/edit/src/text_regions/styled.rs +++ b/crates/edit/src/text_regions/styled.rs @@ -824,7 +824,8 @@ pub(crate) fn rewrite_styled_block( // from the initial state. let largest = block.lines.iter().filter_map(|i| lines.get(*i)).map(|l| l.size).chain(style.ranges.iter().filter_map(|r| r.size)).fold(first.size, f64::max); - let pitch = source_leading.max(row_step(style, largest, source_leading)); + // At least one line of the largest glyphs, so a larger range reaches up as far as it draws. + let pitch = source_leading.max(row_step(style, largest, source_leading)).max(largest * 1.2); let widest = measured.iter().copied().fold(width, f64::max); // This path lays rows out along user-space baselines, so the paragraph is upright here. let reach = crate::Reach { rect: block.rect, offset: [dx, dy], width: widest, rows: rows.len(), pitch, upright: true }; From ffa08fa00d664898222ba63fd04ffc45703fe634 Mon Sep 17 00:00:00 2001 From: KenM76 Date: Fri, 30 Oct 2026 03:37:21 -0400 Subject: [PATCH 11/53] fix(edit): reach one line above the paragraph's top (#3973) A larger range on the first line of a multi-line paragraph rises from the first line's baseline, not from the old box's bottom, so the reach now extends one pitch above the moved box's top. Name the 1.2 em line floor. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SqmHYx8FiWM5PcsoiErgp9 --- crates/edit/src/place.rs | 9 ++------- crates/edit/src/tests.rs | 6 +++--- crates/edit/src/text_regions/styled.rs | 3 ++- 3 files changed, 7 insertions(+), 11 deletions(-) diff --git a/crates/edit/src/place.rs b/crates/edit/src/place.rs index 0d2f138a41..b55e71a6ff 100644 --- a/crates/edit/src/place.rs +++ b/crates/edit/src/place.rs @@ -375,7 +375,7 @@ pub(crate) struct Reach { impl Reach { /// The old box and the moved one, grown to `width` either way (alignment may grow a line left - /// or right), down to `rows` lines and up to one `pitch` above its bottom (a larger size); turned text grows by the lines' diagonal every way. + /// or right), down to `rows` lines and up one `pitch` above its top (a larger size on the first line, however many lines the old box had); turned text grows by the lines' diagonal every way. /// Unbounded when a number is not finite. pub(crate) fn bounds(&self) -> [f64; 4] { const UNBOUNDED: [f64; 4] = [f64::NEG_INFINITY, f64::NEG_INFINITY, f64::INFINITY, f64::INFINITY]; @@ -387,12 +387,7 @@ impl Reach { let rows = f64::from(u32::try_from(self.rows).unwrap_or(u32::MAX)); let (width, height) = (self.width.abs(), self.pitch.abs() * rows); let grown = if self.upright { - [ - moved[0].min(moved[2] - width), - moved[1].min(moved[3] - height), - moved[2].max(moved[0] + width), - moved[3].max(moved[1] + self.pitch.abs()), - ] + [moved[0].min(moved[2] - width), moved[1].min(moved[3] - height), moved[2].max(moved[0] + width), moved[3] + self.pitch.abs()] } else { let d = width.hypot(height); [moved[0] - d, moved[1] - d, moved[2] + d, moved[3] + d] diff --git a/crates/edit/src/tests.rs b/crates/edit/src/tests.rs index dead2dcef0..7ee78e882b 100644 --- a/crates/edit/src/tests.rs +++ b/crates/edit/src/tests.rs @@ -4622,13 +4622,13 @@ fn an_ocr_rewrite_goes_on_top_of_paint_over_where_it_lands() { #[test] fn a_rewrites_reach_covers_its_move_width_and_lines() { let reach = |offset, width, rows, upright| crate::Reach { rect: [10.0, 90.0, 30.0, 100.0], offset, width, rows, pitch: 10.0, upright }.bounds(); - assert_eq!(reach([0.0, 0.0], 20.0, 1, true), [10.0, 90.0, 30.0, 100.0]); + assert_eq!(reach([0.0, 0.0], 20.0, 1, true), [10.0, 90.0, 30.0, 110.0]); assert_eq!(reach([5.0, -50.0], 20.0, 1, true), [10.0, 40.0, 35.0, 100.0]); - assert_eq!(reach([0.0, 0.0], 50.0, 3, true), [-20.0, 70.0, 60.0, 100.0]); + assert_eq!(reach([0.0, 0.0], 50.0, 3, true), [-20.0, 70.0, 60.0, 110.0]); assert_eq!(reach([0.0, 0.0], 30.0, 4, false), [-40.0, 40.0, 80.0, 150.0]); assert_eq!(reach([f64::NAN, 0.0], 20.0, 1, true)[0], f64::NEG_INFINITY); let larger = crate::Reach { rect: [10.0, 90.0, 30.0, 100.0], offset: [0.0, 0.0], width: 20.0, rows: 1, pitch: 24.0, upright: true }; - assert_eq!(larger.bounds(), [10.0, 76.0, 30.0, 114.0], "a larger size grows the line up as well as down"); + assert_eq!(larger.bounds(), [10.0, 76.0, 30.0, 124.0], "a larger size grows the line up as well as down"); } /// Revealed OCR text may have to go on top of the page; a page whose marked content nests too deep diff --git a/crates/edit/src/text_regions/styled.rs b/crates/edit/src/text_regions/styled.rs index 7fe7a6a066..943bece7af 100644 --- a/crates/edit/src/text_regions/styled.rs +++ b/crates/edit/src/text_regions/styled.rs @@ -825,7 +825,8 @@ pub(crate) fn rewrite_styled_block( let largest = block.lines.iter().filter_map(|i| lines.get(*i)).map(|l| l.size).chain(style.ranges.iter().filter_map(|r| r.size)).fold(first.size, f64::max); // At least one line of the largest glyphs, so a larger range reaches up as far as it draws. - let pitch = source_leading.max(row_step(style, largest, source_leading)).max(largest * 1.2); + const LINE_EM: f64 = 1.2; + let pitch = source_leading.max(row_step(style, largest, source_leading)).max(largest * LINE_EM); let widest = measured.iter().copied().fold(width, f64::max); // This path lays rows out along user-space baselines, so the paragraph is upright here. let reach = crate::Reach { rect: block.rect, offset: [dx, dy], width: widest, rows: rows.len(), pitch, upright: true }; From e93e8d50ba7082c4c6e60c4dd41791280bbb2506 Mon Sep 17 00:00:00 2001 From: KenM76 Date: Mon, 2 Nov 2026 13:15:43 -0500 Subject: [PATCH 12/53] fix(edit): keep a soft mask anchored when the edit is lifted over the scan (#3973) A soft mask is placed by the CTM in force when its `gs` runs (ISO 32000-2 11.6.5.2). The paragraph lifted over the scan re-selects the graphics state after its own `cm`, so a mask selected before a translation landed translated with it and could hide the new text. A re-emitted state that carries a dictionary /SMask now records its anchor CTM and is selected there: `cm` to the anchor, `gs`, `cm` back. The test edits the same paragraph with and without a mask and requires the same ink; it fails with the re-anchoring disabled (the mask lands 100 pt right and the text vanishes). Both rewrite paths. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SqmHYx8FiWM5PcsoiErgp9 --- crates/edit/src/text/interp.rs | 54 ++++++++++++++++++++------ crates/edit/src/text/rewrite.rs | 4 +- crates/edit/src/text_regions/styled.rs | 3 +- crates/engine/tests/it/ocr_edit.rs | 53 +++++++++++++++++++++++++ 4 files changed, 101 insertions(+), 13 deletions(-) diff --git a/crates/edit/src/text/interp.rs b/crates/edit/src/text/interp.rs index a9eb8a0c5f..61f82509b1 100644 --- a/crates/edit/src/text/interp.rs +++ b/crates/edit/src/text/interp.rs @@ -18,7 +18,7 @@ pub(super) struct TextState { /// graphics state the text was painted with that [`TextState::ops`] doesn't restate. pub(super) stroke: Vec, pub(super) line_width: Option, - ext_gstate: Vec>, + ext_gstate: Vec, /// The dash pattern and line cap in force (see [`Ts`]): compared, not restated. pub(super) dash: Dash, pub(super) line_cap: i64, @@ -74,13 +74,17 @@ impl TextState { /// /// A `gs` has no inverse, so a selection `current` has and this state doesn't stays in force: /// run these inside a `q`/`Q` that starts from `current`, one per state drawn. - pub(crate) fn paint_ops(&self, current: &TextState) -> Vec { + /// + /// `ctm` is the CTM where they run. A soft mask is placed by the CTM its `gs` ran at + /// (ISO 32000-2 §11.6.5.2), so one selected at another CTM is selected there and the CTM + /// set back. + pub(crate) fn paint_ops(&self, current: &TextState, ctm: [f64; 6]) -> Vec { let n = pdfcraft_content::num; let mut v = Vec::new(); // The selections both share stay in force; the ones after them are selected again in order. let shared = self.ext_gstate.iter().zip(¤t.ext_gstate).take_while(|(a, b)| a == b).count(); - for name in self.ext_gstate.iter().skip(shared) { - v.push(Op::new("gs", vec![Object::Name(name.clone())])); + for sel in self.ext_gstate.iter().skip(shared) { + sel.push_ops(&mut v, ctm); } // A reselected ExtGState may set `/LW`, so the width (already the effective one) follows it. let reselected = !v.is_empty(); @@ -242,7 +246,7 @@ pub(super) struct Ts { /// The line width a `w` or an ExtGState's `/LW` set (`None`: the initial 1.0). pub(super) line_width: Option, /// The ExtGState resources selected with `gs`, oldest first (see [`select_gs`]). - pub(super) ext_gstate: Vec>, + pub(super) ext_gstate: Vec, /// The dash pattern a `d` or an ExtGState's `/D` set (initially solid: `[] 0`). pub(super) dash: Dash, /// The line cap a `J` or an ExtGState's `/LC` set (initially 0, butt). @@ -275,12 +279,38 @@ const MAX_GS: usize = 32; /// Record `gs /name` in `list`. A `gs` sets every parameter its dictionary names to an absolute /// value, so a later selection of the same dictionary overrides the earlier one completely: the /// earlier entry goes, which keeps replaying the list equivalent to replaying the stream. -fn select_gs(list: &mut Vec>, name: &[u8]) { - list.retain(|n| n != name); +fn select_gs(list: &mut Vec, name: &[u8], anchor: Option<[f64; 6]>) { + list.retain(|s| s.name != name); if list.len() >= MAX_GS { list.remove(0); } - list.push(name.to_vec()); + list.push(GsSelection { name: name.to_vec(), anchor }); +} + +/// One `gs` selection: the ExtGState's resource name, and the CTM it ran at when the dictionary +/// has a soft mask, which that CTM places (ISO 32000-2 §11.6.5.2). +#[derive(Clone, Debug, PartialEq)] +pub(crate) struct GsSelection { + name: Vec, + anchor: Option<[f64; 6]>, +} + +impl GsSelection { + /// Select this ExtGState from where the CTM is `ctm`: at its anchor, then back to `ctm`. Both + /// moves need an invertible matrix; without one it is selected where it runs. + fn push_ops(&self, v: &mut Vec, ctm: [f64; 6]) { + let gs = Op::new("gs", vec![Object::Name(self.name.clone())]); + let moves = self.anchor.filter(|a| *a != ctm).and_then(|a| { + let (anchor, here) = (Matrix(a), Matrix(ctm)); + Some((anchor.then(&here.invert()?), here.then(&anchor.invert()?))) + }); + let Some((to, back)) = moves else { + v.push(gs); + return; + }; + let cm = |m: Matrix| Op::new("cm", m.0.iter().map(|x| pdfcraft_content::num(*x)).collect()); + v.extend([cm(to), gs, cm(back)]); + } } /// The text-showing operators of one stream with their text, font, box and baseline. @@ -898,10 +928,12 @@ pub(super) fn interpret( b"Ts" => ts.rise = op.num(0).unwrap_or(0.0), b"Tr" => ts.render_mode = op.operands.first().and_then(Object::as_int).unwrap_or(-1), b"gs" => { - if let Some(name) = op.name(0) { - select_gs(&mut ts.ext_gstate, name); - } let params = op.name(0).and_then(|name| states.get(name)).map(|p| doc.resolve(p)); + if let Some(name) = op.name(0) { + let masked = + params.as_deref().and_then(Object::as_dict).and_then(|p| p.get(b"SMask")).is_some_and(|m| doc.resolve(m).as_dict().is_some()); + select_gs(&mut ts.ext_gstate, name, masked.then_some(ts.ctm.0)); + } if let Some(params) = params.as_deref().and_then(Object::as_dict) { ts.gs.apply(doc, params); // `/LW` sets the line width as `w` does (ISO 32000-2 §8.4.5). diff --git a/crates/edit/src/text/rewrite.rs b/crates/edit/src/text/rewrite.rs index 6bf3d89784..d27d23e0b5 100644 --- a/crates/edit/src/text/rewrite.rs +++ b/crates/edit/src/text/rewrite.rs @@ -702,11 +702,13 @@ fn block_ops(plan: &BlockPlan<'_>, writer: &LineWriter<'_>, wrapped: &[(String, let mut ops = vec![Op::new("q", vec![])]; ops.extend(reveal_ops(plan.block, plan.lines, plan.style, plan.emit.ctm)?); // Moved: a translation inside that state. + let mut ctm = Matrix(plan.emit.ctm); if let Some((ax, ay)) = plan.movement { ops.push(Op::new("cm", vec![n(1.0), n(0.0), n(0.0), n(1.0), n(ax), n(ay)])); + ctm = Matrix([1.0, 0.0, 0.0, 1.0, ax, ay]).then(&ctm); } // Painted as the paragraph was, including what its text object set before the first show. - ops.extend(plan.state.paint_ops(plan.after)); + ops.extend(plan.state.paint_ops(plan.after, ctm.0)); ops.push(Op::new("BT", vec![])); let mut state = plan.state.clone(); state.word_spacing = 0.0; diff --git a/crates/edit/src/text_regions/styled.rs b/crates/edit/src/text_regions/styled.rs index 5ddb1064fb..7a9204a8c0 100644 --- a/crates/edit/src/text_regions/styled.rs +++ b/crates/edit/src/text_regions/styled.rs @@ -1001,6 +1001,7 @@ pub(crate) fn rewrite_styled_block( .ok_or_else(|| EditError::Invalid("the paragraph is drawn in a space it cannot be edited in".into()))?; let (tx, ty) = back.apply(dx, dy); block_ops.push(Op::new("cm", vec![n(1.0), n(0.0), n(0.0), n(1.0), n(tx), n(ty)])); + let run_ctm = Matrix([1.0, 0.0, 0.0, 1.0, tx, ty]).then(&Matrix(o.ctm)).0; // A new row can land one leading above the next paragraph and look like the same // paragraph. When every row is in this rewrite, a paragraph marker keeps them in one // scope so the neighbour stays out (#2517). A row left in the source stream cannot share @@ -1098,7 +1099,7 @@ pub(crate) fn rewrite_styled_block( // object (`gs`, stroke, line width, `Tr`; #3508/#3509) from the state the rewrite is // drawn in: a `gs` has no inverse, so one run's ExtGState would stay for the next. block_ops.push(Op::new("q", vec![])); - block_ops.extend(state.paint_ops(drawn_from)); + block_ops.extend(state.paint_ops(drawn_from, run_ctm)); if let Some((_, shaped, _)) = shaped_runs.get(&(row_index, pos)) { let resource = glyph_resources.get(&(row_index, pos)).ok_or_else(|| EditError::Invalid("shaped font is missing".into()))?; let mut properties = Dict::new(); diff --git a/crates/engine/tests/it/ocr_edit.rs b/crates/engine/tests/it/ocr_edit.rs index a51ce8ddcb..53ca309326 100644 --- a/crates/engine/tests/it/ocr_edit.rs +++ b/crates/engine/tests/it/ocr_edit.rs @@ -216,6 +216,11 @@ fn ocr_under_scan(text: &str) -> std::sync::Arc> { /// A page whose `text` content is followed by the opaque 200 × 100 scan image. fn under_scan(text: &str) -> std::sync::Arc> { + under_scan_with(text, |_| pdfcraft_cos::Dict::new()) +} + +/// As [`under_scan`], with the page's ExtGState resources made by `ext_gstates`. +fn under_scan_with(text: &str, ext_gstates: impl FnOnce(&mut pdfcraft_cos::Document) -> pdfcraft_cos::Dict) -> std::sync::Arc> { use pdfcraft_cos::{Dict, Document, Object, SaveOptions, Stream}; let (w, h) = (200u32, 100u32); let mut rgb = Vec::new(); @@ -248,6 +253,7 @@ fn under_scan(text: &str) -> std::sync::Arc> { let mut res = Dict::new(); res.set(b"Font".to_vec(), Object::Dict(fonts)); res.set(b"XObject".to_vec(), Object::Dict(xobjects)); + res.set(b"ExtGState".to_vec(), Object::Dict(ext_gstates(&mut d))); let mut p = Dict::new(); p.set(b"Type".to_vec(), Object::name("Page")); p.set(b"Parent".to_vec(), Object::Ref(pages)); @@ -343,3 +349,50 @@ fn an_ocr_phrase_split_from_visible_text_under_the_scan_shows_its_edit() { assert_edit_shows(&s, id, &format!("{edit:?} split")); } } + +/// `/SM`: a luminosity soft mask that is white over 30..110 × 30..70 in the space it is selected +/// in, black elsewhere. +fn soft_mask_state(d: &mut pdfcraft_cos::Document) -> pdfcraft_cos::Dict { + use pdfcraft_cos::{Dict, Object, Stream}; + let mut group = Dict::new(); + group.set(b"S".to_vec(), Object::name("Transparency")); + group.set(b"CS".to_vec(), Object::name("DeviceRGB")); + let mut form = Dict::new(); + form.set(b"Type".to_vec(), Object::name("XObject")); + form.set(b"Subtype".to_vec(), Object::name("Form")); + form.set(b"BBox".to_vec(), Object::Array([0, 0, 200, 100].map(Object::Int).to_vec())); + form.set(b"Group".to_vec(), Object::Dict(group)); + let mut mask = Dict::new(); + mask.set(b"Type".to_vec(), Object::name("Mask")); + mask.set(b"S".to_vec(), Object::name("Luminosity")); + mask.set(b"G".to_vec(), Object::Ref(d.add(Object::Stream(Stream::flate(form, b"1 g 30 30 80 40 re f"))))); + let mut state = Dict::new(); + state.set(b"Type".to_vec(), Object::name("ExtGState")); + state.set(b"SMask".to_vec(), Object::Dict(mask)); + let mut states = Dict::new(); + states.set(b"SM".to_vec(), Object::Dict(state)); + states +} + +/// The soft mask is selected before a translation, so it is placed by the CTM before it. The +/// paragraph lifted over the scan selects it again after its own `cm`: there it must be placed +/// as it was, not 100 pt to the right where it would hide the new text, so the edit renders +/// exactly as the same edit on the page without the mask. Both rewrite paths. +#[test] +fn a_soft_mask_selected_before_a_cm_stays_put_when_the_edit_is_lifted() { + for edit in ["Jane", "Jane\nDoe"] { + let mut plain = Session::new(); + let page = under_scan("q 1 0 0 1 100 0 cm BT /F1 20 Tf 3 Tr -58 44 Td (John) Tj ET Q"); + let plain_id = plain.open("plain.pdf", None, page, None).unwrap(); + edit_block(&mut plain, plain_id, 0, edit); + let mut s = Session::new(); + let page = under_scan_with("q /SM gs 1 0 0 1 100 0 cm BT /F1 20 Tf 3 Tr -58 44 Td (John) Tj ET Q", soft_mask_state); + let id = s.open("masked.pdf", None, page, None).unwrap(); + assert_eq!(texts(&s, id), ["John"]); + edit_block(&mut s, id, 0, edit); + assert_eq!(texts(&s, id), [edit.replace('\n', " ")], "{edit:?}"); + assert_edit_shows(&s, id, &format!("{edit:?} masked")); + let ink = |s: &Session, id| word_pixels(s, id).iter().filter(|p| near(**p, INK)).count(); + assert_eq!(ink(&s, id), ink(&plain, plain_id), "{edit:?}: the mask does not hide the new text"); + } +} From 56578a6006ebce754bac22b13389ea1f8319620c Mon Sep 17 00:00:00 2001 From: KenM76 Date: Mon, 2 Nov 2026 13:42:16 -0500 Subject: [PATCH 13/53] M3: test what moves an OCR edit on top: clipped, two images, form (#3973) Review asked for text under and above a scan (already pinned), an image clipped away entirely, two images, and a form holding the image. The clipped case fails when the hit test uses the image's box instead of its visible box; the negative cases fail when any later XObject counts. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SqmHYx8FiWM5PcsoiErgp9 --- crates/edit/src/tests.rs | 30 +++++++++++++++++++++++++++--- 1 file changed, 27 insertions(+), 3 deletions(-) diff --git a/crates/edit/src/tests.rs b/crates/edit/src/tests.rs index deb4d67a7c..2b78b07113 100644 --- a/crates/edit/src/tests.rs +++ b/crates/edit/src/tests.rs @@ -6214,16 +6214,17 @@ fn group_selection_refuses_incomplete_content_and_resource_budget_overruns() { } /// Invisible text moves on top only when something painted after it covers it: an XObject drawn -/// over its box, or any inline image or shading. +/// over its box (an image, or a form holding one), or any inline image or shading. An image +/// clipped away entirely, or drawn elsewhere, does not move it. #[test] fn revealed_text_moves_on_top_only_when_painted_over() { - let on_top = |content: &str| { - let doc = image_page_with(content); + let on_top_of = |doc: Document| { let block = text::text_blocks(&doc, 0).unwrap().remove(0); let ops = pdfcraft_content::parse(&page_content_bytes(&doc, 0)).ops; let at = ops.iter().position(|o| o.is("BT")).unwrap(); revealed_under_paint(&doc, 0, block.invisible, block.rect, &ops, at) }; + let on_top = |content: &str| on_top_of(image_page_with(content)); let scan = "q 200 0 0 100 100 100 cm /Im0 Do Q"; assert!(on_top(&format!("BT /F1 12 Tf 3 Tr 110 150 Td (Under) Tj ET {scan}")), "the scan covers it"); assert!(!on_top(&format!("BT /F1 12 Tf 3 Tr 400 600 Td (Away) Tj ET {scan}")), "an image elsewhere"); @@ -6232,6 +6233,29 @@ fn revealed_text_moves_on_top_only_when_painted_over() { assert!(!on_top("BT /F1 12 Tf 3 Tr 110 150 Td (Path) Tj ET 0 0 600 800 re f"), "a filled path does not move it"); assert!(on_top("BT /F1 12 Tf 3 Tr 400 600 Td (Shade) Tj ET /Sh sh"), "a shading after it"); assert!(on_top("BT /F1 12 Tf 3 Tr 400 600 Td (Inline) Tj ET BI /W 1 /H 1 /CS /G /BPC 8 ID x EI"), "an inline image after it"); + let under = "BT /F1 12 Tf 3 Tr 110 150 Td (Under) Tj ET"; + let elsewhere = "q 50 0 0 50 400 600 cm /Im0 Do Q"; + assert!(!on_top(&format!("{under} q 0 0 10 10 re W n {scan} Q")), "the scan is clipped away entirely"); + assert!(on_top(&format!("{under} {elsewhere} {scan}")), "two images, the second over it"); + assert!(on_top(&format!("{under} {scan} {elsewhere}")), "two images, the first over it"); + assert!(!on_top(&format!("{under} {elsewhere} {elsewhere}")), "two images, neither over it"); + let in_form = |content: &str| { + let page = "<< /Type /Page /Parent 2 0 R /MediaBox [0 0 600 800] /Contents 4 0 R /Resources << /Font << /F1 5 0 R >> /XObject << /Fm0 7 0 R >> >> >>"; + let image = stream("/Type /XObject /Subtype /Image /Width 1 /Height 1 /ColorSpace /DeviceGray /BitsPerComponent 8", "A"); + let form = stream("/Type /XObject /Subtype /Form /BBox [0 0 1 1] /Resources << /XObject << /Im0 6 0 R >> >>", "/Im0 Do"); + let objs = [ + "<< /Type /Catalog /Pages 2 0 R >>".to_string(), + "<< /Type /Pages /Kids [3 0 R] /Count 1 >>".into(), + page.into(), + stream("", content), + "<< /Type /Font /Subtype /Type1 /BaseFont /Helvetica >>".into(), + image, + form, + ]; + on_top_of(build(&objs.iter().map(String::as_str).collect::>())) + }; + assert!(in_form(&format!("{under} q 200 0 0 100 100 100 cm /Fm0 Do Q")), "a form holding the scan, over it"); + assert!(!in_form(&format!("{under} q 50 0 0 50 400 600 cm /Fm0 Do Q")), "a form holding an image elsewhere"); } /// Revealed OCR text is checked where the rewrite lands, not only where it was: an image drawn From fbd2813d7a95be479ab4efb0ee56f29d3136c1b9 Mon Sep 17 00:00:00 2001 From: Andrew Date: Mon, 2 Nov 2026 06:19:05 -0700 Subject: [PATCH 14/53] M5.7: the Fill & Sign text controls use a colour dot and a size flyout The quick bar is too narrow for the "Font size:" label, number box and "Colour" text, which wrapped mid-word. They now work like Draw's controls: a divider, a colour dot that opens the swatches, and an icon that opens a Font size flyout with a slider. Picking a colour also recolours the selected Fill & Sign text, as Draw's colour does for its own marks. --- crates/ui-egui/src/hotbar.rs | 90 +++++++++++-------- crates/ui-egui/src/tools/comments.rs | 2 +- crates/ui-egui/src/tools/comments/menu.rs | 29 +++--- crates/ui-egui/tests/it/fill_sign.rs | 103 +++++++++++++++++++++- 4 files changed, 173 insertions(+), 51 deletions(-) diff --git a/crates/ui-egui/src/hotbar.rs b/crates/ui-egui/src/hotbar.rs index c66b7e30b0..636414502c 100644 --- a/crates/ui-egui/src/hotbar.rs +++ b/crates/ui-egui/src/hotbar.rs @@ -287,6 +287,30 @@ fn core_tool_buttons(ui: &mut egui::Ui, app: &mut PdfCraftApp, cell: f32, popup_ }); } +fn controls_divider(ui: &mut egui::Ui, t: &Tokens, vertical: bool) { + let (r, _) = if vertical { + ui.allocate_exact_size(vec2(crate::canvas::bars::QUICK_BAR_WIDTH, 9.0), Sense::hover()) + } else { + ui.allocate_exact_size(vec2(9.0, crate::canvas::bars::QUICK_BAR_WIDTH), Sense::hover()) + }; + if vertical { + ui.painter().hline(r.x_range().shrink(6.0), r.center().y, Stroke::new(1.0, t.divider)); + } else { + ui.painter().vline(r.center().x, r.y_range().shrink(6.0), Stroke::new(1.0, t.divider)); + } +} + +fn selected_fill_text(app: &PdfCraftApp) -> Option<(usize, usize)> { + let (index, id) = app.active_ids()?; + let (page, annotation) = app.views.get(index)?.comments.selected?; + let doc = app.session.get(id).filter(|doc| doc.allows_annotation())?; + doc.info + .annotations + .iter() + .any(|a| a.page == page && a.index == annotation && a.fill_sign && a.subtype == "FreeText" && !a.locked) + .then_some((page, annotation)) +} + /// Extra controls rendered below/beside the buttons when a comment or fill tool is active. fn active_tool_controls(ui: &mut egui::Ui, app: &mut PdfCraftApp, t: &Tokens, vertical: bool) { let current_fill = match app.quick_tool { @@ -294,47 +318,37 @@ fn active_tool_controls(ui: &mut egui::Ui, app: &mut PdfCraftApp, t: &Tokens, ve _ => None, }; if matches!(current_fill, Some(crate::fill_sign::FillTool::Text | crate::fill_sign::FillTool::Date)) { - if vertical { - ui.label(tl!("Font size:")); - } - let size = ui.add( - egui::DragValue::new(&mut app.fill_text_size) - .range(crate::fill_sign::MIN_TEXT_SIZE..=crate::fill_sign::MAX_TEXT_SIZE) - .speed(0.25) - .suffix(" pt"), - ); - if ((size.changed() && !size.dragged()) || size.drag_stopped()) - && let Some((index, id)) = app.active_ids() - && let Some((page, annotation)) = app.views.get(index).and_then(|view| view.comments.selected) - && app.session.get(id).is_some_and(|doc| { - doc.allows_annotation() - && doc - .info - .annotations - .iter() - .any(|a| a.page == page && a.index == annotation && a.fill_sign && a.subtype == "FreeText" && !a.locked) - }) - { - app.apply_edit(Edit::FontSizeAnnotation { page, index: annotation, size: app.fill_text_size }); - } - ui.menu_button(tl!("Colour"), |ui| { - if let Some(color) = comments::swatch_grid(ui, Some(app.fill_text_color)) { - app.fill_text_color = color; - ui.close(); + controls_divider(ui, t, vertical); + if let Some(color) = comments::quick_bar_colour(ui, app.fill_text_color) { + app.fill_text_color = color; + if let Some((page, index)) = selected_fill_text(app) { + app.apply_edit(Edit::StyleAnnotation { page, index, color: Some(color), opacity: None, width: None, endings: None }); } - }); + } + let resp = icons::button(ui, "sliders-horizontal", 32.0, false, tl!("Font size")); + let resized = egui::Popup::menu(&resp) + .align(egui::RectAlign::RIGHT_START) + .close_behavior(egui::PopupCloseBehavior::CloseOnClickOutside) + .show(|ui| { + ui.set_min_width(180.0); + ui.label(egui::RichText::new(tl!("Font size")).font(theme::semibold(12.0))); + let size = ui.add( + egui::Slider::new(&mut app.fill_text_size, crate::fill_sign::MIN_TEXT_SIZE..=crate::fill_sign::MAX_TEXT_SIZE) + .logarithmic(true) + .suffix(" pt"), + ); + // #1274: the size also applies to the Fill & Sign text that is selected, + // once per change (a typed value, or the end of a drag), so a drag adds + // one undo step and not one per frame. + (size.changed() && !size.dragged()) || size.drag_stopped() + }) + .is_some_and(|popup| popup.inner); + if resized && let Some((page, index)) = selected_fill_text(app) { + app.apply_edit(Edit::FontSizeAnnotation { page, index, size: app.fill_text_size }); + } } if let QuickTool::Comment(tool) = app.quick_tool { - let (r, _) = if vertical { - ui.allocate_exact_size(vec2(crate::canvas::bars::QUICK_BAR_WIDTH, 9.0), Sense::hover()) - } else { - ui.allocate_exact_size(vec2(9.0, crate::canvas::bars::QUICK_BAR_WIDTH), Sense::hover()) - }; - if vertical { - ui.painter().hline(r.x_range().shrink(6.0), r.center().y, Stroke::new(1.0, t.divider)); - } else { - ui.painter().vline(r.center().x, r.y_range().shrink(6.0), Stroke::new(1.0, t.divider)); - } + controls_divider(ui, t, vertical); if let Some(color) = comments::quick_bar_controls(ui, tool, &mut app.comment_prefs) && let Some((index, id)) = app.active_ids() && let Some((page, annotation)) = app.views.get(index).and_then(|view| view.comments.selected) diff --git a/crates/ui-egui/src/tools/comments.rs b/crates/ui-egui/src/tools/comments.rs index cb8bafe642..95053a16fe 100644 --- a/crates/ui-egui/src/tools/comments.rs +++ b/crates/ui-egui/src/tools/comments.rs @@ -41,7 +41,7 @@ pub(crate) use input::{describe, fill_grabs, hit, keys, nudge_keys, page_after_t pub use menu::{CanvasAction, fill_picker, recent_custom_colours, remember_custom_colour, status_badge, swatch_grid}; #[cfg(test)] use menu::{RECENT_CUSTOM, same_colour}; -pub(crate) use menu::{attachment_items, context_menu, quick_bar_controls}; +pub(crate) use menu::{attachment_items, context_menu, quick_bar_colour, quick_bar_controls}; pub(crate) use paint::paint_page; #[cfg(test)] use paint::{changing_points, drawn_shape, edited_in_view}; diff --git a/crates/ui-egui/src/tools/comments/menu.rs b/crates/ui-egui/src/tools/comments/menu.rs index b577fbbd29..0e0da5caf3 100644 --- a/crates/ui-egui/src/tools/comments/menu.rs +++ b/crates/ui-egui/src/tools/comments/menu.rs @@ -286,25 +286,32 @@ pub fn fill_picker(ui: &mut egui::Ui, current: Option) -> Option Option { - let style = prefs.style(tool); - let mut selected_color = None; - if icons::button(ui, "pin", 32.0, prefs.pinned, if prefs.pinned { tl!("Keep tool selected: on") } else { tl!("Keep tool selected") }).clicked() { - prefs.pinned = !prefs.pinned; - } +pub(crate) fn quick_bar_colour(ui: &mut egui::Ui, current: Rgb) -> Option { let (r, resp) = ui.allocate_exact_size(vec2(32.0, 32.0), Sense::click()); - ui.painter().circle_filled(r.center(), 9.0, color32(style.color)); + ui.painter().circle_filled(r.center(), 9.0, color32(current)); ui.painter().circle_stroke(r.center(), 9.0, Stroke::new(1.0, Color32::from_black_alpha(50))); resp.widget_info(|| egui::WidgetInfo::labeled(egui::Role::Button, true, tl!("Colour"))); let resp = resp.on_hover_text(tl!("Colour")); + let mut picked = None; egui::Popup::menu(&resp).align(egui::RectAlign::RIGHT_START).close_behavior(egui::PopupCloseBehavior::CloseOnClickOutside).show(|ui| { - if let Some(c) = swatch_grid(ui, Some(style.color)) { - prefs.set_color(tool, c); - selected_color = Some(c); + if let Some(c) = swatch_grid(ui, Some(current)) { + picked = Some(c); ui.close(); } }); + picked +} + +/// The comment tools' extra quick-bar controls: pin, colour, fill, opacity and thickness. +pub(crate) fn quick_bar_controls(ui: &mut egui::Ui, tool: CommentTool, prefs: &mut CommentPrefs) -> Option { + let style = prefs.style(tool); + if icons::button(ui, "pin", 32.0, prefs.pinned, if prefs.pinned { tl!("Keep tool selected: on") } else { tl!("Keep tool selected") }).clicked() { + prefs.pinned = !prefs.pinned; + } + let selected_color = quick_bar_colour(ui, style.color); + if let Some(c) = selected_color { + prefs.set_color(tool, c); + } if tool.has_fill() { let resp = fill_button(ui, style.fill); egui::Popup::menu(&resp).align(egui::RectAlign::RIGHT_START).close_behavior(egui::PopupCloseBehavior::CloseOnClickOutside).show(|ui| { diff --git a/crates/ui-egui/tests/it/fill_sign.rs b/crates/ui-egui/tests/it/fill_sign.rs index ce21bd9f74..979ba0ff97 100644 --- a/crates/ui-egui/tests/it/fill_sign.rs +++ b/crates/ui-egui/tests/it/fill_sign.rs @@ -164,7 +164,7 @@ fn fill_text_size_controls_new_text_and_survives_settings_restore() { let mut h = harness(); h.state_mut().execute("sign.fill.text"); h.run_steps(2); - h.get_by_label("Font size:"); + h.get_by_label("Font size"); h.state_mut().fill_text_size = 24.0; click(&mut h, 50.0, 250.0); assert_eq!(h.state().views[0].fill_text.as_ref().map(|t| t.size), Some(24.0)); @@ -201,6 +201,8 @@ fn fill_text_size_resizes_the_selected_text() { }; let before = height(&h); let set_size = |h: &mut Harness<'static, PdfCraftApp>, value: &str| { + h.get_by_role_and_label(egui::accesskit::Role::Button, "Font size").click(); + h.run_steps(2); h.query_all_by_role(egui::accesskit::Role::SpinButton) .find(|n| n.value().is_some_and(|v| v.ends_with(" pt"))) .expect("the Font size control") @@ -211,6 +213,8 @@ fn fill_text_size_resizes_the_selected_text() { h.run_steps(1); h.key_press(egui::Key::Enter); h.run_steps(4); + h.get_by_role_and_label(egui::accesskit::Role::Button, "Font size").click(); + h.run_steps(2); }; // Nothing selected: only the size for new text changes. h.state_mut().views[0].comments.selected = None; @@ -1812,3 +1816,100 @@ fn mode_tabs_keep_the_hand_over_a_type_box_scrolled_under_them() { h.run_steps(3); assert_eq!(h.state().mode, pdfcraft_ui_egui::Mode::Convert, "the tab takes the click, not the text box"); } + +fn place_selected_text(h: &mut Harness<'static, PdfCraftApp>, text: &str) { + h.state_mut().execute("sign.fill.text"); + click(h, 50.0, 250.0); + h.event(egui::Event::Text(text.into())); + h.run_steps(1); + h.key_press(egui::Key::Enter); + h.run_steps(3); + let selected = { + let s = h.state(); + s.session + .get(s.views[0].id) + .unwrap() + .info + .annotations + .iter() + .find(|a| a.contents.as_deref() == Some(text)) + .map(|a| (a.page, a.index)) + .unwrap() + }; + h.state_mut().views[0].comments.selected = Some(selected); + h.run_steps(2); +} + +fn placed_height(h: &Harness<'static, PdfCraftApp>, text: &str) -> f32 { + let s = h.state(); + let a = s.session.get(s.views[0].id).unwrap().info.annotations.iter().find(|a| a.contents.as_deref() == Some(text)).unwrap(); + a.rect[3] - a.rect[1] +} + +fn placed_color(h: &Harness<'static, PdfCraftApp>, text: &str) -> Option<[f32; 3]> { + let s = h.state(); + s.session.get(s.views[0].id).unwrap().info.annotations.iter().find(|a| a.contents.as_deref() == Some(text)).unwrap().color +} + +#[test] +fn the_fill_text_size_flyout_opens_from_an_icon_and_leaves_a_stored_size_alone() { + let mut h = harness(); + h.state_mut().fill_text_size = 9.1; + place_selected_text(&mut h, "Keep me"); + let height = placed_height(&h, "Keep me"); + assert_eq!(h.query_all_by_role(egui::accesskit::Role::SpinButton).filter(|n| n.value().is_some_and(|v| v.ends_with(" pt"))).count(), 0); + h.get_by_label("Font size").click(); + h.run_steps(4); + assert!( + h.query_all_by_role(egui::accesskit::Role::SpinButton).any(|n| n.value().is_some_and(|v| v.ends_with(" pt"))), + "the flyout holds the size" + ); + assert_eq!(h.state().fill_text_size, 9.1, "opening the flyout does not round the size"); + assert_eq!(placed_height(&h, "Keep me"), height, "opening the flyout does not resize the selected text"); +} + +#[test] +fn dragging_the_fill_text_size_slider_is_one_undo_step() { + let mut h = harness(); + place_selected_text(&mut h, "Drag me"); + let height = placed_height(&h, "Drag me"); + h.get_by_label("Font size").click(); + h.run_steps(2); + let slider = h.get_by_role(egui::accesskit::Role::Slider).rect(); + let y = (slider.min.y + slider.max.y) / 2.0; + let from = pos2(slider.min.x + 10.0, y); + let to = pos2(slider.max.x - 10.0, y); + h.hover_at(from); + h.run_steps(1); + h.drag_at(from); + h.run_steps(1); + for k in 1..=4 { + h.hover_at(from + (to - from) * (k as f32 / 4.0)); + h.run_steps(1); + } + assert_eq!(placed_height(&h, "Drag me"), height, "nothing is applied before the release"); + h.drop_at(to); + h.run_steps(3); + assert!(placed_height(&h, "Drag me") > height, "the release resizes the selected text"); + h.state_mut().execute("edit.undo"); + h.run_steps(3); + assert_eq!(placed_height(&h, "Drag me"), height, "one undo takes the whole drag back"); +} + +#[test] +fn the_fill_colour_dot_recolours_the_selected_text() { + let mut h = harness(); + place_selected_text(&mut h, "Paint me"); + let before = placed_color(&h, "Paint me"); + h.get_by_label("Colour").click(); + h.run_steps(2); + h.get_by_label("Blue").click(); + h.run_steps(3); + let color = h.state().fill_text_color.map(|v| v as f32); + assert!(color[2] > color[0], "blue was selected: {color:?}"); + assert_ne!(before, Some(color)); + assert_eq!(placed_color(&h, "Paint me"), Some(color)); + h.state_mut().execute("edit.undo"); + h.run_steps(3); + assert_eq!(placed_color(&h, "Paint me"), before, "one undo takes the colour back"); +} From b3047998125db72655e9b9441a7cfc5f29c4dba6 Mon Sep 17 00:00:00 2001 From: Andrew Date: Mon, 2 Nov 2026 06:37:33 -0700 Subject: [PATCH 15/53] M5.7: the Fill & Sign Font size button uses Lucide's a-large-small icon The sliders icon is also Draw's Line thickness and doesn't say "text". Adds one icon to the Lucide set, from lucide-static 1.49.0, with its attribution. --- ATTRIBUTION.md | 3 ++- ATTRIBUTION.toml | 11 +++++++++++ assets/icons/a-large-small.svg | 18 ++++++++++++++++++ crates/ui-egui/src/hotbar.rs | 2 +- crates/ui-egui/src/widgets/icon_data.rs | 1 + crates/ui-egui/src/widgets/icons.rs | 2 +- 6 files changed, 34 insertions(+), 3 deletions(-) create mode 100644 assets/icons/a-large-small.svg diff --git a/ATTRIBUTION.md b/ATTRIBUTION.md index fc88e8b664..6327de11ca 100644 --- a/ATTRIBUTION.md +++ b/ATTRIBUTION.md @@ -4,7 +4,7 @@ Every asset PdfCraft includes, bundles or uses to build its published material, with its author, source and licence. The policy is in [AGENTS.md](AGENTS.md) §1. The machine-readable list, with the SHA-256 of every asset but the translation catalogs, is [ATTRIBUTION.toml](ATTRIBUTION.toml). Licence texts are kept beside the assets and summarised in [NOTICE](NOTICE). -## In this repository (289) +## In this repository (290) | Asset | Title | Author | Licence | Source | Used for | |---|---|---|---|---|---| @@ -293,6 +293,7 @@ Every asset PdfCraft includes, bundles or uses to build its published material, | `assets/icons/superscript.svg` | Lucide icon "superscript" | Lucide Contributors (portions Cole Bemis, Feather) | ISC AND MIT | https://github.com/lucide-icons/lucide (npm lucide-static 1.49.0, icons/superscript.svg) | UI icon, embedded via crates/ui-egui/src/icon_data.rs | | `assets/icons/subscript.svg` | Lucide icon "subscript" | Lucide Contributors (portions Cole Bemis, Feather) | ISC AND MIT | https://github.com/lucide-icons/lucide (npm lucide-static 1.49.0, icons/subscript.svg) | UI icon, embedded via crates/ui-egui/src/icon_data.rs | | `assets/icons/pilcrow.svg` | Lucide icon "pilcrow" | Lucide Contributors (portions Cole Bemis, Feather) | ISC AND MIT | https://github.com/lucide-icons/lucide (npm lucide-static 1.49.0, icons/pilcrow.svg) | UI icon, embedded via crates/ui-egui/src/icon_data.rs | +| `assets/icons/a-large-small.svg` | Lucide icon "a-large-small" | Lucide Contributors (portions Cole Bemis, Feather) | ISC AND MIT | https://github.com/lucide-icons/lucide (npm lucide-static 1.49.0, icons/a-large-small.svg) | UI icon, embedded via crates/ui-egui/src/icon_data.rs | | `assets/icons/move-horizontal.svg` | Lucide icon "move-horizontal" | Lucide Contributors (portions Cole Bemis, Feather) | ISC AND MIT | https://github.com/lucide-icons/lucide (npm lucide-static 1.49.0, icons/move-horizontal.svg) | UI icon, embedded via crates/ui-egui/src/icon_data.rs | | `crates/ui-egui/src/i18n/sq.tsv` | Albanian interface translation catalog | PdfCraft contributors | MIT OR Apache-2.0 | Contributor-original: clean-room translations of PdfCraft's English UI labels, no proprietary localisation resources; 2026-10-22 page-gap labels are contributor-original AI-assisted translations from English, not separately native-reviewed; 2026-10-22 signature-appearance labels are contributor-original AI-assisted translations from English, not separately native-reviewed; 2026-10-22 text-field-table labels are contributor-original AI-assisted translations from English, not separately native-reviewed; 2026-10-22 secured-document notice labels are contributor-original AI-assisted translations from English, not separately native-reviewed; the 2026-10-22 reply-count label is an AI-assisted translation from English, not separately native-reviewed | Albanian UI strings, compiled in by crates/ui-egui/src/i18n.rs | | `crates/ui-egui/src/i18n/th.tsv` | Thai interface translation catalog | PdfCraft contributors | MIT OR Apache-2.0 | Contributor-original: clean-room translations of PdfCraft's English UI labels, no proprietary localisation resources; 2026-10-22 page-gap labels are contributor-original AI-assisted translations from English, not separately native-reviewed; 2026-10-22 secured-document notice labels are contributor-original AI-assisted translations from English, not separately native-reviewed; the 2026-10-22 reply-count label is an AI-assisted translation from English, not separately native-reviewed | Thai UI strings, compiled in by crates/ui-egui/src/i18n.rs | diff --git a/ATTRIBUTION.toml b/ATTRIBUTION.toml index 76f4cf40a8..7c27c8c85d 100644 --- a/ATTRIBUTION.toml +++ b/ATTRIBUTION.toml @@ -3489,6 +3489,17 @@ kind = "icon" usage = "UI icon, embedded via crates/ui-egui/src/icon_data.rs" sha256 = "d344d9f9222f6c1b698c0c82a753a2212d9bd7785f06c00cd6e4910fe2fee8f8" +[[asset]] +path = "assets/icons/a-large-small.svg" +title = "Lucide icon \"a-large-small\"" +author = "Lucide Contributors (portions Cole Bemis, Feather)" +source = "https://github.com/lucide-icons/lucide (npm lucide-static 1.49.0, icons/a-large-small.svg)" +licence = "ISC AND MIT" +licence_file = "assets/icons/LICENSE-lucide.txt" +kind = "icon" +usage = "UI icon, embedded via crates/ui-egui/src/icon_data.rs" +sha256 = "d4985fc71174d108e322c3b25cb6abe2837a115fe3f239a1918645db436743ec" + [[asset]] path = "assets/icons/move-horizontal.svg" title = "Lucide icon \"move-horizontal\"" diff --git a/assets/icons/a-large-small.svg b/assets/icons/a-large-small.svg new file mode 100644 index 0000000000..7716db3cf2 --- /dev/null +++ b/assets/icons/a-large-small.svg @@ -0,0 +1,18 @@ + + + + + + + diff --git a/crates/ui-egui/src/hotbar.rs b/crates/ui-egui/src/hotbar.rs index 636414502c..a6bf36ec9f 100644 --- a/crates/ui-egui/src/hotbar.rs +++ b/crates/ui-egui/src/hotbar.rs @@ -325,7 +325,7 @@ fn active_tool_controls(ui: &mut egui::Ui, app: &mut PdfCraftApp, t: &Tokens, ve app.apply_edit(Edit::StyleAnnotation { page, index, color: Some(color), opacity: None, width: None, endings: None }); } } - let resp = icons::button(ui, "sliders-horizontal", 32.0, false, tl!("Font size")); + let resp = icons::button(ui, "a-large-small", 32.0, false, tl!("Font size")); let resized = egui::Popup::menu(&resp) .align(egui::RectAlign::RIGHT_START) .close_behavior(egui::PopupCloseBehavior::CloseOnClickOutside) diff --git a/crates/ui-egui/src/widgets/icon_data.rs b/crates/ui-egui/src/widgets/icon_data.rs index 72829fe4b8..d58ddcbab6 100644 --- a/crates/ui-egui/src/widgets/icon_data.rs +++ b/crates/ui-egui/src/widgets/icon_data.rs @@ -187,6 +187,7 @@ pub static ICONS: &[(&str, &[u8])] = &[ ("wrench", include_bytes!("../../../../assets/icons/wrench.svg")), ("move-horizontal", include_bytes!("../../../../assets/icons/move-horizontal.svg")), ("pilcrow", include_bytes!("../../../../assets/icons/pilcrow.svg")), + ("a-large-small", include_bytes!("../../../../assets/icons/a-large-small.svg")), ("subscript", include_bytes!("../../../../assets/icons/subscript.svg")), ("superscript", include_bytes!("../../../../assets/icons/superscript.svg")), ("x", include_bytes!("../../../../assets/icons/x.svg")), diff --git a/crates/ui-egui/src/widgets/icons.rs b/crates/ui-egui/src/widgets/icons.rs index 5ad5425e8d..13f50a0d6c 100644 --- a/crates/ui-egui/src/widgets/icons.rs +++ b/crates/ui-egui/src/widgets/icons.rs @@ -99,7 +99,7 @@ mod tests { #[test] fn shared_catalog_preserves_existing_icons_and_fallback_uri() { - assert_eq!(ICONS.len(), 190); + assert_eq!(ICONS.len(), 191); assert!(exists("minimize-2")); for (name, original) in ICONS { let expected = From 84aa947c0f97feb51dbef78b66a402d6aab071fa Mon Sep 17 00:00:00 2001 From: Andrew Date: Mon, 2 Nov 2026 07:08:41 -0700 Subject: [PATCH 16/53] M5.7: the quick bar's comments no longer say Fill & Sign widens it The Text and Date controls now fit the 32 pt column, so the bar stays at its narrowest with every tool. The centring still measures the rendered width. --- crates/ui-egui/src/canvas/bars.rs | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/crates/ui-egui/src/canvas/bars.rs b/crates/ui-egui/src/canvas/bars.rs index dd3fe8c498..fbd4200217 100644 --- a/crates/ui-egui/src/canvas/bars.rs +++ b/crates/ui-egui/src/canvas/bars.rs @@ -315,7 +315,7 @@ mod tests { #[test] fn a_wider_quick_bar_centres_by_its_own_width_and_needs_a_wider_margin() { - // Fill & Sign's Text tool widens the bar to 70 pt: it centres by that, and 72 pt no longer holds it. + // A 70 pt bar centres by that, and 72 pt no longer holds it. assert_eq!(quick_bar_offset(120.0, 70.0), 15.0); assert_eq!(quick_bar_offset(72.0, 70.0), 14.0); } From 546a9d90a5572b2216143ac6a0328dcaedd463ff Mon Sep 17 00:00:00 2001 From: Andrew Date: Mon, 2 Nov 2026 07:28:12 -0700 Subject: [PATCH 17/53] M5.7: the a-large-small attribution names where the icon is embedded icon_data.rs moved to crates/ui-egui/src/widgets/. ATTRIBUTION.md is regenerated with cargo xtask assets --write. --- ATTRIBUTION.md | 2 +- ATTRIBUTION.toml | 2 +- 2 files changed, 2 insertions(+), 2 deletions(-) diff --git a/ATTRIBUTION.md b/ATTRIBUTION.md index 6327de11ca..a3855b5c70 100644 --- a/ATTRIBUTION.md +++ b/ATTRIBUTION.md @@ -293,7 +293,7 @@ Every asset PdfCraft includes, bundles or uses to build its published material, | `assets/icons/superscript.svg` | Lucide icon "superscript" | Lucide Contributors (portions Cole Bemis, Feather) | ISC AND MIT | https://github.com/lucide-icons/lucide (npm lucide-static 1.49.0, icons/superscript.svg) | UI icon, embedded via crates/ui-egui/src/icon_data.rs | | `assets/icons/subscript.svg` | Lucide icon "subscript" | Lucide Contributors (portions Cole Bemis, Feather) | ISC AND MIT | https://github.com/lucide-icons/lucide (npm lucide-static 1.49.0, icons/subscript.svg) | UI icon, embedded via crates/ui-egui/src/icon_data.rs | | `assets/icons/pilcrow.svg` | Lucide icon "pilcrow" | Lucide Contributors (portions Cole Bemis, Feather) | ISC AND MIT | https://github.com/lucide-icons/lucide (npm lucide-static 1.49.0, icons/pilcrow.svg) | UI icon, embedded via crates/ui-egui/src/icon_data.rs | -| `assets/icons/a-large-small.svg` | Lucide icon "a-large-small" | Lucide Contributors (portions Cole Bemis, Feather) | ISC AND MIT | https://github.com/lucide-icons/lucide (npm lucide-static 1.49.0, icons/a-large-small.svg) | UI icon, embedded via crates/ui-egui/src/icon_data.rs | +| `assets/icons/a-large-small.svg` | Lucide icon "a-large-small" | Lucide Contributors (portions Cole Bemis, Feather) | ISC AND MIT | https://github.com/lucide-icons/lucide (npm lucide-static 1.49.0, icons/a-large-small.svg) | UI icon, embedded via crates/ui-egui/src/widgets/icon_data.rs | | `assets/icons/move-horizontal.svg` | Lucide icon "move-horizontal" | Lucide Contributors (portions Cole Bemis, Feather) | ISC AND MIT | https://github.com/lucide-icons/lucide (npm lucide-static 1.49.0, icons/move-horizontal.svg) | UI icon, embedded via crates/ui-egui/src/icon_data.rs | | `crates/ui-egui/src/i18n/sq.tsv` | Albanian interface translation catalog | PdfCraft contributors | MIT OR Apache-2.0 | Contributor-original: clean-room translations of PdfCraft's English UI labels, no proprietary localisation resources; 2026-10-22 page-gap labels are contributor-original AI-assisted translations from English, not separately native-reviewed; 2026-10-22 signature-appearance labels are contributor-original AI-assisted translations from English, not separately native-reviewed; 2026-10-22 text-field-table labels are contributor-original AI-assisted translations from English, not separately native-reviewed; 2026-10-22 secured-document notice labels are contributor-original AI-assisted translations from English, not separately native-reviewed; the 2026-10-22 reply-count label is an AI-assisted translation from English, not separately native-reviewed | Albanian UI strings, compiled in by crates/ui-egui/src/i18n.rs | | `crates/ui-egui/src/i18n/th.tsv` | Thai interface translation catalog | PdfCraft contributors | MIT OR Apache-2.0 | Contributor-original: clean-room translations of PdfCraft's English UI labels, no proprietary localisation resources; 2026-10-22 page-gap labels are contributor-original AI-assisted translations from English, not separately native-reviewed; 2026-10-22 secured-document notice labels are contributor-original AI-assisted translations from English, not separately native-reviewed; the 2026-10-22 reply-count label is an AI-assisted translation from English, not separately native-reviewed | Thai UI strings, compiled in by crates/ui-egui/src/i18n.rs | diff --git a/ATTRIBUTION.toml b/ATTRIBUTION.toml index 7c27c8c85d..f287a4aede 100644 --- a/ATTRIBUTION.toml +++ b/ATTRIBUTION.toml @@ -3497,7 +3497,7 @@ source = "https://github.com/lucide-icons/lucide (npm lucide-static 1.49.0, icon licence = "ISC AND MIT" licence_file = "assets/icons/LICENSE-lucide.txt" kind = "icon" -usage = "UI icon, embedded via crates/ui-egui/src/icon_data.rs" +usage = "UI icon, embedded via crates/ui-egui/src/widgets/icon_data.rs" sha256 = "d4985fc71174d108e322c3b25cb6abe2837a115fe3f239a1918645db436743ec" [[asset]] From 0a8aa374eb1aba1bc9f87b7ec7c554e52437cbc3 Mon Sep 17 00:00:00 2001 From: Andrew Date: Mon, 2 Nov 2026 10:15:33 -0700 Subject: [PATCH 18/53] M5.7: pressing the Fill & Sign size slider without moving it adds no undo step The slider senses drags only, so a press and release counts as a finished drag even when the value did not move. The edit is now skipped when the selected text already has that size, compared to the three decimals the size is saved with. --- crates/ui-egui/src/hotbar.rs | 11 +++++- crates/ui-egui/tests/it/fill_sign.rs | 56 ++++++++++++++++++++++++++++ 2 files changed, 65 insertions(+), 2 deletions(-) diff --git a/crates/ui-egui/src/hotbar.rs b/crates/ui-egui/src/hotbar.rs index a6bf36ec9f..328f85e706 100644 --- a/crates/ui-egui/src/hotbar.rs +++ b/crates/ui-egui/src/hotbar.rs @@ -343,8 +343,15 @@ fn active_tool_controls(ui: &mut egui::Ui, app: &mut PdfCraftApp, t: &Tokens, ve (size.changed() && !size.dragged()) || size.drag_stopped() }) .is_some_and(|popup| popup.inner); - if resized && let Some((page, index)) = selected_fill_text(app) { - app.apply_edit(Edit::FontSizeAnnotation { page, index, size: app.fill_text_size }); + let size = app.fill_text_size; + if resized + && let Some((page, index)) = selected_fill_text(app) + && app + .active_ids() + .and_then(|(_, id)| app.session.get(id)?.comment_props(page, index)?.font_size) + .is_none_or(|stored| format!("{stored:.3}") != format!("{size:.3}")) + { + app.apply_edit(Edit::FontSizeAnnotation { page, index, size }); } } if let QuickTool::Comment(tool) = app.quick_tool { diff --git a/crates/ui-egui/tests/it/fill_sign.rs b/crates/ui-egui/tests/it/fill_sign.rs index 979ba0ff97..931018214a 100644 --- a/crates/ui-egui/tests/it/fill_sign.rs +++ b/crates/ui-egui/tests/it/fill_sign.rs @@ -1896,6 +1896,62 @@ fn dragging_the_fill_text_size_slider_is_one_undo_step() { assert_eq!(placed_height(&h, "Drag me"), height, "one undo takes the whole drag back"); } +#[test] +fn pressing_the_fill_text_size_slider_without_moving_it_adds_no_undo_step() { + let mut h = harness(); + place_selected_text(&mut h, "Hold me"); + let height = placed_height(&h, "Hold me"); + h.get_by_label("Font size").click(); + h.run_steps(2); + let slider = h.get_by_role(egui::accesskit::Role::Slider).rect(); + let at = pos2(slider.min.x + (slider.max.x - slider.min.x) * 0.75, (slider.min.y + slider.max.y) / 2.0); + let press = |h: &mut Harness<'static, PdfCraftApp>| { + h.hover_at(at); + h.run_steps(1); + h.drag_at(at); + h.run_steps(1); + h.drop_at(at); + h.run_steps(3); + }; + press(&mut h); + let resized = placed_height(&h, "Hold me"); + assert!(resized > height, "the first press moves the size: {height} -> {resized}"); + press(&mut h); + assert_eq!(placed_height(&h, "Hold me"), resized); + h.state_mut().execute("edit.undo"); + h.run_steps(3); + assert_eq!(placed_height(&h, "Hold me"), height, "one undo takes back the only change; the second press added no step"); +} + +#[test] +fn typing_a_fill_text_size_that_rounds_to_the_stored_one_adds_no_undo_step() { + let mut h = harness(); + place_selected_text(&mut h, "Round me"); + let height = placed_height(&h, "Round me"); + h.get_by_label("Font size").click(); + h.run_steps(2); + let type_size = |h: &mut Harness<'static, PdfCraftApp>, value: &str| { + h.query_all_by_role(egui::accesskit::Role::SpinButton) + .find(|n| n.value().is_some_and(|v| v.ends_with(" pt"))) + .expect("the Font size control") + .click(); + h.run_steps(2); + h.key_press_modifiers(egui::Modifiers::COMMAND, egui::Key::A); + h.event(egui::Event::Text(value.into())); + h.run_steps(1); + h.key_press(egui::Key::Enter); + h.run_steps(4); + }; + type_size(&mut h, "36.1234"); + let resized = placed_height(&h, "Round me"); + assert!(resized > height, "the first size applies: {height} -> {resized}"); + type_size(&mut h, "36.1231"); + assert_eq!(h.state().fill_text_size, 36.1231); + h.state_mut().execute("edit.undo"); + h.run_steps(3); + assert_eq!(placed_height(&h, "Round me"), height, "one undo takes back the only change; the rounded retype added no step"); +} + #[test] fn the_fill_colour_dot_recolours_the_selected_text() { let mut h = harness(); From e90c92124c1695dc87780f763f70a2556e63b81e Mon Sep 17 00:00:00 2001 From: Andrew Date: Fri, 30 Oct 2026 07:31:31 -0600 Subject: [PATCH 19/53] M5.7: the quick bar's Fill & Sign button shows the tool it picks The button showed a generic pen until clicked, then swapped to Add text's icon. It now works like the Comment, Highlight and Draw groups: it shows the last Fill & Sign tool used (Signature at first) and clicking it picks that tool. The flyout marks the active saved signature or initials card, and the cards get the pointing-hand cursor like the other rows. --- crates/ui-egui/src/app.rs | 1 + crates/ui-egui/src/app/control.rs | 9 ++++-- crates/ui-egui/src/commands.rs | 1 + crates/ui-egui/src/dialogs.rs | 1 + crates/ui-egui/src/hotbar.rs | 16 +++++++++-- crates/ui-egui/src/lib.rs | 1 + crates/ui-egui/src/tools/fill_sign.rs | 13 ++++++--- crates/ui-egui/tests/it/comments.rs | 2 +- crates/ui-egui/tests/it/fill_sign.rs | 40 +++++++++++++++++++++++---- 9 files changed, 69 insertions(+), 15 deletions(-) diff --git a/crates/ui-egui/src/app.rs b/crates/ui-egui/src/app.rs index 6ac8ea8783..bb1748f705 100644 --- a/crates/ui-egui/src/app.rs +++ b/crates/ui-egui/src/app.rs @@ -108,6 +108,7 @@ impl PdfCraftApp { ui_scale_applied: None, flatten_fill_sign_on_save: false, fill_text_size: fill_sign::TEXT_SIZE, + fill_tool: fill_sign::FillTool::Signature, fill_text_color: [0.0; 3], dialog: None, updates: updates::Updates::default(), diff --git a/crates/ui-egui/src/app/control.rs b/crates/ui-egui/src/app/control.rs index 0309042bf1..fee5378b6e 100644 --- a/crates/ui-egui/src/app/control.rs +++ b/crates/ui-egui/src/app/control.rs @@ -333,9 +333,12 @@ impl PdfCraftApp { field if field.starts_with("field-") => QuickTool::Field( prepare::FieldTool::from_command(&format!("form.add.{}", &field[6..])).ok_or_else(|| format!("unknown tool {field}"))?, ), - fill if fill.starts_with("fill-") => QuickTool::Fill( - fill_sign::FillTool::from_command(&format!("sign.fill.{}", &fill[5..])).ok_or_else(|| format!("unknown tool {fill}"))?, - ), + fill if fill.starts_with("fill-") => { + let f = + fill_sign::FillTool::from_command(&format!("sign.fill.{}", &fill[5..])).ok_or_else(|| format!("unknown tool {fill}"))?; + self.fill_tool = f; + QuickTool::Fill(f) + } other => { let t = comments::CommentTool::from_command(&format!("comment.{other}")).ok_or_else(|| format!("unknown tool {other}"))?; self.comment_prefs.group_tool[t.group()] = t; diff --git a/crates/ui-egui/src/commands.rs b/crates/ui-egui/src/commands.rs index f072b7febe..4fdec5af8a 100644 --- a/crates/ui-egui/src/commands.rs +++ b/crates/ui-egui/src/commands.rs @@ -697,6 +697,7 @@ impl PdfCraftApp { } fill if crate::fill_sign::FillTool::from_command(fill).is_some() => { let Some(tool) = crate::fill_sign::FillTool::from_command(fill) else { return Some(false) }; + self.fill_tool = tool; self.quick_tool = crate::QuickTool::Fill(tool); let initials = tool == crate::fill_sign::FillTool::Initials; if (tool == crate::fill_sign::FillTool::Signature && self.signature.is_none()) || (initials && self.initials.is_none()) { diff --git a/crates/ui-egui/src/dialogs.rs b/crates/ui-egui/src/dialogs.rs index 048d519222..8105a22342 100644 --- a/crates/ui-egui/src/dialogs.rs +++ b/crates/ui-egui/src/dialogs.rs @@ -1349,6 +1349,7 @@ fn dialog_signature(ui: &mut egui::Ui, app: &mut PdfCraftApp, req: &mut Pending) app.signature = d.saved(); crate::fill_sign::FillTool::Signature }; + app.fill_tool = tool; app.quick_tool = crate::QuickTool::Fill(tool); app.toast = None; } diff --git a/crates/ui-egui/src/hotbar.rs b/crates/ui-egui/src/hotbar.rs index c66b7e30b0..bbd98be2ef 100644 --- a/crates/ui-egui/src/hotbar.rs +++ b/crates/ui-egui/src/hotbar.rs @@ -239,8 +239,20 @@ fn core_tool_buttons(ui: &mut egui::Ui, app: &mut PdfCraftApp, cell: f32, popup_ QuickTool::Fill(f) => Some(f), _ => None, }; - let shown = current_fill.unwrap_or(crate::fill_sign::FillTool::Text); - let resp = icons::button(ui, if current_fill.is_some() { shown.icon() } else { "pen-line" }, cell, current_fill.is_some(), tl!("Fill & Sign")); + let shown = app.fill_tool; + let tip = match shown { + crate::fill_sign::FillTool::Signature | crate::fill_sign::FillTool::Initials => { + let initials = shown == crate::fill_sign::FillTool::Initials; + let saved = if initials { app.initials.is_some() } else { app.signature.is_some() }; + crate::i18n::fmt( + tl!(if saved { "Place saved {what}" } else { "Add {what}" }), + &[("what", tl!(if initials { "initials" } else { "signature" }))], + ) + } + _ => tl!(shown.label()).to_string(), + }; + let tip = format!("{}: {tip}", tl!("Fill & Sign")); + let resp = icons::button(ui, shown.icon(), cell, current_fill.is_some(), &tip); let r = resp.rect; let tri = [r.right_bottom() + vec2(-4.0, -4.0), r.right_bottom() + vec2(-9.0, -4.0), r.right_bottom() + vec2(-4.0, -9.0)]; ui.painter().add(egui::Shape::convex_polygon(tri.to_vec(), if current_fill.is_some() { t.accent } else { t.text_faint }, Stroke::NONE)); diff --git a/crates/ui-egui/src/lib.rs b/crates/ui-egui/src/lib.rs index 643d5e90ff..ae89f96bcb 100644 --- a/crates/ui-egui/src/lib.rs +++ b/crates/ui-egui/src/lib.rs @@ -331,6 +331,7 @@ pub struct PdfCraftApp { pub flatten_fill_sign_on_save: bool, /// Point size for new Fill & Sign text and dates, kept across launches. pub fill_text_size: f64, + pub fill_tool: fill_sign::FillTool, /// Colour for new Fill & Sign text and dates, kept across launches. pub fill_text_color: [f64; 3], pub dialog: Option, diff --git a/crates/ui-egui/src/tools/fill_sign.rs b/crates/ui-egui/src/tools/fill_sign.rs index 045bb64974..9beaa9b834 100644 --- a/crates/ui-egui/src/tools/fill_sign.rs +++ b/crates/ui-egui/src/tools/fill_sign.rs @@ -373,20 +373,22 @@ pub(crate) fn signature_entries(ui: &mut egui::Ui, app: &mut crate::PdfCraftApp, } continue; }; - ui.horizontal(|ui| { + let on = app.quick_tool == crate::QuickTool::Fill(if i == 0 { FillTool::Signature } else { FillTool::Initials }); + let wash = ui.painter().add(egui::Shape::Noop); + let row = ui.horizontal(|ui| { let (rect, response) = ui.allocate_exact_size(vec2((ui.available_width() - 68.0).max(140.0), 52.0), Sense::click()); let use_label = label("Use {what}"); - response.widget_info(|| egui::WidgetInfo::labeled(egui::Role::Button, true, use_label.clone())); + response.widget_info(|| egui::WidgetInfo::selected(egui::Role::Button, true, on, use_label.clone())); let painter = ui.painter_at(rect); painter.rect_filled(rect, CornerRadius::same(4), Color32::WHITE); painter.rect_stroke( rect, CornerRadius::same(4), - Stroke::new(1.0, if response.hovered() { t.accent } else { t.border }), + Stroke::new(1.0, if on || response.hovered() { t.accent } else { t.border }), egui::StrokeKind::Inside, ); paint_saved(ui, rect.shrink(8.0), saved, &mut app.saved_signature_previews[i], &format!("saved-{what}")); - if response.on_hover_text(label("Place saved {what}")).clicked() { + if crate::theme::hand(response).on_hover_text(label("Place saved {what}")).clicked() { command = Some(use_id); } if crate::icons::button(ui, "pencil", 26.0, false, &label("Change {what}")).clicked() { @@ -396,6 +398,9 @@ pub(crate) fn signature_entries(ui: &mut egui::Ui, app: &mut crate::PdfCraftApp, command = Some(remove_id); } }); + if on { + ui.painter().set(wash, egui::Shape::rect_filled(row.response.rect.expand(3.0), CornerRadius::same(6), t.accent_soft)); + } ui.add_space(4.0); } command diff --git a/crates/ui-egui/tests/it/comments.rs b/crates/ui-egui/tests/it/comments.rs index 939de58dff..51b9eabce1 100644 --- a/crates/ui-egui/tests/it/comments.rs +++ b/crates/ui-egui/tests/it/comments.rs @@ -3280,7 +3280,7 @@ fn hotbar_docked_left_does_not_override_quick_tools() { }); h.run_steps(5); // Controls in the docked bar share one layout, so none may overlap (#5890). - let labels = ["Drag hotbar", "Select (V)", "Hand (H)", "Add a sticky note", "Highlight", "Draw", "Fill & Sign", "Hotbar settings"]; + let labels = ["Drag hotbar", "Select (V)", "Hand (H)", "Add a sticky note", "Highlight", "Draw", "Fill & Sign: Add signature", "Hotbar settings"]; let rects: Vec<_> = labels.iter().map(|&label| h.get_by_label(label).rect()).collect(); for (i, a) in rects.iter().enumerate() { for (j, b) in rects.iter().enumerate().skip(i + 1) { diff --git a/crates/ui-egui/tests/it/fill_sign.rs b/crates/ui-egui/tests/it/fill_sign.rs index ce21bd9f74..503080e71c 100644 --- a/crates/ui-egui/tests/it/fill_sign.rs +++ b/crates/ui-egui/tests/it/fill_sign.rs @@ -2,7 +2,7 @@ use egui::{Pos2, pos2}; use egui_kittest::Harness; -use egui_kittest::kittest::Queryable; +use egui_kittest::kittest::{NodeT, Queryable}; use pdfcraft_ui_egui::fill_sign::{FillTool, SavedSig}; use pdfcraft_ui_egui::{Dialog, PdfCraftApp, QuickTool}; @@ -423,9 +423,9 @@ fn changing_saved_signatures_and_initials_preserves_placed_marks() { let original = format!("{:?}", h.state().session.get(h.state().views[0].id).unwrap().info.annotations); // The quick toolbar exposes replacement without changing the normal placement action. - h.get_by_label("Fill & Sign").click(); + h.get_by_label("Fill & Sign: Place saved signature").click(); h.run_steps(2); - h.get_by_label("Fill & Sign").click(); + h.get_by_label("Fill & Sign: Place saved signature").click(); h.run_steps(2); h.get_by_label("Change signature").click(); h.run_steps(2); @@ -445,9 +445,9 @@ fn changing_saved_signatures_and_initials_preserves_placed_marks() { click(&mut h, 40.0, 200.0); assert_eq!(items(&h).len(), 2); - h.get_by_label("Fill & Sign").click(); + h.get_by_label("Fill & Sign: Place saved signature").click(); h.run_steps(2); - h.get_by_label("Fill & Sign").click(); + h.get_by_label("Fill & Sign: Place saved signature").click(); h.run_steps(2); h.get_by_label("Change initials").click(); h.run_steps(2); @@ -1812,3 +1812,33 @@ fn mode_tabs_keep_the_hand_over_a_type_box_scrolled_under_them() { h.run_steps(3); assert_eq!(h.state().mode, pdfcraft_ui_egui::Mode::Convert, "the tab takes the click, not the text box"); } + +#[test] +fn the_quick_bar_fill_button_shows_the_tool_it_picks() { + let mut h = harness(); + h.get_by_label("Fill & Sign: Add signature").click(); + h.run_steps(2); + assert_eq!(h.state().quick_tool, QuickTool::Fill(FillTool::Signature)); + assert_eq!(h.state().dialog, Some(Dialog::Signature)); + + h.state_mut().dialog = None; + h.state_mut().signature = Some(SavedSig::Typed("Ada Lovelace".into())); + h.state_mut().quick_tool = QuickTool::Select; + h.run_steps(2); + h.get_by_label("Fill & Sign: Place saved signature").click(); + h.run_steps(2); + assert_eq!(h.state().quick_tool, QuickTool::Fill(FillTool::Signature)); + assert_eq!(h.state().dialog, None); + h.get_by_label("Fill & Sign: Place saved signature").click(); + h.run_steps(2); + assert_eq!(h.get_by_label("Use signature").accesskit_node().toggled(), Some(egui::accesskit::Toggled::True), "the active tool's card is marked"); + h.key_press(egui::Key::Escape); + h.run_steps(2); + + h.state_mut().execute("sign.fill.check"); + h.state_mut().quick_tool = QuickTool::Select; + h.run_steps(2); + h.get_by_label("Fill & Sign: Checkmark").click(); + h.run_steps(2); + assert_eq!(h.state().quick_tool, QuickTool::Fill(FillTool::Check), "the button keeps the last Fill & Sign tool"); +} From edee301ebf8f719a5574949e17919c1bb8cddd16 Mon Sep 17 00:00:00 2001 From: Andrew Date: Fri, 30 Oct 2026 07:55:28 -0600 Subject: [PATCH 20/53] M5.7: the Fill & Sign button follows the active tool and names it Tools restored without a command (leaving full screen, the temporary tool, the control channel) left the remembered Fill & Sign tool stale, so the button could show Checkmark while a click placed a signature. The bar now takes the remembered tool from the active one, which replaces the dialog's and control channel's own updates. While a Fill & Sign tool is active the button opens the flyout, so its label names the tool instead of describing placing or adding a signature. --- crates/ui-egui/src/app/control.rs | 9 +++------ crates/ui-egui/src/dialogs.rs | 1 - crates/ui-egui/src/hotbar.rs | 5 ++++- crates/ui-egui/tests/it/fill_sign.rs | 14 +++++++++++--- 4 files changed, 18 insertions(+), 11 deletions(-) diff --git a/crates/ui-egui/src/app/control.rs b/crates/ui-egui/src/app/control.rs index fee5378b6e..0309042bf1 100644 --- a/crates/ui-egui/src/app/control.rs +++ b/crates/ui-egui/src/app/control.rs @@ -333,12 +333,9 @@ impl PdfCraftApp { field if field.starts_with("field-") => QuickTool::Field( prepare::FieldTool::from_command(&format!("form.add.{}", &field[6..])).ok_or_else(|| format!("unknown tool {field}"))?, ), - fill if fill.starts_with("fill-") => { - let f = - fill_sign::FillTool::from_command(&format!("sign.fill.{}", &fill[5..])).ok_or_else(|| format!("unknown tool {fill}"))?; - self.fill_tool = f; - QuickTool::Fill(f) - } + fill if fill.starts_with("fill-") => QuickTool::Fill( + fill_sign::FillTool::from_command(&format!("sign.fill.{}", &fill[5..])).ok_or_else(|| format!("unknown tool {fill}"))?, + ), other => { let t = comments::CommentTool::from_command(&format!("comment.{other}")).ok_or_else(|| format!("unknown tool {other}"))?; self.comment_prefs.group_tool[t.group()] = t; diff --git a/crates/ui-egui/src/dialogs.rs b/crates/ui-egui/src/dialogs.rs index 8105a22342..048d519222 100644 --- a/crates/ui-egui/src/dialogs.rs +++ b/crates/ui-egui/src/dialogs.rs @@ -1349,7 +1349,6 @@ fn dialog_signature(ui: &mut egui::Ui, app: &mut PdfCraftApp, req: &mut Pending) app.signature = d.saved(); crate::fill_sign::FillTool::Signature }; - app.fill_tool = tool; app.quick_tool = crate::QuickTool::Fill(tool); app.toast = None; } diff --git a/crates/ui-egui/src/hotbar.rs b/crates/ui-egui/src/hotbar.rs index bbd98be2ef..88c45bbd1e 100644 --- a/crates/ui-egui/src/hotbar.rs +++ b/crates/ui-egui/src/hotbar.rs @@ -239,9 +239,12 @@ fn core_tool_buttons(ui: &mut egui::Ui, app: &mut PdfCraftApp, cell: f32, popup_ QuickTool::Fill(f) => Some(f), _ => None, }; + if let Some(f) = current_fill { + app.fill_tool = f; + } let shown = app.fill_tool; let tip = match shown { - crate::fill_sign::FillTool::Signature | crate::fill_sign::FillTool::Initials => { + crate::fill_sign::FillTool::Signature | crate::fill_sign::FillTool::Initials if current_fill.is_none() => { let initials = shown == crate::fill_sign::FillTool::Initials; let saved = if initials { app.initials.is_some() } else { app.signature.is_some() }; crate::i18n::fmt( diff --git a/crates/ui-egui/tests/it/fill_sign.rs b/crates/ui-egui/tests/it/fill_sign.rs index 503080e71c..bd8bd30939 100644 --- a/crates/ui-egui/tests/it/fill_sign.rs +++ b/crates/ui-egui/tests/it/fill_sign.rs @@ -425,7 +425,7 @@ fn changing_saved_signatures_and_initials_preserves_placed_marks() { // The quick toolbar exposes replacement without changing the normal placement action. h.get_by_label("Fill & Sign: Place saved signature").click(); h.run_steps(2); - h.get_by_label("Fill & Sign: Place saved signature").click(); + h.get_by_label("Fill & Sign: Sign").click(); h.run_steps(2); h.get_by_label("Change signature").click(); h.run_steps(2); @@ -447,7 +447,7 @@ fn changing_saved_signatures_and_initials_preserves_placed_marks() { h.get_by_label("Fill & Sign: Place saved signature").click(); h.run_steps(2); - h.get_by_label("Fill & Sign: Place saved signature").click(); + h.get_by_label("Fill & Sign: Sign").click(); h.run_steps(2); h.get_by_label("Change initials").click(); h.run_steps(2); @@ -1829,7 +1829,7 @@ fn the_quick_bar_fill_button_shows_the_tool_it_picks() { h.run_steps(2); assert_eq!(h.state().quick_tool, QuickTool::Fill(FillTool::Signature)); assert_eq!(h.state().dialog, None); - h.get_by_label("Fill & Sign: Place saved signature").click(); + h.get_by_label("Fill & Sign: Sign").click(); h.run_steps(2); assert_eq!(h.get_by_label("Use signature").accesskit_node().toggled(), Some(egui::accesskit::Toggled::True), "the active tool's card is marked"); h.key_press(egui::Key::Escape); @@ -1841,4 +1841,12 @@ fn the_quick_bar_fill_button_shows_the_tool_it_picks() { h.get_by_label("Fill & Sign: Checkmark").click(); h.run_steps(2); assert_eq!(h.state().quick_tool, QuickTool::Fill(FillTool::Check), "the button keeps the last Fill & Sign tool"); + + h.state_mut().quick_tool = QuickTool::Fill(FillTool::Initials); + h.run_steps(2); + h.state_mut().quick_tool = QuickTool::Select; + h.run_steps(2); + h.get_by_label("Fill & Sign: Add initials").click(); + h.run_steps(2); + assert_eq!(h.state().quick_tool, QuickTool::Fill(FillTool::Initials), "a tool restored without a command still updates the button"); } From 2e699f489a093a4edb8e57cd7f6bd8c10cf455f6 Mon Sep 17 00:00:00 2001 From: Andrew Date: Mon, 2 Nov 2026 05:33:39 -0700 Subject: [PATCH 21/53] M5.7: a test that the Fill & Sign button keeps Date across documents Picks Date from the flyout, deselects it, and checks that the button still shows Date and that a click picks it, in a newly opened document and again after every document has been closed. --- crates/ui-egui/tests/it/fill_sign.rs | 33 ++++++++++++++++++++++++++++ 1 file changed, 33 insertions(+) diff --git a/crates/ui-egui/tests/it/fill_sign.rs b/crates/ui-egui/tests/it/fill_sign.rs index bd8bd30939..77e4420ba2 100644 --- a/crates/ui-egui/tests/it/fill_sign.rs +++ b/crates/ui-egui/tests/it/fill_sign.rs @@ -1850,3 +1850,36 @@ fn the_quick_bar_fill_button_shows_the_tool_it_picks() { h.run_steps(2); assert_eq!(h.state().quick_tool, QuickTool::Fill(FillTool::Initials), "a tool restored without a command still updates the button"); } + +#[test] +fn the_quick_bar_fill_button_keeps_date_after_deselecting_and_across_documents() { + let mut h = harness(); + h.get_by_label("Fill & Sign: Add signature").click_secondary(); + h.run_steps(2); + h.get_by_label("Date").click(); + h.run_steps(2); + assert_eq!(h.state().quick_tool, QuickTool::Fill(FillTool::Date)); + h.get_by_label("Select (V)").click(); + h.run_steps(2); + assert_eq!(h.state().quick_tool, QuickTool::Select); + + h.state_mut().open_bytes("second.pdf", None, FIXTURE.to_vec()).unwrap(); + h.state_mut().active = Some(1); + h.run_steps(2); + h.get_by_label("Fill & Sign: Date").click(); + h.run_steps(2); + assert_eq!(h.state().quick_tool, QuickTool::Fill(FillTool::Date), "a click picks the shown tool, not the default"); + assert_eq!(h.state().dialog, None, "the default tool's dialog stays closed"); + h.get_by_label("Select (V)").click(); + h.run_steps(2); + + h.state_mut().close_tab(1); + h.state_mut().close_tab(0); + h.run_steps(2); + h.state_mut().open_bytes("third.pdf", None, FIXTURE.to_vec()).unwrap(); + h.state_mut().active = Some(0); + h.run_steps(2); + h.get_by_label("Fill & Sign: Date").click(); + h.run_steps(2); + assert_eq!(h.state().quick_tool, QuickTool::Fill(FillTool::Date), "the tool outlives closing every document"); +} From 6028bdab7c13b0d11c5495f1eb91b1a10f898f41 Mon Sep 17 00:00:00 2001 From: Andrew Date: Mon, 2 Nov 2026 06:47:43 -0700 Subject: [PATCH 22/53] M5.7: the Fill & Sign button remembers tools where they are selected The quick bar copied the active tool into fill_tool on every draw, so a tool restored on leaving full screen replaced the last selection, and a tool selected while the bar was hidden was forgotten. The signature dialog and the control channel now record the tool when they select it, and the bar only reads: the active tool if there is one, otherwise the remembered one. --- crates/ui-egui/src/app/control.rs | 9 ++++-- crates/ui-egui/src/dialogs.rs | 1 + crates/ui-egui/src/hotbar.rs | 5 +-- crates/ui-egui/tests/it/fill_sign.rs | 46 ++++++++++++++++++++++++++-- 4 files changed, 51 insertions(+), 10 deletions(-) diff --git a/crates/ui-egui/src/app/control.rs b/crates/ui-egui/src/app/control.rs index 0309042bf1..443b970148 100644 --- a/crates/ui-egui/src/app/control.rs +++ b/crates/ui-egui/src/app/control.rs @@ -333,9 +333,12 @@ impl PdfCraftApp { field if field.starts_with("field-") => QuickTool::Field( prepare::FieldTool::from_command(&format!("form.add.{}", &field[6..])).ok_or_else(|| format!("unknown tool {field}"))?, ), - fill if fill.starts_with("fill-") => QuickTool::Fill( - fill_sign::FillTool::from_command(&format!("sign.fill.{}", &fill[5..])).ok_or_else(|| format!("unknown tool {fill}"))?, - ), + fill if fill.starts_with("fill-") => { + let tool = + fill_sign::FillTool::from_command(&format!("sign.fill.{}", &fill[5..])).ok_or_else(|| format!("unknown tool {fill}"))?; + self.fill_tool = tool; + QuickTool::Fill(tool) + } other => { let t = comments::CommentTool::from_command(&format!("comment.{other}")).ok_or_else(|| format!("unknown tool {other}"))?; self.comment_prefs.group_tool[t.group()] = t; diff --git a/crates/ui-egui/src/dialogs.rs b/crates/ui-egui/src/dialogs.rs index 048d519222..8105a22342 100644 --- a/crates/ui-egui/src/dialogs.rs +++ b/crates/ui-egui/src/dialogs.rs @@ -1349,6 +1349,7 @@ fn dialog_signature(ui: &mut egui::Ui, app: &mut PdfCraftApp, req: &mut Pending) app.signature = d.saved(); crate::fill_sign::FillTool::Signature }; + app.fill_tool = tool; app.quick_tool = crate::QuickTool::Fill(tool); app.toast = None; } diff --git a/crates/ui-egui/src/hotbar.rs b/crates/ui-egui/src/hotbar.rs index 88c45bbd1e..2b1ec38a96 100644 --- a/crates/ui-egui/src/hotbar.rs +++ b/crates/ui-egui/src/hotbar.rs @@ -239,10 +239,7 @@ fn core_tool_buttons(ui: &mut egui::Ui, app: &mut PdfCraftApp, cell: f32, popup_ QuickTool::Fill(f) => Some(f), _ => None, }; - if let Some(f) = current_fill { - app.fill_tool = f; - } - let shown = app.fill_tool; + let shown = current_fill.unwrap_or(app.fill_tool); let tip = match shown { crate::fill_sign::FillTool::Signature | crate::fill_sign::FillTool::Initials if current_fill.is_none() => { let initials = shown == crate::fill_sign::FillTool::Initials; diff --git a/crates/ui-egui/tests/it/fill_sign.rs b/crates/ui-egui/tests/it/fill_sign.rs index 77e4420ba2..241726cc9e 100644 --- a/crates/ui-egui/tests/it/fill_sign.rs +++ b/crates/ui-egui/tests/it/fill_sign.rs @@ -1841,14 +1841,54 @@ fn the_quick_bar_fill_button_shows_the_tool_it_picks() { h.get_by_label("Fill & Sign: Checkmark").click(); h.run_steps(2); assert_eq!(h.state().quick_tool, QuickTool::Fill(FillTool::Check), "the button keeps the last Fill & Sign tool"); +} - h.state_mut().quick_tool = QuickTool::Fill(FillTool::Initials); +#[test] +fn a_restored_fill_tool_shows_without_replacing_the_last_selection() { + let mut h = harness(); + h.state_mut().execute("sign.fill.date"); h.run_steps(2); + h.state_mut().apply_full_screen(true); + h.run_steps(2); + h.state_mut().execute("sign.fill.check"); + h.run_steps(2); + h.state_mut().apply_full_screen(false); + h.run_steps(2); + assert_eq!(h.state().quick_tool, QuickTool::Fill(FillTool::Date)); + h.get_by_label("Fill & Sign: Date"); + h.state_mut().quick_tool = QuickTool::Select; h.run_steps(2); - h.get_by_label("Fill & Sign: Add initials").click(); + h.get_by_label("Fill & Sign: Checkmark").click(); h.run_steps(2); - assert_eq!(h.state().quick_tool, QuickTool::Fill(FillTool::Initials), "a tool restored without a command still updates the button"); + assert_eq!(h.state().quick_tool, QuickTool::Fill(FillTool::Check)); +} + +#[test] +fn a_fill_tool_selected_while_the_quick_bar_is_hidden_is_remembered() { + let mut h = harness(); + let mode = h.state().mode; + h.state_mut().set_option("mode", "read").unwrap(); + h.run_steps(2); + h.state_mut().set_option("quick", "fill-date").unwrap(); + h.run_steps(2); + h.state_mut().set_option("quick", "select").unwrap(); + h.state_mut().mode = mode; + h.run_steps(2); + h.get_by_label("Fill & Sign: Date"); + + h.state_mut().initials = Some(SavedSig::Typed("AL".into())); + h.state_mut().set_option("mode", "read").unwrap(); + h.run_steps(2); + h.state_mut().execute("sign.fill.initials.change"); + h.run_steps(2); + h.get_by_label("Apply").click(); + h.run_steps(3); + assert_eq!(h.state().quick_tool, QuickTool::Fill(FillTool::Initials)); + h.state_mut().quick_tool = QuickTool::Select; + h.state_mut().mode = mode; + h.run_steps(2); + h.get_by_label("Fill & Sign: Place saved initials"); } #[test] From ea6d0c6bfbba5ca8985c6880791c8c46cbabd8ba Mon Sep 17 00:00:00 2001 From: KenM76 Date: Tue, 3 Nov 2026 20:46:33 -0500 Subject: [PATCH 23/53] M12: pdfcraft-prc-render, a headless renderer for PRC models New L3 crate over pdfcraft-prc: assemble places every product-tree tessellation into world space with its colours (pure `place` does the transform), Camera fits a view to the model, and render fills depth-buffered triangles with headlight or unlit shading into an RGBA image. Opaque triangles draw first, translucent ones blend over them; perspective views clip at a near plane. Never panics: MAX_RENDER_PIXELS and MAX_ASSEMBLED_TRIANGLES cap the work, every pixel and vertex index is checked, and damage returns RenderError. Single-threaded. Tests: 24 unit tests (two drive the product-tree placement path) plus an end-to-end one that builds a synthetic PRC stream in-test, renders it, and checks the background, the nearer triangle and the farther one by pixel; every truncation of that stream parses and assembles without panicking. Reverting the depth test or the placement transform fails the matching tests. Adapted from pdfcer (MIT, same author) at 912437c3, trimmed to the camera, raster and headlight parts and rewritten to the pdfcraft-prc types. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SqmHYx8FiWM5PcsoiErgp9 --- Cargo.lock | 9 + Cargo.toml | 2 + crates/prc-render/Cargo.toml | 18 ++ crates/prc-render/README.md | 28 ++ crates/prc-render/src/assemble.rs | 126 ++++++++ crates/prc-render/src/assemble/tests.rs | 88 ++++++ crates/prc-render/src/lib.rs | 17 ++ crates/prc-render/src/raster.rs | 226 +++++++++++++++ crates/prc-render/src/render.rs | 365 ++++++++++++++++++++++++ crates/prc-render/src/render/tests.rs | 269 +++++++++++++++++ crates/prc-render/src/vec3.rs | 53 ++++ crates/prc-render/tests/model.rs | 145 ++++++++++ xtask/src/layers.rs | 1 + 13 files changed, 1347 insertions(+) create mode 100644 crates/prc-render/Cargo.toml create mode 100644 crates/prc-render/README.md create mode 100644 crates/prc-render/src/assemble.rs create mode 100644 crates/prc-render/src/assemble/tests.rs create mode 100644 crates/prc-render/src/lib.rs create mode 100644 crates/prc-render/src/raster.rs create mode 100644 crates/prc-render/src/render.rs create mode 100644 crates/prc-render/src/render/tests.rs create mode 100644 crates/prc-render/src/vec3.rs create mode 100644 crates/prc-render/tests/model.rs diff --git a/Cargo.lock b/Cargo.lock index 6e92f0fcf0..bc76adfde5 100644 --- a/Cargo.lock +++ b/Cargo.lock @@ -4560,6 +4560,15 @@ dependencies = [ "thiserror 2.0.21", ] +[[package]] +name = "pdfcraft-prc-render" +version = "0.18.0" +dependencies = [ + "flate2", + "pdfcraft-prc", + "thiserror 2.0.21", +] + [[package]] name = "pdfcraft-preflight" version = "0.18.0" diff --git a/Cargo.toml b/Cargo.toml index 4e1dbffaea..1515269906 100644 --- a/Cargo.toml +++ b/Cargo.toml @@ -59,6 +59,8 @@ pdfcraft-create = { path = "crates/create", version = "0.18.0" } pdfcraft-forms = { path = "crates/forms", version = "0.18.0" } pdfcraft-xfa = { path = "crates/xfa", version = "0.18.0" } pdfcraft-render = { path = "crates/render", version = "0.18.0" } +pdfcraft-prc = { path = "crates/prc", version = "0.18.0" } +pdfcraft-prc-render = { path = "crates/prc-render", version = "0.18.0" } pdfcraft-engine = { path = "crates/engine", version = "0.18.0" } pdfcraft-platform = { path = "crates/platform", version = "0.18.0" } pdfcraft-ui-egui = { path = "crates/ui-egui", version = "0.18.0" } diff --git a/crates/prc-render/Cargo.toml b/crates/prc-render/Cargo.toml new file mode 100644 index 0000000000..11ccd3346f --- /dev/null +++ b/crates/prc-render/Cargo.toml @@ -0,0 +1,18 @@ +[package] +name = "pdfcraft-prc-render" +description = "Headless CPU renderer for PRC 3D models: camera, depth-buffered triangle fill, headlight shading" +version.workspace = true +edition.workspace = true +license.workspace = true +rust-version.workspace = true +repository.workspace = true + +[lints] +workspace = true + +[dependencies] +pdfcraft-prc.workspace = true +thiserror.workspace = true + +[dev-dependencies] +flate2.workspace = true diff --git a/crates/prc-render/README.md b/crates/prc-render/README.md new file mode 100644 index 0000000000..43730040f2 --- /dev/null +++ b/crates/prc-render/README.md @@ -0,0 +1,28 @@ +# pdfcraft-prc-render + +A headless CPU renderer for PRC 3D models read by `pdfcraft-prc`: placed meshes in world space, a fitted camera, a +depth-buffered triangle fill and headlight shading, into a straight-RGBA image. + +- **Layer:** L3. Depends on `pdfcraft-prc` only. +- **Licence:** MIT OR Apache-2.0. +- `#![forbid(unsafe_code)]`. Never panics on input: image size, triangle count and every index are checked, and + damage returns `RenderError`. + +## API + +```rust +assemble(&PrcFile) -> Result, RenderError> // every placement's mesh moved into the world and coloured +place(&TriangleMesh, &Matrix, colour, triangle_colours) -> WorldMesh +Camera::fit(&Bounds, direction, up, perspective, aspect) / Camera::fit_meshes(&[WorldMesh], ...) +render(&[WorldMesh], &Camera, &RenderOptions) -> Result +``` + +- With no product tree, each tessellated mesh is drawn where it lies, in the default colour. +- A triangle's own colour wins over its mesh's, which wins over `RenderOptions::colour`. +- Opaque triangles are drawn first, then translucent ones blended over them; alpha 0 is not drawn. +- `Lighting::Headlight` (default) shades by the angle to the view direction; `Lighting::Unlit` draws flat colour. +- Perspective triangles are clipped at a near plane; orthographic views are not clipped. +- `MAX_RENDER_PIXELS` (2^24, about 16 MP) bounds the image: at most 64 MB of RGBA plus 128 MB of depth. + `MAX_ASSEMBLED_TRIANGLES` (2^24) bounds one `assemble` call. Neither bounds time: a model of many large + triangles at full size is a long CPU render, so callers on a UI thread should render small or off-thread. +- Not yet: shadows, ambient occlusion, specular light, scene lights, textures, wires and markups. diff --git a/crates/prc-render/src/assemble.rs b/crates/prc-render/src/assemble.rs new file mode 100644 index 0000000000..3f8fa54caf --- /dev/null +++ b/crates/prc-render/src/assemble.rs @@ -0,0 +1,126 @@ +//! A parsed PRC file's tessellations placed in the world, ready for [`crate::render`]. + +use crate::RenderError; +use crate::raster::channel; +use pdfcraft_prc::{Matrix, PrcFile, Rgba, Tessellation, TriangleMesh}; + +/// The most triangles [`assemble`] places, over every placement: a tessellation many +/// occurrences draw is copied once per occurrence. +pub const MAX_ASSEMBLED_TRIANGLES: usize = 1 << 24; + +/// A triangle mesh in world coordinates with the colours it draws in. +#[derive(Debug, Clone, Default, PartialEq)] +pub struct WorldMesh { + /// The triangles, already placed. + pub mesh: TriangleMesh, + /// The mesh's straight-RGBA colour; `None` takes [`crate::RenderOptions::colour`], opaque. + pub colour: Option<[u8; 4]>, + /// Per triangle, its own colour when its face is styled; `None` (the whole list or an entry) + /// takes [`Self::colour`]. + pub triangle_colours: Option>>, +} + +impl WorldMesh { + /// `mesh`, uncoloured. + #[must_use] + pub fn new(mesh: TriangleMesh) -> Self { + WorldMesh { mesh, colour: None, triangle_colours: None } + } + + /// The colour triangle `t` draws in, before the render options' default. + pub(crate) fn colour_of(&self, t: usize) -> Option<[u8; 4]> { + self.triangle_colours.as_ref().and_then(|c| c.get(t).copied().flatten()).or(self.colour) + } +} + +/// `mesh` moved into the world by `matrix`, drawn in `colour` and, per triangle, `triangle_colours`. +#[must_use] +pub fn place(mesh: &TriangleMesh, matrix: &Matrix, colour: Option<[u8; 4]>, triangle_colours: Option>>) -> WorldMesh { + WorldMesh { mesh: mesh.transformed(matrix), colour, triangle_colours } +} + +/// Every triangle mesh `prc` draws, in world coordinates and coloured as its product tree says +/// ([`PrcFile::placements`]). A file with no tree draws each file structure's meshes where they +/// are, uncoloured. Wires, markups and compressed tessellations are not drawn. +/// +/// # Errors +/// [`RenderError::Model`] when a tessellation section or the tree cannot be read; +/// [`RenderError::TooManyTriangles`] past [`MAX_ASSEMBLED_TRIANGLES`]. +pub fn assemble(prc: &PrcFile) -> Result, RenderError> { + let tessellations = prc.file_structures.iter().map(|fs| fs.tessellations()).collect::, _>>()?; + let mut budget = Budget(0); + let mut out = Vec::new(); + if !prc.stores_tree() { + for mesh in tessellations.iter().flatten().filter_map(as_mesh) { + budget.spend(mesh)?; + out.push(WorldMesh::new(mesh.clone())); + } + return Ok(out); + } + let items = prc.placements()?.into_iter().map(|p| Item { + file_structure: p.file_structure, + tessellation: p.tessellation, + matrix: p.matrix, + colour: p.colour, + triangle_colours: Box::new(move |mesh: &TriangleMesh| p.triangle_colours(mesh)), + }); + place_items(&tessellations, items, &mut budget, &mut out)?; + Ok(out) +} + +/// One placement as [`place_items`] reads it: a tessellation by file structure and index, where +/// it goes and how it is coloured. +pub(crate) struct Item<'a> { + pub(crate) file_structure: usize, + pub(crate) tessellation: usize, + pub(crate) matrix: Matrix, + pub(crate) colour: Option, + pub(crate) triangle_colours: Box>, +} + +/// Per triangle of a mesh, the colour its face's style resolves to. +type FaceColours<'a> = dyn Fn(&TriangleMesh) -> Option>> + 'a; + +/// Each item's mesh moved into the world and coloured, appended to `out`. Items naming no +/// triangle mesh are skipped. +pub(crate) fn place_items<'a>( + tessellations: &[Vec], + items: impl IntoIterator>, + budget: &mut Budget, + out: &mut Vec, +) -> Result<(), RenderError> { + for p in items { + let Some(mesh) = tessellations.get(p.file_structure).and_then(|t| t.get(p.tessellation)).and_then(as_mesh) else { + continue; + }; + budget.spend(mesh)?; + let triangle_colours = (p.triangle_colours)(mesh).map(|c| c.into_iter().map(|c| c.map(bytes)).collect()); + out.push(place(mesh, &p.matrix, p.colour.map(bytes), triangle_colours)); + } + Ok(()) +} + +/// Triangles placed so far. +pub(crate) struct Budget(pub(crate) usize); + +impl Budget { + fn spend(&mut self, mesh: &TriangleMesh) -> Result<(), RenderError> { + self.0 = self.0.checked_add(mesh.triangles.len()).filter(|&n| n <= MAX_ASSEMBLED_TRIANGLES).ok_or(RenderError::TooManyTriangles)?; + Ok(()) + } +} + +fn as_mesh(t: &Tessellation) -> Option<&TriangleMesh> { + match t { + Tessellation::Mesh(m) => Some(m), + _ => None, + } +} + +/// A 0–1 colour as straight-RGBA bytes, each channel clamped. +fn bytes(c: Rgba) -> [u8; 4] { + [c.r, c.g, c.b, c.a].map(|v| channel(v.clamp(0.0, 1.0) * 255.0)) +} + +#[cfg(test)] +mod tests; diff --git a/crates/prc-render/src/assemble/tests.rs b/crates/prc-render/src/assemble/tests.rs new file mode 100644 index 0000000000..87c133f4da --- /dev/null +++ b/crates/prc-render/src/assemble/tests.rs @@ -0,0 +1,88 @@ +#![allow(clippy::unwrap_used)] + +use super::*; + +fn triangle() -> TriangleMesh { + let mut m = TriangleMesh::default(); + m.positions = vec![[0.0, 0.0, 0.0], [1.0, 0.0, 0.0], [0.0, 1.0, 0.0]]; + m.triangles = vec![[0, 1, 2]]; + m +} + +#[test] +fn a_placement_moves_its_mesh_into_the_world() { + let mut shift = Matrix::IDENTITY; + shift.0[0][3] = 5.0; + shift.0[2][3] = -2.0; + let placed = place(&triangle(), &shift, Some([1, 2, 3, 255]), None); + assert_eq!(placed.mesh.positions, [[5.0, 0.0, -2.0], [6.0, 0.0, -2.0], [5.0, 1.0, -2.0]]); + assert_eq!(placed.colour, Some([1, 2, 3, 255])); + // A mirror keeps the triangle facing the same way round. + let mut mirror = Matrix::IDENTITY; + mirror.0[0][0] = -1.0; + assert_eq!(place(&triangle(), &mirror, None, None).mesh.triangles, [[0, 2, 1]]); +} + +#[test] +fn a_triangle_colour_wins_over_the_mesh_colour() { + let mut m = place(&triangle(), &Matrix::IDENTITY, Some([9, 9, 9, 255]), Some(vec![Some([1, 1, 1, 255]), None])); + m.mesh.triangles.push([0, 2, 1]); + m.mesh.triangles.push([1, 2, 0]); + assert_eq!(m.colour_of(0), Some([1, 1, 1, 255])); + assert_eq!(m.colour_of(1), Some([9, 9, 9, 255]), "an unstyled face takes the mesh's colour"); + assert_eq!(m.colour_of(2), Some([9, 9, 9, 255]), "past the list too"); + assert_eq!(WorldMesh::new(triangle()).colour_of(0), None); +} + +#[test] +fn colours_are_clamped_into_bytes() { + let [r, g, b, a] = [0.0, 1.0, 0.5, 2.0]; + let mut c = Rgba::default(); + (c.r, c.g, c.b, c.a) = (r, g, b, a); + assert_eq!(bytes(c), [0, 255, 128, 255]); + c.r = f64::NAN; + c.g = -3.0; + assert_eq!(bytes(c)[..2], [0, 0]); +} + +#[test] +fn the_triangle_budget_is_shared_and_refuses_past_the_ceiling() { + let mut b = Budget(MAX_ASSEMBLED_TRIANGLES - 1); + assert!(b.spend(&triangle()).is_ok()); + assert_eq!(b.spend(&triangle()), Err(RenderError::TooManyTriangles)); + let mut b = Budget(usize::MAX); + assert_eq!(b.spend(&triangle()), Err(RenderError::TooManyTriangles)); +} + +fn item<'a>(fs: usize, tess: usize, x: f64, colour: Option) -> Item<'a> { + let mut matrix = Matrix::IDENTITY; + matrix.0[0][3] = x; + Item { file_structure: fs, tessellation: tess, matrix, colour, triangle_colours: Box::new(|_| None) } +} + +#[test] +fn each_placement_draws_its_tessellation_where_and_in_the_colour_the_tree_says() { + let tessellations = vec![vec![Tessellation::Mesh(triangle())]]; + let mut red = Rgba::default(); + red.r = 1.0; + red.a = 1.0; + // Two occurrences of one tessellation at different places, one coloured; one naming nothing. + let items = [item(0, 0, 3.0, Some(red)), item(0, 0, -4.0, None), item(1, 0, 0.0, None), item(0, 7, 0.0, None)]; + let (mut budget, mut out) = (Budget(0), Vec::new()); + place_items(&tessellations, items, &mut budget, &mut out).unwrap(); + assert_eq!(out.len(), 2, "items naming no mesh are skipped"); + assert_eq!(out[0].mesh.positions, [[3.0, 0.0, 0.0], [4.0, 0.0, 0.0], [3.0, 1.0, 0.0]]); + assert_eq!(out[0].colour, Some([255, 0, 0, 255])); + assert_eq!(out[1].mesh.positions, [[-4.0, 0.0, 0.0], [-3.0, 0.0, 0.0], [-4.0, 1.0, 0.0]]); + assert_eq!(out[1].colour, None); + assert_eq!(budget.0, 2, "each occurrence spends its triangles"); +} + +#[test] +fn a_placement_past_the_triangle_budget_is_an_error() { + let tessellations = vec![vec![Tessellation::Mesh(triangle())]]; + let (mut budget, mut out) = (Budget(MAX_ASSEMBLED_TRIANGLES), Vec::new()); + let r = place_items(&tessellations, [item(0, 0, 0.0, None)], &mut budget, &mut out); + assert!(matches!(r, Err(RenderError::TooManyTriangles))); + assert!(out.is_empty()); +} diff --git a/crates/prc-render/src/lib.rs b/crates/prc-render/src/lib.rs new file mode 100644 index 0000000000..da0ed41230 --- /dev/null +++ b/crates/prc-render/src/lib.rs @@ -0,0 +1,17 @@ +//! Headless CPU rendering of PRC 3D models (ISO 14739-1) to an RGBA image. +//! +//! Layer L3: depends only on `pdfcraft-prc`. [`assemble`] turns a parsed [`pdfcraft_prc::PrcFile`] +//! into world-space meshes; [`render`] draws them from a [`Camera`] with one depth-tested sample per +//! pixel centre and a light at the camera. Single-threaded and GUI-free, so it runs unchanged on wasm32. +//! The image size is capped by [`MAX_RENDER_PIXELS`] and the assembled model by +//! [`MAX_ASSEMBLED_TRIANGLES`]; bad input is an error, never a panic. +#![forbid(unsafe_code)] +#![deny(clippy::unwrap_used, clippy::expect_used, clippy::panic, clippy::unimplemented, clippy::todo, clippy::unreachable)] + +mod assemble; +mod raster; +mod render; +mod vec3; + +pub use assemble::{MAX_ASSEMBLED_TRIANGLES, WorldMesh, assemble, place}; +pub use render::{Bounds, Camera, Image, Lighting, MAX_RENDER_PIXELS, Projection, RenderError, RenderOptions, render}; diff --git a/crates/prc-render/src/raster.rs b/crates/prc-render/src/raster.rs new file mode 100644 index 0000000000..dddb3b394d --- /dev/null +++ b/crates/prc-render/src/raster.rs @@ -0,0 +1,226 @@ +//! Projection, near-plane clipping and the depth-tested triangle fill behind [`crate::render`]. + +use crate::render::{Camera, Projection, RenderError}; +use crate::vec3::{length, sub}; + +/// One view-space corner: x, y, z (z along the view direction), then its red, green and blue shade. +pub(crate) type ViewCorner = [f64; 6]; + +/// One screen-space corner: pixel x, pixel y and depth key, and its shade. +#[derive(Debug, Clone, Copy)] +pub(crate) struct Corner { + pub(crate) at: [f64; 3], + pub(crate) shade: [f64; 3], +} + +pub(crate) enum View { + /// `focal` = 1 / tan(fov_y / 2); `near` = the clip distance. + Perspective { + focal: f64, + aspect: f64, + near: f64, + }, + Orthographic { + half_height: f64, + aspect: f64, + }, +} + +impl View { + /// The projection for `camera` at width/height `aspect`; refuses a field of view outside + /// (0, 180) degrees or a non-positive orthographic height. + pub(crate) fn new(camera: &Camera, aspect: f64) -> Result { + match camera.projection { + Projection::Perspective { fov_y } => { + if !(fov_y > 0.0 && fov_y < 180.0) { + return Err(RenderError::Camera("the field of view is not between 0 and 180 degrees")); + } + let distance = length(sub(camera.target, camera.eye)); + Ok(View::Perspective { focal: 1.0 / (fov_y.to_radians() / 2.0).tan(), aspect, near: (distance * 1e-4).max(f64::MIN_POSITIVE) }) + } + Projection::Orthographic { height } => { + if !(height.is_finite() && height > 0.0) { + return Err(RenderError::Camera("the orthographic height is not positive")); + } + Ok(View::Orthographic { half_height: height / 2.0, aspect }) + } + } + } + + /// The part of `tri` in front of the near plane (Sutherland–Hodgman against one plane): + /// 0, 3 or 4 corners, the shade interpolated along each cut edge. + pub(crate) fn clip(&self, tri: [ViewCorner; 3]) -> Vec { + let View::Perspective { near, .. } = *self else { + return tri.to_vec(); + }; + let mut out = Vec::with_capacity(4); + for (k, &p) in tri.iter().enumerate() { + let q = tri.get((k + 1) % 3).copied().unwrap_or(p); + let z = |v: ViewCorner| v.get(2).copied().unwrap_or(0.0); + let (p_in, q_in) = (z(p) >= near, z(q) >= near); + if p_in { + out.push(p); + } + if p_in != q_in { + let t = (near - z(p)) / (z(q) - z(p)); + let mut cut = p; + for (c, d) in cut.iter_mut().zip(q) { + *c += t * (d - *c); + } + out.push(cut); + } + } + out + } + + /// The screen corner of view corner `v`: pixel x, pixel y and a depth key that is linear + /// in screen space and larger for nearer points. + pub(crate) fn project(&self, [x, y, z, r, g, b]: ViewCorner, width: u32, height: u32) -> Corner { + let (nx, ny, key) = match *self { + View::Perspective { focal, aspect, .. } => (x * focal / (z * aspect), y * focal / z, 1.0 / z), + View::Orthographic { half_height, aspect } => (x / (half_height * aspect), y / half_height, -z), + }; + Corner { at: [(nx + 1.0) / 2.0 * f64::from(width), (1.0 - ny) / 2.0 * f64::from(height), key], shade: [r, g, b] } + } +} + +/// The image being drawn: `rgba` holds `width * height` straight-RGBA pixels and `depth` their +/// depth keys, row by row from the top. +pub(crate) struct Target<'b> { + pub(crate) width: u32, + pub(crate) height: u32, + pub(crate) rgba: &'b mut [u8], + pub(crate) depth: &'b mut [f64], +} + +impl Target<'_> { + /// Fill the screen triangle of `corners` with straight-RGBA `colour`, depth-tested at pixel + /// centres, the shade interpolated across it. An opaque pixel replaces the pixel and its + /// depth; a translucent one is blended over it, leaving the depth. + pub(crate) fn fill(&mut self, corners: [Corner; 3], colour: [u8; 4]) { + let [a, b, c] = corners; + let area = edge(a.at, b.at, c.at); + if !area.is_finite() || area.abs() < 1e-12 { + return; + } + let Some((x0, x1, y0, y1)) = self.bounds(&corners) else { + return; + }; + for y in y0..=y1 { + for x in x0..=x1 { + let p = [f64::from(x) + 0.5, f64::from(y) + 0.5, 0.0]; + let w = [edge(b.at, c.at, p), edge(c.at, a.at, p), edge(a.at, b.at, p)].map(|e| e / area); + if w.iter().any(|&k| k < 0.0) { + continue; + } + let mix = |f: &dyn Fn(&Corner) -> f64| w.iter().zip(&corners).map(|(k, c)| k * f(c)).sum::(); + let key = mix(&|c| at(c.at, 2)); + let Some(i) = pixel_index(self.width, x, y) else { + continue; + }; + if self.depth.get(i).is_none_or(|&d| key <= d) { + continue; + } + let shade = if a.shade == b.shade && b.shade == c.shade { a.shade } else { [0, 1, 2].map(|j| mix(&|c| at(c.shade, j))) }; + self.blend(i, key, shaded(colour, shade)); + } + } + } + + /// The pixel rows and columns `corners` may cover, clipped to the image; `None` when off it. + fn bounds(&self, corners: &[Corner; 3]) -> Option<(u32, u32, u32, u32)> { + let lo = |i: usize| corners.iter().map(|c| at(c.at, i)).fold(f64::INFINITY, f64::min).floor().max(0.0); + let hi = |i: usize, limit: u32| corners.iter().map(|c| at(c.at, i)).fold(f64::NEG_INFINITY, f64::max).ceil().min(f64::from(limit) - 1.0); + let (x0, x1) = (lo(0), hi(0, self.width)); + let (y0, y1) = (lo(1), hi(1, self.height)); + // Each bound is finite and within 0..limit here, so the casts are exact. + (x0 <= x1 && y0 <= y1).then_some((x0 as u32, x1 as u32, y0 as u32, y1 as u32)) + } + + /// Write straight-RGBA `rgba` at pixel `i`: replacing it and its depth when opaque, blended + /// over it otherwise. + fn blend(&mut self, i: usize, key: f64, rgba: [u8; 4]) { + let opaque = rgba[3] == 255; + if opaque && let Some(d) = self.depth.get_mut(i) { + *d = key; + } + let Some(px) = i.checked_mul(4).and_then(|s| self.rgba.get_mut(s..s.checked_add(4)?)) else { + return; + }; + if opaque { + px.copy_from_slice(&rgba); + } else if let [pr, pg, pb, pa] = px { + let [sr, sg, sb, sa] = rgba; + let src = f64::from(sa) / 255.0; + let dst = f64::from(*pa) / 255.0 * (1.0 - src); + let out = src + dst; + if out <= 0.0 { + return; + } + for (p, s) in [pr, pg, pb].into_iter().zip([sr, sg, sb]) { + *p = channel((f64::from(s) * src + f64::from(*p) * dst) / out); + } + *pa = channel(out * 255.0); + } + } +} + +/// Pixel `x, y`'s index in an image `width` wide; `None` on overflow. +pub(crate) fn pixel_index(width: u32, x: u32, y: u32) -> Option { + usize::try_from(y).ok()?.checked_mul(usize::try_from(width).ok()?)?.checked_add(usize::try_from(x).ok()?) +} + +/// `v` rounded and clamped to a colour channel; NaN is 0. +pub(crate) fn channel(v: f64) -> u8 { + // Clamped to 0..=255 first, so the cast is exact (NaN casts to 0). + v.round().clamp(0.0, 255.0) as u8 +} + +/// `colour`'s RGB scaled channel by channel by `shade`, saturating at 255; alpha kept. +fn shaded([r, g, b, a]: [u8; 4], [sr, sg, sb]: [f64; 3]) -> [u8; 4] { + let [r, g, b] = [(r, sr), (g, sg), (b, sb)].map(|(v, s)| channel(f64::from(v) * s)); + [r, g, b, a] +} + +/// Twice the signed area of (a, b, p) in the xy plane. +pub(crate) fn edge(a: [f64; 3], b: [f64; 3], p: [f64; 3]) -> f64 { + // Two triangles sharing an edge meet it in opposite orders; evaluating it in one canonical + // order makes their values exact negatives, so a pixel centre on the edge cannot round out + // of both and leave a crack. + if (at(a, 0), at(a, 1)) > (at(b, 0), at(b, 1)) { + return -edge(b, a, p); + } + (at(b, 0) - at(a, 0)) * (at(p, 1) - at(a, 1)) - (at(b, 1) - at(a, 1)) * (at(p, 0) - at(a, 0)) +} + +/// Component `i` of `v`, 0.0 out of range. +pub(crate) fn at(v: [f64; 3], i: usize) -> f64 { + v.get(i).copied().unwrap_or(0.0) +} + +#[cfg(test)] +mod tests { + use super::*; + + #[test] + fn an_edge_seen_from_either_triangle_is_exactly_negated() { + // So no pixel centre on a shared edge is rejected by both. + let vals = [0.1, 0.7, 1.3, 2.9, 3.3, 5.5]; + for (a, b, p) in [([0.1, 0.1], [0.7, 0.7], [0.1, 2.9]), ([5.5, 0.7], [1.3, 3.3], [2.9, 0.1])] + .into_iter() + .chain(vals.iter().map(|&v| ([v, 0.1], [0.7, v], [v, v]))) + { + let [a, b, p] = [a, b, p].map(|[x, y]| [x, y, 0.0]); + assert_eq!(edge(a, b, p), -edge(b, a, p), "{a:?} {b:?} {p:?}"); + } + } + + #[test] + fn pixel_indices_and_channels_never_overflow() { + assert_eq!(pixel_index(10, 3, 2), Some(23)); + assert_eq!(channel(f64::NAN), 0); + assert_eq!(channel(1e300), 255); + assert_eq!(channel(-4.0), 0); + assert_eq!(channel(127.5), 128); + } +} diff --git a/crates/prc-render/src/render.rs b/crates/prc-render/src/render.rs new file mode 100644 index 0000000000..146266434b --- /dev/null +++ b/crates/prc-render/src/render.rs @@ -0,0 +1,365 @@ +//! The camera, the render options and [`render`]: world meshes to an RGBA image. +//! +//! One sample per pixel centre, no antialiasing. Faces are lit on both sides because a +//! producer's winding is not trusted to face outward. + +use crate::WorldMesh; +use crate::raster::{Corner, Target, View, ViewCorner, at}; +use crate::vec3::{cross, dot, length, normalize, scale, sub}; +use pdfcraft_prc::TriangleMesh; + +/// The largest image [`render`] draws, in pixels (width × height). +pub const MAX_RENDER_PIXELS: u64 = 16 * 1024 * 1024; + +/// [`Camera::fit`]'s framed extent over the model's: 5% clear on each side. +const FIT_MARGIN: f64 = 1.1; + +/// How the camera projects the scene onto the image. +#[derive(Debug, Clone, Copy, PartialEq)] +pub enum Projection { + /// A pinhole camera with this vertical field of view, in degrees (exclusive range 0–180). + Perspective { + /// Vertical field of view, degrees. + fov_y: f64, + }, + /// A parallel projection showing this much of the scene vertically, in model units. + Orthographic { + /// Visible height, model units. + height: f64, + }, +} + +/// Where the camera is and where it looks, in world coordinates. +#[derive(Debug, Clone, Copy, PartialEq)] +pub struct Camera { + /// The camera position. + pub eye: [f64; 3], + /// The point at the image centre. + pub target: [f64; 3], + /// The direction that appears upward; must not be parallel to the view direction. + pub up: [f64; 3], + /// The projection. + pub projection: Projection, +} + +/// An axis-aligned box around every point a set of meshes draws. +#[derive(Debug, Clone, Copy, PartialEq)] +pub struct Bounds { + /// The smallest x, y and z. + pub min: [f64; 3], + /// The largest x, y and z. + pub max: [f64; 3], +} + +impl Bounds { + /// The box around every finite vertex the meshes' triangles use, or `None` when there is none. + #[must_use] + pub fn of(meshes: &[WorldMesh]) -> Option { + let mut out: Option = None; + for p in vertices(meshes) { + let b = out.get_or_insert(Bounds { min: p, max: p }); + for ((lo, hi), c) in b.min.iter_mut().zip(b.max.iter_mut()).zip(p) { + *lo = lo.min(c); + *hi = hi.max(c); + } + } + out + } + + /// The box's centre. + #[must_use] + pub fn centre(&self) -> [f64; 3] { + [0, 1, 2].map(|i| (at(self.min, i) + at(self.max, i)) / 2.0) + } + + /// Half the box's diagonal: the radius of a sphere holding it. + #[must_use] + pub fn radius(&self) -> f64 { + length(sub(self.max, self.min)) / 2.0 + } +} + +/// Every finite vertex the meshes' triangles use. +fn vertices(meshes: &[WorldMesh]) -> impl Iterator + Clone + '_ { + meshes.iter().flat_map(|m| { + let mesh = &m.mesh; + mesh.triangles + .iter() + .flatten() + .filter_map(|&i| mesh.positions.get(usize::try_from(i).ok()?).copied()) + .filter(|p| p.iter().all(|c| c.is_finite())) + }) +} + +impl Camera { + /// A camera looking along `direction` that fits `bounds` in an image of the given aspect + /// ratio (width / height), with a 30° perspective or an orthographic projection: the box's + /// eight corners framed from this direction, about 5% clear on each side of the tighter axis. + /// + /// # Errors + /// [`RenderError::Camera`] when `direction` is zero or parallel to `up`, or `aspect` is not positive. + pub fn fit(bounds: &Bounds, direction: [f64; 3], up: [f64; 3], perspective: bool, aspect: f64) -> Result { + let corners = (0..8).map(|k| [0, 1, 2].map(|i| if k >> i & 1 == 0 { at(bounds.min, i) } else { at(bounds.max, i) })); + Camera::frame(bounds, corners, direction, up, (perspective, aspect)) + } + + /// As [`Camera::fit`], framing every vertex the meshes' triangles use rather than their box, + /// so an oblique view is not padded by empty box corners. + /// + /// # Errors + /// As [`Camera::fit`], and when the meshes have no finite vertex. + pub fn fit_meshes(meshes: &[WorldMesh], direction: [f64; 3], up: [f64; 3], perspective: bool, aspect: f64) -> Result { + let bounds = Bounds::of(meshes).ok_or(RenderError::Camera("the model has no vertex"))?; + Camera::frame(&bounds, vertices(meshes), direction, up, (perspective, aspect)) + } + + /// Frames `points` (which `bounds` holds) from `direction`, aimed at the middle of their projection. + fn frame( + bounds: &Bounds, + points: impl Iterator + Clone, + direction: [f64; 3], + up: [f64; 3], + (perspective, aspect): (bool, f64), + ) -> Result { + if !(aspect.is_finite() && aspect > 0.0) { + return Err(RenderError::Camera("the aspect ratio is not positive")); + } + let dir = normalize(direction).ok_or(RenderError::Camera("the view direction is zero"))?; + let centre = bounds.centre(); + let radius = match bounds.radius() { + r if r.is_finite() && r > 0.0 => r, + _ => 1.0, + }; + let [right, upward, forward] = + Camera { eye: sub(centre, dir), target: centre, up, projection: Projection::Orthographic { height: 1.0 } }.basis()?; + let view = |p: [f64; 3]| { + let r = sub(p, centre); + [dot(r, right), dot(r, upward), dot(r, forward)] + }; + let ((x0, x1), (y0, y1)) = points + .clone() + .map(view) + .fold(((f64::INFINITY, f64::NEG_INFINITY), (f64::INFINITY, f64::NEG_INFINITY)), |((a0, a1), (b0, b1)), [x, y, _]| { + ((a0.min(x), a1.max(x)), (b0.min(y), b1.max(y))) + }); + if !(x0 <= x1 && y0 <= y1) { + return Err(RenderError::Camera("the model has no vertex")); + } + let (mx, my) = ((x0 + x1) / 2.0, (y0 + y1) / 2.0); + let target = [0, 1, 2].map(|i| at(centre, i) + mx * at(right, i) + my * at(upward, i)); + let floor = radius * 1e-3; + let (projection, distance) = if perspective { + let fov_y: f64 = 30.0; + let ty = (fov_y.to_radians() / 2.0).tan() / FIT_MARGIN; + let tx = ty * aspect; + // A point at lateral (x, y) from the target and depth f past it is in view when + // |x| <= (d + f)·tx and |y| <= (d + f)·ty. + let d = points.map(view).fold(floor, |d, [x, y, f]| { + let need = ((x - mx).abs() / tx).max((y - my).abs() / ty); + d.max(need - f).max(floor - f) + }); + (Projection::Perspective { fov_y }, d) + } else { + let height = ((y1 - y0).max((x1 - x0) / aspect) * FIT_MARGIN).max(floor); + (Projection::Orthographic { height }, 2.0 * radius) + }; + let camera = Camera { eye: sub(target, scale(dir, distance)), target, up, projection }; + camera.basis()?; + Ok(camera) + } + + /// Right, up and forward unit vectors. + pub(crate) fn basis(&self) -> Result<[[f64; 3]; 3], RenderError> { + let forward = normalize(sub(self.target, self.eye)).ok_or(RenderError::Camera("the eye and the target coincide"))?; + let right = normalize(cross(forward, self.up)).ok_or(RenderError::Camera("the up direction is zero or parallel to the view direction"))?; + Ok([right, cross(right, forward), forward]) + } +} + +/// How surfaces are lit. +#[derive(Debug, Clone, Copy, PartialEq, Eq, Default)] +#[non_exhaustive] +pub enum Lighting { + /// One light at the camera: a surface's colour times `0.3 + 0.7·|n·v|`, where `v` points from + /// the surface to the eye (orthographic: against the view direction). + #[default] + Headlight, + /// No shading: every surface in its own colour. + Unlit, +} + +/// What [`render`] draws. +#[derive(Debug, Clone, Copy, PartialEq, Eq)] +pub struct RenderOptions { + /// Image width, pixels. + pub width: u32, + /// Image height, pixels. + pub height: u32, + /// Background colour, straight RGBA. + pub background: [u8; 4], + /// The colour of a surface with none of its own. + pub colour: [u8; 3], + /// How surfaces are lit. + pub lighting: Lighting, +} + +impl Default for RenderOptions { + /// 1024 × 768, opaque white background, light grey surface, the headlight. + fn default() -> Self { + RenderOptions { width: 1024, height: 768, background: [255, 255, 255, 255], colour: [190, 192, 200], lighting: Lighting::Headlight } + } +} + +/// A rendered image: `rgba` holds `width * height` straight-alpha pixels, row by row from the top. +#[derive(Debug, Clone, PartialEq, Eq)] +pub struct Image { + /// Width, pixels. + pub width: u32, + /// Height, pixels. + pub height: u32, + /// The pixels. + pub rgba: Vec, +} + +/// Why [`render`], [`Camera::fit`] or [`crate::assemble`] refused. +#[derive(Debug, Clone, PartialEq, Eq, thiserror::Error)] +#[non_exhaustive] +pub enum RenderError { + /// The image is empty or larger than [`MAX_RENDER_PIXELS`]. + #[error("a {width}x{height} image is empty or larger than the render ceiling")] + Size { + /// Requested width. + width: u32, + /// Requested height. + height: u32, + }, + /// The camera cannot form a view. + #[error("invalid camera: {0}")] + Camera(&'static str), + /// The model could not be read. + #[error("unreadable model: {0}")] + Model(#[from] pdfcraft_prc::PrcError), + /// The placed model has more than [`crate::MAX_ASSEMBLED_TRIANGLES`] triangles. + #[error("the placed model has more triangles than the assembly ceiling")] + TooManyTriangles, +} + +/// Draw `meshes` as seen by `camera`. +/// +/// Each triangle is filled in its own colour ([`WorldMesh::triangle_colours`]), else its mesh's +/// ([`WorldMesh::colour`]), else [`RenderOptions::colour`], and lit per [`RenderOptions::lighting`]; +/// nearer surfaces hide farther ones. Opaque triangles are drawn first; translucent ones are then +/// blended over them, hidden by nearer opaque surfaces but not by each other, in mesh order; fully +/// transparent ones are not drawn. Triangles with a non-finite or out-of-range vertex are skipped. +/// A perspective camera clips what lies behind it. +/// +/// # Errors +/// [`RenderError::Size`] for an empty or over-ceiling image; [`RenderError::Camera`] for a camera +/// that cannot form a view (eye on the target, `up` parallel to the view, a field of view outside +/// 0–180°, a non-positive orthographic height). +pub fn render(meshes: &[WorldMesh], camera: &Camera, options: &RenderOptions) -> Result { + let (width, height) = (options.width, options.height); + let size = RenderError::Size { width, height }; + let pixels = u64::from(width) * u64::from(height); + if pixels == 0 || pixels > MAX_RENDER_PIXELS { + return Err(size); + } + let pixels = usize::try_from(pixels).map_err(|_| size)?; + let view = View::new(camera, f64::from(width) / f64::from(height))?; + let scene = Scene { eye: camera.eye, basis: camera.basis()?, view, lighting: options.lighting }; + let mut rgba = options.background.repeat(pixels); + let mut depth = vec![f64::NEG_INFINITY; pixels]; + let mut target = Target { width, height, rgba: &mut rgba, depth: &mut depth }; + let [r, g, b] = options.colour; + for opaque in [true, false] { + for m in meshes { + for t in 0..m.mesh.triangles.len() { + let colour = m.colour_of(t).unwrap_or([r, g, b, 255]); + if (colour[3] == 255) == opaque && colour[3] != 0 { + scene.draw(&mut target, &m.mesh, t, colour); + } + } + } + } + Ok(Image { width, height, rgba }) +} + +/// The camera a frame is drawn from, with its basis and projection. +struct Scene { + eye: [f64; 3], + /// Right, up and forward. + basis: [[f64; 3]; 3], + view: View, + lighting: Lighting, +} + +impl Scene { + /// The shade at world point `p` with normal `n`. + fn shade(&self, p: [f64; 3], n: [f64; 3]) -> [f64; 3] { + if self.lighting == Lighting::Unlit { + return [1.0; 3]; + } + let towards = match self.view { + View::Perspective { .. } => normalize(sub(self.eye, p)), + View::Orthographic { .. } => Some(scale(at3(&self.basis, 2), -1.0)), + }; + [0.3 + 0.7 * towards.map_or(1.0, |v| dot(n, v).abs().min(1.0)); 3] + } + + /// Draw triangle `t` of `mesh` into `target` in straight-RGBA `colour`. + fn draw(&self, target: &mut Target, mesh: &TriangleMesh, t: usize, colour: [u8; 4]) { + let Some(world) = mesh.triangles.get(t).and_then(|tri| corners(mesh, tri)) else { + return; + }; + let Some(face) = normalize(cross(sub(at3(&world, 1), at3(&world, 0)), sub(at3(&world, 2), at3(&world, 0)))) else { + return; + }; + let stored = stored_normals(mesh, t); + let normal = |k: usize| stored.and_then(|n| n.get(k).copied().flatten()).unwrap_or(face); + // Without stored normals the triangle is flat: one shade, lit along the view ray to its first corner. + let flat = stored.is_none().then(|| self.shade(at3(&world, 0), normal(0))); + let [right, up, forward] = self.basis; + let view_corners = [0, 1, 2].map(|k| -> ViewCorner { + let p = at3(&world, k); + let [r, g, b] = flat.unwrap_or_else(|| self.shade(p, normal(k))); + let d = sub(p, self.eye); + [dot(d, right), dot(d, up), dot(d, forward), r, g, b] + }); + let screen: Vec = self.view.clip(view_corners).into_iter().map(|v| self.view.project(v, target.width, target.height)).collect(); + if let Some((&a, rest)) = screen.split_first() { + for pair in rest.windows(2) { + if let [b, c] = pair { + target.fill([a, *b, *c], colour); + } + } + } + } +} + +/// Triangle `t`'s stored corner normals, unit length; a corner whose normal is zero or not finite +/// is `None`, and so is the whole when the mesh stores none. +fn stored_normals(mesh: &TriangleMesh, t: usize) -> Option<[Option<[f64; 3]>; 3]> { + let slots = mesh.triangle_normals.get(t)?; + Some(slots.map(|i| mesh.normals.get(usize::try_from(i).ok()?).copied().and_then(normalize))) +} + +/// The three corners of a triangle, or `None` when one is missing or not finite. +fn corners(mesh: &TriangleMesh, tri: &[u32; 3]) -> Option<[[f64; 3]; 3]> { + let mut out = [[0.0; 3]; 3]; + for (slot, &i) in out.iter_mut().zip(tri) { + let p = *mesh.positions.get(usize::try_from(i).ok()?)?; + if !p.iter().all(|c| c.is_finite()) { + return None; + } + *slot = p; + } + Some(out) +} + +/// Corner `i` of a triangle, the origin when out of range. +fn at3(v: &[[f64; 3]; 3], i: usize) -> [f64; 3] { + v.get(i).copied().unwrap_or([0.0; 3]) +} + +#[cfg(test)] +mod tests; diff --git a/crates/prc-render/src/render/tests.rs b/crates/prc-render/src/render/tests.rs new file mode 100644 index 0000000000..008f177491 --- /dev/null +++ b/crates/prc-render/src/render/tests.rs @@ -0,0 +1,269 @@ +#![allow(clippy::unwrap_used, clippy::panic, clippy::unreachable)] + +use super::*; + +fn mesh(positions: Vec<[f64; 3]>, triangles: Vec<[u32; 3]>) -> TriangleMesh { + let mut m = TriangleMesh::default(); + m.positions = positions; + m.triangles = triangles; + m +} + +fn quad_mesh(z: f64, half: f64, tilt: f64) -> TriangleMesh { + mesh(vec![[-half, -half, z - tilt], [half, -half, z + tilt], [half, half, z + tilt], [-half, half, z - tilt]], vec![[0, 1, 2], [0, 2, 3]]) +} + +fn quad(z: f64, half: f64, tilt: f64) -> WorldMesh { + WorldMesh::new(quad_mesh(z, half, tilt)) +} + +fn painted(m: WorldMesh, colour: Option<[u8; 4]>) -> WorldMesh { + WorldMesh { colour, ..m } +} + +fn ortho(height: f64) -> Camera { + Camera { eye: [0.0, 0.0, 10.0], target: [0.0, 0.0, 0.0], up: [0.0, 1.0, 0.0], projection: Projection::Orthographic { height } } +} + +fn small() -> RenderOptions { + RenderOptions { width: 40, height: 40, ..RenderOptions::default() } +} + +fn pixel(image: &Image, x: u32, y: u32) -> [u8; 4] { + let i = (y * image.width + x) as usize * 4; + image.rgba[i..i + 4].try_into().unwrap() +} + +#[test] +fn a_facing_square_fills_the_middle_and_leaves_the_corners() { + let image = render(&[quad(0.0, 1.0, 0.0)], &ortho(4.0), &small()).unwrap(); + assert_eq!(pixel(&image, 20, 20), [190, 192, 200, 255]); + assert_eq!(pixel(&image, 1, 1), [255, 255, 255, 255]); + // Half the view height is covered: pixels 10..30. + assert_eq!(pixel(&image, 10, 20)[0], 190); + assert_eq!(pixel(&image, 9, 20)[0], 255); +} + +#[test] +fn stored_normals_shade_smoothly_across_a_flat_square() { + let mut m = quad_mesh(0.0, 1.0, 0.0); + let s = 3f64.sqrt() / 2.0; + // The left corners lean 60° away from the viewer; the right face it. + m.normals = vec![[s, 0.0, 0.5], [0.0, 0.0, 2.0]]; + m.triangle_normals = vec![[0, 1, 1], [0, 1, 0]]; + let image = render(&[WorldMesh::new(m.clone())], &ortho(4.0), &small()).unwrap(); + let red = |x| pixel(&image, x, 20)[0]; + assert!(red(11) < red(20) && red(20) < red(29), "{} {} {}", red(11), red(20), red(29)); + // Shade 0.3 + 0.7 cos θ at the edges: 0.65 left, 1.0 right. + assert!((i32::from(red(10)) - 128).abs() <= 4, "{}", red(10)); + assert!(red(29) >= 186); + m.normals.clear(); + m.triangle_normals.clear(); + let flat = render(&[WorldMesh::new(m)], &ortho(4.0), &small()).unwrap(); + assert_eq!(pixel(&flat, 11, 20), pixel(&flat, 29, 20)); +} + +#[test] +fn the_nearer_surface_hides_the_farther_whatever_the_order() { + let far = quad(-1.0, 2.0, 1.0); + let near = quad(1.0, 0.5, 0.0); + for meshes in [[far.clone(), near.clone()], [near, far]] { + let image = render(&meshes, &ortho(8.0), &small()).unwrap(); + assert_eq!(pixel(&image, 20, 20), [190, 192, 200, 255]); + assert!(pixel(&image, 20, 12)[0] < 190, "the tilted far square is darker"); + } +} + +const RED: Option<[u8; 4]> = Some([255, 0, 0, 255]); +const GLASS: Option<[u8; 4]> = Some([0, 0, 255, 102]); + +/// Glass (40% blue) in front of red blends whichever is listed first, glass behind red is hidden, +/// a clear mesh draws nothing, and a mesh with no colour takes the options' colour. +#[test] +fn meshes_draw_in_their_own_colours_and_glass_blends() { + let unlit = RenderOptions { lighting: Lighting::Unlit, ..small() }; + let at_middle = |meshes: &[(f64, Option<[u8; 4]>)]| { + let meshes: Vec = meshes.iter().map(|&(z, c)| painted(quad(z, 1.0, 0.0), c)).collect(); + pixel(&render(&meshes, &ortho(4.0), &unlit).unwrap(), 20, 20) + }; + assert_eq!(at_middle(&[(1.0, GLASS), (-1.0, RED)]), [153, 0, 102, 255]); + assert_eq!(at_middle(&[(-1.0, RED), (1.0, GLASS)]), [153, 0, 102, 255]); + assert_eq!(at_middle(&[(-1.0, GLASS), (1.0, RED)]), [255, 0, 0, 255]); + assert_eq!(at_middle(&[(1.0, Some([0, 0, 255, 0])), (-1.0, RED)]), [255, 0, 0, 255]); + assert_eq!(at_middle(&[(1.0, None)]), [190, 192, 200, 255]); + assert_eq!(at_middle(&[(2.0, GLASS), (1.0, GLASS), (-1.0, RED)]), [92, 0, 163, 255], "glass does not hide glass"); + let clear = RenderOptions { background: [0, 0, 0, 0], ..unlit }; + let image = render(&[painted(quad(1.0, 1.0, 0.0), GLASS)], &ortho(4.0), &clear).unwrap(); + assert_eq!(pixel(&image, 20, 20), [0, 0, 255, 102], "glass over nothing"); +} + +#[test] +fn a_triangle_colour_overrides_its_mesh_colour() { + let mut m = painted(quad(0.0, 1.0, 0.0), RED); + // The lower-right triangle is green; the upper-left keeps the mesh's red. + m.triangle_colours = Some(vec![Some([0, 255, 0, 255]), None]); + let unlit = RenderOptions { lighting: Lighting::Unlit, ..small() }; + let image = render(&[m], &ortho(4.0), &unlit).unwrap(); + assert_eq!(pixel(&image, 25, 25), [0, 255, 0, 255]); + assert_eq!(pixel(&image, 14, 14), [255, 0, 0, 255]); +} + +#[test] +fn each_edge_bounds_a_triangle() { + let tri = WorldMesh::new(mesh(vec![[-1.0, -1.0, 0.0], [1.0, -1.0, 0.0], [0.0, 1.0, 0.0]], vec![[0, 1, 2]])); + let image = render(&[tri], &ortho(4.0), &small()).unwrap(); + assert_ne!(pixel(&image, 20, 25)[0], 255); + // Just past each edge: below the base, left and right of the apex. + for (x, y) in [(20, 31), (13, 15), (27, 15)] { + assert_eq!(pixel(&image, x, y)[0], 255, "({x},{y})"); + } +} + +#[test] +fn a_face_turned_away_is_still_drawn() { + let mut back = quad_mesh(0.0, 1.0, 0.0); + for t in &mut back.triangles { + t.swap(1, 2); + } + let image = render(&[WorldMesh::new(back)], &ortho(4.0), &small()).unwrap(); + assert_eq!(pixel(&image, 20, 20), [190, 192, 200, 255]); +} + +#[test] +fn perspective_clips_what_is_behind_the_camera() { + let camera = Camera { projection: Projection::Perspective { fov_y: 60.0 }, ..ortho(1.0) }; + let image = render(&[quad(20.0, 1.0, 0.0)], &camera, &small()).unwrap(); + assert!(image.rgba.chunks(4).all(|p| p == [255, 255, 255, 255])); + // A floor running from behind the camera to beyond the target. + let floor = WorldMesh::new(mesh(vec![[-5.0, -1.0, 20.0], [5.0, -1.0, 20.0], [0.0, -1.0, -20.0]], vec![[0, 1, 2]])); + let image = render(&[floor], &camera, &small()).unwrap(); + assert_ne!(pixel(&image, 20, 38), [255, 255, 255, 255]); + assert_eq!(pixel(&image, 20, 2), [255, 255, 255, 255]); +} + +#[test] +fn a_fitted_camera_frames_the_model() { + // Depth matters in perspective: the nearer square looks larger. + let meshes = [quad(3.0, 3.0, 0.0), quad(-3.0, 3.0, 0.0)]; + let bounds = Bounds::of(&meshes).unwrap(); + for perspective in [true, false] { + let camera = Camera::fit(&bounds, [0.0, 0.0, -1.0], [0.0, 1.0, 0.0], perspective, 2.0).unwrap(); + let image = render(&meshes, &camera, &RenderOptions { width: 80, height: 40, ..RenderOptions::default() }).unwrap(); + assert_ne!(pixel(&image, 40, 20)[0], 255); + // Nothing drawn touches the image border. + for x in 0..80 { + assert_eq!(pixel(&image, x, 0)[0], 255); + assert_eq!(pixel(&image, x, 39)[0], 255); + } + } +} + +#[test] +fn a_mesh_fit_ignores_empty_box_corners() { + // A thin strip along the box's diagonal: seen across the diagonal, the box is as tall as it + // is wide, the strip is not. + let meshes = [WorldMesh::new(mesh(vec![[0.0, 0.0, 0.0], [10.0, 10.0, 0.0], [10.0, 10.01, 0.0]], vec![[0, 1, 2]]))]; + let (dir, up) = ([0.0, 0.0, -1.0], [1.0, -1.0, 0.0]); + let height = |c: Camera| match c.projection { + Projection::Orthographic { height } => height, + Projection::Perspective { .. } => unreachable!(), + }; + let boxed = height(Camera::fit(&Bounds::of(&meshes).unwrap(), dir, up, false, 2.0).unwrap()); + let tight = height(Camera::fit_meshes(&meshes, dir, up, false, 2.0).unwrap()); + assert!((boxed / tight - 2.0).abs() < 0.01, "{boxed} vs {tight}"); + let near = Camera::fit_meshes(&meshes, dir, up, true, 2.0).unwrap(); + let far = Camera::fit(&Bounds::of(&meshes).unwrap(), dir, up, true, 2.0).unwrap(); + assert!(length(sub(near.eye, near.target)) < 0.6 * length(sub(far.eye, far.target))); + + // Lopsided in its box: the limiting vertex sits exactly on the margin, measured from the + // re-centred target. Narrow (width limits) and wide (height limits). + let tri = WorldMesh::new(mesh(vec![[0.0, 0.0, 0.0], [2.0, 0.0, 0.0], [1.0, 2.0, 1.0]], vec![[0, 1, 2]])); + for aspect in [0.5, 3.0] { + let c = Camera::fit_meshes(std::slice::from_ref(&tri), [0.0, 0.3, -1.0], [1.0, 1.0, 0.0], true, aspect).unwrap(); + let [right, upward, forward] = c.basis().unwrap(); + let ty = 15f64.to_radians().tan() / FIT_MARGIN; + let fill = tri + .mesh + .positions + .iter() + .map(|&p| { + let r = sub(p, c.eye); + (dot(r, right).abs() / (ty * aspect)).max(dot(r, upward).abs() / ty) / dot(r, forward) + }) + .fold(0.0, f64::max); + assert!((fill - 1.0).abs() < 1e-6, "aspect {aspect}: {fill}"); + } +} + +#[test] +fn a_fitted_camera_fills_the_tighter_axis() { + let drawn = |image: &Image, horizontal: bool, line: u32| { + let n = if horizontal { image.width } else { image.height }; + (0..n) + .filter(|&i| { + let (x, y) = if horizontal { (i, line) } else { (line, i) }; + pixel(image, x, y)[0] != 255 + }) + .count() + }; + // Height-limited: a 6 x 6 square in an 80 x 40 image. + let square = [quad(0.0, 3.0, 0.0)]; + // Width-limited and off-origin: a 10 x 1 strip in a 40 x 40 image. + let strip = [WorldMesh::new(mesh(vec![[0.0, 0.0, 0.0], [10.0, 0.0, 0.0], [10.0, 1.0, 0.0], [0.0, 1.0, 0.0]], vec![[0, 1, 2], [0, 2, 3]]))]; + for perspective in [true, false] { + let fit = + |meshes: &[WorldMesh], aspect| Camera::fit(&Bounds::of(meshes).unwrap(), [0.0, 0.0, -1.0], [0.0, 1.0, 0.0], perspective, aspect).unwrap(); + let wide = RenderOptions { width: 80, height: 40, ..RenderOptions::default() }; + let rows = drawn(&render(&square, &fit(&square, 2.0), &wide).unwrap(), false, 40); + assert!((34..=37).contains(&rows), "{perspective}: {rows} of 40 rows"); + let image = render(&strip, &fit(&strip, 1.0), &small()).unwrap(); + let cols = drawn(&image, true, 20); + assert!((34..=37).contains(&cols), "{perspective}: {cols} of 40 columns"); + // Centred: the margins differ by at most a pixel. + let left = (0..40).take_while(|&x| pixel(&image, x, 20)[0] == 255).count(); + assert!(left.abs_diff(40 - cols - left) <= 1, "{perspective}: left {left}"); + } +} + +#[test] +fn bad_sizes_and_cameras_are_refused() { + let m = [quad(0.0, 1.0, 0.0)]; + let size = |width, height| RenderOptions { width, height, ..RenderOptions::default() }; + assert!(matches!(render(&m, &ortho(1.0), &size(0, 10)), Err(RenderError::Size { .. }))); + assert!(matches!(render(&m, &ortho(1.0), &size(10_000, 10_000)), Err(RenderError::Size { .. }))); + let bad = [ + Camera { eye: [0.0; 3], target: [0.0; 3], ..ortho(1.0) }, + Camera { up: [0.0, 0.0, 1.0], ..ortho(1.0) }, + Camera { projection: Projection::Perspective { fov_y: 180.0 }, ..ortho(1.0) }, + Camera { projection: Projection::Perspective { fov_y: f64::NAN }, ..ortho(1.0) }, + ortho(0.0), + ortho(f64::INFINITY), + ]; + for camera in bad { + assert!(matches!(render(&m, &camera, &small()), Err(RenderError::Camera(_))), "{camera:?}"); + } + let b = Bounds::of(&m).unwrap(); + assert!(Camera::fit(&b, [0.0; 3], [0.0, 1.0, 0.0], true, 1.0).is_err()); + assert!(Camera::fit(&b, [0.0, 0.0, 1.0], [0.0, 1.0, 0.0], true, 0.0).is_err()); + assert!(Camera::fit_meshes(&[], [0.0, 0.0, 1.0], [0.0, 1.0, 0.0], true, 1.0).is_err()); +} + +#[test] +fn non_finite_and_missing_vertices_are_skipped() { + let m = WorldMesh::new(mesh(vec![[0.0, 0.0, 0.0], [f64::NAN, 0.0, 0.0], [0.0, 1.0, 0.0]], vec![[0, 1, 2], [0, 2, 9], [0, 2, u32::MAX]])); + let image = render(std::slice::from_ref(&m), &ortho(4.0), &small()).unwrap(); + assert!(image.rgba.chunks(4).all(|p| p == [255, 255, 255, 255])); + assert_eq!(Bounds::of(&[m]).unwrap().max, [0.0, 1.0, 0.0]); +} + +#[test] +fn unlit_draws_the_plain_colour_and_the_headlight_shades_by_angle() { + let lit = |lighting, tilt| { + let image = render(&[quad(0.0, 1.0, tilt)], &ortho(4.0), &RenderOptions { lighting, ..small() }).unwrap(); + pixel(&image, 20, 20) + }; + assert_eq!(lit(Lighting::Headlight, 0.0), [190, 192, 200, 255]); + // Tilted 45°: 0.3 + 0.7 cos 45° ≈ 0.795 of the colour. + assert_eq!(lit(Lighting::Headlight, 1.0), [151, 153, 159, 255]); + assert_eq!(lit(Lighting::Unlit, 1.0), [190, 192, 200, 255]); +} diff --git a/crates/prc-render/src/vec3.rs b/crates/prc-render/src/vec3.rs new file mode 100644 index 0000000000..90af0f3ba2 --- /dev/null +++ b/crates/prc-render/src/vec3.rs @@ -0,0 +1,53 @@ +//! Three-component vector arithmetic for the camera and the rasterizer. + +/// A point or direction, `[x, y, z]`. +pub(crate) type Vec3 = [f64; 3]; + +/// `a - b`. +pub(crate) fn sub([ax, ay, az]: Vec3, [bx, by, bz]: Vec3) -> Vec3 { + [ax - bx, ay - by, az - bz] +} + +/// `a` times `s`. +pub(crate) fn scale(a: Vec3, s: f64) -> Vec3 { + a.map(|c| c * s) +} + +/// The dot product `a · b`. +pub(crate) fn dot([ax, ay, az]: Vec3, [bx, by, bz]: Vec3) -> f64 { + ax * bx + ay * by + az * bz +} + +/// The right-handed cross product `a × b`. +pub(crate) fn cross([ax, ay, az]: Vec3, [bx, by, bz]: Vec3) -> Vec3 { + [ay * bz - az * by, az * bx - ax * bz, ax * by - ay * bx] +} + +/// The Euclidean length of `a`. +pub(crate) fn length(a: Vec3) -> f64 { + dot(a, a).sqrt() +} + +/// `a` scaled to unit length; `None` when its length is zero or not finite. +pub(crate) fn normalize(a: Vec3) -> Option { + let l = length(a); + (l.is_finite() && l > 0.0).then(|| scale(a, 1.0 / l)) +} + +#[cfg(test)] +mod tests { + use super::*; + + #[test] + fn cross_follows_the_right_hand_rule() { + assert_eq!(cross([1.0, 0.0, 0.0], [0.0, 1.0, 0.0]), [0.0, 0.0, 1.0]); + assert_eq!(cross([0.0, 1.0, 0.0], [1.0, 0.0, 0.0]), [0.0, 0.0, -1.0]); + } + + #[test] + fn normalize_rejects_degenerate_vectors() { + assert_eq!(normalize([0.0, 4.0, 0.0]), Some([0.0, 1.0, 0.0])); + assert_eq!(normalize([0.0; 3]), None); + assert_eq!(normalize([f64::NAN, 0.0, 0.0]), None); + } +} diff --git a/crates/prc-render/tests/model.rs b/crates/prc-render/tests/model.rs new file mode 100644 index 0000000000..80e84252c6 --- /dev/null +++ b/crates/prc-render/tests/model.rs @@ -0,0 +1,145 @@ +//! End to end: a synthetic PRC stream parsed, assembled and rendered to pixels. +#![allow(clippy::unwrap_used, clippy::expect_used, clippy::panic)] + +use pdfcraft_prc::PrcFile; +use pdfcraft_prc_render::{Camera, Image, RenderOptions, assemble, render}; +use std::io::Write as _; + +/// An MSB-first bit sink: the inverse of the PRC bit reader, for the few values these tests write. +#[derive(Default)] +struct W { + bits: Vec, +} + +impl W { + fn put(&mut self, v: u64, n: u32) -> &mut Self { + for i in (0..n).rev() { + self.bits.push((v >> i) & 1 == 1); + } + self + } + + fn bit(&mut self, b: bool) -> &mut Self { + self.bits.push(b); + self + } + + /// An `UnsignedInteger`: bytes low first, each after a 1 bit, then a 0 bit. + fn uint(&mut self, mut v: u32) -> &mut Self { + while v != 0 { + self.bit(true).put(u64::from(v & 0xff), 8); + v >>= 8; + } + self.bit(false) + } + + fn uints(&mut self, v: &[u32]) -> &mut Self { + self.uint(u32::try_from(v.len()).unwrap()); + for &x in v { + self.uint(x); + } + self + } + + /// A `Double` from the few the code table stores whole: 0, ±0.5, ±1, ±2. + fn double(&mut self, v: f64) -> &mut Self { + let negative = v < 0.0; + match v.abs() { + 0.0 => self.put(0b01, 2), + 0.5 => self.put(0x777, 11).bit(negative), + 1.0 => self.put(0b0000, 4).bit(negative), + // The exponent row of [2, 4) with no mantissa bytes. + 2.0 => self.put(0x1b, 5).bit(negative).bit(false), + _ => panic!("{v} has no short code here"), + } + } + + fn bytes(&self) -> Vec { + self.bits.chunks(8).map(|c| c.iter().enumerate().fold(0u8, |a, (i, &b)| a | (u8::from(b) << (7 - i)))).collect() + } +} + +/// A tessellation section of one `TESS_3D` (authoring version 8137) with explicit normals and +/// triangles of `(normal, point)` coordinate indices. +fn tess_section(points: &[[f64; 3]], normals: &[[f64; 3]], corners: &[(u32, u32)]) -> Vec { + let mut w = W::default(); + w.uint(305).uint(0).bit(true).uint(1); + w.uint(172).bit(false).uint(u32::try_from(points.len() * 3).unwrap()); + points.iter().flatten().for_each(|&c| _ = w.double(c)); + // Faces present; no loop flag; normals not recalculated. + w.bit(true).bit(false).bit(false); + w.uint(u32::try_from(normals.len() * 3).unwrap()); + normals.iter().flatten().for_each(|&c| _ = w.double(c)); + w.uint(0).uints(&corners.iter().flat_map(|&(n, p)| [n, p]).collect::>()); + // One face of triangles with a normal per corner. + w.uint(1).uint(174).uint(0).uint(0).uint(0).uint(0x2).uint(0).uints(&[u32::try_from(corners.len() / 3).unwrap()]); + w.uint(0).bit(false); + w.uint(0).uint(0); + w.bytes() +} + +/// A one-file-structure PRC stream whose tessellation section is `tess`, with no product tree. +fn prc_file(tess: &[u8]) -> Vec { + const HEADER_LEN: usize = 107; + let zlib = |data: &[u8]| { + let mut e = flate2::write::ZlibEncoder::new(Vec::new(), flate2::Compression::default()); + e.write_all(data).unwrap(); + e.finish().unwrap() + }; + let u32s = |out: &mut Vec, vs: &[u32]| vs.iter().for_each(|v| out.extend_from_slice(&v.to_le_bytes())); + let mut globals = W::default(); + globals.uint(0); + let sections: [&[u8]; 5] = [&globals.bytes(), b"", tess, b"", b""]; + let mut body = b"PRC".to_vec(); + u32s(&mut body, &[8137, 8137, 5, 6, 7, 8, 0, 0, 0, 0, 0]); + let mut offs = vec![HEADER_LEN]; + for s in sections { + offs.push(HEADER_LEN + body.len()); + body.extend(zlib(s)); + } + let mf_start = HEADER_LEN + body.len(); + body.extend(zlib(b"")); + let mf_end = HEADER_LEN + body.len(); + let mut out = b"PRC".to_vec(); + u32s(&mut out, &[8137, 8137, 1, 2, 3, 4, 0, 0, 0, 0, 1]); + u32s(&mut out, &[5, 6, 7, 8, 0, 6]); + u32s(&mut out, &offs.iter().map(|&o| u32::try_from(o).unwrap()).collect::>()); + u32s(&mut out, &[u32::try_from(mf_start).unwrap(), u32::try_from(mf_end).unwrap(), 0]); + out.extend(body); + out +} + +fn pixel(image: &Image, x: u32, y: u32) -> [u8; 4] { + let i = (y * image.width + x) as usize * 4; + image.rgba[i..i + 4].try_into().unwrap() +} + +#[test] +fn a_synthetic_prc_renders_its_nearer_triangle_over_the_farther() { + // Listed first, a small triangle at z = 1 facing the viewer; behind it at z = -1, a larger + // one whose stored normal lies across the view, so the headlight shades it darkest. + let points = [[-1., -1., 1.], [1., -1., 1.], [0., 1., 1.], [-2., -2., -1.], [2., -2., -1.], [0., 2., -1.]]; + let normals = [[0., 0., 1.], [1., 0., 0.]]; + let corners = [(0, 0), (0, 3), (0, 6), (3, 9), (3, 12), (3, 15)]; + let prc = PrcFile::parse(&prc_file(&tess_section(&points, &normals, &corners))).unwrap(); + let meshes = assemble(&prc).unwrap(); + assert_eq!(meshes.len(), 1); + assert_eq!(meshes[0].mesh.triangles.len(), 2); + let camera = Camera::fit_meshes(&meshes, [0.0, 0.0, -1.0], [0.0, 1.0, 0.0], false, 1.0).unwrap(); + let image = render(&meshes, &camera, &RenderOptions { width: 40, height: 40, ..RenderOptions::default() }).unwrap(); + assert_eq!(pixel(&image, 1, 1), [255, 255, 255, 255], "background corner"); + assert_eq!(pixel(&image, 20, 20), [190, 192, 200, 255], "the nearer triangle, lit head on"); + // Below the near triangle only the far one: 0.3 of the surface colour. + assert_eq!(pixel(&image, 20, 34), [57, 58, 60, 255], "the farther triangle, lit edge on"); +} + +#[test] +fn damaged_streams_are_errors_not_panics() { + let points = [[0., 0., 0.], [1., 0., 0.], [0., 1., 0.]]; + let bytes = prc_file(&tess_section(&points, &[[0., 0., 1.]], &[(0, 0), (0, 3), (0, 6)])); + for end in 0..bytes.len() { + if let Ok(prc) = PrcFile::parse(&bytes[..end]) { + let _ = assemble(&prc); + } + } +} diff --git a/xtask/src/layers.rs b/xtask/src/layers.rs index 4f59c1dc94..923b7121ab 100644 --- a/xtask/src/layers.rs +++ b/xtask/src/layers.rs @@ -45,6 +45,7 @@ pub const TABLE: &[(&str, Class)] = &[ ("ops", Class::Layer(2)), // L3 core services ("render", Class::Layer(3)), + ("prc-render", Class::Layer(3)), ("text", Class::Layer(3)), ("annot", Class::Layer(3)), ("forms", Class::Layer(3)), From ce393b276be374a8cb27fbc0f6ce574862c3b4a1 Mon Sep 17 00:00:00 2001 From: KenM76 Date: Tue, 3 Nov 2026 21:41:05 -0500 Subject: [PATCH 24/53] M12: prc-render review fixes (missing tessellations, vertex ceiling, seams, perspective shade) - assemble: a placement naming a tessellation its file structure does not have is now PrcError::Malformed instead of a silently missing part; wires, markups and compressed tessellations are still skipped. - assemble: positions, normals and texture coordinates are charged against a new MAX_ASSEMBLED_VERTICES (2^24) with RenderError::TooManyVertices, so a one-triangle mesh with many vertices placed many times cannot outgrow the triangle ceiling's memory bound. A refused mesh spends nothing. - raster: a field of view too small for a finite focal length is a Camera error, not a background-only image. - raster: a pixel centre exactly on an edge belongs to one triangle only (the side the triangle lies on, +x first, then +y), so translucent quads no longer blend their diagonal twice. - raster: under a perspective camera the shade is interpolated with the 1/z depth key, so it follows the surface rather than the screen. - vec3: normalize divides by the largest component first, so a cross product above ~1e154 (or a subnormal one) still normalizes. - render docs and README state the near plane: 1/10,000 of the eye-to-target distance. Each new test fails with its fix reverted. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SqmHYx8FiWM5PcsoiErgp9 --- crates/prc-render/README.md | 4 +- crates/prc-render/src/assemble.rs | 38 ++++++++++---- crates/prc-render/src/assemble/tests.rs | 42 ++++++++++++--- crates/prc-render/src/lib.rs | 4 +- crates/prc-render/src/raster.rs | 68 ++++++++++++++++++++++--- crates/prc-render/src/render.rs | 13 +++-- crates/prc-render/src/render/tests.rs | 1 + crates/prc-render/src/vec3.rs | 15 +++++- 8 files changed, 154 insertions(+), 31 deletions(-) diff --git a/crates/prc-render/README.md b/crates/prc-render/README.md index 43730040f2..c3d4db867e 100644 --- a/crates/prc-render/README.md +++ b/crates/prc-render/README.md @@ -21,8 +21,8 @@ render(&[WorldMesh], &Camera, &RenderOptions) -> Result - A triangle's own colour wins over its mesh's, which wins over `RenderOptions::colour`. - Opaque triangles are drawn first, then translucent ones blended over them; alpha 0 is not drawn. - `Lighting::Headlight` (default) shades by the angle to the view direction; `Lighting::Unlit` draws flat colour. -- Perspective triangles are clipped at a near plane; orthographic views are not clipped. +- Perspective triangles are clipped at a near plane 1/10,000 of the eye-to-target distance from the eye; orthographic views are not clipped. - `MAX_RENDER_PIXELS` (2^24, about 16 MP) bounds the image: at most 64 MB of RGBA plus 128 MB of depth. - `MAX_ASSEMBLED_TRIANGLES` (2^24) bounds one `assemble` call. Neither bounds time: a model of many large + `MAX_ASSEMBLED_TRIANGLES` and `MAX_ASSEMBLED_VERTICES` (2^24 each) bound one `assemble` call. None bounds time: a model of many large triangles at full size is a long CPU render, so callers on a UI thread should render small or off-thread. - Not yet: shadows, ambient occlusion, specular light, scene lights, textures, wires and markups. diff --git a/crates/prc-render/src/assemble.rs b/crates/prc-render/src/assemble.rs index 3f8fa54caf..7fbb9bd914 100644 --- a/crates/prc-render/src/assemble.rs +++ b/crates/prc-render/src/assemble.rs @@ -8,6 +8,9 @@ use pdfcraft_prc::{Matrix, PrcFile, Rgba, Tessellation, TriangleMesh}; /// occurrences draw is copied once per occurrence. pub const MAX_ASSEMBLED_TRIANGLES: usize = 1 << 24; +/// The most positions, normals and texture coordinates [`assemble`] copies, over every placement. +pub const MAX_ASSEMBLED_VERTICES: usize = 1 << 24; + /// A triangle mesh in world coordinates with the colours it draws in. #[derive(Debug, Clone, Default, PartialEq)] pub struct WorldMesh { @@ -44,11 +47,12 @@ pub fn place(mesh: &TriangleMesh, matrix: &Matrix, colour: Option<[u8; 4]>, tria /// are, uncoloured. Wires, markups and compressed tessellations are not drawn. /// /// # Errors -/// [`RenderError::Model`] when a tessellation section or the tree cannot be read; -/// [`RenderError::TooManyTriangles`] past [`MAX_ASSEMBLED_TRIANGLES`]. +/// [`RenderError::Model`] when a tessellation section or the tree cannot be read, or the tree +/// places a tessellation its file structure does not have; [`RenderError::TooManyTriangles`] past +/// [`MAX_ASSEMBLED_TRIANGLES`] and [`RenderError::TooManyVertices`] past [`MAX_ASSEMBLED_VERTICES`]. pub fn assemble(prc: &PrcFile) -> Result, RenderError> { let tessellations = prc.file_structures.iter().map(|fs| fs.tessellations()).collect::, _>>()?; - let mut budget = Budget(0); + let mut budget = Budget::default(); let mut out = Vec::new(); if !prc.stores_tree() { for mesh in tessellations.iter().flatten().filter_map(as_mesh) { @@ -81,8 +85,9 @@ pub(crate) struct Item<'a> { /// Per triangle of a mesh, the colour its face's style resolves to. type FaceColours<'a> = dyn Fn(&TriangleMesh) -> Option>> + 'a; -/// Each item's mesh moved into the world and coloured, appended to `out`. Items naming no -/// triangle mesh are skipped. +/// Each item's mesh moved into the world and coloured, appended to `out`. Items naming a wire, +/// markup or compressed tessellation are skipped; an item naming no tessellation at all is +/// [`PrcError::Malformed`](pdfcraft_prc::PrcError::Malformed). pub(crate) fn place_items<'a>( tessellations: &[Vec], items: impl IntoIterator>, @@ -90,7 +95,11 @@ pub(crate) fn place_items<'a>( out: &mut Vec, ) -> Result<(), RenderError> { for p in items { - let Some(mesh) = tessellations.get(p.file_structure).and_then(|t| t.get(p.tessellation)).and_then(as_mesh) else { + let Some(tessellation) = tessellations.get(p.file_structure).and_then(|t| t.get(p.tessellation)) else { + let what = format!("tessellation {} of file structure {}, which has none by that index", p.tessellation, p.file_structure); + return Err(pdfcraft_prc::PrcError::Malformed(what).into()); + }; + let Some(mesh) = as_mesh(tessellation) else { continue; }; budget.spend(mesh)?; @@ -100,12 +109,23 @@ pub(crate) fn place_items<'a>( Ok(()) } -/// Triangles placed so far. -pub(crate) struct Budget(pub(crate) usize); +/// Triangles and vertex entries (positions, normals, texture coordinates) placed so far. +#[derive(Debug, Default)] +pub(crate) struct Budget { + pub(crate) triangles: usize, + pub(crate) vertices: usize, +} impl Budget { fn spend(&mut self, mesh: &TriangleMesh) -> Result<(), RenderError> { - self.0 = self.0.checked_add(mesh.triangles.len()).filter(|&n| n <= MAX_ASSEMBLED_TRIANGLES).ok_or(RenderError::TooManyTriangles)?; + let triangles = + self.triangles.checked_add(mesh.triangles.len()).filter(|&n| n <= MAX_ASSEMBLED_TRIANGLES).ok_or(RenderError::TooManyTriangles)?; + let vertices = [mesh.positions.len(), mesh.normals.len(), mesh.uvs.len()] + .into_iter() + .try_fold(self.vertices, usize::checked_add) + .filter(|&n| n <= MAX_ASSEMBLED_VERTICES) + .ok_or(RenderError::TooManyVertices)?; + (self.triangles, self.vertices) = (triangles, vertices); Ok(()) } } diff --git a/crates/prc-render/src/assemble/tests.rs b/crates/prc-render/src/assemble/tests.rs index 87c133f4da..6947e78033 100644 --- a/crates/prc-render/src/assemble/tests.rs +++ b/crates/prc-render/src/assemble/tests.rs @@ -47,13 +47,29 @@ fn colours_are_clamped_into_bytes() { #[test] fn the_triangle_budget_is_shared_and_refuses_past_the_ceiling() { - let mut b = Budget(MAX_ASSEMBLED_TRIANGLES - 1); + let mut b = Budget { triangles: MAX_ASSEMBLED_TRIANGLES - 1, vertices: 0 }; assert!(b.spend(&triangle()).is_ok()); assert_eq!(b.spend(&triangle()), Err(RenderError::TooManyTriangles)); - let mut b = Budget(usize::MAX); + let mut b = Budget { triangles: usize::MAX, vertices: 0 }; assert_eq!(b.spend(&triangle()), Err(RenderError::TooManyTriangles)); } +#[test] +fn the_vertex_budget_counts_positions_normals_and_uvs() { + let mut heavy = triangle(); + heavy.normals = vec![[0.0, 0.0, 1.0]; 2]; + heavy.uvs = vec![[0.0; 2]; 4]; + let mut b = Budget::default(); + b.spend(&heavy).unwrap(); + assert_eq!((b.triangles, b.vertices), (1, 9)); + // One triangle over many vertices: the triangle ceiling alone would let it through. + let mut b = Budget { triangles: 0, vertices: MAX_ASSEMBLED_VERTICES - 8 }; + assert_eq!(b.spend(&heavy), Err(RenderError::TooManyVertices)); + assert_eq!((b.triangles, b.vertices), (0, MAX_ASSEMBLED_VERTICES - 8), "a refused mesh spends nothing"); + let mut b = Budget { triangles: 0, vertices: usize::MAX }; + assert_eq!(b.spend(&triangle()), Err(RenderError::TooManyVertices)); +} + fn item<'a>(fs: usize, tess: usize, x: f64, colour: Option) -> Item<'a> { let mut matrix = Matrix::IDENTITY; matrix.0[0][3] = x; @@ -66,22 +82,32 @@ fn each_placement_draws_its_tessellation_where_and_in_the_colour_the_tree_says() let mut red = Rgba::default(); red.r = 1.0; red.a = 1.0; - // Two occurrences of one tessellation at different places, one coloured; one naming nothing. - let items = [item(0, 0, 3.0, Some(red)), item(0, 0, -4.0, None), item(1, 0, 0.0, None), item(0, 7, 0.0, None)]; - let (mut budget, mut out) = (Budget(0), Vec::new()); + // Two occurrences of one tessellation at different places, one coloured. + let items = [item(0, 0, 3.0, Some(red)), item(0, 0, -4.0, None)]; + let (mut budget, mut out) = (Budget::default(), Vec::new()); place_items(&tessellations, items, &mut budget, &mut out).unwrap(); - assert_eq!(out.len(), 2, "items naming no mesh are skipped"); + assert_eq!(out.len(), 2); assert_eq!(out[0].mesh.positions, [[3.0, 0.0, 0.0], [4.0, 0.0, 0.0], [3.0, 1.0, 0.0]]); assert_eq!(out[0].colour, Some([255, 0, 0, 255])); assert_eq!(out[1].mesh.positions, [[-4.0, 0.0, 0.0], [-3.0, 0.0, 0.0], [-4.0, 1.0, 0.0]]); assert_eq!(out[1].colour, None); - assert_eq!(budget.0, 2, "each occurrence spends its triangles"); + assert_eq!(budget.triangles, 2, "each occurrence spends its triangles"); +} + +#[test] +fn a_placement_naming_a_missing_tessellation_is_malformed() { + let tessellations = vec![vec![Tessellation::Mesh(triangle())]]; + for (fs, tess) in [(1, 0), (0, 7)] { + let (mut budget, mut out) = (Budget::default(), Vec::new()); + let r = place_items(&tessellations, [item(fs, tess, 0.0, None)], &mut budget, &mut out); + assert!(matches!(r, Err(RenderError::Model(pdfcraft_prc::PrcError::Malformed(_)))), "{fs}/{tess}: {r:?}"); + } } #[test] fn a_placement_past_the_triangle_budget_is_an_error() { let tessellations = vec![vec![Tessellation::Mesh(triangle())]]; - let (mut budget, mut out) = (Budget(MAX_ASSEMBLED_TRIANGLES), Vec::new()); + let (mut budget, mut out) = (Budget { triangles: MAX_ASSEMBLED_TRIANGLES, vertices: 0 }, Vec::new()); let r = place_items(&tessellations, [item(0, 0, 0.0, None)], &mut budget, &mut out); assert!(matches!(r, Err(RenderError::TooManyTriangles))); assert!(out.is_empty()); diff --git a/crates/prc-render/src/lib.rs b/crates/prc-render/src/lib.rs index da0ed41230..4044f110bb 100644 --- a/crates/prc-render/src/lib.rs +++ b/crates/prc-render/src/lib.rs @@ -4,7 +4,7 @@ //! into world-space meshes; [`render`] draws them from a [`Camera`] with one depth-tested sample per //! pixel centre and a light at the camera. Single-threaded and GUI-free, so it runs unchanged on wasm32. //! The image size is capped by [`MAX_RENDER_PIXELS`] and the assembled model by -//! [`MAX_ASSEMBLED_TRIANGLES`]; bad input is an error, never a panic. +//! [`MAX_ASSEMBLED_TRIANGLES`] and [`MAX_ASSEMBLED_VERTICES`]; bad input is an error, never a panic. #![forbid(unsafe_code)] #![deny(clippy::unwrap_used, clippy::expect_used, clippy::panic, clippy::unimplemented, clippy::todo, clippy::unreachable)] @@ -13,5 +13,5 @@ mod raster; mod render; mod vec3; -pub use assemble::{MAX_ASSEMBLED_TRIANGLES, WorldMesh, assemble, place}; +pub use assemble::{MAX_ASSEMBLED_TRIANGLES, MAX_ASSEMBLED_VERTICES, WorldMesh, assemble, place}; pub use render::{Bounds, Camera, Image, Lighting, MAX_RENDER_PIXELS, Projection, RenderError, RenderOptions, render}; diff --git a/crates/prc-render/src/raster.rs b/crates/prc-render/src/raster.rs index dddb3b394d..cced24db90 100644 --- a/crates/prc-render/src/raster.rs +++ b/crates/prc-render/src/raster.rs @@ -13,6 +13,9 @@ pub(crate) struct Corner { pub(crate) shade: [f64; 3], } +/// The perspective near plane, as a fraction of the eye-to-target distance. +pub(crate) const NEAR_FRACTION: f64 = 1e-4; + pub(crate) enum View { /// `focal` = 1 / tan(fov_y / 2); `near` = the clip distance. Perspective { @@ -35,8 +38,13 @@ impl View { if !(fov_y > 0.0 && fov_y < 180.0) { return Err(RenderError::Camera("the field of view is not between 0 and 180 degrees")); } + let focal = 1.0 / (fov_y.to_radians() / 2.0).tan(); + // A field of view too small to measure leaves no finite focal length. + if !focal.is_finite() { + return Err(RenderError::Camera("the field of view is too small")); + } let distance = length(sub(camera.target, camera.eye)); - Ok(View::Perspective { focal: 1.0 / (fov_y.to_radians() / 2.0).tan(), aspect, near: (distance * 1e-4).max(f64::MIN_POSITIVE) }) + Ok(View::Perspective { focal, aspect, near: (distance * NEAR_FRACTION).max(f64::MIN_POSITIVE) }) } Projection::Orthographic { height } => { if !(height.is_finite() && height > 0.0) { @@ -87,6 +95,8 @@ impl View { /// The image being drawn: `rgba` holds `width * height` straight-RGBA pixels and `depth` their /// depth keys, row by row from the top. pub(crate) struct Target<'b> { + /// Depth keys are 1/z (a perspective view), so the shade is interpolated perspective-correct. + pub(crate) perspective: bool, pub(crate) width: u32, pub(crate) height: u32, pub(crate) rgba: &'b mut [u8], @@ -95,11 +105,13 @@ pub(crate) struct Target<'b> { impl Target<'_> { /// Fill the screen triangle of `corners` with straight-RGBA `colour`, depth-tested at pixel - /// centres, the shade interpolated across it. An opaque pixel replaces the pixel and its - /// depth; a translucent one is blended over it, leaving the depth. + /// centres, the shade interpolated across it. A pixel centre on an edge two triangles share + /// belongs to one of them only. An opaque pixel replaces the pixel and its depth; a + /// translucent one is blended over it, leaving the depth. pub(crate) fn fill(&mut self, corners: [Corner; 3], colour: [u8; 4]) { let [a, b, c] = corners; let area = edge(a.at, b.at, c.at); + let edges = [(b.at, c.at), (c.at, a.at), (a.at, b.at)]; if !area.is_finite() || area.abs() < 1e-12 { return; } @@ -109,8 +121,8 @@ impl Target<'_> { for y in y0..=y1 { for x in x0..=x1 { let p = [f64::from(x) + 0.5, f64::from(y) + 0.5, 0.0]; - let w = [edge(b.at, c.at, p), edge(c.at, a.at, p), edge(a.at, b.at, p)].map(|e| e / area); - if w.iter().any(|&k| k < 0.0) { + let w = edges.map(|(from, to)| edge(from, to, p) / area); + if w.iter().zip(edges).any(|(&k, (from, to))| k < 0.0 || (k == 0.0 && !owns(from, to, area))) { continue; } let mix = |f: &dyn Fn(&Corner) -> f64| w.iter().zip(&corners).map(|(k, c)| k * f(c)).sum::(); @@ -121,7 +133,14 @@ impl Target<'_> { if self.depth.get(i).is_none_or(|&d| key <= d) { continue; } - let shade = if a.shade == b.shade && b.shade == c.shade { a.shade } else { [0, 1, 2].map(|j| mix(&|c| at(c.shade, j))) }; + let shade = if a.shade == b.shade && b.shade == c.shade { + a.shade + } else if self.perspective && key > 0.0 { + // Screen weights times 1/z, renormalised, are the surface's own weights. + [0, 1, 2].map(|j| mix(&|c| at(c.shade, j) * at(c.at, 2)) / key) + } else { + [0, 1, 2].map(|j| mix(&|c| at(c.shade, j))) + }; self.blend(i, key, shaded(colour, shade)); } } @@ -193,6 +212,15 @@ pub(crate) fn edge(a: [f64; 3], b: [f64; 3], p: [f64; 3]) -> f64 { (at(b, 0) - at(a, 0)) * (at(p, 1) - at(a, 1)) - (at(b, 1) - at(a, 1)) * (at(p, 0) - at(a, 0)) } +/// Whether a pixel centre exactly on the edge `from`-`to` of a triangle of signed doubled area +/// `area` is the triangle's: the side the triangle grows towards, +x first, then +y. Two +/// triangles sharing the edge lie on opposite sides of it, so exactly one owns it. +fn owns(from: [f64; 3], to: [f64; 3], area: f64) -> bool { + let s = area.signum(); + let (gx, gy) = (-(at(to, 1) - at(from, 1)) * s, (at(to, 0) - at(from, 0)) * s); + gx > 0.0 || (gx == 0.0 && gy > 0.0) +} + /// Component `i` of `v`, 0.0 out of range. pub(crate) fn at(v: [f64; 3], i: usize) -> f64 { v.get(i).copied().unwrap_or(0.0) @@ -223,4 +251,32 @@ mod tests { assert_eq!(channel(-4.0), 0); assert_eq!(channel(127.5), 128); } + + fn corner(x: f64, y: f64, key: f64, shade: f64) -> Corner { + Corner { at: [x, y, key], shade: [shade; 3] } + } + + #[test] + fn a_pixel_on_an_edge_two_triangles_share_is_blended_once() { + let (mut rgba, mut depth) = (vec![0; 64], vec![f64::NEG_INFINITY; 16]); + let mut t = Target { perspective: false, width: 4, height: 4, rgba: &mut rgba, depth: &mut depth }; + let [a, b, c, d] = [(0.0, 0.0), (4.0, 0.0), (4.0, 4.0), (0.0, 4.0)].map(|(x, y)| corner(x, y, 1.0, 1.0)); + t.fill([a, b, c], [255, 0, 0, 128]); + t.fill([a, c, d], [255, 0, 0, 128]); + // The diagonal runs through four pixel centres; each must look like the rest. + for px in rgba.chunks(4) { + assert_eq!(px, [255, 0, 0, 128]); + } + } + + #[test] + fn a_perspective_shade_is_interpolated_on_the_surface_not_the_screen() { + let (mut rgba, mut depth) = (vec![0; 64], vec![f64::NEG_INFINITY; 16]); + let mut t = Target { perspective: true, width: 4, height: 4, rgba: &mut rgba, depth: &mut depth }; + // The x = 4 corner is three times as far away as the others, and the only lit one. + t.fill([corner(0.0, 0.0, 1.0, 0.0), corner(4.0, 0.0, 1.0 / 3.0, 1.0), corner(0.0, 4.0, 1.0, 0.0)], [240, 240, 240, 255]); + // At screen (1.5, 0.5) the weights are 0.5, 0.375, 0.125: on the surface the far corner + // counts 0.125 / 0.75 = 1/6, where on the screen it would count 0.375. + assert_eq!(rgba.get(4..8), Some(&[40, 40, 40, 255][..])); + } } diff --git a/crates/prc-render/src/render.rs b/crates/prc-render/src/render.rs index 146266434b..d9072a3055 100644 --- a/crates/prc-render/src/render.rs +++ b/crates/prc-render/src/render.rs @@ -242,6 +242,10 @@ pub enum RenderError { /// The placed model has more than [`crate::MAX_ASSEMBLED_TRIANGLES`] triangles. #[error("the placed model has more triangles than the assembly ceiling")] TooManyTriangles, + /// The placed model has more than [`crate::MAX_ASSEMBLED_VERTICES`] positions, normals and + /// texture coordinates. + #[error("the placed model has more vertices than the assembly ceiling")] + TooManyVertices, } /// Draw `meshes` as seen by `camera`. @@ -251,12 +255,14 @@ pub enum RenderError { /// nearer surfaces hide farther ones. Opaque triangles are drawn first; translucent ones are then /// blended over them, hidden by nearer opaque surfaces but not by each other, in mesh order; fully /// transparent ones are not drawn. Triangles with a non-finite or out-of-range vertex are skipped. -/// A perspective camera clips what lies behind it. +/// A perspective camera clips at a near plane 1/10,000 of the way from the eye to the target, so +/// what lies behind the eye, or in front of it but nearer than that, is not drawn; an orthographic +/// camera does not clip. /// /// # Errors /// [`RenderError::Size`] for an empty or over-ceiling image; [`RenderError::Camera`] for a camera /// that cannot form a view (eye on the target, `up` parallel to the view, a field of view outside -/// 0–180°, a non-positive orthographic height). +/// 0–180° or too small for a finite focal length, a non-positive orthographic height). pub fn render(meshes: &[WorldMesh], camera: &Camera, options: &RenderOptions) -> Result { let (width, height) = (options.width, options.height); let size = RenderError::Size { width, height }; @@ -266,10 +272,11 @@ pub fn render(meshes: &[WorldMesh], camera: &Camera, options: &RenderOptions) -> } let pixels = usize::try_from(pixels).map_err(|_| size)?; let view = View::new(camera, f64::from(width) / f64::from(height))?; + let perspective = matches!(view, View::Perspective { .. }); let scene = Scene { eye: camera.eye, basis: camera.basis()?, view, lighting: options.lighting }; let mut rgba = options.background.repeat(pixels); let mut depth = vec![f64::NEG_INFINITY; pixels]; - let mut target = Target { width, height, rgba: &mut rgba, depth: &mut depth }; + let mut target = Target { perspective, width, height, rgba: &mut rgba, depth: &mut depth }; let [r, g, b] = options.colour; for opaque in [true, false] { for m in meshes { diff --git a/crates/prc-render/src/render/tests.rs b/crates/prc-render/src/render/tests.rs index 008f177491..8d0764315d 100644 --- a/crates/prc-render/src/render/tests.rs +++ b/crates/prc-render/src/render/tests.rs @@ -236,6 +236,7 @@ fn bad_sizes_and_cameras_are_refused() { Camera { up: [0.0, 0.0, 1.0], ..ortho(1.0) }, Camera { projection: Projection::Perspective { fov_y: 180.0 }, ..ortho(1.0) }, Camera { projection: Projection::Perspective { fov_y: f64::NAN }, ..ortho(1.0) }, + Camera { projection: Projection::Perspective { fov_y: 1e-310 }, ..ortho(1.0) }, ortho(0.0), ortho(f64::INFINITY), ]; diff --git a/crates/prc-render/src/vec3.rs b/crates/prc-render/src/vec3.rs index 90af0f3ba2..8ebf844087 100644 --- a/crates/prc-render/src/vec3.rs +++ b/crates/prc-render/src/vec3.rs @@ -28,8 +28,14 @@ pub(crate) fn length(a: Vec3) -> f64 { dot(a, a).sqrt() } -/// `a` scaled to unit length; `None` when its length is zero or not finite. +/// `a` scaled to unit length; `None` when its length is zero or not finite. `a` is first divided +/// by its largest component, so squaring cannot overflow or underflow. pub(crate) fn normalize(a: Vec3) -> Option { + let big = a.iter().fold(0.0_f64, |m, c| m.max(c.abs())); + if !(big.is_finite() && big > 0.0) { + return None; + } + let a = a.map(|c| c / big); let l = length(a); (l.is_finite() && l > 0.0).then(|| scale(a, 1.0 / l)) } @@ -49,5 +55,12 @@ mod tests { assert_eq!(normalize([0.0, 4.0, 0.0]), Some([0.0, 1.0, 0.0])); assert_eq!(normalize([0.0; 3]), None); assert_eq!(normalize([f64::NAN, 0.0, 0.0]), None); + assert_eq!(normalize([f64::INFINITY, 0.0, 0.0]), None); + } + + #[test] + fn normalize_keeps_huge_and_tiny_vectors() { + assert_eq!(normalize([1e200, 0.0, 0.0]), Some([1.0, 0.0, 0.0])); + assert_eq!(normalize([0.0, -3e-320, 0.0]), Some([0.0, -1.0, 0.0])); } } From ae1b28cf1dc9d8e133344ff43161111a4b685555 Mon Sep 17 00:00:00 2001 From: KenM76 Date: Tue, 3 Nov 2026 21:44:55 -0500 Subject: [PATCH 25/53] M12: model3d_render, a 3D annotation's PRC model drawn to a PNG pdfcraft-annot reads a page's 3D annotations and the model each shows: the /3DD stream, or the stream a /3DRef names (at most four deep, so a cycle ends), decoded within 256 MB and typed PRC or U3D by /Subtype. pdfcraft-engine exposes both on Document. The automation tool model3d_render picks a page's Nth 3D model, parses PRC with pdfcraft-prc, assembles and frames it with pdfcraft-prc-render from a chosen direction (default an isometric-style view, Z up), and returns the image as a PNG. U3D gives an error until a decoder lands. parity misc.3d-view moves to partial. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SqmHYx8FiWM5PcsoiErgp9 --- Cargo.lock | 2 + crates/annot/src/lib.rs | 2 + crates/annot/src/model3d.rs | 86 ++++++++++++++ crates/annot/src/model3d/tests.rs | 84 ++++++++++++++ crates/automation/Cargo.toml | 2 + crates/automation/src/dispatch.rs | 1 + crates/automation/src/lib.rs | 1 + crates/automation/src/model3d_render.rs | 106 ++++++++++++++++++ crates/automation/src/tools.rs | 7 ++ crates/automation/tests/it/main.rs | 1 + crates/automation/tests/it/model3d_render.rs | 112 +++++++++++++++++++ crates/engine/src/lib.rs | 1 + crates/engine/src/model3d.rs | 19 ++++ parity/acrobat-features.toml | 8 +- 14 files changed, 428 insertions(+), 4 deletions(-) create mode 100644 crates/annot/src/model3d.rs create mode 100644 crates/annot/src/model3d/tests.rs create mode 100644 crates/automation/src/model3d_render.rs create mode 100644 crates/automation/tests/it/model3d_render.rs create mode 100644 crates/engine/src/model3d.rs diff --git a/Cargo.lock b/Cargo.lock index bc76adfde5..72be32bc2b 100644 --- a/Cargo.lock +++ b/Cargo.lock @@ -4277,6 +4277,8 @@ dependencies = [ "pdfcraft-model", "pdfcraft-organize", "pdfcraft-platform", + "pdfcraft-prc", + "pdfcraft-prc-render", "pdfcraft-render", "pdfcraft-xfa", "png 0.18.1", diff --git a/crates/annot/src/lib.rs b/crates/annot/src/lib.rs index 7bd0894c66..0c83538ff8 100644 --- a/crates/annot/src/lib.rs +++ b/crates/annot/src/lib.rs @@ -22,6 +22,7 @@ mod create; mod draw; mod info; mod lookup; +mod model3d; mod modify; mod types; @@ -36,6 +37,7 @@ use lookup::{annots, page_ref}; pub use lookup::{ has_visible_fill_sign, is_fill_sign, isolate_annotation, isolate_annotations, page_refs, picture_rotation, set_annots, signature_image, }; +pub use model3d::{MAX_MODEL_BYTES, Model3d, Model3dKind, model_3d, model_3d_annotations}; pub(crate) use modify::n; pub use modify::{ add_reply, add_text_replacement, deletable, delete_annotation, erase_ink, fillable, goes_with, lift_annotation, move_annotation, set_contents, diff --git a/crates/annot/src/model3d.rs b/crates/annot/src/model3d.rs new file mode 100644 index 0000000000..66a755b862 --- /dev/null +++ b/crates/annot/src/model3d.rs @@ -0,0 +1,86 @@ +//! Reading the model a 3D annotation shows (ISO 32000-2 §13.6.3 and §13.6.4): its `/3DD` is a +//! 3D stream, or a 3D reference dictionary naming the stream another annotation shares. + +use super::lookup::{annots, page_ref}; +use super::{AnnotError, Dict, Document, Object}; + +/// The largest decoded 3D stream [`model_3d`] reads, in bytes. +pub const MAX_MODEL_BYTES: usize = 256 << 20; + +/// How many `/3DRef` dictionaries [`model_3d`] follows before giving up: the standard allows one, +/// so a longer chain is damage, and a cycle ends here too. +const MAX_REF_DEPTH: usize = 4; + +/// The format of a 3D stream, from its `/Subtype`. +#[derive(Debug, Clone, Copy, PartialEq, Eq)] +pub enum Model3dKind { + /// Product Representation Compact (ISO 14739-1). + Prc, + /// Universal 3D (ECMA-363). + U3d, +} + +/// A 3D annotation's model: its format and the decoded stream. +#[derive(Debug, Clone, PartialEq)] +pub struct Model3d { + pub kind: Model3dKind, + pub data: Vec, +} + +/// The `/Annots` positions of the 3D annotations on 0-based `page`, in order. +/// +/// # Errors +/// [`AnnotError::NoPageTree`] or [`AnnotError::NoSuchPage`]. +pub fn model_3d_annotations(doc: &Document, page: usize) -> Result, AnnotError> { + let list = annots(doc, page_ref(doc, page)?); + Ok(list.iter().enumerate().filter(|(_, a)| doc.dict(a).is_some_and(|d| d.name(b"Subtype") == Some(b"3D"))).map(|(i, _)| i).collect()) +} + +/// The model of the 3D annotation at `/Annots` position `index` on 0-based `page`, decoded. +/// +/// # Errors +/// [`AnnotError::NoSuchAnnotation`] when there is no annotation there; [`AnnotError::Invalid`] +/// when it is not a 3D annotation, has no readable 3D stream, names a format other than PRC or +/// U3D, or decodes past [`MAX_MODEL_BYTES`]. +pub fn model_3d(doc: &Document, page: usize, index: usize) -> Result { + let list = annots(doc, page_ref(doc, page)?); + let entry = list.get(index).ok_or(AnnotError::NoSuchAnnotation { page, index })?; + let annot = doc.dict(entry).ok_or(AnnotError::NoSuchAnnotation { page, index })?; + let at = || format!("annotation {} on page {}", index.saturating_add(1), page.saturating_add(1)); + if annot.name(b"Subtype") != Some(b"3D") { + return Err(AnnotError::Invalid(format!("{} is not a 3D annotation", at()))); + } + let stream = stream_of(doc, &annot).ok_or_else(|| AnnotError::Invalid(format!("{} has no 3D stream", at())))?; + let Object::Stream(s) = stream.as_ref() else { + return Err(AnnotError::Invalid(format!("{} has no 3D stream", at()))); + }; + let kind = match s.dict.name(b"Subtype") { + Some(b"PRC") => Model3dKind::Prc, + Some(b"U3D") => Model3dKind::U3d, + other => { + let name = other.map_or_else(|| "none".to_owned(), |n| String::from_utf8_lossy(n).into_owned()); + return Err(AnnotError::Invalid(format!("{}'s 3D stream is of an unknown format ({name})", at()))); + } + }; + let data = doc + .with_direct_parms(s) + .decoded_within(MAX_MODEL_BYTES) + .map_err(|e| AnnotError::Invalid(format!("{}'s 3D stream could not be decoded: {e}", at())))?; + Ok(Model3d { kind, data }) +} + +/// The 3D stream an annotation's `/3DD` names, through at most [`MAX_REF_DEPTH`] `/3DRef`s. +fn stream_of(doc: &Document, annot: &Dict) -> Option> { + let mut obj = doc.resolve(annot.get(b"3DD")?); + for _ in 0..=MAX_REF_DEPTH { + match obj.as_ref() { + Object::Stream(_) => return Some(obj), + Object::Dict(d) if d.name(b"Type") == Some(b"3DRef") => obj = doc.resolve(d.get(b"3DD")?), + _ => return None, + } + } + None +} + +#[cfg(test)] +mod tests; diff --git a/crates/annot/src/model3d/tests.rs b/crates/annot/src/model3d/tests.rs new file mode 100644 index 0000000000..a58e24c30c --- /dev/null +++ b/crates/annot/src/model3d/tests.rs @@ -0,0 +1,84 @@ +use std::sync::Arc; + +use super::*; + +/// A one-page document whose page has `annots` as its `/Annots`, plus `extra` objects from 4 on. +fn doc(annots: &str, extra: &[Vec]) -> Document { + let mut objs: Vec> = vec![ + b"<< /Type /Catalog /Pages 2 0 R >>".to_vec(), + b"<< /Type /Pages /Kids [3 0 R] /Count 1 /MediaBox [0 0 612 792] >>".to_vec(), + format!("<< /Type /Page /Parent 2 0 R /Annots {annots} >>").into_bytes(), + ]; + objs.extend(extra.iter().cloned()); + let mut out = b"%PDF-1.7\n".to_vec(); + let mut offsets = Vec::new(); + for (i, o) in objs.iter().enumerate() { + offsets.push(out.len()); + out.extend_from_slice(format!("{} 0 obj\n", i + 1).as_bytes()); + out.extend_from_slice(o); + out.extend_from_slice(b"\nendobj\n"); + } + let xref = out.len(); + out.extend_from_slice(format!("xref\n0 {}\n0000000000 65535 f \n", objs.len() + 1).as_bytes()); + for o in offsets { + out.extend_from_slice(format!("{o:010} 00000 n \n").as_bytes()); + } + out.extend_from_slice(format!("trailer\n<< /Size {} /Root 1 0 R >>\nstartxref\n{xref}\n%%EOF\n", objs.len() + 1).as_bytes()); + Document::open(Arc::new(out)).expect("opens") +} + +fn stream(dict: &str, body: &[u8]) -> Vec { + let mut s = format!("<< {dict} /Length {} >>\nstream\n", body.len()).into_bytes(); + s.extend_from_slice(body); + s.extend_from_slice(b"\nendstream"); + s +} + +fn annot(dd: &str) -> Vec { + format!("<< /Type /Annot /Subtype /3D /Rect [0 0 100 100] /3DD {dd} >>").into_bytes() +} + +#[test] +fn a_plain_3d_stream_is_read_through_its_filters() { + let d = doc("[4 0 R]", &[annot("5 0 R"), stream("/Type /3D /Subtype /PRC /Filter /ASCIIHexDecode", b"505243>")]); + assert_eq!(model_3d(&d, 0, 0), Ok(Model3d { kind: Model3dKind::Prc, data: b"PRC".to_vec() })); +} + +#[test] +fn a_3dref_leads_to_the_stream_it_names() { + let d = doc("[4 0 R]", &[annot("<< /Type /3DRef /3DD 5 0 R >>"), stream("/Type /3D /Subtype /PRC", b"PRC")]); + assert_eq!(model_3d(&d, 0, 0).map(|m| m.data), Ok(b"PRC".to_vec())); +} + +#[test] +fn a_u3d_stream_is_read_and_says_so() { + let d = doc("[4 0 R]", &[annot("5 0 R"), stream("/Type /3D /Subtype /U3D", b"U3D")]); + assert_eq!(model_3d(&d, 0, 0).map(|m| m.kind), Ok(Model3dKind::U3d)); +} + +#[test] +fn a_3d_annotation_without_a_stream_is_invalid() { + let d = doc("[<< /Type /Annot /Subtype /3D /Rect [0 0 1 1] >>]", &[]); + assert!(matches!(model_3d(&d, 0, 0), Err(AnnotError::Invalid(m)) if m.contains("no 3D stream"))); +} + +#[test] +fn an_unknown_format_is_invalid() { + let d = doc("[4 0 R]", &[annot("5 0 R"), stream("/Type /3D /Subtype /STL", b"x")]); + assert!(matches!(model_3d(&d, 0, 0), Err(AnnotError::Invalid(m)) if m.contains("unknown format (STL)"))); +} + +#[test] +fn a_cycle_of_3drefs_ends() { + let d = doc("[4 0 R]", &[annot("5 0 R"), b"<< /Type /3DRef /3DD 6 0 R >>".to_vec(), b"<< /Type /3DRef /3DD 5 0 R >>".to_vec()]); + assert!(matches!(model_3d(&d, 0, 0), Err(AnnotError::Invalid(m)) if m.contains("no 3D stream"))); +} + +#[test] +fn another_annotation_is_not_a_3d_one() { + let d = doc("[<< /Type /Annot /Subtype /Square /Rect [0 0 1 1] >> 4 0 R]", &[annot("5 0 R"), stream("/Subtype /PRC", b"PRC")]); + assert!(matches!(model_3d(&d, 0, 0), Err(AnnotError::Invalid(m)) if m.contains("not a 3D annotation"))); + assert_eq!(model_3d(&d, 0, 2), Err(AnnotError::NoSuchAnnotation { page: 0, index: 2 })); + assert_eq!(model_3d_annotations(&d, 0), Ok(vec![1])); + assert_eq!(model_3d_annotations(&d, 1), Err(AnnotError::NoSuchPage(1))); +} diff --git a/crates/automation/Cargo.toml b/crates/automation/Cargo.toml index a6e43de1f0..a825b1fe52 100644 --- a/crates/automation/Cargo.toml +++ b/crates/automation/Cargo.toml @@ -14,6 +14,8 @@ pdfcraft-cos = { workspace = true } pdfcraft-render = { workspace = true } pdfcraft-organize = { workspace = true } pdfcraft-platform = { workspace = true } +pdfcraft-prc = { workspace = true } +pdfcraft-prc-render = { workspace = true } serde_json = { workspace = true } sha2 = "0.11" tempfile = "3.27" diff --git a/crates/automation/src/dispatch.rs b/crates/automation/src/dispatch.rs index f3c51d22e0..8269963598 100644 --- a/crates/automation/src/dispatch.rs +++ b/crates/automation/src/dispatch.rs @@ -69,6 +69,7 @@ impl Automation { self.apply(&a, edit)? } "page_render" => return self.page_render(&a), + "model3d_render" => return self.model3d_render(&a), "comment_image_preview" => return self.comment_image_preview(&a), "text_extract" => self.text_extract(&a)?, "text_word_count" => self.text_word_count(&a)?, diff --git a/crates/automation/src/lib.rs b/crates/automation/src/lib.rs index 5bcb032fce..eb58b52e75 100644 --- a/crates/automation/src/lib.rs +++ b/crates/automation/src/lib.rs @@ -27,6 +27,7 @@ mod links; #[cfg(feature = "mcp")] pub mod mcp; mod measure; +mod model3d_render; mod printing; mod reading; mod redact; diff --git a/crates/automation/src/model3d_render.rs b/crates/automation/src/model3d_render.rs new file mode 100644 index 0000000000..7f789118a5 --- /dev/null +++ b/crates/automation/src/model3d_render.rs @@ -0,0 +1,106 @@ +//! `model3d_render`: a page's 3D model (ISO 32000-2 §13.6) drawn to a PNG, without a viewer. + +use pdfcraft_engine::model3d::Model3dKind; +use pdfcraft_prc::PrcFile; +use pdfcraft_prc_render::{Camera, RenderOptions, assemble, render}; + +use super::Automation; +use super::dispatch::Content; +use super::error::{Result, ToolError, failed}; +use super::json::Args; +use super::pages::encode_png; + +/// Default and largest side of the image, in pixels. +pub(crate) const DEFAULT_SIDE: u32 = 512; +pub(crate) const MAX_SIDE: u32 = 4096; + +/// The default view: from the front, right and above, as an isometric view looks. +const DEFAULT_DIRECTION: [f64; 3] = [-1.0, 1.0, -1.0]; + +impl Automation { + pub(crate) fn model3d_render(&mut self, a: &Args) -> Result> { + let page = self.page(a)?; + let width = side(a, "width")?; + let height = side(a, "height")?; + let direction = direction(a)?; + let perspective = a.opt_bool("perspective")?.unwrap_or(true); + let doc = self.doc(a)?; + let found = doc.model_3d_annotations(page).map_err(failed)?; + let index = match a.opt_int("model")? { + None => *found.first().ok_or_else(|| failed(format!("page {} has no 3D model", page.saturating_add(1))))?, + Some(n) => { + let pos = usize::try_from(n).ok().and_then(|n| n.checked_sub(1)).ok_or_else(|| Args::wrong("model", "a positive integer"))?; + *found.get(pos).ok_or_else(|| { + ToolError::InvalidArgs(format!("page {} has {} 3D models, not {n}", page.saturating_add(1), found.len())) + })? + } + }; + let model = doc.model_3d(page, index).map_err(failed)?; + if model.kind == Model3dKind::U3d { + return Err(failed("this 3D model is in U3D, which can't be drawn yet; only PRC models can")); + } + let prc = PrcFile::parse(&model.data).map_err(|e| failed(format!("the 3D model could not be read: {e}")))?; + let meshes = assemble(&prc).map_err(failed)?; + let aspect = f64::from(width) / f64::from(height); + let camera = Camera::fit_meshes(&meshes, direction, up_for(direction), perspective, aspect).map_err(failed)?; + let image = render(&meshes, &camera, &RenderOptions { width, height, ..RenderOptions::default() }).map_err(failed)?; + let data = encode_png(image.width, image.height, &premultiplied(image.rgba))?; + Ok(vec![Content::Png { data, width: image.width, height: image.height }]) + } +} + +fn side(a: &Args, key: &str) -> Result { + match a.opt_int(key)? { + None => Ok(DEFAULT_SIDE), + Some(n) => u32::try_from(n).ok().filter(|n| (1..=MAX_SIDE).contains(n)).ok_or_else(|| Args::wrong(key, &format!("between 1 and {MAX_SIDE}"))), + } +} + +/// `direction`: three finite numbers, not all zero, from the eye towards the model. +fn direction(a: &Args) -> Result<[f64; 3]> { + let Some(v) = a.get("direction") else { return Ok(DEFAULT_DIRECTION) }; + let wrong = || Args::wrong("direction", "three numbers [x, y, z], not all zero"); + let items = v.as_array().filter(|x| x.len() == 3).ok_or_else(wrong)?; + let mut out = [0.0; 3]; + for (o, x) in out.iter_mut().zip(items) { + *o = x.as_f64().filter(|f| f.is_finite()).ok_or_else(wrong)?; + } + if out.iter().all(|c| *c == 0.0) { + return Err(wrong()); + } + Ok(out) +} + +/// Z is up, as CAD models are drawn, unless the view looks along it. +fn up_for([x, y, z]: [f64; 3]) -> [f64; 3] { + if x.hypot(y) <= z.abs() * 1e-6 { [0.0, 1.0, 0.0] } else { [0.0, 0.0, 1.0] } +} + +/// Straight RGBA to the premultiplied RGBA `encode_png` takes. +fn premultiplied(mut rgba: Vec) -> Vec { + for px in rgba.chunks_exact_mut(4) { + if let [r, g, b, alpha] = px { + let a = u16::from(*alpha); + for c in [r, g, b] { + *c = u8::try_from((u16::from(*c) * a + 127) / 255).unwrap_or(u8::MAX); + } + } + } + rgba +} + +#[cfg(test)] +mod tests { + use super::*; + + #[test] + fn premultiplying_scales_colour_by_alpha() { + assert_eq!(premultiplied(vec![200, 100, 0, 255, 200, 100, 0, 128, 9, 9, 9, 0]), vec![200, 100, 0, 255, 100, 50, 0, 128, 0, 0, 0, 0]); + } + + #[test] + fn a_view_along_z_keeps_y_up() { + assert_eq!(up_for([0.0, 0.0, -1.0]), [0.0, 1.0, 0.0]); + assert_eq!(up_for(DEFAULT_DIRECTION), [0.0, 0.0, 1.0]); + } +} diff --git a/crates/automation/src/tools.rs b/crates/automation/src/tools.rs index 4672e49886..0d8a3a6c97 100644 --- a/crates/automation/src/tools.rs +++ b/crates/automation/src/tools.rs @@ -181,6 +181,13 @@ pub fn tools() -> Vec { "hide_text_block": { "type": "integer", "minimum": 1, "description": "Render an inline editor background without this paragraph (number from text_paragraphs, resolved as text_edit does: refused once an edit merged that paragraph into a neighbour). Does not modify the document." }, "hide_added_text": { "type": "integer", "minimum": 1, "description": "Render a retyping background without this added text item (added index from object_list). Mutually exclusive with hide_text_block. Does not modify the document." } }), &["doc", "page"])), + t("model3d_render", "Render a 3D model", "Draw a page's 3D model (a 3D annotation's PRC stream) to a PNG, framed to fill the image, with headlight shading on a white background. model is its 1-based position among the page's 3D models (default 1). direction is the view direction [x, y, z] from the eye towards the model (default [-1, 1, -1]); Z is up unless the view looks along Z. perspective defaults to true. U3D models can't be drawn yet and give an error. Does not modify the document.") + .ro() + .with(schema(json!({ "doc": doc(), "page": { "type": "integer", "minimum": 1 }, "model": { "type": "integer", "minimum": 1 }, + "width": { "type": "integer", "minimum": 1, "maximum": 4096 }, "height": { "type": "integer", "minimum": 1, "maximum": 4096 }, + "direction": { "type": "array", "items": { "type": "number" }, "minItems": 3, "maxItems": 3 }, + "perspective": { "type": "boolean" } + }), &["doc", "page"])), t("object_list", "List editable objects", "List existing paragraphs, Image/Form artwork and added content once, with source kind/index references, displayed rectangles and document generation. Indexes change when content changes; use generation in object_move to reject stale references.") .ro().with(schema(json!({"doc":doc(),"page":{"type":"integer","minimum":1}}), &["doc","page"])), t("object_move", "Move objects together", "Atomically translate 1–1000 mixed references from object_list by [dx, dy] in displayed points (right/down). Preserves glyph codes, fonts, image data and relative spacing; success is one undo step. Text used as a clipping path and vertical text are refused. A bad/stale reference refuses the entire move.") diff --git a/crates/automation/tests/it/main.rs b/crates/automation/tests/it/main.rs index d36dbe93a9..d4946b6d73 100644 --- a/crates/automation/tests/it/main.rs +++ b/crates/automation/tests/it/main.rs @@ -8,6 +8,7 @@ mod batch; mod button_caption; mod form_bulk; mod hostile_args; +mod model3d_render; mod optimize_content_boundaries; mod redact_after_stray_q_form; mod redact_text_after_q; diff --git a/crates/automation/tests/it/model3d_render.rs b/crates/automation/tests/it/model3d_render.rs new file mode 100644 index 0000000000..47e757906a --- /dev/null +++ b/crates/automation/tests/it/model3d_render.rs @@ -0,0 +1,112 @@ +//! `model3d_render` draws a page's 3D model headlessly: a synthetic PDF holding a synthetic PRC. + +use std::path::PathBuf; + +use pdfcraft_automation::{Automation, Content}; +use serde_json::{Value, json}; + +/// One triangle at z = 0 facing +z, with corners (-1,-1), (1,-1) and (0,1), as a PRC file of one +/// file structure with no product tree, in hexadecimal for the stream's `/ASCIIHexDecode`. +const TRIANGLE_PRC_HEX: &str = "505243c91f0000c91f00000100000002000000030000000400000000000000000000000000000000000000010000000500000006000000070000000800000000000000060000006b0000009a000000a3000000ab000000cd000000d5000000dd000000e500000000000000505243c91f0000c91f0000050000000600000007000000080000000000000000000000000000000000000000000000789c63000000010001789c030000000001789c9b71c087758da2422b43084fe3140e038713464cd759381a162800006ed70828789c030000000001789c030000000001789c030000000001"; + +/// A one-page PDF whose page holds a 3D annotation over a stream of `/Subtype /{subtype}`. +fn pdf(subtype: &str, hex: &str) -> Vec { + let stream = format!("<< /Type /3D /Subtype /{subtype} /Filter /ASCIIHexDecode /Length {} >>\nstream\n{hex}>\nendstream", hex.len() + 1); + let objs: Vec = vec![ + "<< /Type /Catalog /Pages 2 0 R >>".into(), + "<< /Type /Pages /Kids [3 0 R] /Count 1 /MediaBox [0 0 200 200] >>".into(), + "<< /Type /Page /Parent 2 0 R /Annots [<< /Type /Annot /Subtype /Square /Rect [0 0 10 10] >> 4 0 R] >>".into(), + "<< /Type /Annot /Subtype /3D /Rect [20 20 180 180] /3DD 5 0 R >>".into(), + stream, + ]; + let mut out = b"%PDF-1.7\n".to_vec(); + let mut offsets = Vec::new(); + for (i, o) in objs.iter().enumerate() { + offsets.push(out.len()); + out.extend_from_slice(format!("{} 0 obj\n{o}\nendobj\n", i + 1).as_bytes()); + } + let xref = out.len(); + out.extend_from_slice(format!("xref\n0 {}\n0000000000 65535 f \n", objs.len() + 1).as_bytes()); + for o in offsets { + out.extend_from_slice(format!("{o:010} 00000 n \n").as_bytes()); + } + out.extend_from_slice(format!("trailer\n<< /Size {} /Root 1 0 R >>\nstartxref\n{xref}\n%%EOF\n", objs.len() + 1).as_bytes()); + out +} + +/// A fresh folder with `prc.pdf` and `u3d.pdf` in `in`, `outside.pdf` beside it, and an +/// automation rooted at `in`. +fn setup(test: &str) -> (PathBuf, Automation) { + let dir = std::env::temp_dir().join(format!("pdfcraft-model3d-{test}-{}", std::process::id())); + let _ = std::fs::remove_dir_all(&dir); + std::fs::create_dir_all(dir.join("in")).unwrap(); + std::fs::write(dir.join("in/prc.pdf"), pdf("PRC", TRIANGLE_PRC_HEX)).unwrap(); + std::fs::write(dir.join("in/u3d.pdf"), pdf("U3D", "553344")).unwrap(); + std::fs::write(dir.join("outside.pdf"), pdf("PRC", TRIANGLE_PRC_HEX)).unwrap(); + let a = Automation::new().with_root(dir.join("in")).unwrap(); + (dir, a) +} + +fn open(a: &mut Automation, path: &str) -> u64 { + match a.call("doc_open", &json!({ "path": path })).unwrap().remove(0) { + Content::Json(v) => v["doc"].as_u64().unwrap(), + other => panic!("expected JSON: {other:?}"), + } +} + +/// The RGB of pixel (x, y) of a PNG. +fn rgb(data: &[u8], x: usize, y: usize) -> [u8; 3] { + let mut reader = png::Decoder::new(std::io::Cursor::new(data)).read_info().unwrap(); + let mut pixels = vec![0; reader.output_buffer_size().unwrap()]; + let info = reader.next_frame(&mut pixels).unwrap(); + let channels = info.line_size / info.width as usize; + let at = y * info.line_size + x * channels; + [pixels[at], pixels[at + 1], pixels[at + 2]] +} + +#[test] +fn a_prc_model_is_drawn_over_the_background() { + let (_dir, mut a) = setup("prc"); + let doc = open(&mut a, "prc.pdf"); + let args = json!({ "doc": doc, "page": 1, "width": 64, "height": 48, "direction": [0, 0, -1], "perspective": false }); + let out = a.call("model3d_render", &args).unwrap(); + let Content::Png { data, width, height } = &out[0] else { panic!("expected a PNG") }; + assert_eq!((*width, *height), (64, 48)); + assert_eq!(rgb(data, 0, 0), [255, 255, 255], "the corner is background"); + assert_eq!(rgb(data, 32, 24), [190, 192, 200], "the centre is the triangle, lit head on"); + // In perspective the headlight meets the face at a slight angle, so only not-background holds. + let out = a.call("model3d_render", &json!({ "doc": doc, "page": 1, "width": 64, "height": 48, "direction": [0, 0, -1] })).unwrap(); + let Content::Png { data, .. } = &out[0] else { panic!("expected a PNG") }; + assert_ne!(rgb(data, 32, 24), [255, 255, 255], "the centre is drawn in perspective too"); + // The same model named explicitly, and the default isometric view, both draw. + assert!(a.call("model3d_render", &json!({ "doc": doc, "page": 1, "model": 1 })).is_ok()); +} + +#[test] +fn a_u3d_model_or_a_missing_one_is_a_clear_error() { + let (_dir, mut a) = setup("u3d"); + let doc = open(&mut a, "u3d.pdf"); + let err = a.call("model3d_render", &json!({ "doc": doc, "page": 1 })).unwrap_err().to_string(); + assert!(err.contains("U3D"), "{err}"); + let doc = open(&mut a, "prc.pdf"); + let err = a.call("model3d_render", &json!({ "doc": doc, "page": 1, "model": 2 })).unwrap_err().to_string(); + assert!(err.contains("has 1 3D models, not 2"), "{err}"); + for bad in [json!({ "direction": [0, 0, 0] }), json!({ "direction": [1, 2] }), json!({ "width": 0 }), json!({ "height": 5000 })] { + let mut args = json!({ "doc": doc, "page": 1 }); + args.as_object_mut().unwrap().extend(bad.as_object().unwrap().clone()); + assert!(a.call("model3d_render", &args).is_err(), "{bad}"); + } +} + +#[test] +fn a_model_outside_the_root_is_never_opened_to_be_drawn() { + let (dir, mut a) = setup("root"); + let outside = dir.join("outside.pdf"); + for path in [Value::from("../outside.pdf"), Value::from(outside.to_string_lossy().into_owned())] { + assert!(a.call("doc_open", &json!({ "path": path })).is_err(), "{path} lies outside the root"); + } + // The tool takes no path of its own: only a document opened inside the root reaches it. + assert!(a.call("model3d_render", &json!({ "doc": 1, "page": 1 })).is_err()); + let doc = open(&mut a, "prc.pdf"); + assert!(a.call("model3d_render", &json!({ "doc": doc, "page": 1 })).is_ok()); +} diff --git a/crates/engine/src/lib.rs b/crates/engine/src/lib.rs index fe139164be..3359bad37c 100644 --- a/crates/engine/src/lib.rs +++ b/crates/engine/src/lib.rs @@ -35,6 +35,7 @@ mod history; pub mod js; mod layers; pub mod links; +pub mod model3d; pub mod ocr; mod open; pub mod optimizer; diff --git a/crates/engine/src/model3d.rs b/crates/engine/src/model3d.rs new file mode 100644 index 0000000000..b95a332a22 --- /dev/null +++ b/crates/engine/src/model3d.rs @@ -0,0 +1,19 @@ +//! The 3D models on a page (ISO 32000-2 §13.6), read so a renderer can draw them. + +pub use pdfcraft_annot::{MAX_MODEL_BYTES, Model3d, Model3dKind}; + +use crate::Document; + +impl Document { + /// The `/Annots` positions of the 3D annotations on 0-based `page`, in order. + pub fn model_3d_annotations(&self, page: usize) -> Result, String> { + let editor = self.editor.as_ref().ok_or("the document can't be read")?; + pdfcraft_annot::model_3d_annotations(&editor.cos, page).map_err(|e| e.to_string()) + } + + /// The decoded model of the 3D annotation at `/Annots` position `index` on 0-based `page`. + pub fn model_3d(&self, page: usize, index: usize) -> Result { + let editor = self.editor.as_ref().ok_or("the document can't be read")?; + pdfcraft_annot::model_3d(&editor.cos, page, index).map_err(|e| e.to_string()) + } +} diff --git a/parity/acrobat-features.toml b/parity/acrobat-features.toml index 8b7ac909a0..21e208c003 100644 --- a/parity/acrobat-features.toml +++ b/parity/acrobat-features.toml @@ -9896,11 +9896,11 @@ title = "View 3D (U3D/PRC) content" area = "N" tier = "P3" milestone = "M12" -status = "planned" +status = "partial" commands = [] -tools = [] -evidence = [] -notes = "" +tools = ["model3d_render"] +evidence = ["crates/automation/tests/it/model3d_render.rs::a_prc_model_is_drawn_over_the_background", "crates/annot/src/model3d/tests.rs::a_3dref_leads_to_the_stream_it_names", "crates/prc-render/tests/model.rs::a_synthetic_prc_renders_its_nearer_triangle_over_the_farther"] +notes = "headless only: model3d_render draws a page's PRC model (through a /3DRef too) to a PNG, framed from a chosen direction, with headlight shading. Missing: the 3D view in the app (no activation, no navigation), U3D (an error), the annotation's /3DV views, lighting schemes and /RM render modes, materials and textures, compressed tessellation, picking, cross sections, the model tree, and measurements" [[feature]] id = "misc.legacy-multimedia" From fb05b733f98fa3c8e9f14d6a5782766abdc9d387 Mon Sep 17 00:00:00 2001 From: KenM76 Date: Tue, 3 Nov 2026 22:53:15 -0500 Subject: [PATCH 26/53] M12: model_render, and a 3D reference's stream under /3D MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit A 3D reference dictionary holds the shared stream under /3D (ISO 32000-2 §13.6.3), not /3DD; it is followed once, and the caller refuses anything that is not a stream, so a chain or cycle of references ends. The tool is renamed model_render, since MCP tool names here are lowercase letters and underscores only, and to match model_add. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SqmHYx8FiWM5PcsoiErgp9 --- crates/annot/src/model3d.rs | 19 ++++++------------- crates/annot/src/model3d/tests.rs | 10 +++++++--- crates/automation/README.md | 9 +++++++++ crates/automation/src/dispatch.rs | 2 +- crates/automation/src/lib.rs | 2 +- .../{model3d_render.rs => model_render.rs} | 10 +++++----- crates/automation/src/tools.rs | 2 +- crates/automation/tests/it/main.rs | 2 +- .../it/{model3d_render.rs => model_render.rs} | 18 +++++++++--------- parity/acrobat-features.toml | 6 +++--- 10 files changed, 43 insertions(+), 37 deletions(-) rename crates/automation/src/{model3d_render.rs => model_render.rs} (91%) rename crates/automation/tests/it/{model3d_render.rs => model_render.rs} (86%) diff --git a/crates/annot/src/model3d.rs b/crates/annot/src/model3d.rs index 66a755b862..60d2df0f2a 100644 --- a/crates/annot/src/model3d.rs +++ b/crates/annot/src/model3d.rs @@ -7,10 +7,6 @@ use super::{AnnotError, Dict, Document, Object}; /// The largest decoded 3D stream [`model_3d`] reads, in bytes. pub const MAX_MODEL_BYTES: usize = 256 << 20; -/// How many `/3DRef` dictionaries [`model_3d`] follows before giving up: the standard allows one, -/// so a longer chain is damage, and a cycle ends here too. -const MAX_REF_DEPTH: usize = 4; - /// The format of a 3D stream, from its `/Subtype`. #[derive(Debug, Clone, Copy, PartialEq, Eq)] pub enum Model3dKind { @@ -69,17 +65,14 @@ pub fn model_3d(doc: &Document, page: usize, index: usize) -> Result Option> { - let mut obj = doc.resolve(annot.get(b"3DD")?); - for _ in 0..=MAX_REF_DEPTH { - match obj.as_ref() { - Object::Stream(_) => return Some(obj), - Object::Dict(d) if d.name(b"Type") == Some(b"3DRef") => obj = doc.resolve(d.get(b"3DD")?), - _ => return None, - } + let obj = doc.resolve(annot.get(b"3DD")?); + match obj.as_ref() { + Object::Dict(d) if d.name(b"Type") == Some(b"3DRef") => Some(doc.resolve(d.get(b"3D")?)), + _ => Some(obj), } - None } #[cfg(test)] diff --git a/crates/annot/src/model3d/tests.rs b/crates/annot/src/model3d/tests.rs index a58e24c30c..90addb1bb6 100644 --- a/crates/annot/src/model3d/tests.rs +++ b/crates/annot/src/model3d/tests.rs @@ -46,7 +46,7 @@ fn a_plain_3d_stream_is_read_through_its_filters() { #[test] fn a_3dref_leads_to_the_stream_it_names() { - let d = doc("[4 0 R]", &[annot("<< /Type /3DRef /3DD 5 0 R >>"), stream("/Type /3D /Subtype /PRC", b"PRC")]); + let d = doc("[4 0 R]", &[annot("<< /Type /3DRef /3D 5 0 R >>"), stream("/Type /3D /Subtype /PRC", b"PRC")]); assert_eq!(model_3d(&d, 0, 0).map(|m| m.data), Ok(b"PRC".to_vec())); } @@ -69,8 +69,12 @@ fn an_unknown_format_is_invalid() { } #[test] -fn a_cycle_of_3drefs_ends() { - let d = doc("[4 0 R]", &[annot("5 0 R"), b"<< /Type /3DRef /3DD 6 0 R >>".to_vec(), b"<< /Type /3DRef /3DD 5 0 R >>".to_vec()]); +fn a_3dref_is_followed_once_and_only_to_a_stream() { + // A reference naming another reference, or one that names itself, is damage: no stream. + let d = doc("[4 0 R]", &[annot("5 0 R"), b"<< /Type /3DRef /3D 6 0 R >>".to_vec(), b"<< /Type /3DRef /3D 5 0 R >>".to_vec()]); + assert!(matches!(model_3d(&d, 0, 0), Err(AnnotError::Invalid(m)) if m.contains("no 3D stream"))); + // The stream is under `/3D`; a `/3DD` inside the reference is not the standard's key. + let d = doc("[4 0 R]", &[annot("<< /Type /3DRef /3DD 5 0 R >>"), stream("/Type /3D /Subtype /PRC", b"PRC")]); assert!(matches!(model_3d(&d, 0, 0), Err(AnnotError::Invalid(m)) if m.contains("no 3D stream"))); } diff --git a/crates/automation/README.md b/crates/automation/README.md index 8ffa2d0b3e..d218fbcda6 100644 --- a/crates/automation/README.md +++ b/crates/automation/README.md @@ -250,6 +250,15 @@ choices and values over 4096 UTF-8 bytes return an error. Example: `form_set_props doc=1 field=agree export_value=Accepted widget=1`. +## Rendering a 3D model + +`model_render` draws a page's 3D model (a 3D annotation's PRC stream, directly or through a 3D +reference) to a PNG, without the viewer. `model` picks the page's Nth 3D model (default 1); +`width` and `height` are 1 to 4096 (default 512); `direction` is the view direction from the eye +(default `[-1, 1, -1]`, with Z up unless the view looks along Z); `perspective` defaults to true. +The model is framed to fill the image and shaded by a headlight on white. U3D models are refused +with an error, and the annotation's own views, lighting and render modes are not applied yet. + ## Viewer navigation history A running PdfCraft instance launched with `--control FILE` exposes its per-tab history through diff --git a/crates/automation/src/dispatch.rs b/crates/automation/src/dispatch.rs index 8269963598..f694506adf 100644 --- a/crates/automation/src/dispatch.rs +++ b/crates/automation/src/dispatch.rs @@ -69,7 +69,7 @@ impl Automation { self.apply(&a, edit)? } "page_render" => return self.page_render(&a), - "model3d_render" => return self.model3d_render(&a), + "model_render" => return self.model_render(&a), "comment_image_preview" => return self.comment_image_preview(&a), "text_extract" => self.text_extract(&a)?, "text_word_count" => self.text_word_count(&a)?, diff --git a/crates/automation/src/lib.rs b/crates/automation/src/lib.rs index eb58b52e75..ddbfd45786 100644 --- a/crates/automation/src/lib.rs +++ b/crates/automation/src/lib.rs @@ -27,7 +27,7 @@ mod links; #[cfg(feature = "mcp")] pub mod mcp; mod measure; -mod model3d_render; +mod model_render; mod printing; mod reading; mod redact; diff --git a/crates/automation/src/model3d_render.rs b/crates/automation/src/model_render.rs similarity index 91% rename from crates/automation/src/model3d_render.rs rename to crates/automation/src/model_render.rs index 7f789118a5..145103ffe9 100644 --- a/crates/automation/src/model3d_render.rs +++ b/crates/automation/src/model_render.rs @@ -1,4 +1,4 @@ -//! `model3d_render`: a page's 3D model (ISO 32000-2 §13.6) drawn to a PNG, without a viewer. +//! `model_render`: a page's 3D model (ISO 32000-2 §13.6) drawn to a PNG, without a viewer. use pdfcraft_engine::model3d::Model3dKind; use pdfcraft_prc::PrcFile; @@ -18,7 +18,7 @@ pub(crate) const MAX_SIDE: u32 = 4096; const DEFAULT_DIRECTION: [f64; 3] = [-1.0, 1.0, -1.0]; impl Automation { - pub(crate) fn model3d_render(&mut self, a: &Args) -> Result> { + pub(crate) fn model_render(&mut self, a: &Args) -> Result> { let page = self.page(a)?; let width = side(a, "width")?; let height = side(a, "height")?; @@ -30,9 +30,9 @@ impl Automation { None => *found.first().ok_or_else(|| failed(format!("page {} has no 3D model", page.saturating_add(1))))?, Some(n) => { let pos = usize::try_from(n).ok().and_then(|n| n.checked_sub(1)).ok_or_else(|| Args::wrong("model", "a positive integer"))?; - *found.get(pos).ok_or_else(|| { - ToolError::InvalidArgs(format!("page {} has {} 3D models, not {n}", page.saturating_add(1), found.len())) - })? + *found + .get(pos) + .ok_or_else(|| ToolError::InvalidArgs(format!("page {} has {} 3D models, not {n}", page.saturating_add(1), found.len())))? } }; let model = doc.model_3d(page, index).map_err(failed)?; diff --git a/crates/automation/src/tools.rs b/crates/automation/src/tools.rs index 0d8a3a6c97..7944896b27 100644 --- a/crates/automation/src/tools.rs +++ b/crates/automation/src/tools.rs @@ -181,7 +181,7 @@ pub fn tools() -> Vec { "hide_text_block": { "type": "integer", "minimum": 1, "description": "Render an inline editor background without this paragraph (number from text_paragraphs, resolved as text_edit does: refused once an edit merged that paragraph into a neighbour). Does not modify the document." }, "hide_added_text": { "type": "integer", "minimum": 1, "description": "Render a retyping background without this added text item (added index from object_list). Mutually exclusive with hide_text_block. Does not modify the document." } }), &["doc", "page"])), - t("model3d_render", "Render a 3D model", "Draw a page's 3D model (a 3D annotation's PRC stream) to a PNG, framed to fill the image, with headlight shading on a white background. model is its 1-based position among the page's 3D models (default 1). direction is the view direction [x, y, z] from the eye towards the model (default [-1, 1, -1]); Z is up unless the view looks along Z. perspective defaults to true. U3D models can't be drawn yet and give an error. Does not modify the document.") + t("model_render", "Render a 3D model", "Draw a page's 3D model (a 3D annotation's PRC stream) to a PNG, framed to fill the image, with headlight shading on a white background. model is its 1-based position among the page's 3D models (default 1). direction is the view direction [x, y, z] from the eye towards the model (default [-1, 1, -1]); Z is up unless the view looks along Z. perspective defaults to true. U3D models can't be drawn yet and give an error. Does not modify the document.") .ro() .with(schema(json!({ "doc": doc(), "page": { "type": "integer", "minimum": 1 }, "model": { "type": "integer", "minimum": 1 }, "width": { "type": "integer", "minimum": 1, "maximum": 4096 }, "height": { "type": "integer", "minimum": 1, "maximum": 4096 }, diff --git a/crates/automation/tests/it/main.rs b/crates/automation/tests/it/main.rs index d4946b6d73..85524f13a2 100644 --- a/crates/automation/tests/it/main.rs +++ b/crates/automation/tests/it/main.rs @@ -8,7 +8,7 @@ mod batch; mod button_caption; mod form_bulk; mod hostile_args; -mod model3d_render; +mod model_render; mod optimize_content_boundaries; mod redact_after_stray_q_form; mod redact_text_after_q; diff --git a/crates/automation/tests/it/model3d_render.rs b/crates/automation/tests/it/model_render.rs similarity index 86% rename from crates/automation/tests/it/model3d_render.rs rename to crates/automation/tests/it/model_render.rs index 47e757906a..615a5346a9 100644 --- a/crates/automation/tests/it/model3d_render.rs +++ b/crates/automation/tests/it/model_render.rs @@ -1,4 +1,4 @@ -//! `model3d_render` draws a page's 3D model headlessly: a synthetic PDF holding a synthetic PRC. +//! `model_render` draws a page's 3D model headlessly: a synthetic PDF holding a synthetic PRC. use std::path::PathBuf; @@ -69,32 +69,32 @@ fn a_prc_model_is_drawn_over_the_background() { let (_dir, mut a) = setup("prc"); let doc = open(&mut a, "prc.pdf"); let args = json!({ "doc": doc, "page": 1, "width": 64, "height": 48, "direction": [0, 0, -1], "perspective": false }); - let out = a.call("model3d_render", &args).unwrap(); + let out = a.call("model_render", &args).unwrap(); let Content::Png { data, width, height } = &out[0] else { panic!("expected a PNG") }; assert_eq!((*width, *height), (64, 48)); assert_eq!(rgb(data, 0, 0), [255, 255, 255], "the corner is background"); assert_eq!(rgb(data, 32, 24), [190, 192, 200], "the centre is the triangle, lit head on"); // In perspective the headlight meets the face at a slight angle, so only not-background holds. - let out = a.call("model3d_render", &json!({ "doc": doc, "page": 1, "width": 64, "height": 48, "direction": [0, 0, -1] })).unwrap(); + let out = a.call("model_render", &json!({ "doc": doc, "page": 1, "width": 64, "height": 48, "direction": [0, 0, -1] })).unwrap(); let Content::Png { data, .. } = &out[0] else { panic!("expected a PNG") }; assert_ne!(rgb(data, 32, 24), [255, 255, 255], "the centre is drawn in perspective too"); // The same model named explicitly, and the default isometric view, both draw. - assert!(a.call("model3d_render", &json!({ "doc": doc, "page": 1, "model": 1 })).is_ok()); + assert!(a.call("model_render", &json!({ "doc": doc, "page": 1, "model": 1 })).is_ok()); } #[test] fn a_u3d_model_or_a_missing_one_is_a_clear_error() { let (_dir, mut a) = setup("u3d"); let doc = open(&mut a, "u3d.pdf"); - let err = a.call("model3d_render", &json!({ "doc": doc, "page": 1 })).unwrap_err().to_string(); + let err = a.call("model_render", &json!({ "doc": doc, "page": 1 })).unwrap_err().to_string(); assert!(err.contains("U3D"), "{err}"); let doc = open(&mut a, "prc.pdf"); - let err = a.call("model3d_render", &json!({ "doc": doc, "page": 1, "model": 2 })).unwrap_err().to_string(); + let err = a.call("model_render", &json!({ "doc": doc, "page": 1, "model": 2 })).unwrap_err().to_string(); assert!(err.contains("has 1 3D models, not 2"), "{err}"); for bad in [json!({ "direction": [0, 0, 0] }), json!({ "direction": [1, 2] }), json!({ "width": 0 }), json!({ "height": 5000 })] { let mut args = json!({ "doc": doc, "page": 1 }); args.as_object_mut().unwrap().extend(bad.as_object().unwrap().clone()); - assert!(a.call("model3d_render", &args).is_err(), "{bad}"); + assert!(a.call("model_render", &args).is_err(), "{bad}"); } } @@ -106,7 +106,7 @@ fn a_model_outside_the_root_is_never_opened_to_be_drawn() { assert!(a.call("doc_open", &json!({ "path": path })).is_err(), "{path} lies outside the root"); } // The tool takes no path of its own: only a document opened inside the root reaches it. - assert!(a.call("model3d_render", &json!({ "doc": 1, "page": 1 })).is_err()); + assert!(a.call("model_render", &json!({ "doc": 1, "page": 1 })).is_err()); let doc = open(&mut a, "prc.pdf"); - assert!(a.call("model3d_render", &json!({ "doc": doc, "page": 1 })).is_ok()); + assert!(a.call("model_render", &json!({ "doc": doc, "page": 1 })).is_ok()); } diff --git a/parity/acrobat-features.toml b/parity/acrobat-features.toml index 21e208c003..26ca5e9df8 100644 --- a/parity/acrobat-features.toml +++ b/parity/acrobat-features.toml @@ -9898,9 +9898,9 @@ tier = "P3" milestone = "M12" status = "partial" commands = [] -tools = ["model3d_render"] -evidence = ["crates/automation/tests/it/model3d_render.rs::a_prc_model_is_drawn_over_the_background", "crates/annot/src/model3d/tests.rs::a_3dref_leads_to_the_stream_it_names", "crates/prc-render/tests/model.rs::a_synthetic_prc_renders_its_nearer_triangle_over_the_farther"] -notes = "headless only: model3d_render draws a page's PRC model (through a /3DRef too) to a PNG, framed from a chosen direction, with headlight shading. Missing: the 3D view in the app (no activation, no navigation), U3D (an error), the annotation's /3DV views, lighting schemes and /RM render modes, materials and textures, compressed tessellation, picking, cross sections, the model tree, and measurements" +tools = ["model_render"] +evidence = ["crates/automation/tests/it/model_render.rs::a_prc_model_is_drawn_over_the_background", "crates/annot/src/model3d/tests.rs::a_3dref_leads_to_the_stream_it_names", "crates/prc-render/tests/model.rs::a_synthetic_prc_renders_its_nearer_triangle_over_the_farther"] +notes = "headless only: model_render draws a page's PRC model (through a /3DRef too) to a PNG, framed from a chosen direction, with headlight shading. Missing: the 3D view in the app (no activation, no navigation), U3D (an error), the annotation's /3DV views, lighting schemes and /RM render modes, materials and textures, compressed tessellation, picking, cross sections, the model tree, and measurements" [[feature]] id = "misc.legacy-multimedia" From ca21d4fb7bb763e3dc49d6d9b8f05b5de7a8f8e7 Mon Sep 17 00:00:00 2001 From: KenM76 Date: Tue, 3 Nov 2026 23:06:54 -0500 Subject: [PATCH 27/53] M12: prc-render: scale a triangle's edges before their cross product normalize already scales by the largest component, but the face normal was the normalized cross product of two edges, and edges near 1e160 overflow that product to infinity first, so the triangle was dropped. unit_cross divides both edges by their largest component before the cross product. a_huge_square_is_drawn fails with the old line (white instead of [190,192,200]). Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SqmHYx8FiWM5PcsoiErgp9 --- crates/prc-render/src/render.rs | 4 ++-- crates/prc-render/src/render/tests.rs | 8 ++++++++ crates/prc-render/src/vec3.rs | 19 +++++++++++++++++++ 3 files changed, 29 insertions(+), 2 deletions(-) diff --git a/crates/prc-render/src/render.rs b/crates/prc-render/src/render.rs index d9072a3055..3670878e17 100644 --- a/crates/prc-render/src/render.rs +++ b/crates/prc-render/src/render.rs @@ -5,7 +5,7 @@ use crate::WorldMesh; use crate::raster::{Corner, Target, View, ViewCorner, at}; -use crate::vec3::{cross, dot, length, normalize, scale, sub}; +use crate::vec3::{cross, dot, length, normalize, scale, sub, unit_cross}; use pdfcraft_prc::TriangleMesh; /// The largest image [`render`] draws, in pixels (width × height). @@ -318,7 +318,7 @@ impl Scene { let Some(world) = mesh.triangles.get(t).and_then(|tri| corners(mesh, tri)) else { return; }; - let Some(face) = normalize(cross(sub(at3(&world, 1), at3(&world, 0)), sub(at3(&world, 2), at3(&world, 0)))) else { + let Some(face) = unit_cross(sub(at3(&world, 1), at3(&world, 0)), sub(at3(&world, 2), at3(&world, 0))) else { return; }; let stored = stored_normals(mesh, t); diff --git a/crates/prc-render/src/render/tests.rs b/crates/prc-render/src/render/tests.rs index 8d0764315d..ec8104a260 100644 --- a/crates/prc-render/src/render/tests.rs +++ b/crates/prc-render/src/render/tests.rs @@ -268,3 +268,11 @@ fn unlit_draws_the_plain_colour_and_the_headlight_shades_by_angle() { assert_eq!(lit(Lighting::Headlight, 1.0), [151, 153, 159, 255]); assert_eq!(lit(Lighting::Unlit, 1.0), [190, 192, 200, 255]); } + +#[test] +fn a_huge_square_is_drawn() { + // Its edges' cross product (about 1e320) overflows unless they are scaled first. + let image = render(&[quad(0.0, 1e160, 0.0)], &ortho(4e160), &small()).unwrap(); + assert_eq!(pixel(&image, 20, 20), [190, 192, 200, 255]); + assert_eq!(pixel(&image, 1, 1), [255, 255, 255, 255]); +} diff --git a/crates/prc-render/src/vec3.rs b/crates/prc-render/src/vec3.rs index 8ebf844087..7a673e0055 100644 --- a/crates/prc-render/src/vec3.rs +++ b/crates/prc-render/src/vec3.rs @@ -40,6 +40,17 @@ pub(crate) fn normalize(a: Vec3) -> Option { (l.is_finite() && l > 0.0).then(|| scale(a, 1.0 / l)) } +/// The unit normal of the plane spanned by `a` and `b` (`a × b`, normalized); `None` when they are +/// parallel, zero or not finite. Both are first divided by their largest component, so the cross +/// product of two huge edges cannot overflow before it is normalized. +pub(crate) fn unit_cross(a: Vec3, b: Vec3) -> Option { + let big = a.iter().chain(b.iter()).fold(0.0_f64, |m, c| m.max(c.abs())); + if !(big.is_finite() && big > 0.0) { + return None; + } + normalize(cross(a.map(|c| c / big), b.map(|c| c / big))) +} + #[cfg(test)] mod tests { use super::*; @@ -63,4 +74,12 @@ mod tests { assert_eq!(normalize([1e200, 0.0, 0.0]), Some([1.0, 0.0, 0.0])); assert_eq!(normalize([0.0, -3e-320, 0.0]), Some([0.0, -1.0, 0.0])); } + + #[test] + fn unit_cross_survives_edges_whose_product_overflows() { + assert_eq!(unit_cross([1e160, 0.0, 0.0], [0.0, 1e160, 0.0]), Some([0.0, 0.0, 1.0])); + assert_eq!(unit_cross([1e-170, 0.0, 0.0], [0.0, 1e-170, 0.0]), Some([0.0, 0.0, 1.0])); + assert_eq!(unit_cross([2.0, 0.0, 0.0], [4.0, 0.0, 0.0]), None); + assert_eq!(unit_cross([f64::INFINITY, 0.0, 0.0], [0.0, 1.0, 0.0]), None); + } } From d3821f35747a25c8822e307501f2c46e12127a6b Mon Sep 17 00:00:00 2001 From: KenM76 Date: Tue, 3 Nov 2026 23:47:10 -0500 Subject: [PATCH 28/53] M12: model_render premultiplies through as_chunks_mut CI's clippy (1.99) flags chunks_exact_mut with a constant size. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SqmHYx8FiWM5PcsoiErgp9 --- crates/automation/src/model_render.rs | 10 ++++------ 1 file changed, 4 insertions(+), 6 deletions(-) diff --git a/crates/automation/src/model_render.rs b/crates/automation/src/model_render.rs index 145103ffe9..dbfabc6708 100644 --- a/crates/automation/src/model_render.rs +++ b/crates/automation/src/model_render.rs @@ -78,12 +78,10 @@ fn up_for([x, y, z]: [f64; 3]) -> [f64; 3] { /// Straight RGBA to the premultiplied RGBA `encode_png` takes. fn premultiplied(mut rgba: Vec) -> Vec { - for px in rgba.chunks_exact_mut(4) { - if let [r, g, b, alpha] = px { - let a = u16::from(*alpha); - for c in [r, g, b] { - *c = u8::try_from((u16::from(*c) * a + 127) / 255).unwrap_or(u8::MAX); - } + for [r, g, b, alpha] in rgba.as_chunks_mut::<4>().0 { + let a = u16::from(*alpha); + for c in [r, g, b] { + *c = u8::try_from((u16::from(*c) * a + 127) / 255).unwrap_or(u8::MAX); } } rgba From c336c59cfb69342f2e12ca4cb2e489b0b3d88142 Mon Sep 17 00:00:00 2001 From: KenM76 Date: Wed, 4 Nov 2026 00:03:28 -0500 Subject: [PATCH 29/53] M12: prc-render: scale-safe lengths, a render work ceiling, fit fixes - vec3: length divides by the largest component before squaring, so a camera 1e200 away keeps a finite near plane; unit_cross normalizes each edge first, so edges of very different sizes keep their normal. - render: MAX_RENDER_WORK (2^31) caps the pixel centres one render tests, charged per triangle's clipped bounding box before scanning; past it, RenderError::TooMuchWork. - Bounds::centre halves each end before adding, finite for any finite box. Bounds::of and fit_meshes use only corners of triangles render draws. Camera::frame builds its basis from the direction, not from points near the model, so a small model far from the origin fits. Each new test fails with its fix reverted. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SqmHYx8FiWM5PcsoiErgp9 --- crates/prc-render/README.md | 5 ++- crates/prc-render/src/lib.rs | 7 ++-- crates/prc-render/src/raster.rs | 37 ++++++++++++++----- crates/prc-render/src/render.rs | 53 +++++++++++++++------------ crates/prc-render/src/render/tests.rs | 34 ++++++++++++++++- crates/prc-render/src/vec3.rs | 44 ++++++++++++++++------ 6 files changed, 131 insertions(+), 49 deletions(-) diff --git a/crates/prc-render/README.md b/crates/prc-render/README.md index c3d4db867e..000d673179 100644 --- a/crates/prc-render/README.md +++ b/crates/prc-render/README.md @@ -23,6 +23,7 @@ render(&[WorldMesh], &Camera, &RenderOptions) -> Result - `Lighting::Headlight` (default) shades by the angle to the view direction; `Lighting::Unlit` draws flat colour. - Perspective triangles are clipped at a near plane 1/10,000 of the eye-to-target distance from the eye; orthographic views are not clipped. - `MAX_RENDER_PIXELS` (2^24, about 16 MP) bounds the image: at most 64 MB of RGBA plus 128 MB of depth. - `MAX_ASSEMBLED_TRIANGLES` and `MAX_ASSEMBLED_VERTICES` (2^24 each) bound one `assemble` call. None bounds time: a model of many large - triangles at full size is a long CPU render, so callers on a UI thread should render small or off-thread. + `MAX_ASSEMBLED_TRIANGLES` and `MAX_ASSEMBLED_VERTICES` (2^24 each) bound one `assemble` call. `MAX_RENDER_WORK` (2^31) bounds + the pixel centres one `render` tests, summed over each triangle's clipped bounding box; past it the render returns + `RenderError::TooMuchWork`. That is still seconds of CPU at the top, so callers on a UI thread should render small or off-thread. - Not yet: shadows, ambient occlusion, specular light, scene lights, textures, wires and markups. diff --git a/crates/prc-render/src/lib.rs b/crates/prc-render/src/lib.rs index 4044f110bb..9cfd49efbe 100644 --- a/crates/prc-render/src/lib.rs +++ b/crates/prc-render/src/lib.rs @@ -3,8 +3,9 @@ //! Layer L3: depends only on `pdfcraft-prc`. [`assemble`] turns a parsed [`pdfcraft_prc::PrcFile`] //! into world-space meshes; [`render`] draws them from a [`Camera`] with one depth-tested sample per //! pixel centre and a light at the camera. Single-threaded and GUI-free, so it runs unchanged on wasm32. -//! The image size is capped by [`MAX_RENDER_PIXELS`] and the assembled model by -//! [`MAX_ASSEMBLED_TRIANGLES`] and [`MAX_ASSEMBLED_VERTICES`]; bad input is an error, never a panic. +//! The image size is capped by [`MAX_RENDER_PIXELS`], the drawing work by [`MAX_RENDER_WORK`] and the +//! assembled model by [`MAX_ASSEMBLED_TRIANGLES`] and [`MAX_ASSEMBLED_VERTICES`]; bad input is an +//! error, never a panic. #![forbid(unsafe_code)] #![deny(clippy::unwrap_used, clippy::expect_used, clippy::panic, clippy::unimplemented, clippy::todo, clippy::unreachable)] @@ -14,4 +15,4 @@ mod render; mod vec3; pub use assemble::{MAX_ASSEMBLED_TRIANGLES, MAX_ASSEMBLED_VERTICES, WorldMesh, assemble, place}; -pub use render::{Bounds, Camera, Image, Lighting, MAX_RENDER_PIXELS, Projection, RenderError, RenderOptions, render}; +pub use render::{Bounds, Camera, Image, Lighting, MAX_RENDER_PIXELS, MAX_RENDER_WORK, Projection, RenderError, RenderOptions, render}; diff --git a/crates/prc-render/src/raster.rs b/crates/prc-render/src/raster.rs index cced24db90..e30105352c 100644 --- a/crates/prc-render/src/raster.rs +++ b/crates/prc-render/src/raster.rs @@ -101,23 +101,29 @@ pub(crate) struct Target<'b> { pub(crate) height: u32, pub(crate) rgba: &'b mut [u8], pub(crate) depth: &'b mut [f64], + /// Pixel centres the fills may still test. + pub(crate) work: u64, } impl Target<'_> { /// Fill the screen triangle of `corners` with straight-RGBA `colour`, depth-tested at pixel /// centres, the shade interpolated across it. A pixel centre on an edge two triangles share /// belongs to one of them only. An opaque pixel replaces the pixel and its depth; a - /// translucent one is blended over it, leaving the depth. - pub(crate) fn fill(&mut self, corners: [Corner; 3], colour: [u8; 4]) { + /// translucent one is blended over it, leaving the depth. The clipped bounding box's pixel + /// centres are charged to [`Self::work`] first: [`RenderError::TooMuchWork`], with nothing + /// drawn, when too few are left. + pub(crate) fn fill(&mut self, corners: [Corner; 3], colour: [u8; 4]) -> Result<(), RenderError> { let [a, b, c] = corners; let area = edge(a.at, b.at, c.at); let edges = [(b.at, c.at), (c.at, a.at), (a.at, b.at)]; if !area.is_finite() || area.abs() < 1e-12 { - return; + return Ok(()); } let Some((x0, x1, y0, y1)) = self.bounds(&corners) else { - return; + return Ok(()); }; + let span = |lo: u32, hi: u32| u64::from(hi.saturating_sub(lo)).saturating_add(1); + self.work = self.work.checked_sub(span(x0, x1).saturating_mul(span(y0, y1))).ok_or(RenderError::TooMuchWork)?; for y in y0..=y1 { for x in x0..=x1 { let p = [f64::from(x) + 0.5, f64::from(y) + 0.5, 0.0]; @@ -144,6 +150,7 @@ impl Target<'_> { self.blend(i, key, shaded(colour, shade)); } } + Ok(()) } /// The pixel rows and columns `corners` may cover, clipped to the image; `None` when off it. @@ -259,10 +266,10 @@ mod tests { #[test] fn a_pixel_on_an_edge_two_triangles_share_is_blended_once() { let (mut rgba, mut depth) = (vec![0; 64], vec![f64::NEG_INFINITY; 16]); - let mut t = Target { perspective: false, width: 4, height: 4, rgba: &mut rgba, depth: &mut depth }; + let mut t = Target { perspective: false, width: 4, height: 4, rgba: &mut rgba, depth: &mut depth, work: u64::MAX }; let [a, b, c, d] = [(0.0, 0.0), (4.0, 0.0), (4.0, 4.0), (0.0, 4.0)].map(|(x, y)| corner(x, y, 1.0, 1.0)); - t.fill([a, b, c], [255, 0, 0, 128]); - t.fill([a, c, d], [255, 0, 0, 128]); + t.fill([a, b, c], [255, 0, 0, 128]).unwrap(); + t.fill([a, c, d], [255, 0, 0, 128]).unwrap(); // The diagonal runs through four pixel centres; each must look like the rest. for px in rgba.chunks(4) { assert_eq!(px, [255, 0, 0, 128]); @@ -272,11 +279,23 @@ mod tests { #[test] fn a_perspective_shade_is_interpolated_on_the_surface_not_the_screen() { let (mut rgba, mut depth) = (vec![0; 64], vec![f64::NEG_INFINITY; 16]); - let mut t = Target { perspective: true, width: 4, height: 4, rgba: &mut rgba, depth: &mut depth }; + let mut t = Target { perspective: true, width: 4, height: 4, rgba: &mut rgba, depth: &mut depth, work: u64::MAX }; // The x = 4 corner is three times as far away as the others, and the only lit one. - t.fill([corner(0.0, 0.0, 1.0, 0.0), corner(4.0, 0.0, 1.0 / 3.0, 1.0), corner(0.0, 4.0, 1.0, 0.0)], [240, 240, 240, 255]); + t.fill([corner(0.0, 0.0, 1.0, 0.0), corner(4.0, 0.0, 1.0 / 3.0, 1.0), corner(0.0, 4.0, 1.0, 0.0)], [240, 240, 240, 255]).unwrap(); // At screen (1.5, 0.5) the weights are 0.5, 0.375, 0.125: on the surface the far corner // counts 0.125 / 0.75 = 1/6, where on the screen it would count 0.375. assert_eq!(rgba.get(4..8), Some(&[40, 40, 40, 255][..])); } + + #[test] + fn fills_stop_when_the_work_allowance_is_spent() { + let (mut rgba, mut depth) = (vec![0; 64], vec![f64::NEG_INFINITY; 16]); + // Each fill's box is the whole 4x4 image: 16 pixel centres. + let mut t = Target { perspective: false, width: 4, height: 4, rgba: &mut rgba, depth: &mut depth, work: 40 }; + let tri = [corner(0.0, 0.0, 1.0, 1.0), corner(4.0, 0.0, 1.0, 1.0), corner(0.0, 4.0, 1.0, 1.0)]; + assert_eq!(t.fill(tri, [255, 0, 0, 255]), Ok(())); + assert_eq!(t.fill(tri, [255, 0, 0, 255]), Ok(())); + assert_eq!(t.fill(tri, [255, 0, 0, 255]), Err(RenderError::TooMuchWork)); + assert_eq!(t.work, 8); + } } diff --git a/crates/prc-render/src/render.rs b/crates/prc-render/src/render.rs index 3670878e17..9d7da84ecd 100644 --- a/crates/prc-render/src/render.rs +++ b/crates/prc-render/src/render.rs @@ -11,6 +11,10 @@ use pdfcraft_prc::TriangleMesh; /// The largest image [`render`] draws, in pixels (width × height). pub const MAX_RENDER_PIXELS: u64 = 16 * 1024 * 1024; +/// The most pixel centres one [`render`] tests, summed over each drawn triangle's bounding box +/// clipped to the image: 128 full layers of the largest image. +pub const MAX_RENDER_WORK: u64 = 1 << 31; + /// [`Camera::fit`]'s framed extent over the model's: 5% clear on each side. const FIT_MARGIN: f64 = 1.1; @@ -66,10 +70,10 @@ impl Bounds { out } - /// The box's centre. + /// The box's centre, finite for any finite box. #[must_use] pub fn centre(&self) -> [f64; 3] { - [0, 1, 2].map(|i| (at(self.min, i) + at(self.max, i)) / 2.0) + [0, 1, 2].map(|i| at(self.min, i) / 2.0 + at(self.max, i) / 2.0) } /// Half the box's diagonal: the radius of a sphere holding it. @@ -79,16 +83,9 @@ impl Bounds { } } -/// Every finite vertex the meshes' triangles use. +/// Every corner of the triangles [`render`] draws: those whose three corners exist and are finite. fn vertices(meshes: &[WorldMesh]) -> impl Iterator + Clone + '_ { - meshes.iter().flat_map(|m| { - let mesh = &m.mesh; - mesh.triangles - .iter() - .flatten() - .filter_map(|&i| mesh.positions.get(usize::try_from(i).ok()?).copied()) - .filter(|p| p.iter().all(|c| c.is_finite())) - }) + meshes.iter().flat_map(|m| m.mesh.triangles.iter().filter_map(|tri| corners(&m.mesh, tri)).flatten()) } impl Camera { @@ -130,8 +127,7 @@ impl Camera { r if r.is_finite() && r > 0.0 => r, _ => 1.0, }; - let [right, upward, forward] = - Camera { eye: sub(centre, dir), target: centre, up, projection: Projection::Orthographic { height: 1.0 } }.basis()?; + let [right, upward, forward] = axes(dir, up)?; let view = |p: [f64; 3]| { let r = sub(p, centre); [dot(r, right), dot(r, upward), dot(r, forward)] @@ -171,11 +167,16 @@ impl Camera { /// Right, up and forward unit vectors. pub(crate) fn basis(&self) -> Result<[[f64; 3]; 3], RenderError> { let forward = normalize(sub(self.target, self.eye)).ok_or(RenderError::Camera("the eye and the target coincide"))?; - let right = normalize(cross(forward, self.up)).ok_or(RenderError::Camera("the up direction is zero or parallel to the view direction"))?; - Ok([right, cross(right, forward), forward]) + axes(forward, self.up) } } +/// Right, up and `forward` (a unit vector) for a camera whose up direction is `up`. +fn axes(forward: [f64; 3], up: [f64; 3]) -> Result<[[f64; 3]; 3], RenderError> { + let right = normalize(cross(forward, up)).ok_or(RenderError::Camera("the up direction is zero or parallel to the view direction"))?; + Ok([right, cross(right, forward), forward]) +} + /// How surfaces are lit. #[derive(Debug, Clone, Copy, PartialEq, Eq, Default)] #[non_exhaustive] @@ -246,6 +247,9 @@ pub enum RenderError { /// texture coordinates. #[error("the placed model has more vertices than the assembly ceiling")] TooManyVertices, + /// Drawing would test more than [`MAX_RENDER_WORK`] pixel centres. + #[error("the render would test more pixels than the work ceiling")] + TooMuchWork, } /// Draw `meshes` as seen by `camera`. @@ -262,7 +266,8 @@ pub enum RenderError { /// # Errors /// [`RenderError::Size`] for an empty or over-ceiling image; [`RenderError::Camera`] for a camera /// that cannot form a view (eye on the target, `up` parallel to the view, a field of view outside -/// 0–180° or too small for a finite focal length, a non-positive orthographic height). +/// 0–180° or too small for a finite focal length, a non-positive orthographic height); +/// [`RenderError::TooMuchWork`] past [`MAX_RENDER_WORK`]. pub fn render(meshes: &[WorldMesh], camera: &Camera, options: &RenderOptions) -> Result { let (width, height) = (options.width, options.height); let size = RenderError::Size { width, height }; @@ -276,14 +281,14 @@ pub fn render(meshes: &[WorldMesh], camera: &Camera, options: &RenderOptions) -> let scene = Scene { eye: camera.eye, basis: camera.basis()?, view, lighting: options.lighting }; let mut rgba = options.background.repeat(pixels); let mut depth = vec![f64::NEG_INFINITY; pixels]; - let mut target = Target { perspective, width, height, rgba: &mut rgba, depth: &mut depth }; + let mut target = Target { perspective, width, height, rgba: &mut rgba, depth: &mut depth, work: MAX_RENDER_WORK }; let [r, g, b] = options.colour; for opaque in [true, false] { for m in meshes { for t in 0..m.mesh.triangles.len() { let colour = m.colour_of(t).unwrap_or([r, g, b, 255]); if (colour[3] == 255) == opaque && colour[3] != 0 { - scene.draw(&mut target, &m.mesh, t, colour); + scene.draw(&mut target, &m.mesh, t, colour)?; } } } @@ -313,13 +318,14 @@ impl Scene { [0.3 + 0.7 * towards.map_or(1.0, |v| dot(n, v).abs().min(1.0)); 3] } - /// Draw triangle `t` of `mesh` into `target` in straight-RGBA `colour`. - fn draw(&self, target: &mut Target, mesh: &TriangleMesh, t: usize, colour: [u8; 4]) { + /// Draw triangle `t` of `mesh` into `target` in straight-RGBA `colour`; refused once the + /// target's work allowance is spent. + fn draw(&self, target: &mut Target, mesh: &TriangleMesh, t: usize, colour: [u8; 4]) -> Result<(), RenderError> { let Some(world) = mesh.triangles.get(t).and_then(|tri| corners(mesh, tri)) else { - return; + return Ok(()); }; let Some(face) = unit_cross(sub(at3(&world, 1), at3(&world, 0)), sub(at3(&world, 2), at3(&world, 0))) else { - return; + return Ok(()); }; let stored = stored_normals(mesh, t); let normal = |k: usize| stored.and_then(|n| n.get(k).copied().flatten()).unwrap_or(face); @@ -336,10 +342,11 @@ impl Scene { if let Some((&a, rest)) = screen.split_first() { for pair in rest.windows(2) { if let [b, c] = pair { - target.fill([a, *b, *c], colour); + target.fill([a, *b, *c], colour)?; } } } + Ok(()) } } diff --git a/crates/prc-render/src/render/tests.rs b/crates/prc-render/src/render/tests.rs index ec8104a260..24a2d6b936 100644 --- a/crates/prc-render/src/render/tests.rs +++ b/crates/prc-render/src/render/tests.rs @@ -254,7 +254,8 @@ fn non_finite_and_missing_vertices_are_skipped() { let m = WorldMesh::new(mesh(vec![[0.0, 0.0, 0.0], [f64::NAN, 0.0, 0.0], [0.0, 1.0, 0.0]], vec![[0, 1, 2], [0, 2, 9], [0, 2, u32::MAX]])); let image = render(std::slice::from_ref(&m), &ortho(4.0), &small()).unwrap(); assert!(image.rgba.chunks(4).all(|p| p == [255, 255, 255, 255])); - assert_eq!(Bounds::of(&[m]).unwrap().max, [0.0, 1.0, 0.0]); + // No triangle can be drawn, so none frames the model. + assert_eq!(Bounds::of(&[m]), None); } #[test] @@ -276,3 +277,34 @@ fn a_huge_square_is_drawn() { assert_eq!(pixel(&image, 20, 20), [190, 192, 200, 255]); assert_eq!(pixel(&image, 1, 1), [255, 255, 255, 255]); } + +#[test] +fn a_camera_very_far_away_still_draws() { + let camera = Camera { eye: [0.0, 0.0, 1e200], target: [0.0; 3], up: [0.0, 1.0, 0.0], projection: Projection::Perspective { fov_y: 30.0 } }; + let image = render(&[quad(0.0, 1e199, 0.0)], &camera, &small()).unwrap(); + // The headlight meets the square almost head-on at the centre pixel. + let [r, _, _, a] = pixel(&image, 20, 20); + assert!((185..=190).contains(&r) && a == 255, "{:?}", pixel(&image, 20, 20)); +} + +#[test] +fn the_centre_of_a_box_near_the_largest_float_is_finite() { + let b = Bounds { min: [1e308; 3], max: [1.5e308; 3] }; + assert!(b.centre().iter().all(|c| c.is_finite()), "{:?}", b.centre()); +} + +#[test] +fn a_triangle_that_is_not_drawn_does_not_widen_the_box() { + // The second triangle has a non-finite corner, so its finite far corner is not drawn either. + let m = mesh(vec![[0.0; 3], [1.0, 0.0, 0.0], [0.0, 1.0, 0.0], [f64::NAN, 0.0, 0.0], [1e6, 0.0, 0.0]], vec![[0, 1, 2], [3, 4, 0]]); + let b = Bounds::of(&[WorldMesh::new(m)]).unwrap(); + assert_eq!(b.max, [1.0, 1.0, 0.0]); +} + +#[test] +fn a_small_model_far_from_the_origin_can_be_fitted() { + let (o, d) = (1e16, 4.0); + let m = mesh(vec![[o, o, o], [o + d, o, o], [o, o + d, o]], vec![[0, 1, 2]]); + let camera = Camera::fit_meshes(&[WorldMesh::new(m)], [1.0, 1.0, 1.0], [0.0, 1.0, 0.0], false, 1.0); + assert!(camera.is_ok(), "{camera:?}"); +} diff --git a/crates/prc-render/src/vec3.rs b/crates/prc-render/src/vec3.rs index 7a673e0055..3bb438438a 100644 --- a/crates/prc-render/src/vec3.rs +++ b/crates/prc-render/src/vec3.rs @@ -23,32 +23,39 @@ pub(crate) fn cross([ax, ay, az]: Vec3, [bx, by, bz]: Vec3) -> Vec3 { [ay * bz - az * by, az * bx - ax * bz, ax * by - ay * bx] } -/// The Euclidean length of `a`. +/// The Euclidean length of `a`, taken with `a` divided by its largest component so squaring +/// cannot overflow or underflow; infinite or NaN when a component is. pub(crate) fn length(a: Vec3) -> f64 { - dot(a, a).sqrt() + let big = largest(a); + if !(big.is_finite() && big > 0.0) { + return dot(a, a).sqrt(); + } + let a = a.map(|c| c / big); + big * dot(a, a).sqrt() +} + +/// The largest component of `a` by magnitude; NaN components are passed over. +fn largest(a: Vec3) -> f64 { + a.iter().fold(0.0_f64, |m, c| m.max(c.abs())) } /// `a` scaled to unit length; `None` when its length is zero or not finite. `a` is first divided /// by its largest component, so squaring cannot overflow or underflow. pub(crate) fn normalize(a: Vec3) -> Option { - let big = a.iter().fold(0.0_f64, |m, c| m.max(c.abs())); + let big = largest(a); if !(big.is_finite() && big > 0.0) { return None; } let a = a.map(|c| c / big); - let l = length(a); + let l = dot(a, a).sqrt(); (l.is_finite() && l > 0.0).then(|| scale(a, 1.0 / l)) } /// The unit normal of the plane spanned by `a` and `b` (`a × b`, normalized); `None` when they are -/// parallel, zero or not finite. Both are first divided by their largest component, so the cross -/// product of two huge edges cannot overflow before it is normalized. +/// parallel, zero or not finite. Each is normalized first, so neither huge edges nor edges of +/// very different sizes overflow or underflow the cross product. pub(crate) fn unit_cross(a: Vec3, b: Vec3) -> Option { - let big = a.iter().chain(b.iter()).fold(0.0_f64, |m, c| m.max(c.abs())); - if !(big.is_finite() && big > 0.0) { - return None; - } - normalize(cross(a.map(|c| c / big), b.map(|c| c / big))) + normalize(cross(normalize(a)?, normalize(b)?)) } #[cfg(test)] @@ -82,4 +89,19 @@ mod tests { assert_eq!(unit_cross([2.0, 0.0, 0.0], [4.0, 0.0, 0.0]), None); assert_eq!(unit_cross([f64::INFINITY, 0.0, 0.0], [0.0, 1.0, 0.0]), None); } + + #[test] + fn unit_cross_survives_edges_of_very_different_sizes() { + assert_eq!(unit_cross([1e-170, 0.0, 0.0], [0.0, 1e160, 0.0]), Some([0.0, 0.0, 1.0])); + } + + #[test] + fn length_survives_huge_and_tiny_vectors() { + let near = |l: f64, want: f64| (l - want).abs() <= want * 1e-15; + assert!(near(length([3e200, 4e200, 0.0]), 5e200)); + assert!(near(length([0.0, 3e-200, 4e-200]), 5e-200)); + assert_eq!(length([0.0; 3]), 0.0); + assert!(length([f64::NAN, 1.0, 0.0]).is_nan()); + assert_eq!(length([f64::INFINITY, 1.0, 0.0]), f64::INFINITY); + } } From 36d24eae6e78b1c7d154eebdf86ae411a98fca4f Mon Sep 17 00:00:00 2001 From: Hasan Date: Mon, 2 Nov 2026 18:04:58 +0500 Subject: [PATCH 30/53] M14: Preferences can turn off the update check at start (#5737, #5460) MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit The desktop app asks for a newer release on its first frame, and nothing could stop it, though the README says PdfCraft works offline. A preference, on by default so nothing changes for anyone who leaves it, keeps the app off the network until Help ▸ Check for updates is chosen. It has its own settings key: check_updates_at_start was an opt-in, off unless chosen, so settings written then would turn this off unasked. --- ROADMAP.md | 2 + crates/ui-egui/src/app.rs | 1 + crates/ui-egui/src/app/control.rs | 3 + crates/ui-egui/src/app/prefs.rs | 6 ++ crates/ui-egui/src/dialogs/preferences.rs | 2 + crates/ui-egui/src/lib.rs | 3 + crates/ui-egui/src/platform/updates.rs | 6 +- crates/ui-egui/tests/it/updates.rs | 76 +++++++++++++++++++++++ docs/roadmap.md | 1 + parity/acrobat-features.toml | 4 +- 10 files changed, 100 insertions(+), 4 deletions(-) diff --git a/ROADMAP.md b/ROADMAP.md index 5bfe1fc817..0a03866be3 100644 --- a/ROADMAP.md +++ b/ROADMAP.md @@ -231,6 +231,8 @@ Ranked; detail and the full milestone table in [docs/roadmap.md](docs/roadmap.md - **2026-11-02 (M10, #5743):** macOS print jobs name their title and owner (`com.apple.print.JobInfo.PMJobName`, `PMJobOwner`), as the system print dialog does, because a Kyocera filter crashes on a job without them. Checked by a test and through `doc_print` against a stand-in `lp`; not yet on a physical printer. +- **2026-11-02 (M14, #5737, #5460):** Preferences can turn off the update check at start. It stays on by default, and Help ▸ Check for updates still asks. + - **2026-11-01 (M7, #5657):** Header/footer, watermark and background additions and replacements stage the complete call before committing. Synthetic later-page failures preserve exact incremental bytes, object count, dirty state and allocation cursor; success controls retain other marks and shared/inherited resources. Estimates unchanged. - **2026-11-01 (M14, #5572):** Preferences ▸ Documents and view sets how many recent files to keep, from 0 to 50 with 12 by default. At 0 Home has no Recent section and File ▸ Open Recent is disabled. The list is cut when Preferences closes, so the number can pass through 0 on its way to another. diff --git a/crates/ui-egui/src/app.rs b/crates/ui-egui/src/app.rs index 6ac8ea8783..d2cec1a29f 100644 --- a/crates/ui-egui/src/app.rs +++ b/crates/ui-egui/src/app.rs @@ -128,6 +128,7 @@ impl PdfCraftApp { last_session: Default::default(), quit_session: None, reload_changed_files: true, + update_check_at_start: true, show_secured_notice: true, always_default_view: false, #[cfg(not(target_arch = "wasm32"))] diff --git a/crates/ui-egui/src/app/control.rs b/crates/ui-egui/src/app/control.rs index 0309042bf1..403af6f4c6 100644 --- a/crates/ui-egui/src/app/control.rs +++ b/crates/ui-egui/src/app/control.rs @@ -47,6 +47,9 @@ impl PdfCraftApp { ("show_page_edge_shadows", _) => { self.show_page_edge_shadows = value.parse::().map_err(|_| "show_page_edge_shadows must be true or false")?; } + ("update_check_at_start", _) => { + self.update_check_at_start = value.parse::().map_err(|_| "update_check_at_start must be true or false")?; + } ("scroll_speed", _) => { self.scroll_speed = scroll_speed::ScrollSpeed::parse(value).ok_or("scroll_speed must be slow, normal, or fast")?; } diff --git a/crates/ui-egui/src/app/prefs.rs b/crates/ui-egui/src/app/prefs.rs index 08a20afb45..0df7d6f30b 100644 --- a/crates/ui-egui/src/app/prefs.rs +++ b/crates/ui-egui/src/app/prefs.rs @@ -147,6 +147,9 @@ impl PdfCraftApp { // Set outside json! above, which is at serde_json's macro recursion limit. v["hide_empty_highlight_hover"] = serde_json::Value::Bool(self.comment_prefs.hide_empty_highlight_hover); v["show_page_edge_shadows"] = self.show_page_edge_shadows.into(); + // Not `check_updates_at_start`: that key was an opt-in of its own, off unless chosen, and + // settings written then would turn this off unasked. + v["update_check_at_start"] = self.update_check_at_start.into(); v["hotbar"] = serde_json::to_value(&self.hotbar).unwrap_or_default(); v["comment_tool_pinned"] = serde_json::Value::Bool(self.comment_prefs.pinned); v["show_secured_notice"] = serde_json::Value::Bool(self.show_secured_notice); @@ -219,6 +222,9 @@ impl PdfCraftApp { if let Some(on) = v["reload_changed_files"].as_bool() { self.reload_changed_files = on; } + if let Some(on) = v["update_check_at_start"].as_bool() { + self.update_check_at_start = on; + } if let Some(on) = v["show_secured_notice"].as_bool() { self.show_secured_notice = on; } diff --git a/crates/ui-egui/src/dialogs/preferences.rs b/crates/ui-egui/src/dialogs/preferences.rs index 9e3182e416..ea985f3207 100644 --- a/crates/ui-egui/src/dialogs/preferences.rs +++ b/crates/ui-egui/src/dialogs/preferences.rs @@ -659,6 +659,8 @@ fn card_general(ui: &mut egui::Ui, app: &mut PdfCraftApp, t: &Tokens, ctx: &mut ctx.checkbox(ui, &mut app.restore_last_view, tl!("Restore the last view (page, zoom and page display) when a document is opened again"), t); #[cfg(not(target_arch = "wasm32"))] ctx.checkbox(ui, &mut app.reload_changed_files, tl!("Reload documents when their files change on disk"), t); + #[cfg(not(target_arch = "wasm32"))] + ctx.checkbox(ui, &mut app.update_check_at_start, tl!("Check for updates when PdfCraft starts"), t); ctx.checkbox(ui, &mut app.show_secured_notice, tl!("Show a notice when a document is secured"), t); #[cfg(not(target_arch = "wasm32"))] diff --git a/crates/ui-egui/src/lib.rs b/crates/ui-egui/src/lib.rs index 643d5e90ff..cb6c99bbe5 100644 --- a/crates/ui-egui/src/lib.rs +++ b/crates/ui-egui/src/lib.rs @@ -370,6 +370,9 @@ pub struct PdfCraftApp { quit_session: Option, /// Preferences: reload a document when its file changes on disk (#431). pub reload_changed_files: bool, + /// Preferences: ask for a newer release when the desktop app starts (#5737). Off, the app + /// asks only from Help ▸ Check for updates. + pub update_check_at_start: bool, /// Preferences: tell the user, above the pages, that a document is secured (its security /// settings restrict changes). Off only hides that notice (#4035); kept in the settings. pub show_secured_notice: bool, diff --git a/crates/ui-egui/src/platform/updates.rs b/crates/ui-egui/src/platform/updates.rs index 6b7ea291fa..622006cd6e 100644 --- a/crates/ui-egui/src/platform/updates.rs +++ b/crates/ui-egui/src/platform/updates.rs @@ -21,7 +21,8 @@ pub(crate) struct Updates { updater: Updater, /// The Updates dialog is showing. pub(crate) open: bool, - /// Ask for a newer release at start; the desktop app turns it on, so tests never reach the network. + /// Ask for a newer release at start, unless Preferences says not to; the desktop app turns + /// it on, so tests never reach the network. #[cfg(not(target_arch = "wasm32"))] startup_check: bool, #[cfg(not(target_arch = "wasm32"))] @@ -73,10 +74,11 @@ impl PdfCraftApp { #[cfg(not(target_arch = "wasm32"))] { let Some(ctx) = self.ctx.clone() else { return }; + let at_start = self.update_check_at_start; let updates = &mut self.updates; if std::mem::take(&mut updates.check_requested) { updates.updater.check_now(&ctx); - } else if updates.startup_check { + } else if updates.startup_check && at_start { updates.updater.check_at_start(&ctx); } updates.updater.tick(&ctx); diff --git a/crates/ui-egui/tests/it/updates.rs b/crates/ui-egui/tests/it/updates.rs index 3ea55e434e..469a42f805 100644 --- a/crates/ui-egui/tests/it/updates.rs +++ b/crates/ui-egui/tests/it/updates.rs @@ -100,6 +100,82 @@ fn nothing_is_asked_until_the_user_checks() { h.get_by_label_contains("PdfCraft 99.0.0 is available"); } +/// The desktop app's start, with `settings` restored, and the count of what the source was asked. +fn started_with(settings: &'static str) -> (Harness<'static, PdfCraftApp>, Arc) { + let calls = Arc::new(std::sync::atomic::AtomicUsize::new(0)); + let counted = calls.clone(); + let h = Harness::builder().with_size(egui::vec2(1200.0, 800.0)).build_eframe(move |_cc| { + let mut app = PdfCraftApp::new(); + app.set_option("language", "en").unwrap(); + app.set_startup_update_check(true); + app.restore(settings); + let counted = counted.clone(); + app.set_update_source(Arc::new(move || { + counted.fetch_add(1, std::sync::atomic::Ordering::SeqCst); + Ok(Some(release("v99.0.0"))) + })); + app + }); + (h, calls) +} + +/// Run frames for `time`, or until the source has been asked. The check at start shows nothing +/// while it runs, so `settle` can't wait for it. +fn run_until_asked(h: &mut Harness<'static, PdfCraftApp>, calls: &std::sync::atomic::AtomicUsize, time: std::time::Duration) { + let deadline = std::time::Instant::now() + time; + while calls.load(std::sync::atomic::Ordering::SeqCst) == 0 && std::time::Instant::now() < deadline { + h.run_steps(2); + std::thread::sleep(std::time::Duration::from_millis(5)); + } +} + +#[test] +fn the_preference_keeps_the_check_at_start_from_reaching_the_network() { + use std::sync::atomic::Ordering::SeqCst; + assert!(PdfCraftApp::new().update_check_at_start, "on by default"); + + // The same start with the preference left on does ask, so a count of none below means + // the preference held the check back and not that it hadn't run yet. + let (mut h, calls) = started_with("{}"); + run_until_asked(&mut h, &calls, std::time::Duration::from_secs(30)); + assert_eq!(calls.load(SeqCst), 1, "on, the app asks at start"); + + let (mut h, calls) = started_with(r#"{"update_check_at_start": false}"#); + run_until_asked(&mut h, &calls, std::time::Duration::from_millis(500)); + assert_eq!(calls.load(SeqCst), 0, "nothing is asked at start"); + + let mut restarted = PdfCraftApp::new(); + restarted.restore(&h.state().persist()); + assert!(!restarted.update_check_at_start, "the choice is kept"); + + // Help ▸ Check for updates still asks. + h.state_mut().execute("help.check_updates"); + settle(&mut h); + assert_eq!(calls.load(SeqCst), 1); + h.get_by_label_contains("PdfCraft 99.0.0 is available"); +} + +#[test] +fn preferences_turn_the_check_at_start_off() { + let (mut h, _calls) = started_with(r#"{"update_check_at_start": true}"#); + h.state_mut().execute("app.preferences"); + h.run_steps(3); + h.get_by_label("Check for updates when PdfCraft starts").click(); + h.run_steps(3); + assert!(!h.state().update_check_at_start); + // The control channel's option, as an agent sets it. + h.state_mut().set_option("update_check_at_start", "true").unwrap(); + assert!(h.state().update_check_at_start); + assert!(h.state_mut().set_option("update_check_at_start", "maybe").is_err()); +} + +#[test] +fn settings_from_the_old_opt_in_leave_the_check_at_start_on() { + let mut app = PdfCraftApp::new(); + app.restore(r#"{"check_updates_at_start": false}"#); + assert!(app.update_check_at_start, "a key of the removed option decides nothing"); +} + #[test] fn the_check_at_start_offers_the_prompt_and_later_dismisses_it() { let mut h = Harness::builder().with_size(egui::vec2(1200.0, 800.0)).build_eframe(|_cc| { diff --git a/docs/roadmap.md b/docs/roadmap.md index f376f7cb00..1c99dddeb8 100644 --- a/docs/roadmap.md +++ b/docs/roadmap.md @@ -307,6 +307,7 @@ Estimates unchanged. | 2026-11-03 | patch | M5 #5850 review: indirect typewriter intent and composer edits preserve outer whitespace; movement regression commits by outside click; estimates unchanged. | | 2026-11-03 | patch | M5 #5850: Enter adds Fill & Sign line breaks; outside clicks commit, Escape cancels, and saved multiline appearances retain their text and height; estimates unchanged. | | 2026-11-02 | patch | M10 #5743: macOS print jobs carry the job name and owner some vendors' filters need; not yet checked on a physical printer | +| 2026-11-02 | patch | M14 #5737, #5460: Preferences turns off the update check at start, on by default | | 2026-11-01 | patch | M7 #5657: header/footer, watermark and background operations commit all selected pages together; later-page errors preserve marks, content, shared/inherited resources and pending edits; estimates unchanged. | | 2026-10-31 | patch | #1577: reconcile optional page shadows with current viewer/organizer and settings modules; CI validation pending. | | 2026-10-31 | patch | M3 #1354: reconcile the launch override with split app/file modules and the structured open result; preserve headless help/version exits. CI validation pending. | diff --git a/parity/acrobat-features.toml b/parity/acrobat-features.toml index 8b7ac909a0..00b3dae2c2 100644 --- a/parity/acrobat-features.toml +++ b/parity/acrobat-features.toml @@ -1778,8 +1778,8 @@ milestone = "M3" status = "partial" commands = ["help.check_updates"] tools = [] -evidence = ["crates/ui-egui/tests/it/updates.rs::a_newer_release_is_offered_for_download", "crates/ui-egui/tests/it/updates.rs::an_up_to_date_or_failed_check_says_so", "crates/ui-egui/tests/it/updates.rs::the_check_at_start_offers_the_prompt_and_later_dismisses_it", "crates/ui-egui/tests/it/updates.rs::nothing_is_asked_until_the_user_checks"] -notes = "asks GitHub Releases through craft-update at start (desktop app only) and on request, offers the release page, and updates and restarts in place where the install type allows it; an install waits while documents are unsaved" +evidence = ["crates/ui-egui/tests/it/updates.rs::a_newer_release_is_offered_for_download", "crates/ui-egui/tests/it/updates.rs::an_up_to_date_or_failed_check_says_so", "crates/ui-egui/tests/it/updates.rs::the_check_at_start_offers_the_prompt_and_later_dismisses_it", "crates/ui-egui/tests/it/updates.rs::nothing_is_asked_until_the_user_checks", "crates/ui-egui/tests/it/updates.rs::the_preference_keeps_the_check_at_start_from_reaching_the_network", "crates/ui-egui/tests/it/updates.rs::settings_from_the_old_opt_in_leave_the_check_at_start_on"] +notes = "asks GitHub Releases through craft-update at start (desktop app only) and on request, offers the release page, and updates and restarts in place where the install type allows it; an install waits while documents are unsaved; Preferences ▸ Check for updates when PdfCraft starts (update_check_at_start, on by default) turns the check at start off, leaving only the request from the Help menu" [[feature]] id = "view.single-key-accelerators" From 1c4ef83491d414b09eba4a033836815b066dcd87 Mon Sep 17 00:00:00 2001 From: Hasan Date: Mon, 2 Nov 2026 21:04:08 +0500 Subject: [PATCH 31/53] M14: decide the update check at start once, on the first frame (#5737) Review of #5758: with the preference off, the startup request stayed pending, so turning the preference on later in the session sent a network request at once. The request is now consumed on the first poll whatever the preference says, and a later change waits for the next start. --- crates/ui-egui/src/platform/updates.rs | 5 +++-- crates/ui-egui/tests/it/updates.rs | 11 +++++++++++ parity/acrobat-features.toml | 4 ++-- 3 files changed, 16 insertions(+), 4 deletions(-) diff --git a/crates/ui-egui/src/platform/updates.rs b/crates/ui-egui/src/platform/updates.rs index 622006cd6e..bcf6783cdb 100644 --- a/crates/ui-egui/src/platform/updates.rs +++ b/crates/ui-egui/src/platform/updates.rs @@ -74,11 +74,12 @@ impl PdfCraftApp { #[cfg(not(target_arch = "wasm32"))] { let Some(ctx) = self.ctx.clone() else { return }; - let at_start = self.update_check_at_start; + // Settled on the first frame: a preference turned on later waits for the next start. + let at_start = std::mem::take(&mut self.updates.startup_check) && self.update_check_at_start; let updates = &mut self.updates; if std::mem::take(&mut updates.check_requested) { updates.updater.check_now(&ctx); - } else if updates.startup_check && at_start { + } else if at_start { updates.updater.check_at_start(&ctx); } updates.updater.tick(&ctx); diff --git a/crates/ui-egui/tests/it/updates.rs b/crates/ui-egui/tests/it/updates.rs index 469a42f805..e6cc09d7c5 100644 --- a/crates/ui-egui/tests/it/updates.rs +++ b/crates/ui-egui/tests/it/updates.rs @@ -155,6 +155,17 @@ fn the_preference_keeps_the_check_at_start_from_reaching_the_network() { h.get_by_label_contains("PdfCraft 99.0.0 is available"); } +/// The setting is about starting: turning it on asks nothing until the next start. +#[test] +fn turning_the_check_at_start_on_waits_for_the_next_start() { + use std::sync::atomic::Ordering::SeqCst; + let (mut h, calls) = started_with(r#"{"update_check_at_start": false}"#); + h.run_steps(3); + h.state_mut().set_option("update_check_at_start", "true").unwrap(); + run_until_asked(&mut h, &calls, std::time::Duration::from_millis(500)); + assert_eq!(calls.load(SeqCst), 0, "nothing is asked this session"); +} + #[test] fn preferences_turn_the_check_at_start_off() { let (mut h, _calls) = started_with(r#"{"update_check_at_start": true}"#); diff --git a/parity/acrobat-features.toml b/parity/acrobat-features.toml index 00b3dae2c2..4aef4fdcc6 100644 --- a/parity/acrobat-features.toml +++ b/parity/acrobat-features.toml @@ -1778,8 +1778,8 @@ milestone = "M3" status = "partial" commands = ["help.check_updates"] tools = [] -evidence = ["crates/ui-egui/tests/it/updates.rs::a_newer_release_is_offered_for_download", "crates/ui-egui/tests/it/updates.rs::an_up_to_date_or_failed_check_says_so", "crates/ui-egui/tests/it/updates.rs::the_check_at_start_offers_the_prompt_and_later_dismisses_it", "crates/ui-egui/tests/it/updates.rs::nothing_is_asked_until_the_user_checks", "crates/ui-egui/tests/it/updates.rs::the_preference_keeps_the_check_at_start_from_reaching_the_network", "crates/ui-egui/tests/it/updates.rs::settings_from_the_old_opt_in_leave_the_check_at_start_on"] -notes = "asks GitHub Releases through craft-update at start (desktop app only) and on request, offers the release page, and updates and restarts in place where the install type allows it; an install waits while documents are unsaved; Preferences ▸ Check for updates when PdfCraft starts (update_check_at_start, on by default) turns the check at start off, leaving only the request from the Help menu" +evidence = ["crates/ui-egui/tests/it/updates.rs::a_newer_release_is_offered_for_download", "crates/ui-egui/tests/it/updates.rs::an_up_to_date_or_failed_check_says_so", "crates/ui-egui/tests/it/updates.rs::the_check_at_start_offers_the_prompt_and_later_dismisses_it", "crates/ui-egui/tests/it/updates.rs::nothing_is_asked_until_the_user_checks", "crates/ui-egui/tests/it/updates.rs::turning_the_check_at_start_on_waits_for_the_next_start", "crates/ui-egui/tests/it/updates.rs::the_preference_keeps_the_check_at_start_from_reaching_the_network", "crates/ui-egui/tests/it/updates.rs::settings_from_the_old_opt_in_leave_the_check_at_start_on"] +notes = "asks GitHub Releases through craft-update at start (desktop app only) and on request, offers the release page, and updates and restarts in place where the install type allows it; an install waits while documents are unsaved; Preferences ▸ Check for updates when PdfCraft starts (update_check_at_start, on by default) turns the check at start off, leaving only the request from the Help menu; turning it on takes effect at the next start" [[feature]] id = "view.single-key-accelerators" From ff227b770ca6e436ee0853a2585739a8be9eb1e9 Mon Sep 17 00:00:00 2001 From: kronbii <145912020+Kronbii@users.noreply.github.com> Date: Thu, 29 Oct 2026 13:15:30 +0200 Subject: [PATCH 32/53] M2: area-average shrunk scans and smooth enlarged ones Scanned pages looked thin and broken at screen size. An image that does not ask for /Interpolate (most scans) was pre-shrunk with nearest-neighbour since about one sample in ten and dropped most strokes of its text, and the nearest sampler then read the shrunk raster again at a fractional offset. - Shrinking (vendored hayro patch 21): non-interpolated images outside Type 3 glyphs are area-averaged (area_shrink: each device pixel the coverage-weighted mean of the samples it covers, 16.16 fixed point whose weights sum to exactly one, premultiplied when masked) to their snapped device rectangle and drawn 1:1. /Interpolate governs smoothing when an image is enlarged (ISO 32000-2 8.9.5.3); averaging adds nothing between samples and drops none. PicScale's Box/Area filters were not usable: shrinking, they average a window four target pixels wide. - Enlarging (patch 28): an image of page-scan size (512+ samples on its shorter side) is drawn bilinearly when enlarged, so a zoomed scan has no staircase edges; icons, QR codes and pixel art keep their hard pixels. - A snapped image whose shrink failed keeps the full transform instead of being drawn at its source size (latent in the interpolated path too). Measured on a 13-page 300 dpi scan at 96, 120 and 192 dpi: every rendered sample is within 1 of the exact area average (it was up to 255 off, 17-23% of samples by more than 1). Bilinear zoom costs nothing measurable (four 1024 px tiles at 400%: 50 ms, nearest 48 ms; bicubic was 267 ms). Review fixes: - a tile snaps a shrunk image's half-pixel edges where the whole page does - a scan shrunk one way is smoothed the way it is stretched - a smaller soft mask doesn't let transparent colour through the area shrink Co-Authored-By: Claude Opus 5.5 --- ROADMAP.md | 3 +- crates/render/src/raster/tests.rs | 214 +++++++++++++--- docs/roadmap.md | 3 +- parity/acrobat-features.toml | 4 +- vendor/README.md | 2 +- vendor/hayro/src/renderer.rs | 411 ++++++++++++++++++++++++++++-- 6 files changed, 571 insertions(+), 66 deletions(-) diff --git a/ROADMAP.md b/ROADMAP.md index 5bfe1fc817..ae64d4f352 100644 --- a/ROADMAP.md +++ b/ROADMAP.md @@ -23,7 +23,7 @@ **Stage: alpha** · next: beta, ~25% (ready for real work 50% → 75%) and ~400–700 h away -> **Last reviewed:** 2026-10-29 | **Last updated:** 2026-10-29 | **Change:** minor (M7 #2517: a narrowed then widened paragraph keeps the typed spaces the wrap consumed, after save and reopen; estimates unchanged) +> **Last reviewed:** 2026-10-29 | **Last updated:** 2026-10-29 | **Change:** minor (M2: scans keep their strokes at screen size and are smoothed when zoomed; M7 #2517: a narrowed then widened paragraph keeps the typed spaces the wrap consumed, after save and reopen; estimates unchanged) > **Last reviewed:** 2026-10-27 | **Last updated:** 2026-10-27 | **Change:** minor (M10: OCR model downloads use verified ONNX pairs with legacy discovery fallback; estimates unchanged) > **Last reviewed:** 2026-10-25 | **Last updated:** 2026-10-25 | **Change:** minor (M14 localization current state: Arabic logical catalog, explicit RTL label direction and mirrored stock chrome; native/browser RTL workflows unqualified. Native release Hans face via LXGW WenKai Lite; estimates unchanged) > **Last reviewed:** 2026-10-23 · **Last updated:** 2026-10-23 · **Change:** minor (M7, #2636 partial: pasting added text boxes and images in the engine and through `content_copy`; estimates unchanged) @@ -257,6 +257,7 @@ Ranked; detail and the full milestone table in [docs/roadmap.md](docs/roadmap.md - **2026-10-26 (M2, PR #4172):** Reconciled current main, preserving its Japanese serif/weight selection and both branches' progress records. Installed CJK viewer and preview fallback remains display-only; estimates unchanged. - **2026-10-29 (M7, #2517):** Widening a paragraph rejoins a wrap with the number of spaces that wrap consumed, including a typed run and none for a Chinese or Japanese word split. The count survives an incremental save and a full rewrite, and a width-only `text_edit` uses it. Estimates unchanged. +- **2026-10-29 (M2):** Scanned pages look like the paper again. An image that doesn't ask for /Interpolate was shrunk by nearest-neighbour (since #635), which kept one sample in N per axis, so the text of a 300 dpi scan at 100% lost most of its strokes; it is now area-averaged to whole device pixels (on a 13-page 300 dpi scan every sample at 96–192 dpi is within 1 of the exact area average; it was up to 255 off). Enlarged, an image of page-scan size is smoothed bilinearly, so a zoomed scan has no staircases; icons, QR codes and pixel art keep hard pixels. Estimates unchanged. - **2026-10-26 (M11, JBIG2 symbols):** Lossless JBIG2 now stores each distinct mark of a black-and-white image once, as a symbol, and places its exact copies, when that is smaller than the generic region: on six pages of rendered 300 dpi text 28% of the generic region's size and 17% of Group 4's, and the same bits back through two independent decoders (hayro-jbig2 and Poppler). Scans, whose letters are seldom identical, gain little until refinement coding; a dictionary shared between pages (`/JBIG2Globals`) is next. Estimates unchanged. - **2026-10-25 (M5, comment replies):** One reply can be deleted from the Comments panel without its thread. On the selected card each reply has its own "…" menu (Edit, Copy text, Delete); Delete takes that reply with its pop-up and the replies to it, as `comment_delete` already did headlessly, and the comment and its other replies stay. The card stays selected, and the panel's editor stays open, on what they were on though the delete moved it in `/Annots`. Undo and Redo close the editor, since either can move what it was open on. A locked reply can't be deleted. Estimates unchanged. - **2026-10-24 (M10, #4158 partial):** Word, HTML, RTF and PowerPoint export keep a block that continues another on its line (an italic case name, a link, a superscript) as a run in its own style inside that paragraph, with the source's word space, instead of a paragraph of its own. Style changes that wrap onto the next line still split paragraphs (the reading blocks keep style boundaries). Estimates unchanged. diff --git a/crates/render/src/raster/tests.rs b/crates/render/src/raster/tests.rs index 9c5d75bfbc..7312f3ddaf 100644 --- a/crates/render/src/raster/tests.rs +++ b/crates/render/src/raster/tests.rs @@ -6436,8 +6436,8 @@ fn draw_minified_image( /// making scans, screenshots and rasterized text look soft. #[test] fn non_interpolated_images_are_not_smoothed_when_minified() { - // Top band K W R G, bottom band B Y M C. Nearest 8×4 → 4×2 sampling picks each block's - // pixel, so every device pixel is one palette colour; Catmull-Rom blends across blocks. + // Top band K W R G, bottom band B Y M C. Shrinking 8×4 → 4×2 averages each 2×2 block, + // so every device pixel is one palette colour; Catmull-Rom blends across blocks. let palette: [[u8; 3]; 8] = [[0, 0, 0], [255, 255, 255], [255, 0, 0], [0, 255, 0], [0, 0, 255], [255, 255, 0], [255, 0, 255], [0, 255, 255]]; let rgb: Vec = block_source(palette).concat(); let exact: Vec<[u8; 4]> = palette.iter().map(|c| [c[0], c[1], c[2], 255]).collect(); @@ -6455,38 +6455,76 @@ fn non_interpolated_images_are_not_smoothed_when_minified() { assert!(pixels(&enabled).iter().any(|px| px[..3].iter().any(|&c| (32..=223).contains(&c))), "/Interpolate true made no mid-tone"); } -/// A shrunk, interpolated image whose edges fall between device pixels is snapped to whole -/// pixels (shrunk once to that size, drawn 1:1). Before, the shrunk raster was sampled a second -/// time at the fractional offset, so every edge of a scan was a blended, soft pixel. +/// A shrunk image whose edges fall between device pixels is snapped to whole pixels (shrunk +/// once to that size, drawn 1:1), interpolated or not. Before, the shrunk raster was sampled a +/// second time at the fractional offset, so every edge of a scan was a blended, soft pixel. #[test] -fn interpolated_minified_image_edges_snap_to_whole_pixels() { - let black = vec![0u8; 40 * 40]; - let content = "q 10 0 0 10 0.4 0.4 cm /Im0 Do Q"; - let image = hex(&black); - let pdf = format!( - "%PDF-1.4\n1 0 obj << /Type /Catalog /Pages 2 0 R >> endobj\n\ +fn minified_image_edges_snap_to_whole_pixels() { + for interpolate in ["/Interpolate true ", ""] { + let black = vec![0u8; 40 * 40]; + let content = "q 10 0 0 10 0.4 0.4 cm /Im0 Do Q"; + let image = hex(&black); + let pdf = format!( + "%PDF-1.4\n1 0 obj << /Type /Catalog /Pages 2 0 R >> endobj\n\ 2 0 obj << /Type /Pages /Kids [3 0 R] /Count 1 >> endobj\n\ 3 0 obj << /Type /Page /Parent 2 0 R /MediaBox [0 0 11 11] /Contents 4 0 R /Resources << /XObject << /Im0 5 0 R >> >> >> endobj\n\ 4 0 obj << /Length {} >> stream\n{content}\nendstream endobj\n\ - 5 0 obj << /Type /XObject /Subtype /Image /Width 40 /Height 40 /ColorSpace /DeviceGray /BitsPerComponent 8 /Interpolate true /Filter /ASCIIHexDecode /Length {} >> stream\n{image}>\nendstream endobj\n\ + 5 0 obj << /Type /XObject /Subtype /Image /Width 40 /Height 40 /ColorSpace /DeviceGray /BitsPerComponent 8 {interpolate}/Filter /ASCIIHexDecode /Length {} >> stream\n{image}>\nendstream endobj\n\ trailer << /Root 1 0 R >>\n%%EOF\n", - content.len(), - image.len() + 1 - ); - let mut renderer = PageRenderer::new(Arc::new(pdf.into_bytes()), RenderConfig::default()); - let page = renderer.render(RenderRequest { page: 0, kind: RequestKind::Pixels, scale: 1.0, ..Default::default() }); - assert!(page.error.is_none(), "{:?}", page.error); - assert_eq!((page.width, page.height), (11, 11)); - let gray = |x: usize, y: usize| page.rgba[(y * 11 + x) * 4]; - // Device box 0.4..10.4 × 0.6..10.6 snaps to columns 0..10 and rows 1..11. - for y in 1..11 { - for x in 0..10 { - assert_eq!(gray(x, y), 0, "pixel ({x}, {y}) inside the snapped image"); + content.len(), + image.len() + 1 + ); + let mut renderer = PageRenderer::new(Arc::new(pdf.into_bytes()), RenderConfig::default()); + let page = renderer.render(RenderRequest { page: 0, kind: RequestKind::Pixels, scale: 1.0, ..Default::default() }); + assert!(page.error.is_none(), "{:?}", page.error); + assert_eq!((page.width, page.height), (11, 11)); + let gray = |x: usize, y: usize| page.rgba[(y * 11 + x) * 4]; + // Device box 0.4..10.4 × 0.6..10.6 snaps to columns 0..10 and rows 1..11. + for y in 1..11 { + for x in 0..10 { + assert_eq!(gray(x, y), 0, "{interpolate:?}: pixel ({x}, {y}) inside the snapped image"); + } + assert_eq!(gray(10, y), 255, "{interpolate:?}: column 10 is outside the snapped image (row {y})"); + } + for x in 0..11 { + assert_eq!(gray(x, 0), 255, "{interpolate:?}: row 0 is outside the snapped image (column {x})"); } - assert_eq!(gray(10, y), 255, "column 10 is outside the snapped image (row {y})"); } - for x in 0..11 { - assert_eq!(gray(x, 0), 255, "row 0 is outside the snapped image (column {x})"); +} + +/// A tile draws exactly the pixels of the whole page under it, also where a shrunk image's +/// edge falls on a half pixel. The edges were snapped by rounding halves away from zero, so +/// the tile at x = 5, which moves the left edge from 2.5 to -2.5, snapped it one pixel left of +/// where the whole page did and shifted every column of the image. +#[test] +fn a_tile_snaps_half_pixel_image_edges_as_the_whole_page_does() { + for interpolate in ["/Interpolate true ", ""] { + let ramp: Vec = (0..40 * 40).map(|i| (i % 40 * 6) as u8).collect(); + let content = "q 10 0 0 10 2.5 2.5 cm /Im0 Do Q"; + let image = hex(&ramp); + let pdf = format!( + "%PDF-1.4\n1 0 obj << /Type /Catalog /Pages 2 0 R >> endobj\n\ + 2 0 obj << /Type /Pages /Kids [3 0 R] /Count 1 >> endobj\n\ + 3 0 obj << /Type /Page /Parent 2 0 R /MediaBox [0 0 20 20] /Contents 4 0 R /Resources << /XObject << /Im0 5 0 R >> >> >> endobj\n\ + 4 0 obj << /Length {} >> stream\n{content}\nendstream endobj\n\ + 5 0 obj << /Type /XObject /Subtype /Image /Width 40 /Height 40 /ColorSpace /DeviceGray /BitsPerComponent 8 {interpolate}/Filter /ASCIIHexDecode /Length {} >> stream\n{image}>\nendstream endobj\n\ + trailer << /Root 1 0 R >>\n%%EOF\n", + content.len(), + image.len() + 1 + ); + let mut renderer = PageRenderer::new(Arc::new(pdf.into_bytes()), RenderConfig::default()); + let req = RenderRequest { page: 0, kind: RequestKind::Pixels, scale: 1.0, ..Default::default() }; + let full = renderer.render(req); + let tile = Tile { x: 5, y: 5, w: 10, h: 10 }; + let part = renderer.render(RenderRequest { tile: Some(tile), ..req }); + assert!(full.error.is_none() && part.error.is_none(), "{:?} {:?}", full.error, part.error); + assert_eq!((full.width, part.width, part.height), (20, 10, 10)); + for y in 0..10usize { + for x in 0..10usize { + let whole = full.rgba[((y + 5) * 20 + x + 5) * 4]; + assert_eq!(part.rgba[(y * 10 + x) * 4], whole, "{interpolate:?}: pixel ({x}, {y}) of the tile"); + } + } } } @@ -6570,28 +6608,132 @@ fn non_interpolated_gray_images_are_not_smoothed_when_minified() { } #[test] -fn non_integer_minification_samples_the_same_columns() { - // Ten columns drawn six pixels wide: centre sampling reads columns 0, 2, 4, 5, 7 and 9. +fn non_integer_minification_averages_the_columns_each_pixel_covers() { + // Ten columns drawn six pixels wide: each pixel covers 10/6 columns. Pixel 0 is 60% of + // column 0 and 40% of column 1 (0.6 × 10 + 0.4 × 20 = 14), pixel 1 is 20% of column 1, + // all of column 2's 60% and 20% of column 3 (30), and so on. Nearest-neighbour read whole + // columns 0, 2, 4, 5, 7 and 9 instead, dropping the four others. let ramp = [10u8, 20, 30, 40, 50, 60, 70, 80, 90, 100]; let page = draw_minified_image("DeviceGray", (10, 1), None, &ramp, None, (6, 1)); let grays: Vec = page.rgba.as_chunks::<4>().0.iter().map(|px| px[0]).collect(); - assert_eq!(grays, [10, 30, 50, 60, 80, 100]); + assert_eq!(grays, [14, 30, 46, 64, 80, 96]); } #[test] -fn non_integer_minification_samples_colour_and_mask_from_the_same_columns() { - // Ten gray columns with a soft mask, drawn six pixels wide: columns 0, 2, 4, 5, 7 and 9 are - // sampled, so the mask must pick the same columns as the colours. +fn non_integer_minification_averages_colour_and_mask_together() { + // Ten gray columns with a soft mask, drawn six pixels wide. Colours are averaged + // premultiplied, over the same columns as the mask, and composited over the white page. + // Pixel 0: 60% of column 0 (gray 10, opaque) and 40% of column 1 (transparent) is + // premultiplied gray 6 at alpha 153, so 6 + 255 × (1 − 153/255) = 108. Pixels 1 and 4 + // cover only transparent columns and show the page. let ramp = [10u8, 20, 30, 40, 50, 60, 70, 80, 90, 100]; let mask = [255u8, 0, 0, 0, 255, 255, 0, 0, 0, 255]; let page = draw_minified_image("DeviceGray", (10, 1), None, &ramp, Some(&mask), (6, 1)); - // Opaque columns keep their gray; transparent columns show the white page. assert_eq!( page.rgba.as_chunks::<4>().0.to_vec(), - vec![[10, 10, 10, 255], [255, 255, 255, 255], [50, 50, 50, 255], [60, 60, 60, 255], [255, 255, 255, 255], [100, 100, 100, 255]] + vec![[108, 108, 108, 255], [255, 255, 255, 255], [132, 132, 132, 255], [138, 138, 138, 255], [255, 255, 255, 255], [162, 162, 162, 255]] ); } +/// A scan shown at screen size: a stroke one sample wide in an image drawn at a third of its +/// resolution. Nearest-neighbour read columns 1, 4 and 7 and lost the stroke in column 3, which +/// is how the text of 300 dpi scans broke up at 100% zoom. Area averaging keeps it as the +/// third of pixel 1 it covers. +#[test] +fn non_interpolated_minification_keeps_thin_strokes() { + let mut stroke = [255u8; 9]; + stroke[3] = 0; + let page = draw_minified_image("DeviceGray", (9, 1), None, &stroke, None, (3, 1)); + let grays: Vec = page.rgba.as_chunks::<4>().0.iter().map(|px| px[0]).collect(); + assert_eq!(grays, [255, 170, 255]); +} + +/// A zoomed scan: an image of page-scan size drawn three times its resolution without +/// `/Interpolate`. Nearest-neighbour turned each sample into a 3 × 3 block, so the edges of +/// scanned text were staircases; it is smoothed (bilinearly) instead, as browsers draw it. +#[test] +fn large_non_interpolated_images_are_smoothed_when_enlarged() { + let side = 512usize; + let half: Vec = (0..side * side).map(|i| if i % side < side / 2 { 0 } else { 255 }).collect(); + let page = draw_minified_image("DeviceGray", (512, 512), None, &half, None, (1536, 1536)); + let row = 768 * 1536; + let edge: Vec = (760..776).map(|x| page.rgba[(row + x) * 4]).collect(); + assert!(edge.iter().any(|v| (32..=223).contains(v)), "a hard edge, not a smoothed one: {edge:?}"); + assert_eq!((page.rgba[(row + 700) * 4], page.rgba[(row + 836) * 4]), (0, 255), "away from the edge"); +} + +/// Small images (icons, QR codes, pixel art) keep their hard pixels when enlarged. +#[test] +fn small_non_interpolated_images_keep_hard_pixels_when_enlarged() { + let checker: Vec = (0..64).map(|i| if (i / 8 + i % 8) % 2 == 0 { 0 } else { 255 }).collect(); + let page = draw_minified_image("DeviceGray", (8, 8), None, &checker, None, (80, 80)); + assert!(page.rgba.as_chunks::<4>().0.iter().all(|px| px[0] == 0 || px[0] == 255), "a small image was smoothed"); +} + +/// A scan shrunk across and stretched down (a fax shown at screen size, zoomed in): the shrunk +/// side is still averaged, and the stretched side is smoothed as a scan enlarged both ways +/// is. Shrinking one side kept the nearest sampler for the other, so its edges were staircases. +#[test] +fn a_scan_shrunk_one_way_is_smoothed_the_way_it_is_stretched() { + // Top half black; bottom half columns alternating black and white, which averages to gray. + let side = 512usize; + let source: Vec = (0..side * side).map(|i| if i / side < side / 2 || i % 2 == 0 { 0 } else { 255 }).collect(); + let page = draw_minified_image("DeviceGray", (512, 512), None, &source, None, (256, 1536)); + let gray = |x: usize, y: usize| page.rgba[(y * 256 + x) * 4]; + let edge: Vec = (760..776).map(|y| gray(128, y)).collect(); + assert!(edge.iter().any(|v| (16..=112).contains(v)), "a hard edge down the stretched side: {edge:?}"); + assert_eq!(gray(128, 700), 0, "above the edge"); + assert!((0..256).all(|x| gray(x, 836).abs_diff(128) <= 1), "the shrunk side is averaged at 1:1"); +} + +/// An image shrunk across and stretched down (a 204 × 98 dpi fax shown at screen size): the +/// shrunk side is still averaged. The area shrink was asked to enlarge the other side too, +/// refused, and the image fell back to its source resolution and the nearest sampler, which +/// read single columns. +#[test] +fn an_image_shrunk_one_way_and_stretched_the_other_is_still_averaged() { + let ramp: Vec = [10u8, 20, 30, 40, 50, 60, 70, 80].repeat(2); + let page = draw_minified_image("DeviceGray", (8, 2), None, &ramp, None, (4, 4)); + let grays: Vec = page.rgba.as_chunks::<4>().0.iter().map(|px| px[0]).collect(); + assert_eq!(grays, [15, 35, 55, 75].repeat(4), "each pixel is the mean of the two columns it covers"); +} + +/// Averaging straight colours would tint an opaque sample with the colour of a transparent +/// neighbour: white beside transparent black, shrunk to one pixel, came out gray over the white +/// page. Premultiplied, the transparent sample adds nothing but its transparency. +#[test] +fn non_interpolated_minification_does_not_bleed_transparent_colour() { + let page = draw_minified_image("DeviceGray", (2, 1), None, &[255, 0], Some(&[255, 0]), (1, 1)); + assert_eq!(page.rgba.as_chunks::<4>().0.to_vec(), vec![[255, 255, 255, 255]]); +} + +/// The same with a soft mask smaller than its image: red, red, blue, blue under a mask that is +/// opaque, then transparent, shrunk to one pixel, is half-covered red over the white page. +/// Colour and mask were averaged apart and multiplied, so the transparent blue came through +/// as purple (191, 127, 191). +#[test] +fn a_smaller_soft_mask_does_not_bleed_transparent_colour_when_minified() { + let content = "q 1 0 0 1 0 0 cm /Im0 Do Q"; + let (image, mask) = ("FF0000FF00000000FF0000FF", "FF00"); + let pdf = format!( + "%PDF-1.4\n1 0 obj << /Type /Catalog /Pages 2 0 R >> endobj\n\ + 2 0 obj << /Type /Pages /Kids [3 0 R] /Count 1 >> endobj\n\ + 3 0 obj << /Type /Page /Parent 2 0 R /MediaBox [0 0 1 1] /Contents 4 0 R /Resources << /XObject << /Im0 5 0 R >> >> >> endobj\n\ + 4 0 obj << /Length {} >> stream\n{content}\nendstream endobj\n\ + 5 0 obj << /Type /XObject /Subtype /Image /Width 4 /Height 1 /ColorSpace /DeviceRGB /BitsPerComponent 8 /SMask 6 0 R /Filter /ASCIIHexDecode /Length {} >> stream\n{image}>\nendstream endobj\n\ + 6 0 obj << /Type /XObject /Subtype /Image /Width 2 /Height 1 /ColorSpace /DeviceGray /BitsPerComponent 8 /Filter /ASCIIHexDecode /Length {} >> stream\n{mask}>\nendstream endobj\n\ + trailer << /Root 1 0 R >>\n%%EOF\n", + content.len(), + image.len() + 1, + mask.len() + 1 + ); + let mut renderer = PageRenderer::new(Arc::new(pdf.into_bytes()), RenderConfig::default()); + let page = renderer.render(RenderRequest { page: 0, kind: RequestKind::Pixels, scale: 1.0, ..Default::default() }); + assert!(page.error.is_none(), "{:?}", page.error); + let [r, g, b, _] = page.rgba.as_chunks::<4>().0.first().copied().unwrap_or_default(); + assert!(r >= 254 && g.abs_diff(128) <= 1 && b == g, "half-covered red over white, got ({r}, {g}, {b})"); +} + #[test] fn non_interpolated_alpha_images_keep_their_colours_and_mask_when_minified() { let palette: [[u8; 3]; 8] = [[0, 0, 0], [255, 255, 255], [255, 0, 0], [0, 255, 0], [0, 0, 255], [255, 255, 0], [255, 0, 255], [0, 255, 255]]; @@ -7716,7 +7858,7 @@ fn stitched_render_partitions_match_a_whole_render() { /// and square caps, a sharp miter, a rotated and a stretched matrix, a hairline, dashes, a /// curve, text, and a page shown rotated. /// -/// Images are skipped the same way (vendored patch (10)): an image on a tile, one crossing +/// Images are skipped the same way (vendored patch (11)): an image on a tile, one crossing /// tile lines, and one fully off the page. #[test] fn skipping_paths_off_the_canvas_changes_no_pixel() { diff --git a/docs/roadmap.md b/docs/roadmap.md index f376f7cb00..d83744afcf 100644 --- a/docs/roadmap.md +++ b/docs/roadmap.md @@ -22,7 +22,7 @@ > **Last reviewed:** 2026-10-26 · **Last updated:** 2026-10-26 · **Change:** patch (M2, PR #4172: document installed-font helpers and regression contracts; behavior and estimates unchanged) > **Last reviewed:** 2026-10-26 · **Last updated:** 2026-10-26 · **Change:** patch (M2, PR #4172: print-preview font policy matches output, with isolated caches and GBK switching regressions; estimates unchanged) > **Last reviewed:** 2026-10-26 · **Last updated:** 2026-10-26 · **Change:** patch (M2, PR #4172: reconcile current main and retain Japanese serif/weight selection alongside display-only CJK fallback; estimates unchanged) -> **Last reviewed:** 2026-10-29 | **Last updated:** 2026-10-29 | **Change:** minor (M7 #2517: a narrowed then widened paragraph keeps the typed spaces the wrap consumed, after save and reopen; estimates unchanged) +> **Last reviewed:** 2026-10-29 | **Last updated:** 2026-10-29 | **Change:** minor (M2: scans keep their strokes at screen size and are smoothed when zoomed; M7 #2517: a narrowed then widened paragraph keeps the typed spaces the wrap consumed, after save and reopen; estimates unchanged) > **Last reviewed:** 2026-10-27 | **Last updated:** 2026-10-27 | **Change:** minor (M10: OCR model downloads use verified ONNX pairs with legacy discovery fallback; estimates unchanged) > **Last reviewed:** 2026-10-25 | **Last updated:** 2026-10-25 | **Change:** minor (M5: one reply in the Comments panel is edited, copied or deleted from its own "…" menu; estimates unchanged) > **Last reviewed:** 2026-10-26 · **Last updated:** 2026-10-26 · **Change:** minor (M11: lossless JBIG2 stores each distinct mark once, as a symbol; estimates unchanged) @@ -321,6 +321,7 @@ Estimates unchanged. | 2026-10-23 | minor | M11: black-and-white images (1-bit, and image masks) are stored losslessly as CCITT Group 4 when that is smaller: PDF Optimizer ▸ Images ▸ Monochrome Images (on by default, so Reduce File Size too), `doc_optimize` `mono`; `pdfcraft_filters::encode_g4`. JBIG2 and monochrome downsampling not yet. Estimates unchanged. | | 2026-10-23 | minor | M6: calculation checks: a formula reading a field the form doesn't have, and a script building the names it reads, are flagged by `form_calculation_order` and on the Calculation order page (a red mark, a card naming the broken formulas, a note for built names); `form_calculation_order` gives each formula and `form_fields` a custom Calculate script. Estimates unchanged. | | 2026-10-29 | minor | M7 #2517: widening a paragraph rejoins a wrap with the spaces that wrap consumed, including a typed run and none for a Chinese or Japanese word split, after an incremental save and a full rewrite. Estimates unchanged. | +| 2026-10-29 | minor | M2: non-interpolated images drawn below their resolution are area-averaged to whole device pixels instead of nearest-neighbour (scanned text kept one sample in N per axis); page-scan-size images are smoothed when enlarged. Estimates unchanged. | | 2026-10-27 | minor | M10: OCR stages SHA-256-verified ONNX models after the upstream RTEN URL brownout; discovery prefers a complete ONNX pair per directory, falls back to a complete RTEN pair, and never mixes formats or directories; language/web limits and estimates unchanged. | | 2026-10-24 | minor | M3, #3970 follow-up: the button release given back after a window move or resize goes before the pointer leaving, so hover doesn't stick. Estimates unchanged. | | 2026-10-23 | minor | M8, #3205 follow-up: phone-pattern redaction includes the +39 country code on labelled Italian geographic numbers with a slash; dates, protocol identifiers and whole-run guards remain. Estimates unchanged. | diff --git a/parity/acrobat-features.toml b/parity/acrobat-features.toml index 8b7ac909a0..367253b4a7 100644 --- a/parity/acrobat-features.toml +++ b/parity/acrobat-features.toml @@ -734,8 +734,8 @@ milestone = "M2" status = "partial" commands = [] tools = ["page_render"] -evidence = ["crates/render/src/raster/tests.rs::renders_a_blue_rectangle", "crates/automation/tests/it/automation.rs::open_inspect_render_and_find", "crates/render/src/raster/tests.rs::user_unit_scales_the_page_and_everything_on_it"] -notes = "hayro bootstrap renderer; own DisplayList devices in M2.6. Pages with /UserUnit are shown at their size in points, as Acrobat reads the value" +evidence = ["crates/render/src/raster/tests.rs::renders_a_blue_rectangle", "crates/automation/tests/it/automation.rs::open_inspect_render_and_find", "crates/render/src/raster/tests.rs::user_unit_scales_the_page_and_everything_on_it", "crates/render/src/raster/tests.rs::non_interpolated_minification_keeps_thin_strokes", "crates/render/src/raster/tests.rs::minified_image_edges_snap_to_whole_pixels", "crates/render/src/raster/tests.rs::large_non_interpolated_images_are_smoothed_when_enlarged", "crates/render/src/raster/tests.rs::small_non_interpolated_images_keep_hard_pixels_when_enlarged"] +notes = "hayro bootstrap renderer; own DisplayList devices in M2.6. Pages with /UserUnit are shown at their size in points, as Acrobat reads the value. An image drawn below its resolution is shrunk once to whole device pixels: by area averaging unless it asks for /Interpolate (then Catmull-Rom), so the text of a 300 dpi scan at 100% keeps its strokes (it was nearest-neighbour, which dropped them). Enlarged, an image of page-scan size (512 samples or more on its shorter side) is smoothed bilinearly whatever its /Interpolate, so a zoomed scan has no staircases; smaller images keep their hard pixels. Not measured against Acrobat" [[feature]] id = "view.render-shadings-patterns" diff --git a/vendor/README.md b/vendor/README.md index 4fda6483af..1a58ad6c1f 100644 --- a/vendor/README.md +++ b/vendor/README.md @@ -40,7 +40,7 @@ Other resource categories retain their existing parent-chain lookup. | Crate | Version | Licence | Patches | Test | |---|---|---|---|---| | hayro-interpret | 0.7.0 | Apache-2.0 OR MIT | (1) select `/AP /N` appearance states via `/AS` (checkboxes, radios, stateful icons) and honour the NoView flag; (2) Type3 Unicode fallback to `/Encoding` glyph names, then printable ASCII; (3) `Type3Glyph::advance_width`; (4) `InterpreterSettings::ocg_overrides` for viewer layer toggles; (5) soft masks whose group lacks `/CS` were dropped (Chrome gradient text); (6) paint nesting cap `MAX_PAINT_NESTING` for tiling patterns and Type 3 glyphs (self-referencing content overflowed the stack); (7) CID width ranges in `/W` and `/W2` clamped to `MAX_CID` (a `0 4294967295 w` range hung); (8) `InterpreterSettings::hide_comments` skips markup annotations but keeps widgets and links (View ▸ Hide all comments); (9) `image_size_ok`: images and masks over 2^28 pixels are not decoded (a fuzzed `/W 4294967295` stencil mask allocated 4 GiB); (10) `MAX_NESTED_PAINTS`: nested paints (forms, Type 3 glyphs and tiling patterns inside one another) share a per-page budget of 50,000 (a glyph showing eight glyphs of its own font, sixteen deep, hung); (11) depends on skrifa 0.47 instead of 0.42 (upstream read-fonts fixed a Type 1 tokenizer that recursed through runs of numbers: a fuzzed font hung or overflowed the stack), with the one API change (`Charset::glyph_id` returns `Option`); (12) annotation appearances whose `/BBox` or `/Matrix` collapses to a line or a point are skipped (mapping them onto `/Rect` divided by zero and handed the device infinite or NaN matrices); (13) a Highlight annotation without a usable normal appearance is drawn from its `/QuadPoints`, `/C` and `/CA` with Multiply blending, as pdfcraft-annot draws highlights (it was not drawn at all); (14) simple-font `/Widths` (`read_widths`): entries only for codes 0–255 (one entry per code below `/FirstChar` was allocated: `4294967295` asked for 34 GB), an entry that isn't a number leaves its code missing instead of ending the array, and a `/LastChar` below `/FirstChar` gives no widths instead of discarding the font for a standard one; (15) the appearance box is mapped onto `/Rect` corner to corner: the translation now accounts for the scale (an appearance whose box didn't start at the origin and differed in size from `/Rect` was drawn shifted); (16) a clip set by `W`/`W*` takes effect after whichever path-painting operator ends the path (`S`, `f`, `B`, … as well as `n`; ISO 32000-2 §8.5.4): it was applied only on `n`, so `re W* S` clipped nothing and the clip stayed pending for a later, unrelated `n`; (17) `InterpreterSettings::cancelled`: once set, interpretation stops before the next content operator (a render pool sets it for every worker when dropped, after an edit or on closing, and for one worker when the watchdog gives up on its render, so renders whose answers nobody can receive stop instead of running to the end); (18) `MAX_PAGE_CONTENT`: the content a page holds at once (its contents, and the forms, tiling patterns and Type 3 glyph procedures being painted) stays within 512 MiB (`decode_content`; once it runs out, `InterpreterWarning::ContentTruncated` is reported (once per page by pdfcraft-render) so a truncated page is not a clean render; each level gives its bytes back when it finishes): forms nest fifty deep and each level kept its content while the next ran, so a form painting itself around a stream inflating to 256 MiB would have held over 12 GB; and Type 0 sampled functions take at most `MAX_SAMPLE_ENTRIES` (2^21) table points and `MAX_SAMPLE_VALUES` (2^23) samples, read no further than `/Size` asks (256 MiB of 1-bit samples asked for over 8 GB); (19) mesh shadings (types 4–7) are sampled only inside the area being drawn (`ShadingPattern::encode_within`, which the vendored hayro renderer calls with that area plus a pixel), within 8 visits per pixel of it plus 4 per triangle (`TriangleSampler`), and patches are triangulated one at a time: points far off the page were sampled over their whole bounding box, up to 65536 × 65536 entries (a fuzzed 147-byte patch stream; a synthetic page passed 6 GB in six seconds), ten thousand overlapping triangles ran for minutes, and every patch became 722 triangles before any limit; meshes hold at most `MAX_MESH_PATCHES` (2^16) patches and `MAX_MESH_TRIANGLES` (2^20) triangles, a lattice needs two to 2^19 vertices per row (`/VerticesPerRow 0` pushed empty rows for ever; one huge row read the whole stream), and a page samples at most `MAX_PAGE_MESH_TRIANGLES` (2^25) mesh triangles in all (a mesh painted a thousand times was triangulated a thousand times); each cut logs a warning; (20) PDF function construction is limited to 64 levels and 10,000 total nodes with cycle detection; calculator parsing and evaluation share procedure depth/work budgets, integer division/bitshift/rotation/index are panic-free, and fixed-size stack errors propagate; (21) a Type 0 font drawn with a substitute that isn't CID-keyed (it isn't embedded, or its program doesn't load) draws `.notdef` for a CID of an Adobe character collection (Japan1, GB1, CNS1, Korea1) that it can't find by Unicode: hayro took the CID for a glyph index of the substitute, which numbers its glyphs its own way, so the character was drawn as an unrelated one (𠮷 as "Ｑ" in BIZ UDPGothic); (22) a Type 0 font drawn with a substitute finds a character given with a variation selector (`/ToUnicode`, or Adobe-Japan1-UCS2, which maps the JIS X 0208 forms of the kanji JIS X 0213:2004 redrew that way: CID 1247 is 噂 U+E0100) through the substitute's variation sequences, else as the character alone: such a CID found no glyph and was drawn as an unrelated one; (23) backport of hayro 0.8's `gradient` module: `EncodedShadingPattern::as_svg_gradient` turns an axial or radial shading into stops of an sRGB gradient (within 0.01 per channel, a stop at every stitching-function bound via `Function::stitching_bounds`) when a native gradient draws it exactly (extended past the box drawn in; a radial one with no background colour); (24) `DecodedImageCache`: decoded image pixels shared across renders through the interpreter cache within a 64 MiB budget (an image-heavy page decoded its thousands of images again per tile and per scale), keyed by every decode input (`DecodedImageKey`: the stream's place and dictionary bytes, its object, mask/stencil/raster output, interpolation, decode target size, preview or final); images under a transfer function and inline images are not cached, an image over the budget is never kept, and a poisoned lock decodes uncached; (25) `Device::actual_text` hands the device a marked content sequence's `/ActualText` (inline or named in `/Properties`), so text extraction keeps the Thai, Indic and Arabic clusters Chrome marks this way (their shaped tone marks and vowels have no Unicode of their own); (26) non-embedded simple fonts drawn with a standard-font stand-in build the Latin Extended-A letters the stand-in lacks (`/Gbreve`, `/Idotaccent`, `/Scedilla`, `/aogonek`, `/ccaron`…) from the base letter and accent it has (`StandardKind::composite_glyph`, `ACCENTED_LETTERS`): PDFium's Foxit stand-ins hold only the Standard/WinAnsi/Mac Roman glyphs, so iText letters with Times New Roman unembedded showed blanks for Ğ, İ, Ş, ğ and ş; the comma-below letters (`/scommaaccent`, `/Tcommaaccent`, `/kcommaaccent`…, found by character, so `/uni0136` and the glyph list's `/Kcedilla` too) are the letter and the font's comma, smaller and under the baseline, since the stand-ins have no `commaaccent` (Romanian ș drew nothing and ț a cedilla in forms laid out from XFA; ģ, whose comma is turned and above, is not built); (27) an embedded TrueType font without a base encoding reads its codes as WinAnsi for text extraction (`EmbeddedKind::code_to_unicode_name`), matching pdftotext's output, instead of Standard: OCR text layers of scans write WinAnsi codes with no `/Encoding`, and Standard has no ü, ö or ß and reads é as Ø (glyph selection unchanged); (28) an image whose pixels hold at most 16 bits of samples (an 8-bit Separation or DeviceN image, a 1-, 2- or 4-bit gray or indexed one) is converted once per possible pixel value through a lookup table instead of once per pixel (`rgb_via_lookup`): a spot-colour tint transform is often a PostScript calculator program, run for every pixel, so a fillable character sheet's 3304 × 4105 DeviceN background took four seconds and 430 MB to draw a page; each entry is converted on its own (`convert_each`), so a tint transform giving the wrong number of components for a value the image never uses cannot shift the colours of the values it does; (29) `Device::begin_annotations`/`end_annotations` bracket the page's annotations, so a device can tell comment paints from document paints (reading colours recolour the document, never a comment, #1548); (30) the Annex D.2 alias codes read as the ink they draw: WinAnsiEncoding 0xA0 and 0xAD, and MacRomanEncoding 0xCA, carry the glyph names `space` and `hyphen` -- the names those tables also give 0x20 and 0x2D -- not `nbspace` and `sfthyphen`, whose glyph list values U+00A0 and U+00AD are invisible formatting characters. Extraction hid drawn ink behind them, so search and redaction matched neither (`XiaoBiaoSong.pdf` in the pdf.js corpus drew 104 soft hyphens; across the 989 files the soft hyphens we extract fall from 117 to 13 and the non-breaking spaces from 30 to 26, what remains coming from `/ToUnicode` maps, which say so explicitly). The base table only, in the text direction only: a `/Differences` name keeps its glyph list value and glyph lookup is untouched, both as pdftotext does; (31) a Type 0 sampled function keeps its samples in one flat array indexed by strides (`Type0::entry`) instead of a hash map keyed by each grid point: the map was built again, one hashed insert per sample, every time a shading using the function was painted (a 2,097,152-sample ramp painted 30 times took 38 s; `bug1721218` in the pdf.js corpus spent most of a 1.5 s page on it); the interpolation is unchanged, so results are bit for bit the same, a sample missing from a short stream still gives no value, and the maximum sample value is computed without overflowing at 32 bits per sample; (32) the decoded-image cache (24) keys a decode by the size the image is drawn at only when a JPX stream decodes it or its soft mask or mask (`decode_reads_target`): JPX is the only filter that decodes a reduced resolution for a size, so a JPEG or Flate scan was decoded again at every zoom step and kept once per zoom level; one decode now serves every size; `InterpreterCache::decoded_image_decodes` counts the cache's decodes; (33) Coons and tensor-product patches copy their vertices' colours less: each point of a patch's 20 × 20 triangle grid is interpolated once (`generate_patch_triangles`; each cell interpolated its four corners itself, 1,444 interpolations a patch for 400 points), a cell's second triangle takes the vertices its first copied, and the mesh sampler moves a patch's triangles into device space instead of copying each vertex (`TriangleSampler::sample_owned`); the same values, 10–16% faster on mesh-gradient pages; (34) `color::tint_transform_runs`: how many colours the calling thread has run through a Separation or DeviceN tint transform (counted once per batch in `convert_f32`), so a test counts patch (28)'s once-per-sample-value runs instead of timing them (its 10 s deadline failed on a busy FreeBSD VM in CI); (35) `EncodedShadingPattern::extent`: where a mesh shading (types 4–7) can paint at all, the pixels its triangles touched when it was encoded (`None` for the types that paint everywhere, and when the shading has a `/Background`), so the renderer samples a texture there alone and skips a mesh with nothing on the canvas; (36) a mesh shading's pixel whose corner no triangle holds strictly (on an edge two triangles share, or on the mesh's side) takes the triangle that covers its centre, edges included (`Triangle::covers`): testing only the corner left a white seam along every shared edge, and the colour is still taken at the corner; (38) the `/DefaultGray`, `/DefaultRGB` and `/DefaultCMYK` colour spaces of a resource dictionary (ISO 32000-2 §8.6.5.6) replace the device space content selects: with `g`/`G`, `rg`/`RG`, `k`/`K` and `cs`/`CS` (by name, or through a `/ColorSpace` resource that is a device space), and as the `/ColorSpace` of an image XObject, an inline image or a shading painted by `sh` or as a shading pattern (`DefaultColorSpaces`, looked up once per content window in the resources and those they inherit, so a form XObject without a default of its own draws in its page's, as Poppler and MuPDF draw it); a default that is a Lab, Indexed, Pattern or device space, has another number of components or is no colour space is ignored and device colour is drawn as before; device spaces inside other spaces (an Indexed base, a Separation or DeviceN alternate, a Pattern's underlying space), shading dictionaries, masks and transparency group spaces are unchanged, and the decoded-image cache (24) keys a decode by the default it was drawn in. hayro drew device colour as it was, so pages that calibrate it this way (common in PDF/A) came out in the wrong colours; (39) knockout groups (#2971): `Device::set_knockout` right after a `/K true` group is pushed; `Device::knockout_image` for an unmasked image before its group is pushed, when `Device::in_knockout_group`: the device clears its placement rectangle only if it will draw the image (on the canvas, not a preview's tiny image, and it decodes), decoded at its draw size so the draw reuses that decode; and, when `Device::knockout_erase_wanted`, a nested group's content drawn once between `begin_knockout_erase` and `end_knockout_erase` before the group itself, which then always gets a layer of its own; (40) an optional content membership dictionary's visibility expression (`/VE`: `/And`, `/Or`, `/Not`, nested up to 32 deep) decides its visibility ahead of `/OCGs` and `/P` (ISO 32000-2 §8.11.2.2); it was ignored, so content under it stayed visible with its layers hidden; a malformed expression, including one naming an object the document doesn't list in `/OCProperties /OCGs`, falls back to `/OCGs` and `/P` (#2986); (41) a transparency group that is neither isolated nor knockout, entered at full opacity with no soft mask and Normal blending, is drawn straight onto the page instead of into a layer of its own (§11.4.7: its elements composite with the backdrop; the layer made a `/BM /Multiply` highlight appearance an opaque block, #1968); (42) text in a Type 3 font is drawn filled whatever the text rendering mode, except mode 3 (invisible): only that mode has any effect on Type 3 fonts (ISO 32000-1 §9.3.6). Stroking ran the glyph procedure a second time, so `2 Tr` made bitmap fonts look bold, `1 Tr` painted an uncoloured glyph in the stroking colour, and `7 Tr` hid it (`7 Tr` since follows 64); (43) Lab colour spaces honour their own `/WhitePoint`: L\*a\*b\* goes to XYZ relative to that white, is Bradford-adapted to D65 and converted to sRGB (ISO 32000-2 §8.6.5.4) instead of through `LAB.icc` as if every white were D50 (a D65 blue came out 45 levels too red; the profile is removed); components normalised to [0, 1] (`manual_scale`, an `/Indexed` palette over Lab) map onto L\* 0..100 and the a\*/b\* `/Range`; (44) a Type 0 font whose `/ToUnicode` is the name `/Identity-H` or `/Identity-V` (not a CMap stream; WSCAD writes it) reads each code as its Unicode value, as Acrobat shows it (`Type0Font::identity_to_unicode`); the identity CMap the name resolved to maps codes to CIDs only, so such pages had no text to select or find; (46) an image XObject's group applies the fill opacity, so the fill alpha is reset to 1 before a stencil mask takes the fill colour as its paint, and a pattern paint drops the opacity it captured at `scn` (`Pattern::without_opacity`) (it was applied twice: a mask at `ca 0.5` was drawn at a quarter, #2975); (47) `path_as_rect`: a five-element path is a rectangle only when its four points are distinct, each step from one to the next runs along a single axis, and the last returns to the first. The length test admits one MoveTo and four LineTo as well as one MoveTo, three LineTo and a ClosePath, and asking only that every corner of the bounding box was visited says nothing about the order or the steps between, so a five-point polyline whose ends reached the corners was painted as that box -- a fourth side the path does not have -- and a clip of that shape was discarded; the coordinates are compared exactly, since the matrix is applied after this check and a `cm` of 10,000 turns `is_nearly_equal`'s 256th of a unit into tens of device units (#3327); (48) an `/ICCBased` colour space is cached by what its profile is read from (`IccProfileKey`): the stream object it names, looked up without reading the stream, and then the stream's dictionary and its own bytes (decrypted, not decoded), so a byte-for-byte copy of a profile in another object is read once, and a profile stream written directly in its array, which reports the object holding the array as every other stream written there does, is told apart by its bytes. hayro keyed it by the stream dictionary's bytes alone, so two different profiles with the same `/N` and `/Length` shared whichever was read first, and colours in the second were drawn through the first. `InterpreterCache::icc_profile_reads` counts the profiles read; (49) `OpenTypeFontBlob::new` also accepts a `post` table version 1.0, 2.0 or 3.0 written with its 16-bit halves swapped (`00 00 00 03`), as ScanSoft PDF Create! embeds every font; such fonts were refused and their text drawn with a substitute (Hebrew came out garbled or missing, #2946); `glyph_names` reads such a table from a copy with the version written correctly, so a 2.0 table's names serve `/Differences`. Any other unknown version is still refused (pdf.js 9462); (50) `CffFontBlob::glyph_index_by_name` looks a glyph name up in a name to glyph map built on first use (shared between clones) instead of searching the whole charset, with a string lookup per glyph, for every character code of a font; the first glyph with a name still wins; (51) at most 4096 graphics states saved at once, holding at most 64 MiB together (each `q` copies the whole state, dash array and clip list included, so short operators multiplied one large state); a `q` past either limit is counted but not copied, and its `Q` still removes the clips added since (from storytold/craft-libs#239); (52) a `/W2` run whose first CID is past `MAX_CID` stops there, as in `/W` (7), checked before the run's metrics are read: a first CID of 4294967295 inserted one entry and then overflowed, discarding the whole `/W2` table (from storytold/craft-libs#254); (55) a Separation or DeviceN tint transform that cannot be read keeps its colour space, and one that cannot be read or evaluated paints the alternate space's initial colour, as evaluation failures already did; that fallback, and a `cs`/`CS` colour space that cannot be read (drawn in DeviceGray as before), are now reported once per content stream as `InterpreterWarning::ColorFallback` (`RenderWarning::ColourFallback` in `pdfcraft-render`) instead of silently (#4260); (56) an annotation whose Invisible flag (bit 1) is set and whose `/Subtype` is none of the standard types of ISO 32000-2 Table 171 (`is_standard_annotation_type`; no `/Subtype` counts as unknown) is not drawn (Table 167), and a `/Popup` annotation's appearance is never painted on the page (§12.5.6.14: a popup has no appearance stream of its own; the viewer shows it as a window for its parent annotation). Both were drawn from their `/AP`; (57) a simple font drawn with a standard-font stand-in because the document names a font it doesn't embed (`StandardKind` with `fallback`) narrows a glyph to a narrower `/Widths` advance but no longer widens it to a wider one: the advance already places the next glyph (ISO 32000-2 §9.2.4; /Widths in Table 109), and widening drew a non-embedded `Arial,Bold` with 600-unit widths as a heavy fixed-pitch face whose `i` and `l` were solid blocks twice their width; an exact standard-14 font still matches its standard metrics both ways; (58) the fallback request for a Type 0 font that is not embedded (`FallbackFontQuery::with_descriptor`) takes its serif, italic and small-cap flags, `/FontWeight`, `/FontFamily` and `/FontStretch` from the descendant CIDFont's font descriptor (ISO 32000-2 §9.7.4.1, §9.7.6.1; §9.8): it read them from the Type 0 dictionary, which has none, so every such request asked for a regular sans-serif face and no resolver could tell a Song or Mincho font from a Hei or Gothic one, or see `/FontWeight 700`; (59) the NoRotate flag (`/F` bit 5) is honoured: the appearance, fitted to `/Rect`, is turned against the page's `/Rotate` about the rectangle's upper-left corner, which stays where it is, so it reads upright as the page is displayed (ISO 32000-2 §12.5.3). Sticky notes, which are saved NoRotate, were drawn turned with the page (#2583). Form fields and links stay in `/Rect`, where PdfCraft takes their click targets from it; (64) text in a Type 3 font follows ISO 32000-2 §9.3.6 in the clipping modes: under `7 Tr` it is not rendered (reported to the device as clipping text, `GlyphDrawMode::Clip`, so extraction keeps it), and in modes 4 to 7 a Type 3 glyph adds nothing to the clipping path (`clip_glyph`), so text shown only in a Type 3 font leaves no clip at `ET`. Since (42) `7 Tr` painted the glyphs in the fill colour (black by default), so a fill meant to follow showed them wherever it didn't cover them; 4 to 6 still render filled; (66) a Separation colour space whose colourant is the special name `All` paints every colourant at once (ISO 32000-2 §8.6.6.4): on screen and in an RGB print raster its tint t is the gray 1 − t, complemented because the device is additive, whatever its alternate space and tint transform say (hayro drew it through the tint transform, so a `/DeviceCMYK` transform giving cyan drew registration marks light cyan); `/None` still paints nothing, any other name still goes through its alternate, and a DeviceN space naming `All` keeps its alternate, since the clause gives the name that meaning in a Separation space; (63) a luminosity soft mask whose group colour space is `/DeviceCMYK` (by name, not an ICC-based or calibrated space) reads a `/DeviceCMYK` fill or stroke colour drawn in it, and its `/BC`, as their gray `1 - min(1, 0.3 c + 0.59 m + 0.11 y + k)` (ISO 32000-2 §10.4.2.3, §11.5.3: device colour converts to gray with no colour calibration; `Context::luminosity_in_device_cmyk`, `color::device_cmyk_to_gray`): it went through the CMYK display profile, where 100% K is a dark grey, so a 100% K mask let about an eighth of the masked paint through. CMYK images, shadings and tiling patterns inside such a mask, and CMYK on the page, still use the profile; (71) a lookup table in an indirect string reads: it was read as a stream first, which uses up a reference (ISO 32000-2 §8.6.6.3); (72) an uncoloured pattern's underlying space given as a name or a reference is used: it was skipped, so a CMYK cyan drew red (ISO 32000-2 §8.7.3.3); (73) `InterpreterSettings::jbig2_work`, set for each page in `interpret_page` and ended with it, and a JBIG2 image or an image with a JBIG2 mask past it left out with `ContentTruncated`; (74) colour spaces nest at most 32 levels, through every way one names another, so a chain of them ends instead of overflowing the stack; an ICC profile's `/Alternate` is read where it's met and isn't kept with the profile, or a page could build a chain past the limit on spaces already read; (75) a soft hyphen (U+00AD) that a CFF substitute for a CID font has no glyph for draws its hyphen, as Annex D.2 reads WinAnsi 0xAD in (30), instead of the CID taken as a glyph index (#5888) | (1) `raster::tests::widget_appearance_states`; (2, 3) `xtask text-oracle` on corpus/pdfjs issue918; (4) `raster::tests::layer_override_hides_content`; (5) `raster::tests::soft_mask_without_group_colour_space_is_applied`; (6) `raster::tests::self_referencing_tiling_pattern_terminates`, `raster::tests::self_referencing_type3_glyph_terminates`; (7) `raster::tests::huge_cid_width_ranges_terminate`; (8) `raster::tests::hiding_comments_keeps_fields`; (9) `raster::tests::absurd_mask_and_fax_sizes_are_refused_before_decoding`; (10) `raster::tests::type3_glyphs_that_fan_out_into_themselves_finish`; (11) `raster::tests::type1_font_programs_with_long_runs_of_numbers_load_quickly`; (12) `raster::tests::degenerate_annotation_appearances_are_skipped`; (13) `raster::tests::highlights_without_an_appearance_are_drawn`; (14) `raster::tests::malformed_simple_font_widths_keep_later_widths_and_the_font`, `raster::tests::huge_simple_font_first_char_terminates`; (15) `raster::tests::annotation_appearance_box_maps_onto_rect`; (16) `raster::tests::clipping_path_applies_after_any_painting_operator`; (17) `raster::tests::dropping_the_pool_cancels_its_renders`, `raster::tests::the_watchdog_stops_the_render_it_gives_up_on`; (18) `raster::tests::page_contents_and_nested_forms_share_one_budget`, `raster::tests::content_painted_again_and_again_keeps_drawing`, `raster::tests::sampled_functions_with_billions_of_samples_are_refused`; (19) `raster::tests::mesh_shadings_far_off_the_page_render`, `raster::tests::many_overlapping_mesh_triangles_finish`, `raster::tests::mesh_shadings_with_many_patches_or_no_rows_finish`; (20) `raster::tests::function_limits_calculator_arithmetic_and_depth`, `raster::tests::function_limits_public_construction_work_and_depth`, `raster::tests::function_limits_stitching_cycles_are_refused_and_shared_children_work`, `raster::tests::function_limits_invalid_calculator_keeps_the_page_renderable`; (21) `raster::tests::cids_a_substitute_lacks_are_not_drawn_as_other_glyphs`; (22) `raster::tests::variation_sequences_draw_their_kanji_in_a_substitute`; (23) `raster::tests::axial_and_radial_shadings_draw_as_native_gradients`; (24) `raster::tests::tiles_match_full_render`, `raster::tests::skipping_paths_off_the_canvas_changes_no_pixel` (both decode through the cache), `raster::tests::decoded_images_with_identical_dictionaries_keep_their_own_pixels`, `raster::tests::decoded_images_under_a_transfer_function_are_not_shared`; (25) `text::tests::actual_text_replaces_the_glyphs_it_marks_in_every_writing_direction`, `text::tests::actual_text_follows_the_pdf_rules_for_marked_content`; (26) `raster::tests::accented_letters_the_standard_stand_ins_lack_are_built_from_their_parts`, `raster::tests::comma_below_letters_in_standard_fonts_draw_a_comma_under_the_letter`; (27) `text::tests::truetype_without_an_encoding_reads_its_codes_as_win_ansi`; (28) `raster::tests::spot_colour_images_run_their_tint_transform_once_per_sample_value`, `raster::tests::images_converted_through_a_table_keep_every_samples_colour`, `raster::tests::a_malformed_tint_for_an_unused_sample_keeps_the_used_colours`; (29) `raster::tests::reading_colors_keep_highlight_annotations_readable`; (30) `text::tests::the_annex_d2_alias_codes_read_as_an_ordinary_hyphen_and_space`; (31) `raster::tests::shadings_with_large_sampled_functions_paint_quickly`, `raster::tests::sampled_functions_interpolate_their_grid_as_before`; (32) `raster::tests::decoded_images_serve_every_size_they_are_drawn_at`; (33) `raster::tests::patch_meshes_keep_their_corner_colours`; (34) `raster::tests::spot_colour_images_run_their_tint_transform_once_per_sample_value`; (35) `mesh_extent_tests::{a_mesh_paints_its_triangles_alone, a_mesh_keeps_the_graphics_state_alpha, a_mesh_off_the_canvas_paints_nothing}` (`cargo test -p hayro`); (36) `raster::tests::mesh_triangles_sharing_an_edge_leave_no_seam`; (38) `raster::tests::default_colour_spaces::device_rgb_operators_draw_in_default_rgb`, `raster::tests::default_colour_spaces::device_cmyk_operators_draw_in_default_cmyk`, `raster::tests::default_colour_spaces::device_gray_operators_draw_in_default_gray`, `raster::tests::default_colour_spaces::device_images_draw_in_the_default_colour_spaces`, `raster::tests::default_colour_spaces::device_shadings_draw_in_the_default_colour_spaces`, `raster::tests::default_colour_spaces::an_image_drawn_with_and_without_a_default_keeps_both_colours`, `raster::tests::default_colour_spaces::forms_inherit_the_default_colour_spaces_they_do_not_override`, `raster::tests::default_colour_spaces::unusable_defaults_leave_device_colour_alone`; (39) `raster::tests::knockout_groups_replace_what_their_elements_cover`; (40) `raster::tests::visibility_expressions_decide_what_a_membership_dictionary_shows`; (41) `raster::tests::a_multiply_highlight_inside_a_transparency_group_keeps_the_text`; (42) `raster::tests::rendering_modes_but_3_and_7_draw_type3_glyphs_as_filled`; (43) `raster::tests::lab_honours_the_space_white_point`, `raster::tests::indexed_palettes_over_lab_keep_their_colours`; (44) `text::tests::identity_to_unicode_names_give_text`; (46) `raster::tests::stencil_masks_are_drawn_at_the_fill_opacity_once`; (47) `raster::tests::a_five_point_polyline_is_not_painted_as_its_bounding_box`, `raster::tests::a_path_that_visits_every_corner_out_of_order_is_not_a_rectangle`, `raster::tests::a_rectangle_is_still_painted_as_a_rectangle`; (48) `raster::tests::icc_profile_cache::icc_profiles_with_identical_dictionaries_draw_their_own_colours`, `raster::tests::icc_profile_cache::icc_profiles_used_again_are_read_once`; (49) `raster::tests::a_font_whose_post_version_has_swapped_halves_is_still_used`; (50) `font::blob::tests::glyph_names_resolve_to_the_first_glyph_with_that_name`; (51) `context::saved_state_tests::repeated_saves_stay_bounded_and_balanced`; (52) `font::cid::tests::w2_run_past_the_largest_cid_keeps_the_rest_of_the_table` (hayro-interpret); (55) `raster::tests::devicen_inks::a_tint_transform_that_cannot_be_evaluated_draws_the_rest_of_the_page_and_warns`, `raster::tests::devicen_inks::a_tint_transform_that_cannot_be_read_falls_back_and_warns`; (56) `raster::tests::invisible_unknown_annotations_and_popups_paint_nothing`; (57) `substitute_widths::a_substitute_keeps_its_glyph_shapes_at_wider_advances`, `substitute_widths::a_substitute_still_narrows_glyphs_to_narrower_advances` (pdfcraft-render `tests/it`); (58) `cid_fallback_query::a_cid_font_fallback_request_carries_its_descendants_style`, `cid_fallback_query::a_cid_font_without_a_descriptor_asks_for_the_default_style` (pdfcraft-render integration tests); (59) `raster::tests::no_rotate_notes_stay_upright_at_their_corner_on_rotated_pages` (hayro-interpret); (64) `type3_text_clip::{clip_only_type3_text_is_not_rendered_and_leaves_no_clip, invisible_type3_text_is_not_rendered, other_modes_render_type3_glyphs_filled_and_never_clip, only_outline_glyphs_shape_a_clip_shared_with_type3_text, self_nesting_type3_text_in_a_clip_mode_stays_bounded}` (`cargo test -p pdfcraft-render --test it`), `raster::tests::rendering_modes_but_3_and_7_draw_type3_glyphs_as_filled`, `text::tests::clipping_mode_type3_text_is_kept_by_the_ocr_filter`; (66) `separation_all::{separation_all_fills_paint_the_complemented_tint_as_gray, separation_all_strokes_paint_the_complemented_tint_as_gray, separation_all_images_paint_each_sample_as_the_complemented_gray, separation_all_shadings_paint_the_complemented_gray, separation_none_paints_nothing, a_named_separation_still_paints_through_its_alternate, a_device_n_space_naming_all_still_paints_through_its_alternate}` (`cargo test -p pdfcraft-render --test it`); (63) `color::device_cmyk_gray_tests::{each_ink_darkens_by_its_iso_weight, ink_past_full_coverage_is_black_and_bad_components_read_as_zero}` (hayro-interpret), `it::luminosity_group_space::{black_ink_in_a_device_cmyk_group_masks_fully_and_no_ink_shows_fully, each_process_ink_in_a_device_cmyk_group_masks_by_its_weight, a_device_cmyk_backdrop_colour_masks_by_its_gray, gray_and_rgb_groups_mask_as_before, device_cmyk_on_the_page_keeps_its_display_colour}` (pdfcraft-render); (71) `raster::tests::an_indexed_lookup_table_is_read_in_every_legal_form` (pdfcraft-render); (72) `raster::tests::an_uncoloured_pattern_uses_its_underlying_space_in_every_form` (pdfcraft-render); (73) `raster::tests::the_jbig2_images_of_a_page_share_one_allowance`, `a_picture_whose_jbig2_mask_is_past_the_pages_allowance_is_not_drawn` and `a_pages_jbig2_allowance_ends_with_the_page` (pdfcraft-render); (74) `raster::tests::{a_chain_of_colour_spaces_ends, colour_spaces_are_read_to_the_limit_of_their_nesting, a_chain_built_on_spaces_already_read_ends, a_profile_read_at_the_limit_keeps_its_alternate_for_a_read_nearer_the_top}` (pdfcraft-render); (75) `raster::tests::a_soft_hyphen_a_substitute_lacks_draws_its_hyphen` (pdfcraft-render) | -| hayro | 0.7.1 | Apache-2.0 OR MIT | (1) `RenderSettings::{x_offset, y_offset}` to render one tile of a large page; (2) images over `MAX_IMAGE_PIXELS` (2^28) are skipped (a fuzzed `/W 4294967295` hung in resampling); (3) `tiling_cell_scale` keeps a tiling pattern's cell pixmap within 3000 px a side whatever its `/XStep`/`/YStep` (a fuzzed step far beyond `/BBox` allocated 2 GB); (4) stroke widths clamped to 4 × (canvas width + height) device pixels (a fuzzed `/LW 9223372036854775807` allocated 10 GB in stroke expansion); (5) `usable_dash_pattern`: a stroke whose dash pattern could cut it into more than `MAX_DASHES_PER_STROKE` (250,000) pieces, counting kurbo's restart of the pattern at every subpath and `ClosePath`, is drawn solid (a fuzzed `/D [[11] 0]` on a line from x = 92234775807 allocated over 5 GB, `[0.000001] 0 d` on a 20 pt line over 10 GB); (6) a dash pattern with a negative or non-finite entry, or a period that isn't positive, is drawn solid (`[-1 -1] 0 d` looped for ever in kurbo); (7) `render_size` and `render_into`: render into a buffer the caller allocates (PdfCraft renders into pixel-aligned memory that the GUI takes over as texture data); `render` is `render_into` on a new pixmap; (8) checked image source/target/RGBA, alpha-premultiplication scratch and intermediate sizes before resampling, finite transforms and exact source/mask buffers; sources wider than u16 may shrink before backend limits, with a 2^20 source-axis cap and conservative 128 MiB filter-planning estimate; an over-limit or failed resize keeps the original image resolution and geometry when the source fits the backend, otherwise skips it, with a logged warning; staging, resampling output/scratch, RGBA conversion and padding buffers use fallible reserves; (9) `may_paint_canvas`: a fill or stroke whose control-point box, grown by how far its paint reaches past the path (half the stroke width times the larger of the miter limit and √2) plus 0.5 user units and 2 pixels, misses the canvas is not drawn, so a tile no longer strokes and fills every path of its page (item 5 of #623); `set_skip_offscreen_paths` turns this off for the regression test; (10) `/Interpolate` governs the pre-resize too (#624): a minified image that does not request interpolation is pre-resized with nearest-neighbour, planned without the Catmull-Rom filter limits (`image_resampling_size` takes the flag); a Type 3 glyph image is sampled bicubically only when interpolated; a minified image with a matching soft mask is nearest-sampled straight into premultiplied RGBA; (11) `may_paint_rect`: an image whose stream-size box, mapped to the device, misses the canvas is skipped before decoding (decoded data can only shrink against that box), so a tile of an image-heavy page no longer decodes, resamples and rasterizes every image on it; (12) preview renders (`InterpreterSettings::preview`: backdrops, thumbnails, grid cells, signature-drag backgrounds) skip images under `PREVIEW_MIN_IMAGE_AREA` device pixels, while final renders always draw them; (13) reading colours (#897): `RenderSettings::color_replacement` takes a `ColorReplacement` (text colour, background, `only_black`, `line_art`) that remaps solid text and line-art paint while rendering, glyph outlines counting as text and stencil-mask ink as line art; images, image alpha masks, soft masks and pattern cells keep their colours; `render_pdf` gains a `color_replacement: Option` argument (pass `None` for true colours); the upstream-only `[dev-dependencies.log]` entry is removed from `Cargo.toml` (log is a normal dependency since the logging patch), so `cargo test -p hayro` runs this crate's unit tests; (14) a shrunk, interpolated image whose axes lie along the device's (upright, mirrored or a quarter turn, as on a page under `/Rotate`) is snapped to whole device pixels: shrunk once to its rounded device rectangle and drawn 1:1 from the rounded corner, instead of a ceil-sized raster resampled again at a fractional scale and offset (`snap_minified_image`; scans looked soft next to Acrobat, and until #5530 a turned one still did, up to 200 grey levels off the same page drawn upright). An image counts as axis-aligned when the terms the snap drops move its far edges by at most a hundredth of a device pixel (`SNAP_SKEW`), not by a tolerance on the coefficients, which let a slight skew of a very wide image through; and the snapped 1:1 transform is used only when the shrink produced the snapped size, so a failed resize keeps the full transform instead of drawing the source raster at its source size; (15) a soft mask of another size than its image is filtered when shrunk, whatever its `/Interpolate` (#1649): such a mask carries the shape on its own (Word saves text as a tiny black image with a glyph-bitmap mask), and nearest-neighbour dropped every stroke thinner than the shrink factor; (16) under a soft mask or a blend mode, axial and radial shadings are sampled instead of drawn as native gradients (#2128): vello_cpu 0.0.8 copies an opaque gradient over fully covered tiles without the mask or blend mode, so PowerPoint's faded gradient fills came out solid; (17) the renderer keeps annotation paints out of the reading-colours mapping between `begin_annotations` and `end_annotations` (#1548); (18) the page renders as one transparency group; under reading colours its background is painted inside it as the initial backdrop, so a Multiply highlight blends with the dark background instead of staying an opaque block, and otherwise the background stays outside, so a blend mode over an empty part of the page keeps its colour instead of blending with the white paper (#2970); (19) `set_paint` samples a mesh shading's texture over its triangles alone (`EncodedShadingPattern::extent`, padded rather than repeated) and skips a mesh with none on the canvas: a BricsCAD plan's 310 one-triangle `sh` each covered the whole tile, 2 s per 1024 px tile at any zoom (the watchdog gave the page up on a busy machine), now about 70 ms; (20) a tiling pattern whose step is too large for a cell pixmap, where what is painted (a fill, or a stroke with its width) can only show the first cell (PowerPoint's `/XStep 2147480100` for a picture fill drawn once), has its `/BBox` drawn at full scale with a transparent border and padded instead of tiled (`shows_one_cell`; a cell whose scale isn't finite and positive, from a `/BBox` too thin to divide by, keeps the old path: `usable_cell_scale`); shrinking the cell to fit the 3000 px limit of (3) left the picture under a pixel, so network diagrams lost their pictures (#3349); (22) knockout groups: while one is innermost, fills, strokes, glyph outlines and shadings are drawn with `Compose::Copy` (weighted by coverage), so each element replaces what earlier elements of the group drew instead of compositing over it; an image or nested group, drawn into a layer of its own, first has what is under its shape cleared (`knockout_image` for an unmasked image's rectangle, only when `draw_image` will draw it, decoded at the size it is drawn; a knockout erase pass that draws a nested group's content with every paint clearing instead of painting, a shading fill's clear clipped to the shading's `/BBox` as its paint is) and then composites over it (#2971); (23) backport of hayro 0.8: shadings that `as_svg_gradient` can represent are painted as a `vello_cpu` linear or two-point radial gradient instead of a texture sampled pixel by pixel through the shading function (InDesign fades are luminosity soft masks of one full-page axial `sh`: 40 on a report page took 15 s at 144 dpi, now under 1 s); without upstream's half-pixel shift, which suits vello_cpu 0.3 but moves 0.0.8's gradients off pixel centres; (25) a tiling cell is drawn at the scale that makes one `/XStep` (`/YStep`) exactly the cell pixmap's whole number of pixels (`tiling_cell_pixels`, at least 1 and at most the cap), so the repeated pixmap puts cells at whole multiples of the step at any device scale; drawing at the unrounded scale repeated the cell up to half a pixel off at every step, and the error grew with distance from the pattern's origin (ISO 32000-2 8.7.3.1) | (1) `raster::tests::tiles_match_full_render`; (2) `raster::tests::absurd_image_dimensions_are_skipped`; (3) `raster::tests::tiling_cells_stay_small_whatever_the_step`; (4) `raster::tests::absurd_predictor_columns_and_line_widths_render`; (5) `raster::tests::dash_patterns_that_cut_a_stroke_into_billions_of_pieces_render`; (6) `raster::tests::invalid_dash_patterns_terminate_and_are_drawn_solid`; (7) `raster::tests::pixels_are_a_plain_renders_bytes_in_a_buffer_a_gui_can_take_over`; (8) `raster::tests::image_resampling_rejects_anisotropic_target_growth`, `raster::tests::image_resampling_bounds_intermediate_before_planning`, `raster::tests::image_resampling_rejects_invalid_sources_and_scales`, `raster::tests::image_resampling_keeps_valid_sizes_and_exact_limits`, `raster::tests::resampling_fallback_preserves_original_geometry_and_pixels`, `raster::tests::image_resampling_wide_sources_shrink_before_backend_side_limits`, `raster::tests::image_resampling_wide_transparent_sources_still_shrink`, `raster::tests::image_resampling_padded_backend_dimensions_stay_checked`, `raster::tests::image_resampling_mismatched_alpha_mask_keeps_pixels_and_placement`; (9) `raster::tests::skipping_paths_off_the_canvas_changes_no_pixel`; (10) `raster::tests::non_interpolated_images_are_not_smoothed_when_minified`, `raster::tests::non_interpolated_gray_images_are_not_smoothed_when_minified`, `raster::tests::non_integer_minification_samples_the_same_columns`, `raster::tests::non_integer_minification_samples_colour_and_mask_from_the_same_columns`, `raster::tests::non_interpolated_alpha_images_keep_their_colours_and_mask_when_minified`, `raster::tests::non_interpolated_type3_glyph_images_are_not_smoothed_when_minified`, `raster::tests::image_resampling_nearest_accepts_wide_sources_the_filter_refuses`, `raster::tests::wide_non_interpolated_sources_draw_after_shrinking`; (11) `raster::tests::skipping_paths_off_the_canvas_changes_no_pixel`; (12) `raster::tests::preview_renders_skip_tiny_images_but_finals_draw_everything`; (13) `color_replacement_tests::{black_maps_to_text_colour, white_maps_to_background, midgrey_interpolates, alpha_copied_verbatim, only_black_threshold, line_art_false_skips_paths_but_not_text}`, `color_replacement_render_tests::{color_replacement_maps_text_and_line_art, color_replacement_maps_stencil_mask_ink}`; (14) `raster::tests::interpolated_minified_image_edges_snap_to_whole_pixels` (`cargo test -p hayro`), `raster::tests::a_turned_or_mirrored_minified_image_snaps_to_whole_pixels`, `rotated_image_snap::a_shrunk_image_on_a_turned_page_is_the_upright_render_turned` (`cargo test -p pdfcraft-render --test it`), `renderer::snap_tests::a_skew_that_moves_a_far_edge_is_not_snapped` (`cargo test -p hayro`); (15) `raster::tests::shrunk_soft_mask_larger_than_its_image_keeps_thin_strokes`; (16) `raster::tests::soft_mask_applies_to_a_shading_pattern_fill`; (17), (18) `raster::tests::reading_colors_keep_highlight_annotations_readable`; (19) `mesh_extent_tests::{a_mesh_paints_its_triangles_alone, a_mesh_keeps_the_graphics_state_alpha, a_mesh_off_the_canvas_paints_nothing}` (`cargo test -p hayro`); (20) `raster::tests::a_tiling_pattern_drawn_once_shows_its_cell`, `raster::tests::a_large_step_still_tiles_where_the_next_cell_shows`, `raster::tests::a_tiling_cell_too_thin_to_scale_keeps_the_old_path`; (22) `raster::tests::knockout_groups_replace_what_their_elements_cover`; (23) `raster::tests::axial_and_radial_shadings_draw_as_native_gradients`; (25) `tiling_cell_placement::{far_cells_of_a_fill_sit_at_whole_steps, far_cells_sit_at_whole_steps_under_a_turned_scaled_matrix, far_cells_of_a_stroke_sit_at_whole_steps, a_cell_pixmap_spans_exactly_one_step}` (`cargo test -p pdfcraft-render --test it`) | +| hayro | 0.7.1 | Apache-2.0 OR MIT | (1) `RenderSettings::{x_offset, y_offset}` to render one tile of a large page; (2) images over `MAX_IMAGE_PIXELS` (2^28) are skipped (a fuzzed `/W 4294967295` hung in resampling); (3) `tiling_cell_scale` keeps a tiling pattern's cell pixmap within 3000 px a side whatever its `/XStep`/`/YStep` (a fuzzed step far beyond `/BBox` allocated 2 GB); (4) stroke widths clamped to 4 × (canvas width + height) device pixels (a fuzzed `/LW 9223372036854775807` allocated 10 GB in stroke expansion); (5) `usable_dash_pattern`: a stroke whose dash pattern could cut it into more than `MAX_DASHES_PER_STROKE` (250,000) pieces, counting kurbo's restart of the pattern at every subpath and `ClosePath`, is drawn solid (a fuzzed `/D [[11] 0]` on a line from x = 92234775807 allocated over 5 GB, `[0.000001] 0 d` on a 20 pt line over 10 GB); (6) a dash pattern with a negative or non-finite entry, or a period that isn't positive, is drawn solid (`[-1 -1] 0 d` looped for ever in kurbo); (7) `render_size` and `render_into`: render into a buffer the caller allocates (PdfCraft renders into pixel-aligned memory that the GUI takes over as texture data); `render` is `render_into` on a new pixmap; (8) checked image source/target/RGBA, alpha-premultiplication scratch and intermediate sizes before resampling, finite transforms and exact source/mask buffers; sources wider than u16 may shrink before backend limits, with a 2^20 source-axis cap and conservative 128 MiB filter-planning estimate; an over-limit or failed resize keeps the original image resolution and geometry when the source fits the backend, otherwise skips it, with a logged warning; staging, resampling output/scratch, RGBA conversion and padding buffers use fallible reserves; (9) `may_paint_canvas`: a fill or stroke whose control-point box, grown by how far its paint reaches past the path (half the stroke width times the larger of the miter limit and √2) plus 0.5 user units and 2 pixels, misses the canvas is not drawn, so a tile no longer strokes and fills every path of its page (item 5 of #623); `set_skip_offscreen_paths` turns this off for the regression test; (10) `/Interpolate` governs the pre-resize too (#624): a minified image that does not request interpolation is pre-resized with nearest-neighbour (since (21), only inside Type 3 glyphs), planned without the Catmull-Rom filter limits (`image_resampling_size` takes the flag); a Type 3 glyph image is sampled bicubically only when interpolated; a minified image with a matching soft mask is nearest-sampled straight into premultiplied RGBA; (11) `may_paint_rect`: an image whose stream-size box, mapped to the device, misses the canvas is skipped before decoding (decoded data can only shrink against that box), so a tile of an image-heavy page no longer decodes, resamples and rasterizes every image on it; (12) preview renders (`InterpreterSettings::preview`: backdrops, thumbnails, grid cells, signature-drag backgrounds) skip images under `PREVIEW_MIN_IMAGE_AREA` device pixels, while final renders always draw them; (13) reading colours (#897): `RenderSettings::color_replacement` takes a `ColorReplacement` (text colour, background, `only_black`, `line_art`) that remaps solid text and line-art paint while rendering, glyph outlines counting as text and stencil-mask ink as line art; images, image alpha masks, soft masks and pattern cells keep their colours; `render_pdf` gains a `color_replacement: Option` argument (pass `None` for true colours); the upstream-only `[dev-dependencies.log]` entry is removed from `Cargo.toml` (log is a normal dependency since the logging patch), so `cargo test -p hayro` runs this crate's unit tests; (14) a shrunk image that is interpolated (or, since (21), area-averaged) and whose axes lie along the device's (upright, mirrored or a quarter turn, as on a page under `/Rotate`) is snapped to whole device pixels: shrunk once to its rounded device rectangle (halves round up, so a tile snaps an edge where the whole page does) and drawn 1:1 from the rounded corner, instead of a ceil-sized raster resampled again at a fractional scale and offset (`snap_minified_image`; scans looked soft next to Acrobat, and until #5530 a turned one still did, up to 200 grey levels off the same page drawn upright). An image counts as axis-aligned when the terms the snap drops move its far edges by at most a hundredth of a device pixel (`SNAP_SKEW`), not by a tolerance on the coefficients, which let a slight skew of a very wide image through; and the snapped 1:1 transform is used only when the shrink produced the snapped size, so a failed resize keeps the full transform instead of drawing the source raster at its source size; (15) a soft mask of another size than its image is filtered when shrunk, whatever its `/Interpolate` (#1649): such a mask carries the shape on its own (Word saves text as a tiny black image with a glyph-bitmap mask), and nearest-neighbour dropped every stroke thinner than the shrink factor; (16) under a soft mask or a blend mode, axial and radial shadings are sampled instead of drawn as native gradients (#2128): vello_cpu 0.0.8 copies an opaque gradient over fully covered tiles without the mask or blend mode, so PowerPoint's faded gradient fills came out solid; (17) the renderer keeps annotation paints out of the reading-colours mapping between `begin_annotations` and `end_annotations` (#1548); (18) the page renders as one transparency group; under reading colours its background is painted inside it as the initial backdrop, so a Multiply highlight blends with the dark background instead of staying an opaque block, and otherwise the background stays outside, so a blend mode over an empty part of the page keeps its colour instead of blending with the white paper (#2970); (19) `set_paint` samples a mesh shading's texture over its triangles alone (`EncodedShadingPattern::extent`, padded rather than repeated) and skips a mesh with none on the canvas: a BricsCAD plan's 310 one-triangle `sh` each covered the whole tile, 2 s per 1024 px tile at any zoom (the watchdog gave the page up on a busy machine), now about 70 ms; (20) a tiling pattern whose step is too large for a cell pixmap, where what is painted (a fill, or a stroke with its width) can only show the first cell (PowerPoint's `/XStep 2147480100` for a picture fill drawn once), has its `/BBox` drawn at full scale with a transparent border and padded instead of tiled (`shows_one_cell`; a cell whose scale isn't finite and positive, from a `/BBox` too thin to divide by, keeps the old path: `usable_cell_scale`); shrinking the cell to fit the 3000 px limit of (3) left the picture under a pixel, so network diagrams lost their pictures (#3349); (21) an image that does not request interpolation and is drawn below its resolution, outside a Type 3 glyph, is shrunk by area averaging (`area_shrink`: each device pixel the coverage-weighted mean of the samples it covers, in 16.16 fixed point whose weights sum to exactly one, premultiplied when the image has a mask; a soft mask no larger than the image is first sampled onto the image's grid by `mask_on_image_grid`, so its colour and mask are averaged together instead of the averaged colour of transparent samples showing under a mask layer at the mask's edges) to its snapped device rectangle, instead of nearest-neighbour: `/Interpolate` asks for smoothing only when an image is enlarged (ISO 32000-2 §8.9.5.3), and nearest-neighbour kept one sample in N per axis, so the text of a 300 dpi scan at 100% lost most of its strokes (sampling them again at a fractional offset dropped more). PicScale's `Box` and `Area` filters were not used: shrinking, both average a window four target pixels wide. On a 13-page 300 dpi scan every rendered sample at 96, 120 and 192 dpi is within 1 of the exact area average (it was up to 255 off, 17-23% of samples by more than 1). A snapped image whose shrink failed keeps the full transform, so it is not drawn at its source size. The shrink never enlarges: a side drawn at or above its resolution (an image shrunk across and stretched down, as a 204 × 98 dpi fax is) keeps its samples for the sampler. Rows are averaged first (contiguous, cheap) and the fewer rows that leaves are then shrunk across with taps worked out once per image, which made the shrink about three times faster (a 300 dpi page at 96 dpi: 25 ms, nearest-neighbour 11 ms). A source with a side over `MAX_RESAMPLING_SOURCE_SIDE` (2^20) keeps (10)'s nearest-neighbour shrink, and `area_shrink` refuses it, so its accumulator row and tap table stay a few times one source row (a 2^27 × 2 image of zeros is a tiny file); (22) knockout groups: while one is innermost, fills, strokes, glyph outlines and shadings are drawn with `Compose::Copy` (weighted by coverage), so each element replaces what earlier elements of the group drew instead of compositing over it; an image or nested group, drawn into a layer of its own, first has what is under its shape cleared (`knockout_image` for an unmasked image's rectangle, only when `draw_image` will draw it, decoded at the size it is drawn; a knockout erase pass that draws a nested group's content with every paint clearing instead of painting, a shading fill's clear clipped to the shading's `/BBox` as its paint is) and then composites over it (#2971); (23) backport of hayro 0.8: shadings that `as_svg_gradient` can represent are painted as a `vello_cpu` linear or two-point radial gradient instead of a texture sampled pixel by pixel through the shading function (InDesign fades are luminosity soft masks of one full-page axial `sh`: 40 on a report page took 15 s at 144 dpi, now under 1 s); without upstream's half-pixel shift, which suits vello_cpu 0.3 but moves 0.0.8's gradients off pixel centres; (25) a tiling cell is drawn at the scale that makes one `/XStep` (`/YStep`) exactly the cell pixmap's whole number of pixels (`tiling_cell_pixels`, at least 1 and at most the cap), so the repeated pixmap puts cells at whole multiples of the step at any device scale; drawing at the unrounded scale repeated the cell up to half a pixel off at every step, and the error grew with distance from the pattern's origin (ISO 32000-2 8.7.3.1); (28) an image of page-scan size (`SMOOTHED_IMAGE_SIDE`: 512 samples or more on its shorter side) that does not request interpolation is drawn with `ImageQuality::Medium` (bilinear, as interpolated images are) when enlarged on either side, also when the other side is shrunk (the area shrink leaves that side at 1:1), outside a Type 3 glyph: scans rarely set `/Interpolate`, and drawn nearest-neighbour every sample of a zoomed scan became a block, so its text had staircase edges; icons, QR codes and pixel art keep their hard pixels. Bicubic (`High`) looked the same on a zoomed 300 dpi scan and drew four 1024 px tiles at 400% in 267 ms against 50 ms | (1) `raster::tests::tiles_match_full_render`; (2) `raster::tests::absurd_image_dimensions_are_skipped`; (3) `raster::tests::tiling_cells_stay_small_whatever_the_step`; (4) `raster::tests::absurd_predictor_columns_and_line_widths_render`; (5) `raster::tests::dash_patterns_that_cut_a_stroke_into_billions_of_pieces_render`; (6) `raster::tests::invalid_dash_patterns_terminate_and_are_drawn_solid`; (7) `raster::tests::pixels_are_a_plain_renders_bytes_in_a_buffer_a_gui_can_take_over`; (8) `raster::tests::image_resampling_rejects_anisotropic_target_growth`, `raster::tests::image_resampling_bounds_intermediate_before_planning`, `raster::tests::image_resampling_rejects_invalid_sources_and_scales`, `raster::tests::image_resampling_keeps_valid_sizes_and_exact_limits`, `raster::tests::resampling_fallback_preserves_original_geometry_and_pixels`, `raster::tests::image_resampling_wide_sources_shrink_before_backend_side_limits`, `raster::tests::image_resampling_wide_transparent_sources_still_shrink`, `raster::tests::image_resampling_padded_backend_dimensions_stay_checked`, `raster::tests::image_resampling_mismatched_alpha_mask_keeps_pixels_and_placement`; (9) `raster::tests::skipping_paths_off_the_canvas_changes_no_pixel`; (10) `raster::tests::non_interpolated_images_are_not_smoothed_when_minified`, `raster::tests::non_interpolated_gray_images_are_not_smoothed_when_minified`, `raster::tests::non_integer_minification_averages_the_columns_each_pixel_covers`, `raster::tests::non_integer_minification_averages_colour_and_mask_together`, `raster::tests::non_interpolated_alpha_images_keep_their_colours_and_mask_when_minified`, `raster::tests::non_interpolated_type3_glyph_images_are_not_smoothed_when_minified`, `raster::tests::image_resampling_nearest_accepts_wide_sources_the_filter_refuses`, `raster::tests::wide_non_interpolated_sources_draw_after_shrinking`; (11) `raster::tests::skipping_paths_off_the_canvas_changes_no_pixel`; (12) `raster::tests::preview_renders_skip_tiny_images_but_finals_draw_everything`; (13) `color_replacement_tests::{black_maps_to_text_colour, white_maps_to_background, midgrey_interpolates, alpha_copied_verbatim, only_black_threshold, line_art_false_skips_paths_but_not_text}`, `color_replacement_render_tests::{color_replacement_maps_text_and_line_art, color_replacement_maps_stencil_mask_ink}`; (14) `raster::tests::minified_image_edges_snap_to_whole_pixels`, `raster::tests::a_tile_snaps_half_pixel_image_edges_as_the_whole_page_does` (`cargo test -p hayro`), `raster::tests::a_turned_or_mirrored_minified_image_snaps_to_whole_pixels`, `rotated_image_snap::a_shrunk_image_on_a_turned_page_is_the_upright_render_turned` (`cargo test -p pdfcraft-render --test it`), `renderer::snap_tests::a_skew_that_moves_a_far_edge_is_not_snapped` (`cargo test -p hayro`); (15) `raster::tests::shrunk_soft_mask_larger_than_its_image_keeps_thin_strokes`; (16) `raster::tests::soft_mask_applies_to_a_shading_pattern_fill`; (17), (18) `raster::tests::reading_colors_keep_highlight_annotations_readable`; (19) `mesh_extent_tests::{a_mesh_paints_its_triangles_alone, a_mesh_keeps_the_graphics_state_alpha, a_mesh_off_the_canvas_paints_nothing}` (`cargo test -p hayro`); (20) `raster::tests::a_tiling_pattern_drawn_once_shows_its_cell`, `raster::tests::a_large_step_still_tiles_where_the_next_cell_shows`, `raster::tests::a_tiling_cell_too_thin_to_scale_keeps_the_old_path`; (21) `raster::tests::non_interpolated_minification_keeps_thin_strokes`, `raster::tests::non_integer_minification_averages_the_columns_each_pixel_covers`, `raster::tests::non_integer_minification_averages_colour_and_mask_together`, `raster::tests::non_interpolated_minification_does_not_bleed_transparent_colour`, `raster::tests::a_smaller_soft_mask_does_not_bleed_transparent_colour_when_minified`, `raster::tests::minified_image_edges_snap_to_whole_pixels`, `raster::tests::an_image_shrunk_one_way_and_stretched_the_other_is_still_averaged`, `raster::tests::wide_non_interpolated_sources_draw_after_shrinking`, `renderer::area_shrink_tests` (`cargo test -p hayro`); (22) `raster::tests::knockout_groups_replace_what_their_elements_cover`; (23) `raster::tests::axial_and_radial_shadings_draw_as_native_gradients`; (25) `tiling_cell_placement::{far_cells_of_a_fill_sit_at_whole_steps, far_cells_sit_at_whole_steps_under_a_turned_scaled_matrix, far_cells_of_a_stroke_sit_at_whole_steps, a_cell_pixmap_spans_exactly_one_step}` (`cargo test -p pdfcraft-render --test it`); (28) `raster::tests::large_non_interpolated_images_are_smoothed_when_enlarged`, `raster::tests::small_non_interpolated_images_keep_hard_pixels_when_enlarged`, `raster::tests::a_scan_shrunk_one_way_is_smoothed_the_way_it_is_stretched` | | hayro-jbig2 | 0.3.0 | Apache-2.0 OR MIT | (1) at most 2^28 pixels per bitmap and page bitmap, and 2^25 cells per halftone grid and grayscale image (each side could be 65535: 4.3 billion pixels decoded one by one, and a 4-byte value per cell); (2) at most 2^26 symbol pixels per symbol dictionary (a fuzzed dictionary decoded for minutes); (3) a symbol dictionary's height classes are capped at one per new symbol and its export runs at `2 × symbols + 1`, each plus `EXTRA_EMPTY_STEPS` (1024): an empty height class or a zero-length export run doesn't advance, and past the end of its data the arithmetic decoder can decode them for ever (one changed byte hung the renderer); (4) a segment header's long-form referred-to segment count is capped at `u16::MAX` and a larger count is refused before its retention or referred-to bytes are read: the 29-bit count went straight to `Vec::with_capacity`, a reservation of about 2 GiB whose failure aborts the process (from storytold/craft-libs#176); (5) what a page's segments hold for later ones (intermediate regions, pattern and symbol dictionaries with their headers and retained contexts, and Huffman tables) is at most 64 MiB, two square bitmaps of 16,384 a side (`MAX_RETAINED_WORDS`), and a dictionary's new and exported symbols are counted against it as they are made: each was bounded on its own, but a few dozen bytes an intermediate region held 33 MB each, and 8 MiB symbols exported again and again killed the process, so a few kilobytes could exhaust memory; (6) a segment may refer to at most 2^20 symbols, checked before any are collected: references may repeat, and 65,535 of them to one dictionary, in a 78 KB stream, asked for 17 GB; (7) a pattern dictionary keeps at most 16 MiB of shifted copies, which only speed up drawing, and none that wouldn't fit what the page has left (`MAX_SHIFTED_WORDS`); (8) a custom Huffman table has at most 2^16 lines, and a stream with its globals at most 2^20 segments and 2^22 references, checked as they are read: both grew before anything counted them, and PDFs of a kilobyte or so reached 5 GB and more behind nested Flate; (9) a page past the bounds of (1) is refused when the stream is opened: a caller sizes its output from the header before the page bitmap is made, and an Indexed image in a 743-byte PDF asked for 137 GB, which aborts; (10) a stream may ask for at most 2^30 units of decoding work, four of the largest pages, about 6 s to 17 s (`MAX_WORK`): decoding a bitmap costs a pixel each and 4 a row, copying one a word each and 1 a row, and each integer an item reads costs 38, what one takes at worst, so one with no pixels pays too, before it is decoded: each bitmap was bounded, but nothing bounded how many there were, and a 795-byte PDF took 265 s; (11) `Image::decode_within` spends from what a run of images may still do, starting with its page's pixels, and `charge_reading` charges reading a stream and its globals before anything is parsed, so an operation over a document can bound what `MAX_WORK` bounds for one image, and images that fail or decode nothing add up too: a file of thousands of images made to reach that limit took as many times as long; (12) adding to a location, a reference offset, a symbol's corner or a Huffman range line near the end of an `i32`, and negating a decoded integer of 2^31, overflowed: a panic wherever overflow is checked. The refinement context, a text region's corners, the integer decoder and the Huffman table wrap, as a release build did, and `Bitmap::combine` saturates, which comes to the same, as a location that overflows is past the end of any page. What is drawn doesn't change | (1) `raster::tests::jbig2_regions_of_billions_of_pixels_are_refused`, `raster::tests::jbig2_halftone_grids_past_their_bound_are_refused`; (2) the nightly `cargo xtask fuzz` finding, which now finishes in about a second; (3) `raster::tests::jbig2_symbol_dictionaries_that_never_advance_terminate`; (4) `segment::tests::test_referred_to_count_is_capped` (`cargo test --manifest-path vendor/hayro-jbig2/Cargo.toml`); (5) `raster::tests::jbig2_regions_held_past_a_pages_bound_are_refused`, `jbig2_symbols_without_pixels_count_towards_a_pages_bound`, `jbig2_retained_contexts_count_towards_a_pages_bound`, `jbig2_huffman_tables_count_towards_a_pages_bound`, `jbig2_new_symbols_count_towards_a_pages_bound`, `jbig2_new_symbols_are_held_to_what_a_page_has_left`, `jbig2_pattern_dictionaries_count_towards_a_pages_bound`, and in the crate (`cargo test --manifest-path vendor/hayro-jbig2/Cargo.toml`) `decode::symbol::tests::exported_symbols_are_counted_before_they_are_copied`, `tests::the_room_left_shrinks_as_segments_are_stored`; (6) `raster::tests::jbig2_segments_referring_to_too_many_symbols_are_refused`; (7) `raster::tests::jbig2_pattern_dictionaries_too_large_to_shift_still_draw`, `jbig2_shifted_patterns_never_refuse_an_image`; (8) `raster::tests::jbig2_huffman_tables_of_too_many_lines_are_refused`, `jbig2_streams_of_too_many_segments_are_refused`, `jbig2_streams_of_too_many_references_are_refused`, and in the crate `tests::the_segments_of_a_stream_and_its_globals_count_together`; (9) `raster::tests::jbig2_pages_past_the_bounds_are_refused_before_anything_is_sized_from_them`, `tests::a_page_past_the_bounds_is_refused_when_opened`; (10) `raster::tests::jbig2_decoding_work_is_counted_item_by_item`; (11) `jbig2::decode::tests::a_run_of_images_spends_its_allowance_down_to_the_step` (pdfcraft-filters); (12) `raster::tests::jbig2_placements_at_the_end_of_an_i32_draw_nothing` | | lopdf | 0.45.0 | MIT | (1) PNG predictor rows longer than the data are refused before their buffers are allocated (a fuzzed `/Columns 4294967295` allocated two 4 GiB rows), and empty frames return without row buffers; (2) predictor row widths use saturating arithmetic or reject overflow; (3) sub-byte TIFF Predictor 2 scratch is bounded by the samples present in the input instead of the declared `/Colors`, aligned with [lopdf#589](https://github.com/J-F-Liu/lopdf/pull/589); (4) the object lookup keeps its cycle history through compressed objects, requires an ordinary in-file container, and caps the chain of indirect lookups at 16 (a chain of new objects, each a `/Length` compressed in a container whose own `/Length` names the next, or a stream whose `/Length` names the next stream, overflowed the stack: a dev build's 2 MiB thread within 800 hops); (5) object-stream members are parsed within their own extent (up to the next distinct offset), an entry whose offset repeats an earlier entry's is skipped (so a repeated object number can resolve to its earlier entry), and offsets are processed in index order; (6) `find_latest_trailer`: a rebuild takes a `/Type /XRef` stream dictionary as the trailer when the file has no `trailer` keyword (the latest in the file wins, as a later `trailer` would), so a concatenated or truncated file keeps its /Info and /ID instead of failing to load; a candidate whose /Root is the catalog beats one whose /Root is another dictionary, and an in-file /Root must match its generation (#2365) | (1) `inspect::tests::png_predictor_rows_longer_than_the_data_are_refused_before_allocating`, `inspect::tests::empty_png_predictor_frames_do_not_allocate_rows`; (2) `inspect::tests::tiff_subbyte_predictor_row_width_overflow_is_refused`; (3) `inspect::tests::tiff_subbyte_predictor_scratch_is_bounded_by_available_samples`, `inspect::tests::tiff_subbyte_predictors_keep_components_rows_and_padding`; (4) `inspect::tests::compressed_object_self_container_does_not_recurse`, `inspect::tests::compressed_object_mutual_containers_do_not_recurse`, `inspect::tests::object_stream_length_cycle_does_not_recurse`, `inspect::tests::compressed_object_container_chain_does_not_recurse`, `inspect::tests::compressed_length_chain_does_not_recurse`, `inspect::tests::plain_stream_length_chain_does_not_recurse`, `inspect::tests::compressed_objects_and_indirect_stream_length_still_load`, `inspect::tests::compressed_length_in_a_container_with_an_indirect_length_still_loads`; (5) `inspect::tests::object_stream_member_extents_bound_owned_strings`, `inspect::tests::object_stream_member_extents_preserve_index_order_and_whitespace`; (6) `inspect::tests::rebuilt_file_keeps_the_info_of_its_cross_reference_stream`, `inspect::tests::rebuilt_file_prefers_a_trailer_whose_root_is_the_catalog` | | hayro-syntax | 0.7.2 | Apache-2.0 OR MIT | (1) Page-tree cycle guard in `resolve_pages` (visited set + depth cap): a `/Kids` loop among compressed objects overflowed the stack and aborted the process; (2) `ccitt_size_ok`: CCITT images wider than 2^20 columns or over 2^28 pixels are not decoded (a fuzzed `/Columns 4294967295` allocated 4 GiB); (3) Flate/LZW predictor row widths are saturating `usize`, a zero width is refused and a row longer than the data is not allocated (a fuzzed `/Columns 9223372036854775807` wrapped to a 2^61-byte allocation and aborted); (4) the inline-image end-of-data check (`EI`) doesn't nest inside its own look-ahead and examines at most 256 candidates and steps (50 images with "EI" in their data ran for over ten minutes); an EI followed by a delimiter (EI%..., EI[...) ends the image like one followed by white space; only a regular character after EI skips the candidate; (5) a stream whose `/Length` is wrong is recovered only when `stream` is the next token — scanning ahead to a later object's `stream` made an indirect appearance-state dictionary (Quartz checkboxes: `/AP /N` → `<< /Yes … /Off … >>`) render as a non-form, so the widget was skipped; (6) JPX rescaling uses an exact integer sample range for 1–32-bit components (32 shifted past u32, and floating-point maxima at 25–31 bits wrapped to zero), and refuses zero-sized rows; (7) `decode_limit`: Flate (through flate2 and the fallback inflater), LZW, RunLength and ASCII85 never write past `MAX_DECODED_STREAM` (256 MiB), or, for the last filter of an image of known size, past twice its pixel bytes (a predictor tag byte per byte), at most 1 GiB; `Stream::decoded_within` for budgeted callers; a page's `/Contents` array shares one `MAX_DECODED_STREAM` (a 2 MB page whose content stream or 8 × 8 image inflated to 2 GB took the renderer past 3 GB in under a second; naming the stream twenty times in `/Contents`, 5 GB); (8) high-precision numbers (jsPDF MediaBox coordinates with over 19 digits) parse without overflowing the u64 mantissa: fractional digits past it are dropped, integer digits past it scale the value (a real, kept finite); (9) large dictionary indexes are built on first use and shared across repeated reads within each PDF (#307); the validated-byte cache is bounded, reader contexts remain separate, repairs and inline/content dictionaries bypass sharing, and small dictionaries stay eager; (10) decoded object-stream slots publish once under contention, cold loading rejects recursive dependencies, and xref repair serializes without holding recursive read guards; this permits one parser per native render pool with private worker caches; (11) `sync.rs`: a poisoned lock is recovered instead of unwrapped (a worker panic no longer cascades into every reader of the same PDF), and xref repair takes a blocking write lock instead of asserting a non-blocking one; (12) `SegmentList` (`util.rs`) returns `None` for an out-of-range index instead of panicking; (13) a page's `/UserUnit` scales `render_dimensions` and `initial_transform`, so the page and everything on it are sized in points, as Acrobat shows them; it is read as Acrobat reads it (the page's own entry, not inherited; a number from 1 up, at most 75,000; anything else counts as 1), and a page without it keeps exactly the transform it had; (14) revision 2–4 passwords are tried as given and then in PDFDocEncoding (`legacy_passwords` in `crypto/mod.rs`, which resolves the upstream `TODO: Convert to PDFDocEncoding`): passwords arrive as UTF-8 text, so a password with characters outside ASCII (`Çok güzel`, `ılık€`) written in PDFDocEncoding, as ISO 32000-2 §7.6.4.3.2 asks and qpdf and Acrobat write it, was always wrong; (15) page content is read a piece at a time (`page::ContentReader`, `Page::content_reader`): plain streams and Flate streams without a predictor are decoded as they are read, with a budget of 64 times their compressed size and never less than `MAX_DECODED_STREAM` (a stream that expands past it is cut and reported); other streams are decoded whole, as before, and share `MAX_DECODED_STREAM`. The content tokenizer reads a window at a time (`UntypedIter::new_window`, `TypedIter::needs_more`, `TypedIter::committed`): an instruction that runs past the window is not decided until more data follows, which `Reader` records by flagging reads and `at_end` past the end of the data, including through the clones the parser makes. The Flate decoder keeps its own 32 KiB history, so it pauses at any output size. `ContentReader::new` reuses raw/decrypted stream bytes for budget accounting and piece creation, and `decoded_raw_within` applies the existing Whole-stream filters and limits to those bytes without decrypting twice; (16) revision 5–6 (AES-256) passwords are tried in their SASLprep form (RFC 4013, then UTF-8, at most 127 bytes, as ISO 32000-2 §7.6.4.3.3 Algorithm 2.A asks and pdfcraft-crypt writes them) and then, if that differs, as given (`aes256_passwords` in `crypto/mod.rs`, which resolves the upstream `TODO: Do the above.`; `stringprep`, already in the tree through pdfcraft-crypt, is an optional dependency enabled by `std`, and without `std` passwords are used as given): the renderer used the password as typed, so one that SASLprep changes (a decomposed `café`, a no-break space, the fullwidth `ａｂ` a Japanese IME types in full-width mode) was wrong, although pdfcraft-cos accepted it: protecting a document with such an open password was rolled back ("the password is incorrect"), and a file with such a password did not open with it.; (17) CCITT `/Rows 0` means the height is not predetermined (ISO 32000-2 Table 11), not zero rows: scanned Group 4 pages written that way rendered blank; (18) The streaming Flate decoder (`lzw_flate::flate::fallback`) copies a back-reference in runs of up to its distance (`extend_from_within`), not byte by byte. The copy reads only bytes already in the output, and a copy whose source does not exist still writes nothing, so the bytes are those a byte-by-byte copy writes. Inflating the A1 sample's content stream (278.6 MB) takes about 14% less time (0.66 s to 0.56 s); the output is identical; (19) CCITT `/Rows` below the image's `/Height` decodes to the height: a plotter driver wrote `/Rows` equal to `/Columns`, so only the first rows were decoded and the rest drawn black (#3308) (20) `stream` followed by spaces or tabs and then the line break (`stream \r\n`, as Microsoft Reporting Services writes it) starts the data after the line break; the content stream was taken to start at the space and the page rendered blank.; (21) a JP2 colour specification naming an enumerated colour space the decoder refuses (CIEJab in a veraPDF fixture, YCbCr, PhotoYCC, YPbPr, bilevel…) no longer fails the image: it is decoded again from a copy whose `colr` names the space the decoder assumes without one, greyscale, sRGB or CMYK by component count, as MuPDF and Poppler draw it (`without_refused_enumerated_colour`; the page drew blank, #3306); (22) an object stream's index of `objnum offset` pairs is parsed once per stream and kept beside its decoded data (`ObjectStreamIndexes`, `XRef::with_object_stream_indexes`), and an offset past `usize` is malformed instead of wrapping: `XRef::get_with` parsed the whole index again for every object it fetched from the stream, so reading all of a stream of N objects took N² steps (a Firefox page whose 65,542 objects share one stream took 37 s to draw and 40 s to extract its text); (23) the inline-image end-of-data look-ahead no longer rejects a candidate `EI` because a byte above 0x7F lies between it and the next one: when the candidate is the end, those bytes are the content stream that follows the image, and its strings carry such bytes (WinAnsi `(Grüße) Tj`, the two-byte codes of any Type 0 font), so the operator check that follows decides alone, bounded as in (4) (a scanned page whose OCR text held an accented word and a later capital "EI" — EINGANG, BEI, KLEIN — drew and extracted nothing after the image: 40 of 275 pages across five real documents); (24) `xref::prev_offset`: a trailer `/Prev` written as a reference `N G R` is read as the offset that object holds when the section's own entries place it (iText's `indirectPrev.pdf`; each section's entries are read before its `Prev` is followed), and otherwise as offset N (tiff2pdf; JHOVE's PDF-HUL-157 files, #2366), as `pdfcraft-cos` and other readers take it, so the earlier sections' pages are seen instead of one; (25) direct arrays and dictionaries share a reader-local nesting cap of 128, counted while skipping and while parsing a dictionary (its lenient repair reads, rather than skips, an object found where a key belongs, so `<< << …` recursed through that read); deeper objects return `None` instead of overflowing the stack (including unused trailer values, trailer entries, content operands and marked-content properties); a chain of 65 direct stitching functions now stops at this cap, before the 64-function guard; (26) an xref stream subsection must be backed by its decoded data: all-zero `/W` widths (an entry reads no bytes) and a declared count longer than the data are refused before any entry is inserted, where a tiny stream with a large `/Size` or `/Index` count inserted that many entries (from storytold/craft-libs#239); a subsection running past object number `i32::MAX` is refused instead of overflowing or turning negative; (27) an xref stream's type field is read big-endian at its full `/W[0]` width, where only its first byte was read and wider fields shifted every later entry, and `/W[0]` over 8 bytes is refused like `/W[1]` (from storytold/craft-libs#241); (28) a content instruction keeps up to 256 operands (`MAX_OPERANDS`) instead of 10, and past that drops the oldest instead of ending the content stream: the eleventh operand of a twelve-ink DeviceN `scn` (ISO 32000-2 §8.6.6.5 allows any number of components) blanked the rest of the page (#4260); (29) a `/JBIG2Globals` stream that is itself JBIG2 isn't decoded: globals are raw segments, and a chain of 16,000 such streams overflowed the stack, which aborts the process; (31) `set_page_jbig2_work`: the JBIG2 images decoded on a thread spend one allowance, set per page, 2^30 steps (`MAX_PAGE_JBIG2_WORK`), so a page of many images made to reach their own limit stops at one, and what they cost to read counts, whether they decode or not; unbounded until a page sets it; (32) a group of array items read together (a tuple: an image's `/Decode`, `/W`, `/W2`, and a function's `/Domain`, `/Range`, `/Encode` and `/Decode`) can start with a reference: it was taken for the whole group and lost, which ended the array, so an image with `/Decode [7 0 R 0]` didn't draw and CID widths fell back to `/DW`; (33) an object that holds only a reference stands for what that names, stored on its own or in an object stream, up to 32 references: read as a number it was the reference's own object number, so a `/Decode` or `/W` entry naming such an object took a wrong value, and as anything else it was lost. Past 32, as any cycle goes, it reads as a missing object. An object stream isn't reached that way, as its checks are of the object named; reading its index counts against the 64 nested loads and reads its dictionary alone, as a chain of them, each with a member of the next for its `/N`, `/First` or `/Length`, ran out of stack; and a comment inside a reference counts as white space; (34) a table rebuilt by scanning the file (`fallback_xref_map`) finds an object by its number when its header names another generation, including one no generation may be (`3 18446744073709551616 obj` in `poppler-85140-0.pdf`, #5580): the header is read leniently only in that scan (`read_object_header`), the object is also filed under generation 0 unless a real `N 0 obj` holds that key (a later header for the number moves the stand-in), `XRef::get_with` matches by object number only in a repaired table, and `XRef::objects` yields each location once | (1) `cos/tests/writer.rs::page_tree_loops_in_object_streams_do_not_crash_readers`; (2) `raster::tests::absurd_mask_and_fax_sizes_are_refused_before_decoding`; (3) `raster::tests::absurd_predictor_columns_and_line_widths_render`; (4) `raster::tests::inline_images_with_ei_in_their_data_parse_in_linear_time`, `content::tests::inline_image_ei_may_be_followed_by_a_delimiter`; (5) `raster::tests::indirect_appearance_state_dict_is_not_another_stream`; (6) `raster::tests::jpx_stream_bit_depth_boundaries_do_not_panic`, `raster::tests::jpx_32_bit_image_does_not_prevent_other_page_content_from_rendering`, `raster::tests::jpx_gray_rgb_and_alpha_keep_eight_bit_samples`, `raster::tests::jpx_streams_reject_out_of_range_bit_depths`; (7) `raster::tests::decompression_bombs_stop_at_the_stream_limit`, `raster::tests::page_contents_and_nested_forms_share_one_budget`; (8) `inspect::tests::jspdf_high_precision_mediabox_does_not_overflow_dimensions`, `object::number::tests::high_precision_float_does_not_overflow_mantissa` (hayro-syntax); (9) `object::dict::lazy_index_tests`, `raster::tests::large_resource_indexes_preserve_pixels_and_text`; (10) `data::tests`, `xref::tests::concurrent_cold_object_stream_reads_all_succeed`, `raster::tests::shared_parser_is_lazy_and_initializes_once_under_contention`; (11) `data::tests::panicking_initializer_wakes_waiters_and_prevents_retries`, `xref::tests::concurrent_cold_object_stream_reads_all_succeed`; (12) `util::tests::segment_list_rejects_out_of_range_indices`; (13) `raster::tests::user_unit_scales_the_page_and_everything_on_it`; (14) `crypto::pdfcraft_password_tests`, `pdfcraft-engine` `tests::legacy_passwords_with_turkish_characters_open`; (15) `content::tests::windows_keep_every_instruction_at_every_split`, `content::tests::windows_keep_every_instruction_in_pieces_of_any_size`, `content::tests::windows_match_the_slice_on_random_content`, `content::tests::windows_drop_blanks_and_keep_a_comment_that_may_go_on`, `page::content_reader_tests::content_past_the_cap_is_read_to_the_end`, `page::content_reader_tests::a_stream_that_expands_past_the_budget_is_cut_there`, `page::content_reader_tests::a_small_expansion_under_the_cap_is_not_cut`, `raster::tests::page_content_past_the_old_cap_renders`, `raster::tests::content_reader_returns_the_same_bytes_in_any_pieces`, `raster::tests::content_bombs_stop_at_the_budget_and_say_so`, `page::content_reader_tests::plain_and_whole_pieces_keep_their_bytes`, `page::content_reader_tests::repeated_content_reference_counts_compressed_bytes_once`, `page::content_reader_tests::whole_cap_still_counts_later_unique_streams`; (16) `crypto::pdfcraft_password_tests::aes256_passwords_are_tried_in_their_saslprep_form_then_as_given`, `crypto::pdfcraft_password_tests::aes256_passwords_open_as_typed_and_in_their_saslprep_form`, `pdfcraft-engine` `tests::aes256_open_passwords_that_saslprep_changes_protect_and_reopen_as_typed`, `tests::aes256_files_whose_passwords_saslprep_changes_open_as_typed`; (17) `raster::tests::group4_fax_with_rows_zero_decodes_to_the_image_height`; (18) `filter::lzw_flate::tests::back_reference_longer_than_its_distance_repeats_its_source`, `back_reference_cut_by_the_limit_resumes_in_the_next_call` and `back_reference_reads_bytes_written_by_an_earlier_call`; each fails when the copy returns the wrong remaining count; (19) `raster::tests::group4_fax_with_rows_below_its_height_decodes_to_the_image_height`; (20) `raster::tests::space_before_the_line_break_after_stream_still_renders`; (21) `raster::tests::a_jp2_image_in_a_refused_enumerated_colour_space_is_drawn`, `filter::jpx::tests::refused_enumerated_colour_spaces_become_the_component_counts_default`, `filter::jpx::tests::what_the_decoder_handles_or_cant_be_read_is_left_alone` (`cargo test --manifest-path vendor/hayro-syntax/Cargo.toml`); (22) `raster::tests::objects_in_one_large_object_stream_are_read_in_linear_time`; (23) `text::tests::text_after_an_inline_image_survives_a_non_ascii_byte_before_a_later_ei`, `content::tests::an_inline_image_ends_at_its_own_ei_when_later_text_holds_a_non_ascii_byte` (hayro-syntax); (24) `xref::tests::prev_as_reference_reaches_the_earlier_section`, `xref::tests::prev_as_reference_to_an_object_holding_the_offset_reads_that_offset`, `pdfcraft-cos` `document::tests::a_prev_written_as_a_reference_still_reaches_the_earlier_sections`, `document::tests::a_prev_reference_to_an_object_holding_the_offset_reads_that_offset`; (25) `raster::tests::deeply_nested_direct_objects_do_not_overflow_the_stack`, `raster::tests::dictionaries_nested_without_keys_in_content_do_not_overflow_the_stack`, `raster::tests::dictionaries_nested_without_keys_in_the_trailer_do_not_overflow_the_stack`, `raster::tests::dictionaries_nested_without_keys_as_marked_content_properties_do_not_overflow_the_stack`, `raster::tests::dictionaries_nested_without_keys_are_read_up_to_the_nesting_cap`, `raster::tests::nesting_at_the_cap_fits_a_512_kib_stack` (native only), `raster::tests::function_limits_depth_guard_refuses_a_chain_of_references`; (26) `xref::tests::xref_stream_entries_need_backing_bytes`; (27) `xref::tests::xref_stream_type_field_uses_the_full_w0_width`, `xref::tests::xref_stream_loader_rejects_oversized_w0` (`cargo test --manifest-path vendor/hayro-syntax/Cargo.toml`); (28) `content::ops::tests::scn_with_twelve_components_keeps_them_all`, `content::ops::tests::operands_past_the_cap_keep_the_latest_in_order` (hayro-syntax), `raster::tests::devicen_inks::a_twelve_ink_devicen_colour_paints`, `raster::tests::devicen_inks::a_four_ink_devicen_colour_paints`, `raster::tests::devicen_inks::an_operator_with_hundreds_of_operands_does_not_end_the_content`; (29) `raster::tests::jbig2_globals_that_are_jbig2_themselves_are_not_decoded`, `raster::tests::jbig2_globals_are_read`, `raster::tests::the_jbig2_images_of_a_page_share_one_allowance`, `raster::tests::an_image_that_fails_spends_the_pages_allowance_too`; (32) `raster::tests::a_decode_array_with_references_in_it_is_read`, `raster::tests::a_functions_arrays_can_start_with_a_reference` (pdfcraft-render), `font::cid::width_tests::a_width_group_can_start_with_a_reference` (hayro-interpret); (33) `raster::tests::a_reference_to_a_reference_is_what_it_names`, `raster::tests::a_decode_array_with_references_in_it_is_read` (pdfcraft-render), `font::cid::width_tests::a_width_group_can_start_with_a_reference` (hayro-interpret), `xref::tests::an_object_stream_isnt_reached_through_a_reference`, `xref::tests::a_chain_of_object_streams_ends_at_the_limit_of_nested_loads`, `xref::tests::a_chain_of_object_streams_takes_no_more_work_for_two_ways_on`, `xref::tests::a_reference_held_in_either_storage_is_followed`, `xref::tests::a_page_a_reference_is_held_to_is_found_once` (hayro-syntax); (34) `raster::tests::a_rebuilt_table_finds_an_object_whose_header_names_another_generation`, `raster::tests::a_rebuilt_tables_generation_0_stand_in_is_not_a_second_object` (pdfcraft-render), `xref::tests::an_object_stream_and_its_first_member_are_both_enumerated` (hayro-syntax) | diff --git a/vendor/hayro/src/renderer.rs b/vendor/hayro/src/renderer.rs index 0b9cfe779e..120a751281 100644 --- a/vendor/hayro/src/renderer.rs +++ b/vendor/hayro/src/renderer.rs @@ -193,6 +193,27 @@ fn image_buffer(len: usize) -> Option> { Some(out) } +/// PdfCraft patch: `mask` sampled onto the grid of a `width` × `height` image, each image sample +/// taking the mask sample under its centre. `None` for an empty side or a buffer over the limits. +fn mask_on_image_grid(mask: &LumaData, width: u32, height: u32) -> Option> { + if width == 0 || height == 0 || mask.width == 0 || mask.height == 0 { + return None; + } + let mut out = image_buffer(source_byte_len(width, height, 1, MAX_IMAGE_PIXELS)?)?; + let mut columns = Vec::new(); + columns.try_reserve_exact(width as usize).ok()?; + columns.extend((0..width).map(|x| nearest_source_index(mask.width, width, x))); + let mask_width = mask.width as usize; + for y in 0..height { + let start = nearest_source_index(mask.height, height, y).checked_mul(mask_width)?; + let row = mask.data.get(start..start.checked_add(mask_width)?)?; + for &x in &columns { + out.push(*row.get(x)?); + } + } + Some(out) +} + /// PdfCraft patch (#624): the source index that PicScale's nearest-neighbour plan copies for /// destination `index` of `dst` samples taken from `src`. It is the same fixed-point centre /// sampling as `ResampleNearestPlan`, so these are the samples a nearest resize would copy. @@ -236,6 +257,177 @@ fn nearest_premultiplied_rgba( Some(out) } +/// PdfCraft patch: the shorter side, in samples, from which an image that did not ask for +/// `/Interpolate` is still smoothed when enlarged. A page scanned at 72 dpi is 612 samples wide; +/// icons, QR codes and pixel art are usually far smaller and keep their hard pixels. +const SMOOTHED_IMAGE_SIDE: u32 = 512; + +/// PdfCraft patch: fixed-point unit of [`area_shrink`]'s weights. +const AREA_ONE: u64 = 1 << 16; + +/// PdfCraft patch: the source samples target sample `t` covers when `src` samples shrink to `dst` +/// (`dst <= src`), with their weights in [`AREA_ONE`] units. Target `t` spans `[t·src, (t+1)·src)` +/// and source `s` spans `[s·dst, (s+1)·dst)`, both in units of 1/(src·dst) of the axis, so every +/// overlap is exact. The weights are differences of one rounded running sum, so they add up to +/// exactly [`AREA_ONE`] and a flat area keeps its exact value. +fn area_taps(src: u32, dst: u32, t: u32) -> impl Iterator { + let (src, dst, t) = (u64::from(src), u64::from(dst), u64::from(t)); + // `t < dst <= src <= 2^28` (MAX_IMAGE_PIXELS), so every product below fits in a u64. + let (lo, hi) = (t * src, (t + 1) * src); + let fixed = move |x: u64| (x * AREA_ONE + src / 2) / src; + (lo / dst..hi.div_ceil(dst)).map(move |s| { + let from = (s * dst).max(lo) - lo; + let to = ((s + 1) * dst).min(hi) - lo; + (s as usize, fixed(to) - fixed(from)) + }) +} + +/// PdfCraft patch: [`area_taps`] of every target sample of one axis, worked out once: for each +/// target, the first source sample it covers and the weights of the samples from there on (the +/// samples a target covers are consecutive). There are at most `src + dst` taps in all. +struct AreaAxis { + first: Vec, + /// Where each target's weights start in `weights`, and where the last one's end. + offsets: Vec, + weights: Vec, +} + +impl AreaAxis { + fn new(src: u32, dst: u32) -> Option { + let n = dst as usize; + let (mut first, mut offsets, mut weights) = (Vec::new(), Vec::new(), Vec::new()); + first.try_reserve_exact(n).ok()?; + offsets.try_reserve_exact(n.checked_add(1)?).ok()?; + weights.try_reserve_exact((src as usize).checked_add(n.checked_mul(2)?)?).ok()?; + offsets.push(0); + for t in 0..dst { + let mut start = None; + for (s, w) in area_taps(src, dst, t) { + start.get_or_insert(s); + // A weight is at most AREA_ONE, so it fits in a u32. + weights.push(w as u32); + } + first.push(start?); + offsets.push(weights.len()); + } + Some(Self { first, offsets, weights }) + } + + /// Target `t`'s first source sample and its weights. + fn get(&self, t: usize) -> Option<(usize, &[u32])> { + let (lo, hi) = (*self.offsets.get(t)?, *self.offsets.get(t.checked_add(1)?)?); + Some((*self.first.get(t)?, self.weights.get(lo..hi)?)) + } +} + +/// PdfCraft patch: shrinks `N`-channel samples from `source` to `target` (no side larger than its +/// source side) by area averaging: each target sample is the mean of the source it covers, +/// weighted by how much of it it covers. Nothing is invented between samples, which is what +/// `/Interpolate` governs (ISO 32000-2 §8.9.5.3, for images drawn above their resolution), and +/// nothing is dropped: nearest-neighbour keeps one sample in N per axis, so a scan shown at screen +/// size lost the strokes of its text. Premultiplied input stays premultiplied. Work memory is one +/// source row of `u32` accumulators, the horizontal taps (at most `sw + 2 × tw`) and a vertically +/// shrunk copy no larger than the source; a side over `MAX_RESAMPLING_SOURCE_SIDE` is refused, so +/// none of it exceeds a few times one source row. `None` if nothing shrinks, the sizes don't fit +/// or a buffer cannot be allocated. +fn area_shrink(data: &[u8], source: (u32, u32), target: (u32, u32)) -> Option> { + let ((sw, sh), (tw, th)) = (source, target); + // A side over MAX_RESAMPLING_SOURCE_SIDE would make the accumulator row and the tap table + // several times larger than the samples (a 2^27 × 2 image of zeros is a tiny file). + if tw == 0 || th == 0 || tw > sw || th > sh || sw > MAX_RESAMPLING_SOURCE_SIDE || sh > MAX_RESAMPLING_SOURCE_SIDE { + return None; + } + let (sw_us, tw_us) = (sw as usize, tw as usize); + if data.len() != source_byte_len(sw, sh, N, MAX_IMAGE_PIXELS)? { + return None; + } + // Weights sum to AREA_ONE (2^16) and samples are at most 255, so every sum fits in a u32. + let round = |acc: u32| ((acc + (AREA_ONE as u32) / 2) >> 16).min(255) as u8; + // Vertical pass first: groups of whole source rows averaged into each target row. Rows are + // contiguous, so this pass is cheap, and it leaves the horizontal pass, which gathers samples + // per target, a third as many rows when a scan shrinks threefold. + let tall = if th == sh { + None + } else { + let row_len = sw_us * N; + let mut rows = image_buffer(source_byte_len(sw, th, N, MAX_IMAGE_PIXELS)?)?; + let mut acc: Vec = Vec::new(); + acc.try_reserve_exact(row_len).ok()?; + acc.resize(row_len, 0); + for t in 0..th { + acc.fill(0); + for (s, w) in area_taps(sh, th, t) { + let row = data.get(s * row_len..(s + 1) * row_len)?; + // A weight is at most AREA_ONE, so it fits in a u32. + let w = w as u32; + for (a, v) in acc.iter_mut().zip(row) { + *a += w * u32::from(*v); + } + } + rows.extend(acc.iter().map(|a| round(*a))); + } + Some(rows) + }; + if tw == sw { + return tall; + } + let rows = tall.as_deref().unwrap_or(data); + // Horizontal pass: each of those rows shrunk to `tw` samples. The taps depend only on the + // widths, so they are worked out once, not again for each row. + let taps = AreaAxis::new(sw, tw)?; + let mut out = image_buffer(image_byte_len(tw, th, N, MAX_IMAGE_PIXELS)?)?; + for row in rows.chunks_exact(sw_us * N) { + for t in 0..tw_us { + let (first, weights) = taps.get(t)?; + let samples = row.get(first * N..(first + weights.len()) * N)?; + let mut acc = [0_u32; N]; + for (w, px) in weights.iter().zip(samples.chunks_exact(N)) { + for (a, v) in acc.iter_mut().zip(px) { + *a += w * u32::from(*v); + } + } + out.extend(acc.map(round)); + } + } + Some(out) +} + +/// PdfCraft patch: [`area_shrink`] for an image's pixel format. If it cannot run, the image keeps +/// its source resolution, as when [`Renderer::resize_image_data`] fails. +fn area_resize( + data: Vec, + source: (u32, u32), + target: (u32, u32), + format: ImagePixelFormat, +) -> (Vec, u32, u32) { + let shrunk = match format { + ImagePixelFormat::Luma => area_shrink::<1>(&data, source, target), + ImagePixelFormat::Rgb => area_shrink::<3>(&data, source, target), + ImagePixelFormat::Rgba => area_shrink::<4>(&data, source, target), + }; + match shrunk { + Some(out) => (out, target.0, target.1), + None => { + log::warn!( + "Image area shrink failed or exceeded its buffer limit; falling back to source data" + ); + (data, source.0, source.1) + } + } +} + +/// PdfCraft patch: premultiplies straight RGBA samples in place. +fn premultiply_rgba(rgba: &mut [u8]) { + let (chunks, _) = rgba.as_chunks_mut::<4>(); + for chunk in chunks { + let [r, g, b, a] = *chunk; + *chunk = AlphaColor::from_rgba8(r, g, b, a) + .premultiply() + .to_rgba8() + .to_u8_array(); + } +} + /// PdfCraft patch: most pieces one dashed stroke may be cut into. Stroke expansion keeps every /// dash, so a pattern that is tiny next to its path asks for billions: a fuzzed `/D [[11] 0]` on /// a line from x = 92234775807 allocated over 5 GB, and `[0.000001] 0 d` on a 20 pt line over @@ -423,9 +615,12 @@ fn snap_minified_image( planned: (u32, u32), ) -> Option<(u32, u32, Affine)> { let [a, b, c, d, e, f] = transform.as_coeffs(); + // Halves round up, not away from zero: a tile's origin moves edges by whole pixels, and only + // then do they snap where the whole page snaps them (-2.5 must go to -2 as 2.5 goes to 3). + let snap = |v: f64| (v + 0.5).floor(); // The device span of a source edge from `origin` across `side` samples of `scale`, rounded. let edge = |origin: f64, scale: f64, side: u32| { - let (p, q) = (origin.round(), (origin + scale * f64::from(side)).round()); + let (p, q) = (snap(origin), snap(origin + scale * f64::from(side))); let (lo, hi) = (p.min(q), p.max(q)); (hi - lo >= 1.0 && lo.abs() < 1e6 && hi.abs() < 1e6).then_some((lo, hi)) }; @@ -935,17 +1130,51 @@ impl Renderer { return; } let interpolate = image_data.interpolate(); - if let Some(a) = &alpha_data { + let requested_resize = x_scale < 1.0 || y_scale < 1.0; + // PdfCraft patch: an image that did not ask for interpolation and is drawn below its + // resolution is shrunk by area averaging (`area_shrink`), not nearest-neighbour, which + // broke up the text of scans shown at screen size. Type 3 glyph images keep the nearest + // shrink of #624: they are drawn with a frame of their own and never snapped. + let area = !interpolate + && requested_resize + && !self.in_type3_glyph + // Wider sources keep #624's bounded nearest-neighbour shrink (`area_shrink` refuses + // them): skipping them, as a refused shrink at source size would, loses the image. + && source_width.max(source_height) <= MAX_RESAMPLING_SOURCE_SIDE; + let mut alpha_data = alpha_data; + if let Some(a) = alpha_data.take() { if source_byte_len(a.width, a.height, 1, MAX_IMAGE_PIXELS) != Some(a.data.len()) { log::warn!("Image has invalid or over-limit alpha dimensions/data and was skipped"); return; } if a.width != source_width || a.height != source_height || a.interpolate != interpolate { - if let Some(alpha) = alpha_data { - self.draw_image_with_alpha_mask(image_data, alpha); + // PdfCraft patch: a soft mask no larger than its area-shrunk image is sampled onto + // the image's grid, so colour and mask are averaged together, premultiplied. Drawn + // as a mask layer over the image instead, the averaged colour (transparent samples + // included) was multiplied by the averaged mask, and the colour of transparent + // samples showed at the mask's edges. A larger mask carries detail the image does + // not (#1649) and keeps the layer. + let on_grid = (area && a.width <= source_width && a.height <= source_height) + .then(|| mask_on_image_grid(&a, source_width, source_height)) + .flatten(); + match on_grid { + Some(data) => { + alpha_data = Some(LumaData { + data, + width: source_width, + height: source_height, + interpolate, + ..a + }); + } + None => { + self.draw_image_with_alpha_mask(image_data, a); + return; + } } - return; + } else { + alpha_data = Some(a); } } let mut quality = if interpolate { @@ -955,7 +1184,6 @@ impl Renderer { }; let has_alpha = alpha_data.is_some(); let mut may_have_transparency = has_alpha || self.force_images_may_have_transparency(); - let requested_resize = x_scale < 1.0 || y_scale < 1.0; let (new_width, new_height) = match image_resampling_size( source_width, source_height, @@ -985,12 +1213,28 @@ impl Renderer { if requested_resize && self.in_type3_glyph && interpolate { quality = ImageQuality::High; } - // PdfCraft patch: an interpolated image that is shrunk, with its axes along the device's - // (upright, mirrored or a quarter turn), lands on whole device pixels. It is shrunk once to - // exactly its snapped device rectangle and drawn 1:1 from the snapped corner. Otherwise the - // shrunk raster (rounded up to a whole size) is sampled a second time at a fractional scale - // and offset, which blurs scans and photos. - let snapped = if interpolate + // PdfCraft patch: an image the size of a page scan is smoothed when drawn above its + // resolution, even without `/Interpolate`: bilinearly, as images that ask for it are. + // Scans rarely ask, and drawn nearest-neighbour each sample of a zoomed scan became a + // block, so the edges of its text were staircases. Bicubic looked the same on a zoomed + // 300 dpi scan and drew its tiles five times slower. Smaller images (icons, QR codes, + // pixel art) keep their hard pixels. A scan shrunk on one side and stretched on the other + // is smoothed too: the area shrink leaves the shrunk side at 1:1, where bilinear adds + // nothing, and the stretched side would otherwise be drawn nearest-neighbour. + if !interpolate + && !self.in_type3_glyph + && (x_scale > 1.0 || y_scale > 1.0) + && source_width.min(source_height) >= SMOOTHED_IMAGE_SIDE + { + quality = ImageQuality::Medium; + } + // PdfCraft patch: an image that is shrunk (filtered or area-averaged), with its axes along + // the device's (upright, mirrored or a quarter turn), lands on whole device pixels. It is + // shrunk once to exactly its snapped device rectangle and drawn 1:1 from the snapped + // corner. Otherwise the shrunk raster (rounded up to a whole size) is sampled a second time + // at a fractional scale and offset, which blurs scans and photos (or, by the nearest + // sampler of a non-interpolated image, drops samples again). + let snapped = if (interpolate || area) && requested_resize && !self.in_type3_glyph && (new_width, new_height) != (source_width, source_height) @@ -1000,13 +1244,22 @@ impl Renderer { None }; let (new_width, new_height) = snapped.map_or((new_width, new_height), |s| (s.0, s.1)); + // PdfCraft patch: an area shrink never enlarges. A side drawn at or above its resolution + // (an image shrunk across and stretched down, as a 204 × 98 dpi fax is, or a snapped side + // rounded past its source) keeps its samples and is enlarged by the sampler; the snap + // then no longer matches, so the full transform places the image. + let (new_width, new_height) = if area { + (new_width.min(source_width), new_height.min(source_height)) + } else { + (new_width, new_height) + }; let needs_resize = (new_width, new_height) != (source_width, source_height); - // PdfCraft patch (#624): a minified image with a soft mask that is not interpolated is - // nearest-sampled straight into premultiplied target RGBA. That skips the full-resolution - // RGBA copy below and the premultiply pass after it. + // PdfCraft patch (#624): a minified Type 3 glyph image with a soft mask that is not + // interpolated is nearest-sampled straight into premultiplied target RGBA. That skips the + // full-resolution RGBA copy below and the premultiply pass after it. let direct = match &alpha_data { - Some(alpha) if needs_resize && !interpolate => nearest_premultiplied_rgba( + Some(alpha) if needs_resize && !interpolate && !area => nearest_premultiplied_rgba( &image_data, &alpha.data, (source_width, source_height), @@ -1014,7 +1267,7 @@ impl Renderer { ), _ => None, }; - let premultiplied = direct.is_some(); + let mut premultiplied = direct.is_some(); // Preserve the single-channel/RGB fast paths. With alpha, expand directly to RGBA; // every source and alpha length was validated before zipping or allocating. let (data, format) = match (direct, image_data, alpha_data) { @@ -1057,6 +1310,15 @@ impl Renderer { }; let (data, mut img_width, mut img_height) = if premultiplied { (data, new_width, new_height) + } else if needs_resize && area { + let mut data = data; + if matches!(format, ImagePixelFormat::Rgba) { + // Averaging straight colours would bleed the colour of transparent samples into + // their neighbours; premultiplied colours count each by its coverage. + premultiply_rgba(&mut data); + premultiplied = true; + } + area_resize(data, (source_width, source_height), (new_width, new_height), format) } else if needs_resize { self.resize_image_data( data, @@ -1103,14 +1365,7 @@ impl Renderer { rgba }; if has_alpha && !premultiplied { - let (chunks, _) = rgba_data.as_chunks_mut::<4>(); - for chunk in chunks { - let [r, g, b, a] = *chunk; - *chunk = AlphaColor::from_rgba8(r, g, b, a) - .premultiply() - .to_rgba8() - .to_u8_array(); - } + premultiply_rgba(&mut rgba_data); } // Type 3 glyphs get a transparent two-pixel frame for interpolation. Check the padded // RGBA allocation and u16 dimensions too, before narrowing or growing either side. @@ -2635,3 +2890,109 @@ mod snap_tests { assert!(snap_minified_image(&turned_noisy, source, (20_000, 200)).is_some()); } } + +/// PdfCraft patch: the area-averaging shrink of non-interpolated images. +#[cfg(test)] +mod area_shrink_tests { + use super::{ + AREA_ONE, AreaAxis, LumaData, MAX_IMAGE_PIXELS, area_shrink, area_taps, mask_on_image_grid, + }; + + fn mask(data: &[u8], width: u32, height: u32) -> LumaData { + LumaData { data: data.to_vec(), width, height, interpolate: false, scale_factors: (1.0, 1.0) } + } + + /// Each image sample takes the mask sample under its centre, at whole and fractional ratios. + #[test] + fn a_mask_is_sampled_onto_the_image_grid() { + assert_eq!(mask_on_image_grid(&mask(&[10, 20], 2, 1), 4, 2), Some(vec![10, 10, 20, 20, 10, 10, 20, 20])); + // Centres of 7 samples over 3: 0.5/7·3 = 0.21 → 0, 1.5/7·3 = 0.64 → 0, 2.5/7·3 = 1.07 → 1, … + assert_eq!(mask_on_image_grid(&mask(&[1, 2, 3], 3, 1), 7, 1), Some(vec![1, 1, 2, 2, 2, 3, 3])); + assert_eq!(mask_on_image_grid(&mask(&[1, 2, 3, 4], 1, 4), 1, 2), Some(vec![2, 4])); + } + + #[test] + fn malformed_masks_are_not_sampled() { + assert_eq!(mask_on_image_grid(&mask(&[], 0, 1), 4, 1), None, "an empty mask"); + assert_eq!(mask_on_image_grid(&mask(&[1, 2], 2, 1), 0, 1), None, "an empty image"); + assert_eq!(mask_on_image_grid(&mask(&[1], 2, 1), 4, 1), None, "mask data shorter than its size"); + } + + /// Every target sample's weights add up to exactly one, and together the targets take every + /// source sample exactly once (each `dst / src` of it, in fixed point within rounding). + #[test] + fn taps_are_exact_partitions_of_unity() { + for src in 1..=40_u32 { + for dst in 1..=src { + let mut taken = vec![0_u64; src as usize]; + for t in 0..dst { + let taps: Vec<_> = area_taps(src, dst, t).collect(); + assert_eq!(taps.iter().map(|(_, w)| w).sum::(), AREA_ONE, "{src} → {dst}, target {t}"); + assert!(taps.iter().all(|(_, w)| *w > 0), "{src} → {dst}, target {t}: an empty tap"); + for (s, w) in taps { + taken[s] += w; + } + } + // Source s is dst/src of one target in all; scaled by src/dst, that is one. + for (s, w) in taken.iter().enumerate() { + let share = (*w * u64::from(src)) as f64 / (AREA_ONE * u64::from(dst)) as f64; + assert!((share - 1.0).abs() < 1e-3, "{src} → {dst}: source {s} weighs {share}"); + } + } + } + } + + /// The table the horizontal pass reads holds exactly the taps of every target, in order. + #[test] + fn the_tap_table_holds_every_targets_taps() { + for src in 1..=40_u32 { + for dst in 1..=src { + let axis = AreaAxis::new(src, dst).expect("table"); + for t in 0..dst { + let (first, weights) = axis.get(t as usize).expect("target"); + let expected: Vec<(usize, u64)> = area_taps(src, dst, t).collect(); + let got: Vec<(usize, u64)> = weights.iter().enumerate().map(|(k, w)| (first + k, u64::from(*w))).collect(); + assert_eq!(got, expected, "{src} → {dst}, target {t}"); + } + assert_eq!(axis.get(dst as usize), None, "{src} → {dst}: no target past the last"); + } + } + } + + /// The largest side an image may have still computes in u64 without overflowing. + #[test] + fn taps_at_the_pixel_limit_do_not_overflow() { + let side = MAX_IMAGE_PIXELS as u32; + for (dst, t) in [(1, 0), (3, 2), (side - 1, side - 2), (side, side - 1)] { + assert_eq!(area_taps(side, dst, t).map(|(_, w)| w).sum::(), AREA_ONE, "{side} → {dst}, target {t}"); + } + } + + #[test] + fn flat_images_keep_their_exact_value() { + for (source, target) in [((7, 5), (3, 2)), ((10, 3), (6, 3)), ((4, 9), (4, 4)), ((13, 11), (1, 1))] { + let data = vec![77_u8; (source.0 * source.1 * 3) as usize]; + let out = area_shrink::<3>(&data, source, target).expect("shrinks"); + assert_eq!(out.len(), (target.0 * target.1 * 3) as usize); + assert!(out.iter().all(|v| *v == 77), "{source:?} → {target:?}: {out:?}"); + } + } + + /// A side over the resampling limit is refused before any work buffer is allocated. + #[test] + fn sides_over_the_resampling_limit_are_refused() { + let wide = (super::MAX_RESAMPLING_SOURCE_SIDE + 1) as usize; + assert_eq!(area_shrink::<1>(&vec![0_u8; wide], (wide as u32, 1), (10, 1)), None, "too wide"); + assert_eq!(area_shrink::<1>(&vec![0_u8; wide], (1, wide as u32), (1, 10)), None, "too tall"); + } + + #[test] + fn malformed_shrinks_are_refused() { + let data = vec![0_u8; 4 * 4]; + assert_eq!(area_shrink::<1>(&data, (4, 4), (4, 4)), None, "nothing shrinks"); + assert_eq!(area_shrink::<1>(&data, (4, 4), (5, 2)), None, "a side grows"); + assert_eq!(area_shrink::<1>(&data, (4, 4), (0, 2)), None, "an empty side"); + assert_eq!(area_shrink::<1>(&data[1..], (4, 4), (2, 2)), None, "data shorter than the source"); + assert_eq!(area_shrink::<3>(&data, (4, 4), (2, 2)), None, "data of another channel count"); + } +} From 5ed2f4ba4f34d5eff689727d468186bc18b53350 Mon Sep 17 00:00:00 2001 From: kronbii <145912020+Kronbii@users.noreply.github.com> Date: Fri, 30 Oct 2026 01:50:48 +0200 Subject: [PATCH 33/53] M2: keep an interpolated soft mask smooth on a shrunk image's grid A soft mask no larger than its area-shrunk image is sampled onto the image's grid, and it was always sampled nearest-neighbour: a mask that asks for /Interpolate became a hard step where main drew a ramp (jpeskov's 64 x 64 image at 32 x 32 under a 2 x 2 mask). An interpolated mask is now blended bilinearly onto the grid, centres aligned; one without /Interpolate is still sampled nearest, which keeps the colour-fringe fix. Co-Authored-By: Claude Opus 5.5 --- crates/render/src/raster/tests.rs | 29 ++++++++++++ vendor/README.md | 2 +- vendor/hayro/src/renderer.rs | 74 ++++++++++++++++++++++++++----- 3 files changed, 93 insertions(+), 12 deletions(-) diff --git a/crates/render/src/raster/tests.rs b/crates/render/src/raster/tests.rs index 7312f3ddaf..f6c0530164 100644 --- a/crates/render/src/raster/tests.rs +++ b/crates/render/src/raster/tests.rs @@ -6734,6 +6734,35 @@ fn a_smaller_soft_mask_does_not_bleed_transparent_colour_when_minified() { assert!(r >= 254 && g.abs_diff(128) <= 1 && b == g, "half-covered red over white, got ({r}, {g}, {b})"); } +/// A smaller soft mask that asks for `/Interpolate` stays smooth when it is sampled onto the grid +/// of an area-shrunk image. Here a 64 × 64 black image is drawn at 32 × 32 under a 2 × 2 mask, +/// opaque on the left and clear on the right: the mask is enlarged while the image shrinks, and +/// sampled nearest-neighbour onto the image's grid the edge became a hard black and white step. +#[test] +fn an_interpolated_smaller_soft_mask_stays_smooth_when_its_image_is_minified() { + let content = "q 32 0 0 32 0 0 cm /Im0 Do Q"; + let (image, mask) = ("00".repeat(64 * 64), "FF00FF00"); + let pdf = format!( + "%PDF-1.4\n1 0 obj << /Type /Catalog /Pages 2 0 R >> endobj\n\ + 2 0 obj << /Type /Pages /Kids [3 0 R] /Count 1 >> endobj\n\ + 3 0 obj << /Type /Page /Parent 2 0 R /MediaBox [0 0 32 32] /Contents 4 0 R /Resources << /XObject << /Im0 5 0 R >> >> >> endobj\n\ + 4 0 obj << /Length {} >> stream\n{content}\nendstream endobj\n\ + 5 0 obj << /Type /XObject /Subtype /Image /Width 64 /Height 64 /ColorSpace /DeviceGray /BitsPerComponent 8 /SMask 6 0 R /Filter /ASCIIHexDecode /Length {} >> stream\n{image}>\nendstream endobj\n\ + 6 0 obj << /Type /XObject /Subtype /Image /Width 2 /Height 2 /ColorSpace /DeviceGray /BitsPerComponent 8 /Interpolate true /Filter /ASCIIHexDecode /Length {} >> stream\n{mask}>\nendstream endobj\n\ + trailer << /Root 1 0 R >>\n%%EOF\n", + content.len(), + image.len() + 1, + mask.len() + 1 + ); + let mut renderer = PageRenderer::new(Arc::new(pdf.into_bytes()), RenderConfig::default()); + let page = renderer.render(RenderRequest { page: 0, kind: RequestKind::Pixels, scale: 1.0, ..Default::default() }); + assert!(page.error.is_none(), "{:?}", page.error); + assert_eq!((page.width, page.height), (32, 32)); + let row: Vec = page.rgba.as_chunks::<4>().0.iter().skip(16 * 32).take(32).map(|[r, ..]| *r).collect(); + let ramp = row.iter().filter(|v| (1..255).contains(*v)).count(); + assert!(ramp >= 8 && row.is_sorted() && row.first() == Some(&0) && row.last() == Some(&255), "a smooth black to white edge, got {row:?}"); +} + #[test] fn non_interpolated_alpha_images_keep_their_colours_and_mask_when_minified() { let palette: [[u8; 3]; 8] = [[0, 0, 0], [255, 255, 255], [255, 0, 0], [0, 255, 0], [0, 0, 255], [255, 255, 0], [255, 0, 255], [0, 255, 255]]; diff --git a/vendor/README.md b/vendor/README.md index 1a58ad6c1f..0d90ad397e 100644 --- a/vendor/README.md +++ b/vendor/README.md @@ -40,7 +40,7 @@ Other resource categories retain their existing parent-chain lookup. | Crate | Version | Licence | Patches | Test | |---|---|---|---|---| | hayro-interpret | 0.7.0 | Apache-2.0 OR MIT | (1) select `/AP /N` appearance states via `/AS` (checkboxes, radios, stateful icons) and honour the NoView flag; (2) Type3 Unicode fallback to `/Encoding` glyph names, then printable ASCII; (3) `Type3Glyph::advance_width`; (4) `InterpreterSettings::ocg_overrides` for viewer layer toggles; (5) soft masks whose group lacks `/CS` were dropped (Chrome gradient text); (6) paint nesting cap `MAX_PAINT_NESTING` for tiling patterns and Type 3 glyphs (self-referencing content overflowed the stack); (7) CID width ranges in `/W` and `/W2` clamped to `MAX_CID` (a `0 4294967295 w` range hung); (8) `InterpreterSettings::hide_comments` skips markup annotations but keeps widgets and links (View ▸ Hide all comments); (9) `image_size_ok`: images and masks over 2^28 pixels are not decoded (a fuzzed `/W 4294967295` stencil mask allocated 4 GiB); (10) `MAX_NESTED_PAINTS`: nested paints (forms, Type 3 glyphs and tiling patterns inside one another) share a per-page budget of 50,000 (a glyph showing eight glyphs of its own font, sixteen deep, hung); (11) depends on skrifa 0.47 instead of 0.42 (upstream read-fonts fixed a Type 1 tokenizer that recursed through runs of numbers: a fuzzed font hung or overflowed the stack), with the one API change (`Charset::glyph_id` returns `Option`); (12) annotation appearances whose `/BBox` or `/Matrix` collapses to a line or a point are skipped (mapping them onto `/Rect` divided by zero and handed the device infinite or NaN matrices); (13) a Highlight annotation without a usable normal appearance is drawn from its `/QuadPoints`, `/C` and `/CA` with Multiply blending, as pdfcraft-annot draws highlights (it was not drawn at all); (14) simple-font `/Widths` (`read_widths`): entries only for codes 0–255 (one entry per code below `/FirstChar` was allocated: `4294967295` asked for 34 GB), an entry that isn't a number leaves its code missing instead of ending the array, and a `/LastChar` below `/FirstChar` gives no widths instead of discarding the font for a standard one; (15) the appearance box is mapped onto `/Rect` corner to corner: the translation now accounts for the scale (an appearance whose box didn't start at the origin and differed in size from `/Rect` was drawn shifted); (16) a clip set by `W`/`W*` takes effect after whichever path-painting operator ends the path (`S`, `f`, `B`, … as well as `n`; ISO 32000-2 §8.5.4): it was applied only on `n`, so `re W* S` clipped nothing and the clip stayed pending for a later, unrelated `n`; (17) `InterpreterSettings::cancelled`: once set, interpretation stops before the next content operator (a render pool sets it for every worker when dropped, after an edit or on closing, and for one worker when the watchdog gives up on its render, so renders whose answers nobody can receive stop instead of running to the end); (18) `MAX_PAGE_CONTENT`: the content a page holds at once (its contents, and the forms, tiling patterns and Type 3 glyph procedures being painted) stays within 512 MiB (`decode_content`; once it runs out, `InterpreterWarning::ContentTruncated` is reported (once per page by pdfcraft-render) so a truncated page is not a clean render; each level gives its bytes back when it finishes): forms nest fifty deep and each level kept its content while the next ran, so a form painting itself around a stream inflating to 256 MiB would have held over 12 GB; and Type 0 sampled functions take at most `MAX_SAMPLE_ENTRIES` (2^21) table points and `MAX_SAMPLE_VALUES` (2^23) samples, read no further than `/Size` asks (256 MiB of 1-bit samples asked for over 8 GB); (19) mesh shadings (types 4–7) are sampled only inside the area being drawn (`ShadingPattern::encode_within`, which the vendored hayro renderer calls with that area plus a pixel), within 8 visits per pixel of it plus 4 per triangle (`TriangleSampler`), and patches are triangulated one at a time: points far off the page were sampled over their whole bounding box, up to 65536 × 65536 entries (a fuzzed 147-byte patch stream; a synthetic page passed 6 GB in six seconds), ten thousand overlapping triangles ran for minutes, and every patch became 722 triangles before any limit; meshes hold at most `MAX_MESH_PATCHES` (2^16) patches and `MAX_MESH_TRIANGLES` (2^20) triangles, a lattice needs two to 2^19 vertices per row (`/VerticesPerRow 0` pushed empty rows for ever; one huge row read the whole stream), and a page samples at most `MAX_PAGE_MESH_TRIANGLES` (2^25) mesh triangles in all (a mesh painted a thousand times was triangulated a thousand times); each cut logs a warning; (20) PDF function construction is limited to 64 levels and 10,000 total nodes with cycle detection; calculator parsing and evaluation share procedure depth/work budgets, integer division/bitshift/rotation/index are panic-free, and fixed-size stack errors propagate; (21) a Type 0 font drawn with a substitute that isn't CID-keyed (it isn't embedded, or its program doesn't load) draws `.notdef` for a CID of an Adobe character collection (Japan1, GB1, CNS1, Korea1) that it can't find by Unicode: hayro took the CID for a glyph index of the substitute, which numbers its glyphs its own way, so the character was drawn as an unrelated one (𠮷 as "Ｑ" in BIZ UDPGothic); (22) a Type 0 font drawn with a substitute finds a character given with a variation selector (`/ToUnicode`, or Adobe-Japan1-UCS2, which maps the JIS X 0208 forms of the kanji JIS X 0213:2004 redrew that way: CID 1247 is 噂 U+E0100) through the substitute's variation sequences, else as the character alone: such a CID found no glyph and was drawn as an unrelated one; (23) backport of hayro 0.8's `gradient` module: `EncodedShadingPattern::as_svg_gradient` turns an axial or radial shading into stops of an sRGB gradient (within 0.01 per channel, a stop at every stitching-function bound via `Function::stitching_bounds`) when a native gradient draws it exactly (extended past the box drawn in; a radial one with no background colour); (24) `DecodedImageCache`: decoded image pixels shared across renders through the interpreter cache within a 64 MiB budget (an image-heavy page decoded its thousands of images again per tile and per scale), keyed by every decode input (`DecodedImageKey`: the stream's place and dictionary bytes, its object, mask/stencil/raster output, interpolation, decode target size, preview or final); images under a transfer function and inline images are not cached, an image over the budget is never kept, and a poisoned lock decodes uncached; (25) `Device::actual_text` hands the device a marked content sequence's `/ActualText` (inline or named in `/Properties`), so text extraction keeps the Thai, Indic and Arabic clusters Chrome marks this way (their shaped tone marks and vowels have no Unicode of their own); (26) non-embedded simple fonts drawn with a standard-font stand-in build the Latin Extended-A letters the stand-in lacks (`/Gbreve`, `/Idotaccent`, `/Scedilla`, `/aogonek`, `/ccaron`…) from the base letter and accent it has (`StandardKind::composite_glyph`, `ACCENTED_LETTERS`): PDFium's Foxit stand-ins hold only the Standard/WinAnsi/Mac Roman glyphs, so iText letters with Times New Roman unembedded showed blanks for Ğ, İ, Ş, ğ and ş; the comma-below letters (`/scommaaccent`, `/Tcommaaccent`, `/kcommaaccent`…, found by character, so `/uni0136` and the glyph list's `/Kcedilla` too) are the letter and the font's comma, smaller and under the baseline, since the stand-ins have no `commaaccent` (Romanian ș drew nothing and ț a cedilla in forms laid out from XFA; ģ, whose comma is turned and above, is not built); (27) an embedded TrueType font without a base encoding reads its codes as WinAnsi for text extraction (`EmbeddedKind::code_to_unicode_name`), matching pdftotext's output, instead of Standard: OCR text layers of scans write WinAnsi codes with no `/Encoding`, and Standard has no ü, ö or ß and reads é as Ø (glyph selection unchanged); (28) an image whose pixels hold at most 16 bits of samples (an 8-bit Separation or DeviceN image, a 1-, 2- or 4-bit gray or indexed one) is converted once per possible pixel value through a lookup table instead of once per pixel (`rgb_via_lookup`): a spot-colour tint transform is often a PostScript calculator program, run for every pixel, so a fillable character sheet's 3304 × 4105 DeviceN background took four seconds and 430 MB to draw a page; each entry is converted on its own (`convert_each`), so a tint transform giving the wrong number of components for a value the image never uses cannot shift the colours of the values it does; (29) `Device::begin_annotations`/`end_annotations` bracket the page's annotations, so a device can tell comment paints from document paints (reading colours recolour the document, never a comment, #1548); (30) the Annex D.2 alias codes read as the ink they draw: WinAnsiEncoding 0xA0 and 0xAD, and MacRomanEncoding 0xCA, carry the glyph names `space` and `hyphen` -- the names those tables also give 0x20 and 0x2D -- not `nbspace` and `sfthyphen`, whose glyph list values U+00A0 and U+00AD are invisible formatting characters. Extraction hid drawn ink behind them, so search and redaction matched neither (`XiaoBiaoSong.pdf` in the pdf.js corpus drew 104 soft hyphens; across the 989 files the soft hyphens we extract fall from 117 to 13 and the non-breaking spaces from 30 to 26, what remains coming from `/ToUnicode` maps, which say so explicitly). The base table only, in the text direction only: a `/Differences` name keeps its glyph list value and glyph lookup is untouched, both as pdftotext does; (31) a Type 0 sampled function keeps its samples in one flat array indexed by strides (`Type0::entry`) instead of a hash map keyed by each grid point: the map was built again, one hashed insert per sample, every time a shading using the function was painted (a 2,097,152-sample ramp painted 30 times took 38 s; `bug1721218` in the pdf.js corpus spent most of a 1.5 s page on it); the interpolation is unchanged, so results are bit for bit the same, a sample missing from a short stream still gives no value, and the maximum sample value is computed without overflowing at 32 bits per sample; (32) the decoded-image cache (24) keys a decode by the size the image is drawn at only when a JPX stream decodes it or its soft mask or mask (`decode_reads_target`): JPX is the only filter that decodes a reduced resolution for a size, so a JPEG or Flate scan was decoded again at every zoom step and kept once per zoom level; one decode now serves every size; `InterpreterCache::decoded_image_decodes` counts the cache's decodes; (33) Coons and tensor-product patches copy their vertices' colours less: each point of a patch's 20 × 20 triangle grid is interpolated once (`generate_patch_triangles`; each cell interpolated its four corners itself, 1,444 interpolations a patch for 400 points), a cell's second triangle takes the vertices its first copied, and the mesh sampler moves a patch's triangles into device space instead of copying each vertex (`TriangleSampler::sample_owned`); the same values, 10–16% faster on mesh-gradient pages; (34) `color::tint_transform_runs`: how many colours the calling thread has run through a Separation or DeviceN tint transform (counted once per batch in `convert_f32`), so a test counts patch (28)'s once-per-sample-value runs instead of timing them (its 10 s deadline failed on a busy FreeBSD VM in CI); (35) `EncodedShadingPattern::extent`: where a mesh shading (types 4–7) can paint at all, the pixels its triangles touched when it was encoded (`None` for the types that paint everywhere, and when the shading has a `/Background`), so the renderer samples a texture there alone and skips a mesh with nothing on the canvas; (36) a mesh shading's pixel whose corner no triangle holds strictly (on an edge two triangles share, or on the mesh's side) takes the triangle that covers its centre, edges included (`Triangle::covers`): testing only the corner left a white seam along every shared edge, and the colour is still taken at the corner; (38) the `/DefaultGray`, `/DefaultRGB` and `/DefaultCMYK` colour spaces of a resource dictionary (ISO 32000-2 §8.6.5.6) replace the device space content selects: with `g`/`G`, `rg`/`RG`, `k`/`K` and `cs`/`CS` (by name, or through a `/ColorSpace` resource that is a device space), and as the `/ColorSpace` of an image XObject, an inline image or a shading painted by `sh` or as a shading pattern (`DefaultColorSpaces`, looked up once per content window in the resources and those they inherit, so a form XObject without a default of its own draws in its page's, as Poppler and MuPDF draw it); a default that is a Lab, Indexed, Pattern or device space, has another number of components or is no colour space is ignored and device colour is drawn as before; device spaces inside other spaces (an Indexed base, a Separation or DeviceN alternate, a Pattern's underlying space), shading dictionaries, masks and transparency group spaces are unchanged, and the decoded-image cache (24) keys a decode by the default it was drawn in. hayro drew device colour as it was, so pages that calibrate it this way (common in PDF/A) came out in the wrong colours; (39) knockout groups (#2971): `Device::set_knockout` right after a `/K true` group is pushed; `Device::knockout_image` for an unmasked image before its group is pushed, when `Device::in_knockout_group`: the device clears its placement rectangle only if it will draw the image (on the canvas, not a preview's tiny image, and it decodes), decoded at its draw size so the draw reuses that decode; and, when `Device::knockout_erase_wanted`, a nested group's content drawn once between `begin_knockout_erase` and `end_knockout_erase` before the group itself, which then always gets a layer of its own; (40) an optional content membership dictionary's visibility expression (`/VE`: `/And`, `/Or`, `/Not`, nested up to 32 deep) decides its visibility ahead of `/OCGs` and `/P` (ISO 32000-2 §8.11.2.2); it was ignored, so content under it stayed visible with its layers hidden; a malformed expression, including one naming an object the document doesn't list in `/OCProperties /OCGs`, falls back to `/OCGs` and `/P` (#2986); (41) a transparency group that is neither isolated nor knockout, entered at full opacity with no soft mask and Normal blending, is drawn straight onto the page instead of into a layer of its own (§11.4.7: its elements composite with the backdrop; the layer made a `/BM /Multiply` highlight appearance an opaque block, #1968); (42) text in a Type 3 font is drawn filled whatever the text rendering mode, except mode 3 (invisible): only that mode has any effect on Type 3 fonts (ISO 32000-1 §9.3.6). Stroking ran the glyph procedure a second time, so `2 Tr` made bitmap fonts look bold, `1 Tr` painted an uncoloured glyph in the stroking colour, and `7 Tr` hid it (`7 Tr` since follows 64); (43) Lab colour spaces honour their own `/WhitePoint`: L\*a\*b\* goes to XYZ relative to that white, is Bradford-adapted to D65 and converted to sRGB (ISO 32000-2 §8.6.5.4) instead of through `LAB.icc` as if every white were D50 (a D65 blue came out 45 levels too red; the profile is removed); components normalised to [0, 1] (`manual_scale`, an `/Indexed` palette over Lab) map onto L\* 0..100 and the a\*/b\* `/Range`; (44) a Type 0 font whose `/ToUnicode` is the name `/Identity-H` or `/Identity-V` (not a CMap stream; WSCAD writes it) reads each code as its Unicode value, as Acrobat shows it (`Type0Font::identity_to_unicode`); the identity CMap the name resolved to maps codes to CIDs only, so such pages had no text to select or find; (46) an image XObject's group applies the fill opacity, so the fill alpha is reset to 1 before a stencil mask takes the fill colour as its paint, and a pattern paint drops the opacity it captured at `scn` (`Pattern::without_opacity`) (it was applied twice: a mask at `ca 0.5` was drawn at a quarter, #2975); (47) `path_as_rect`: a five-element path is a rectangle only when its four points are distinct, each step from one to the next runs along a single axis, and the last returns to the first. The length test admits one MoveTo and four LineTo as well as one MoveTo, three LineTo and a ClosePath, and asking only that every corner of the bounding box was visited says nothing about the order or the steps between, so a five-point polyline whose ends reached the corners was painted as that box -- a fourth side the path does not have -- and a clip of that shape was discarded; the coordinates are compared exactly, since the matrix is applied after this check and a `cm` of 10,000 turns `is_nearly_equal`'s 256th of a unit into tens of device units (#3327); (48) an `/ICCBased` colour space is cached by what its profile is read from (`IccProfileKey`): the stream object it names, looked up without reading the stream, and then the stream's dictionary and its own bytes (decrypted, not decoded), so a byte-for-byte copy of a profile in another object is read once, and a profile stream written directly in its array, which reports the object holding the array as every other stream written there does, is told apart by its bytes. hayro keyed it by the stream dictionary's bytes alone, so two different profiles with the same `/N` and `/Length` shared whichever was read first, and colours in the second were drawn through the first. `InterpreterCache::icc_profile_reads` counts the profiles read; (49) `OpenTypeFontBlob::new` also accepts a `post` table version 1.0, 2.0 or 3.0 written with its 16-bit halves swapped (`00 00 00 03`), as ScanSoft PDF Create! embeds every font; such fonts were refused and their text drawn with a substitute (Hebrew came out garbled or missing, #2946); `glyph_names` reads such a table from a copy with the version written correctly, so a 2.0 table's names serve `/Differences`. Any other unknown version is still refused (pdf.js 9462); (50) `CffFontBlob::glyph_index_by_name` looks a glyph name up in a name to glyph map built on first use (shared between clones) instead of searching the whole charset, with a string lookup per glyph, for every character code of a font; the first glyph with a name still wins; (51) at most 4096 graphics states saved at once, holding at most 64 MiB together (each `q` copies the whole state, dash array and clip list included, so short operators multiplied one large state); a `q` past either limit is counted but not copied, and its `Q` still removes the clips added since (from storytold/craft-libs#239); (52) a `/W2` run whose first CID is past `MAX_CID` stops there, as in `/W` (7), checked before the run's metrics are read: a first CID of 4294967295 inserted one entry and then overflowed, discarding the whole `/W2` table (from storytold/craft-libs#254); (55) a Separation or DeviceN tint transform that cannot be read keeps its colour space, and one that cannot be read or evaluated paints the alternate space's initial colour, as evaluation failures already did; that fallback, and a `cs`/`CS` colour space that cannot be read (drawn in DeviceGray as before), are now reported once per content stream as `InterpreterWarning::ColorFallback` (`RenderWarning::ColourFallback` in `pdfcraft-render`) instead of silently (#4260); (56) an annotation whose Invisible flag (bit 1) is set and whose `/Subtype` is none of the standard types of ISO 32000-2 Table 171 (`is_standard_annotation_type`; no `/Subtype` counts as unknown) is not drawn (Table 167), and a `/Popup` annotation's appearance is never painted on the page (§12.5.6.14: a popup has no appearance stream of its own; the viewer shows it as a window for its parent annotation). Both were drawn from their `/AP`; (57) a simple font drawn with a standard-font stand-in because the document names a font it doesn't embed (`StandardKind` with `fallback`) narrows a glyph to a narrower `/Widths` advance but no longer widens it to a wider one: the advance already places the next glyph (ISO 32000-2 §9.2.4; /Widths in Table 109), and widening drew a non-embedded `Arial,Bold` with 600-unit widths as a heavy fixed-pitch face whose `i` and `l` were solid blocks twice their width; an exact standard-14 font still matches its standard metrics both ways; (58) the fallback request for a Type 0 font that is not embedded (`FallbackFontQuery::with_descriptor`) takes its serif, italic and small-cap flags, `/FontWeight`, `/FontFamily` and `/FontStretch` from the descendant CIDFont's font descriptor (ISO 32000-2 §9.7.4.1, §9.7.6.1; §9.8): it read them from the Type 0 dictionary, which has none, so every such request asked for a regular sans-serif face and no resolver could tell a Song or Mincho font from a Hei or Gothic one, or see `/FontWeight 700`; (59) the NoRotate flag (`/F` bit 5) is honoured: the appearance, fitted to `/Rect`, is turned against the page's `/Rotate` about the rectangle's upper-left corner, which stays where it is, so it reads upright as the page is displayed (ISO 32000-2 §12.5.3). Sticky notes, which are saved NoRotate, were drawn turned with the page (#2583). Form fields and links stay in `/Rect`, where PdfCraft takes their click targets from it; (64) text in a Type 3 font follows ISO 32000-2 §9.3.6 in the clipping modes: under `7 Tr` it is not rendered (reported to the device as clipping text, `GlyphDrawMode::Clip`, so extraction keeps it), and in modes 4 to 7 a Type 3 glyph adds nothing to the clipping path (`clip_glyph`), so text shown only in a Type 3 font leaves no clip at `ET`. Since (42) `7 Tr` painted the glyphs in the fill colour (black by default), so a fill meant to follow showed them wherever it didn't cover them; 4 to 6 still render filled; (66) a Separation colour space whose colourant is the special name `All` paints every colourant at once (ISO 32000-2 §8.6.6.4): on screen and in an RGB print raster its tint t is the gray 1 − t, complemented because the device is additive, whatever its alternate space and tint transform say (hayro drew it through the tint transform, so a `/DeviceCMYK` transform giving cyan drew registration marks light cyan); `/None` still paints nothing, any other name still goes through its alternate, and a DeviceN space naming `All` keeps its alternate, since the clause gives the name that meaning in a Separation space; (63) a luminosity soft mask whose group colour space is `/DeviceCMYK` (by name, not an ICC-based or calibrated space) reads a `/DeviceCMYK` fill or stroke colour drawn in it, and its `/BC`, as their gray `1 - min(1, 0.3 c + 0.59 m + 0.11 y + k)` (ISO 32000-2 §10.4.2.3, §11.5.3: device colour converts to gray with no colour calibration; `Context::luminosity_in_device_cmyk`, `color::device_cmyk_to_gray`): it went through the CMYK display profile, where 100% K is a dark grey, so a 100% K mask let about an eighth of the masked paint through. CMYK images, shadings and tiling patterns inside such a mask, and CMYK on the page, still use the profile; (71) a lookup table in an indirect string reads: it was read as a stream first, which uses up a reference (ISO 32000-2 §8.6.6.3); (72) an uncoloured pattern's underlying space given as a name or a reference is used: it was skipped, so a CMYK cyan drew red (ISO 32000-2 §8.7.3.3); (73) `InterpreterSettings::jbig2_work`, set for each page in `interpret_page` and ended with it, and a JBIG2 image or an image with a JBIG2 mask past it left out with `ContentTruncated`; (74) colour spaces nest at most 32 levels, through every way one names another, so a chain of them ends instead of overflowing the stack; an ICC profile's `/Alternate` is read where it's met and isn't kept with the profile, or a page could build a chain past the limit on spaces already read; (75) a soft hyphen (U+00AD) that a CFF substitute for a CID font has no glyph for draws its hyphen, as Annex D.2 reads WinAnsi 0xAD in (30), instead of the CID taken as a glyph index (#5888) | (1) `raster::tests::widget_appearance_states`; (2, 3) `xtask text-oracle` on corpus/pdfjs issue918; (4) `raster::tests::layer_override_hides_content`; (5) `raster::tests::soft_mask_without_group_colour_space_is_applied`; (6) `raster::tests::self_referencing_tiling_pattern_terminates`, `raster::tests::self_referencing_type3_glyph_terminates`; (7) `raster::tests::huge_cid_width_ranges_terminate`; (8) `raster::tests::hiding_comments_keeps_fields`; (9) `raster::tests::absurd_mask_and_fax_sizes_are_refused_before_decoding`; (10) `raster::tests::type3_glyphs_that_fan_out_into_themselves_finish`; (11) `raster::tests::type1_font_programs_with_long_runs_of_numbers_load_quickly`; (12) `raster::tests::degenerate_annotation_appearances_are_skipped`; (13) `raster::tests::highlights_without_an_appearance_are_drawn`; (14) `raster::tests::malformed_simple_font_widths_keep_later_widths_and_the_font`, `raster::tests::huge_simple_font_first_char_terminates`; (15) `raster::tests::annotation_appearance_box_maps_onto_rect`; (16) `raster::tests::clipping_path_applies_after_any_painting_operator`; (17) `raster::tests::dropping_the_pool_cancels_its_renders`, `raster::tests::the_watchdog_stops_the_render_it_gives_up_on`; (18) `raster::tests::page_contents_and_nested_forms_share_one_budget`, `raster::tests::content_painted_again_and_again_keeps_drawing`, `raster::tests::sampled_functions_with_billions_of_samples_are_refused`; (19) `raster::tests::mesh_shadings_far_off_the_page_render`, `raster::tests::many_overlapping_mesh_triangles_finish`, `raster::tests::mesh_shadings_with_many_patches_or_no_rows_finish`; (20) `raster::tests::function_limits_calculator_arithmetic_and_depth`, `raster::tests::function_limits_public_construction_work_and_depth`, `raster::tests::function_limits_stitching_cycles_are_refused_and_shared_children_work`, `raster::tests::function_limits_invalid_calculator_keeps_the_page_renderable`; (21) `raster::tests::cids_a_substitute_lacks_are_not_drawn_as_other_glyphs`; (22) `raster::tests::variation_sequences_draw_their_kanji_in_a_substitute`; (23) `raster::tests::axial_and_radial_shadings_draw_as_native_gradients`; (24) `raster::tests::tiles_match_full_render`, `raster::tests::skipping_paths_off_the_canvas_changes_no_pixel` (both decode through the cache), `raster::tests::decoded_images_with_identical_dictionaries_keep_their_own_pixels`, `raster::tests::decoded_images_under_a_transfer_function_are_not_shared`; (25) `text::tests::actual_text_replaces_the_glyphs_it_marks_in_every_writing_direction`, `text::tests::actual_text_follows_the_pdf_rules_for_marked_content`; (26) `raster::tests::accented_letters_the_standard_stand_ins_lack_are_built_from_their_parts`, `raster::tests::comma_below_letters_in_standard_fonts_draw_a_comma_under_the_letter`; (27) `text::tests::truetype_without_an_encoding_reads_its_codes_as_win_ansi`; (28) `raster::tests::spot_colour_images_run_their_tint_transform_once_per_sample_value`, `raster::tests::images_converted_through_a_table_keep_every_samples_colour`, `raster::tests::a_malformed_tint_for_an_unused_sample_keeps_the_used_colours`; (29) `raster::tests::reading_colors_keep_highlight_annotations_readable`; (30) `text::tests::the_annex_d2_alias_codes_read_as_an_ordinary_hyphen_and_space`; (31) `raster::tests::shadings_with_large_sampled_functions_paint_quickly`, `raster::tests::sampled_functions_interpolate_their_grid_as_before`; (32) `raster::tests::decoded_images_serve_every_size_they_are_drawn_at`; (33) `raster::tests::patch_meshes_keep_their_corner_colours`; (34) `raster::tests::spot_colour_images_run_their_tint_transform_once_per_sample_value`; (35) `mesh_extent_tests::{a_mesh_paints_its_triangles_alone, a_mesh_keeps_the_graphics_state_alpha, a_mesh_off_the_canvas_paints_nothing}` (`cargo test -p hayro`); (36) `raster::tests::mesh_triangles_sharing_an_edge_leave_no_seam`; (38) `raster::tests::default_colour_spaces::device_rgb_operators_draw_in_default_rgb`, `raster::tests::default_colour_spaces::device_cmyk_operators_draw_in_default_cmyk`, `raster::tests::default_colour_spaces::device_gray_operators_draw_in_default_gray`, `raster::tests::default_colour_spaces::device_images_draw_in_the_default_colour_spaces`, `raster::tests::default_colour_spaces::device_shadings_draw_in_the_default_colour_spaces`, `raster::tests::default_colour_spaces::an_image_drawn_with_and_without_a_default_keeps_both_colours`, `raster::tests::default_colour_spaces::forms_inherit_the_default_colour_spaces_they_do_not_override`, `raster::tests::default_colour_spaces::unusable_defaults_leave_device_colour_alone`; (39) `raster::tests::knockout_groups_replace_what_their_elements_cover`; (40) `raster::tests::visibility_expressions_decide_what_a_membership_dictionary_shows`; (41) `raster::tests::a_multiply_highlight_inside_a_transparency_group_keeps_the_text`; (42) `raster::tests::rendering_modes_but_3_and_7_draw_type3_glyphs_as_filled`; (43) `raster::tests::lab_honours_the_space_white_point`, `raster::tests::indexed_palettes_over_lab_keep_their_colours`; (44) `text::tests::identity_to_unicode_names_give_text`; (46) `raster::tests::stencil_masks_are_drawn_at_the_fill_opacity_once`; (47) `raster::tests::a_five_point_polyline_is_not_painted_as_its_bounding_box`, `raster::tests::a_path_that_visits_every_corner_out_of_order_is_not_a_rectangle`, `raster::tests::a_rectangle_is_still_painted_as_a_rectangle`; (48) `raster::tests::icc_profile_cache::icc_profiles_with_identical_dictionaries_draw_their_own_colours`, `raster::tests::icc_profile_cache::icc_profiles_used_again_are_read_once`; (49) `raster::tests::a_font_whose_post_version_has_swapped_halves_is_still_used`; (50) `font::blob::tests::glyph_names_resolve_to_the_first_glyph_with_that_name`; (51) `context::saved_state_tests::repeated_saves_stay_bounded_and_balanced`; (52) `font::cid::tests::w2_run_past_the_largest_cid_keeps_the_rest_of_the_table` (hayro-interpret); (55) `raster::tests::devicen_inks::a_tint_transform_that_cannot_be_evaluated_draws_the_rest_of_the_page_and_warns`, `raster::tests::devicen_inks::a_tint_transform_that_cannot_be_read_falls_back_and_warns`; (56) `raster::tests::invisible_unknown_annotations_and_popups_paint_nothing`; (57) `substitute_widths::a_substitute_keeps_its_glyph_shapes_at_wider_advances`, `substitute_widths::a_substitute_still_narrows_glyphs_to_narrower_advances` (pdfcraft-render `tests/it`); (58) `cid_fallback_query::a_cid_font_fallback_request_carries_its_descendants_style`, `cid_fallback_query::a_cid_font_without_a_descriptor_asks_for_the_default_style` (pdfcraft-render integration tests); (59) `raster::tests::no_rotate_notes_stay_upright_at_their_corner_on_rotated_pages` (hayro-interpret); (64) `type3_text_clip::{clip_only_type3_text_is_not_rendered_and_leaves_no_clip, invisible_type3_text_is_not_rendered, other_modes_render_type3_glyphs_filled_and_never_clip, only_outline_glyphs_shape_a_clip_shared_with_type3_text, self_nesting_type3_text_in_a_clip_mode_stays_bounded}` (`cargo test -p pdfcraft-render --test it`), `raster::tests::rendering_modes_but_3_and_7_draw_type3_glyphs_as_filled`, `text::tests::clipping_mode_type3_text_is_kept_by_the_ocr_filter`; (66) `separation_all::{separation_all_fills_paint_the_complemented_tint_as_gray, separation_all_strokes_paint_the_complemented_tint_as_gray, separation_all_images_paint_each_sample_as_the_complemented_gray, separation_all_shadings_paint_the_complemented_gray, separation_none_paints_nothing, a_named_separation_still_paints_through_its_alternate, a_device_n_space_naming_all_still_paints_through_its_alternate}` (`cargo test -p pdfcraft-render --test it`); (63) `color::device_cmyk_gray_tests::{each_ink_darkens_by_its_iso_weight, ink_past_full_coverage_is_black_and_bad_components_read_as_zero}` (hayro-interpret), `it::luminosity_group_space::{black_ink_in_a_device_cmyk_group_masks_fully_and_no_ink_shows_fully, each_process_ink_in_a_device_cmyk_group_masks_by_its_weight, a_device_cmyk_backdrop_colour_masks_by_its_gray, gray_and_rgb_groups_mask_as_before, device_cmyk_on_the_page_keeps_its_display_colour}` (pdfcraft-render); (71) `raster::tests::an_indexed_lookup_table_is_read_in_every_legal_form` (pdfcraft-render); (72) `raster::tests::an_uncoloured_pattern_uses_its_underlying_space_in_every_form` (pdfcraft-render); (73) `raster::tests::the_jbig2_images_of_a_page_share_one_allowance`, `a_picture_whose_jbig2_mask_is_past_the_pages_allowance_is_not_drawn` and `a_pages_jbig2_allowance_ends_with_the_page` (pdfcraft-render); (74) `raster::tests::{a_chain_of_colour_spaces_ends, colour_spaces_are_read_to_the_limit_of_their_nesting, a_chain_built_on_spaces_already_read_ends, a_profile_read_at_the_limit_keeps_its_alternate_for_a_read_nearer_the_top}` (pdfcraft-render); (75) `raster::tests::a_soft_hyphen_a_substitute_lacks_draws_its_hyphen` (pdfcraft-render) | -| hayro | 0.7.1 | Apache-2.0 OR MIT | (1) `RenderSettings::{x_offset, y_offset}` to render one tile of a large page; (2) images over `MAX_IMAGE_PIXELS` (2^28) are skipped (a fuzzed `/W 4294967295` hung in resampling); (3) `tiling_cell_scale` keeps a tiling pattern's cell pixmap within 3000 px a side whatever its `/XStep`/`/YStep` (a fuzzed step far beyond `/BBox` allocated 2 GB); (4) stroke widths clamped to 4 × (canvas width + height) device pixels (a fuzzed `/LW 9223372036854775807` allocated 10 GB in stroke expansion); (5) `usable_dash_pattern`: a stroke whose dash pattern could cut it into more than `MAX_DASHES_PER_STROKE` (250,000) pieces, counting kurbo's restart of the pattern at every subpath and `ClosePath`, is drawn solid (a fuzzed `/D [[11] 0]` on a line from x = 92234775807 allocated over 5 GB, `[0.000001] 0 d` on a 20 pt line over 10 GB); (6) a dash pattern with a negative or non-finite entry, or a period that isn't positive, is drawn solid (`[-1 -1] 0 d` looped for ever in kurbo); (7) `render_size` and `render_into`: render into a buffer the caller allocates (PdfCraft renders into pixel-aligned memory that the GUI takes over as texture data); `render` is `render_into` on a new pixmap; (8) checked image source/target/RGBA, alpha-premultiplication scratch and intermediate sizes before resampling, finite transforms and exact source/mask buffers; sources wider than u16 may shrink before backend limits, with a 2^20 source-axis cap and conservative 128 MiB filter-planning estimate; an over-limit or failed resize keeps the original image resolution and geometry when the source fits the backend, otherwise skips it, with a logged warning; staging, resampling output/scratch, RGBA conversion and padding buffers use fallible reserves; (9) `may_paint_canvas`: a fill or stroke whose control-point box, grown by how far its paint reaches past the path (half the stroke width times the larger of the miter limit and √2) plus 0.5 user units and 2 pixels, misses the canvas is not drawn, so a tile no longer strokes and fills every path of its page (item 5 of #623); `set_skip_offscreen_paths` turns this off for the regression test; (10) `/Interpolate` governs the pre-resize too (#624): a minified image that does not request interpolation is pre-resized with nearest-neighbour (since (21), only inside Type 3 glyphs), planned without the Catmull-Rom filter limits (`image_resampling_size` takes the flag); a Type 3 glyph image is sampled bicubically only when interpolated; a minified image with a matching soft mask is nearest-sampled straight into premultiplied RGBA; (11) `may_paint_rect`: an image whose stream-size box, mapped to the device, misses the canvas is skipped before decoding (decoded data can only shrink against that box), so a tile of an image-heavy page no longer decodes, resamples and rasterizes every image on it; (12) preview renders (`InterpreterSettings::preview`: backdrops, thumbnails, grid cells, signature-drag backgrounds) skip images under `PREVIEW_MIN_IMAGE_AREA` device pixels, while final renders always draw them; (13) reading colours (#897): `RenderSettings::color_replacement` takes a `ColorReplacement` (text colour, background, `only_black`, `line_art`) that remaps solid text and line-art paint while rendering, glyph outlines counting as text and stencil-mask ink as line art; images, image alpha masks, soft masks and pattern cells keep their colours; `render_pdf` gains a `color_replacement: Option` argument (pass `None` for true colours); the upstream-only `[dev-dependencies.log]` entry is removed from `Cargo.toml` (log is a normal dependency since the logging patch), so `cargo test -p hayro` runs this crate's unit tests; (14) a shrunk image that is interpolated (or, since (21), area-averaged) and whose axes lie along the device's (upright, mirrored or a quarter turn, as on a page under `/Rotate`) is snapped to whole device pixels: shrunk once to its rounded device rectangle (halves round up, so a tile snaps an edge where the whole page does) and drawn 1:1 from the rounded corner, instead of a ceil-sized raster resampled again at a fractional scale and offset (`snap_minified_image`; scans looked soft next to Acrobat, and until #5530 a turned one still did, up to 200 grey levels off the same page drawn upright). An image counts as axis-aligned when the terms the snap drops move its far edges by at most a hundredth of a device pixel (`SNAP_SKEW`), not by a tolerance on the coefficients, which let a slight skew of a very wide image through; and the snapped 1:1 transform is used only when the shrink produced the snapped size, so a failed resize keeps the full transform instead of drawing the source raster at its source size; (15) a soft mask of another size than its image is filtered when shrunk, whatever its `/Interpolate` (#1649): such a mask carries the shape on its own (Word saves text as a tiny black image with a glyph-bitmap mask), and nearest-neighbour dropped every stroke thinner than the shrink factor; (16) under a soft mask or a blend mode, axial and radial shadings are sampled instead of drawn as native gradients (#2128): vello_cpu 0.0.8 copies an opaque gradient over fully covered tiles without the mask or blend mode, so PowerPoint's faded gradient fills came out solid; (17) the renderer keeps annotation paints out of the reading-colours mapping between `begin_annotations` and `end_annotations` (#1548); (18) the page renders as one transparency group; under reading colours its background is painted inside it as the initial backdrop, so a Multiply highlight blends with the dark background instead of staying an opaque block, and otherwise the background stays outside, so a blend mode over an empty part of the page keeps its colour instead of blending with the white paper (#2970); (19) `set_paint` samples a mesh shading's texture over its triangles alone (`EncodedShadingPattern::extent`, padded rather than repeated) and skips a mesh with none on the canvas: a BricsCAD plan's 310 one-triangle `sh` each covered the whole tile, 2 s per 1024 px tile at any zoom (the watchdog gave the page up on a busy machine), now about 70 ms; (20) a tiling pattern whose step is too large for a cell pixmap, where what is painted (a fill, or a stroke with its width) can only show the first cell (PowerPoint's `/XStep 2147480100` for a picture fill drawn once), has its `/BBox` drawn at full scale with a transparent border and padded instead of tiled (`shows_one_cell`; a cell whose scale isn't finite and positive, from a `/BBox` too thin to divide by, keeps the old path: `usable_cell_scale`); shrinking the cell to fit the 3000 px limit of (3) left the picture under a pixel, so network diagrams lost their pictures (#3349); (21) an image that does not request interpolation and is drawn below its resolution, outside a Type 3 glyph, is shrunk by area averaging (`area_shrink`: each device pixel the coverage-weighted mean of the samples it covers, in 16.16 fixed point whose weights sum to exactly one, premultiplied when the image has a mask; a soft mask no larger than the image is first sampled onto the image's grid by `mask_on_image_grid`, so its colour and mask are averaged together instead of the averaged colour of transparent samples showing under a mask layer at the mask's edges) to its snapped device rectangle, instead of nearest-neighbour: `/Interpolate` asks for smoothing only when an image is enlarged (ISO 32000-2 §8.9.5.3), and nearest-neighbour kept one sample in N per axis, so the text of a 300 dpi scan at 100% lost most of its strokes (sampling them again at a fractional offset dropped more). PicScale's `Box` and `Area` filters were not used: shrinking, both average a window four target pixels wide. On a 13-page 300 dpi scan every rendered sample at 96, 120 and 192 dpi is within 1 of the exact area average (it was up to 255 off, 17-23% of samples by more than 1). A snapped image whose shrink failed keeps the full transform, so it is not drawn at its source size. The shrink never enlarges: a side drawn at or above its resolution (an image shrunk across and stretched down, as a 204 × 98 dpi fax is) keeps its samples for the sampler. Rows are averaged first (contiguous, cheap) and the fewer rows that leaves are then shrunk across with taps worked out once per image, which made the shrink about three times faster (a 300 dpi page at 96 dpi: 25 ms, nearest-neighbour 11 ms). A source with a side over `MAX_RESAMPLING_SOURCE_SIDE` (2^20) keeps (10)'s nearest-neighbour shrink, and `area_shrink` refuses it, so its accumulator row and tap table stay a few times one source row (a 2^27 × 2 image of zeros is a tiny file); (22) knockout groups: while one is innermost, fills, strokes, glyph outlines and shadings are drawn with `Compose::Copy` (weighted by coverage), so each element replaces what earlier elements of the group drew instead of compositing over it; an image or nested group, drawn into a layer of its own, first has what is under its shape cleared (`knockout_image` for an unmasked image's rectangle, only when `draw_image` will draw it, decoded at the size it is drawn; a knockout erase pass that draws a nested group's content with every paint clearing instead of painting, a shading fill's clear clipped to the shading's `/BBox` as its paint is) and then composites over it (#2971); (23) backport of hayro 0.8: shadings that `as_svg_gradient` can represent are painted as a `vello_cpu` linear or two-point radial gradient instead of a texture sampled pixel by pixel through the shading function (InDesign fades are luminosity soft masks of one full-page axial `sh`: 40 on a report page took 15 s at 144 dpi, now under 1 s); without upstream's half-pixel shift, which suits vello_cpu 0.3 but moves 0.0.8's gradients off pixel centres; (25) a tiling cell is drawn at the scale that makes one `/XStep` (`/YStep`) exactly the cell pixmap's whole number of pixels (`tiling_cell_pixels`, at least 1 and at most the cap), so the repeated pixmap puts cells at whole multiples of the step at any device scale; drawing at the unrounded scale repeated the cell up to half a pixel off at every step, and the error grew with distance from the pattern's origin (ISO 32000-2 8.7.3.1); (28) an image of page-scan size (`SMOOTHED_IMAGE_SIDE`: 512 samples or more on its shorter side) that does not request interpolation is drawn with `ImageQuality::Medium` (bilinear, as interpolated images are) when enlarged on either side, also when the other side is shrunk (the area shrink leaves that side at 1:1), outside a Type 3 glyph: scans rarely set `/Interpolate`, and drawn nearest-neighbour every sample of a zoomed scan became a block, so its text had staircase edges; icons, QR codes and pixel art keep their hard pixels. Bicubic (`High`) looked the same on a zoomed 300 dpi scan and drew four 1024 px tiles at 400% in 267 ms against 50 ms | (1) `raster::tests::tiles_match_full_render`; (2) `raster::tests::absurd_image_dimensions_are_skipped`; (3) `raster::tests::tiling_cells_stay_small_whatever_the_step`; (4) `raster::tests::absurd_predictor_columns_and_line_widths_render`; (5) `raster::tests::dash_patterns_that_cut_a_stroke_into_billions_of_pieces_render`; (6) `raster::tests::invalid_dash_patterns_terminate_and_are_drawn_solid`; (7) `raster::tests::pixels_are_a_plain_renders_bytes_in_a_buffer_a_gui_can_take_over`; (8) `raster::tests::image_resampling_rejects_anisotropic_target_growth`, `raster::tests::image_resampling_bounds_intermediate_before_planning`, `raster::tests::image_resampling_rejects_invalid_sources_and_scales`, `raster::tests::image_resampling_keeps_valid_sizes_and_exact_limits`, `raster::tests::resampling_fallback_preserves_original_geometry_and_pixels`, `raster::tests::image_resampling_wide_sources_shrink_before_backend_side_limits`, `raster::tests::image_resampling_wide_transparent_sources_still_shrink`, `raster::tests::image_resampling_padded_backend_dimensions_stay_checked`, `raster::tests::image_resampling_mismatched_alpha_mask_keeps_pixels_and_placement`; (9) `raster::tests::skipping_paths_off_the_canvas_changes_no_pixel`; (10) `raster::tests::non_interpolated_images_are_not_smoothed_when_minified`, `raster::tests::non_interpolated_gray_images_are_not_smoothed_when_minified`, `raster::tests::non_integer_minification_averages_the_columns_each_pixel_covers`, `raster::tests::non_integer_minification_averages_colour_and_mask_together`, `raster::tests::non_interpolated_alpha_images_keep_their_colours_and_mask_when_minified`, `raster::tests::non_interpolated_type3_glyph_images_are_not_smoothed_when_minified`, `raster::tests::image_resampling_nearest_accepts_wide_sources_the_filter_refuses`, `raster::tests::wide_non_interpolated_sources_draw_after_shrinking`; (11) `raster::tests::skipping_paths_off_the_canvas_changes_no_pixel`; (12) `raster::tests::preview_renders_skip_tiny_images_but_finals_draw_everything`; (13) `color_replacement_tests::{black_maps_to_text_colour, white_maps_to_background, midgrey_interpolates, alpha_copied_verbatim, only_black_threshold, line_art_false_skips_paths_but_not_text}`, `color_replacement_render_tests::{color_replacement_maps_text_and_line_art, color_replacement_maps_stencil_mask_ink}`; (14) `raster::tests::minified_image_edges_snap_to_whole_pixels`, `raster::tests::a_tile_snaps_half_pixel_image_edges_as_the_whole_page_does` (`cargo test -p hayro`), `raster::tests::a_turned_or_mirrored_minified_image_snaps_to_whole_pixels`, `rotated_image_snap::a_shrunk_image_on_a_turned_page_is_the_upright_render_turned` (`cargo test -p pdfcraft-render --test it`), `renderer::snap_tests::a_skew_that_moves_a_far_edge_is_not_snapped` (`cargo test -p hayro`); (15) `raster::tests::shrunk_soft_mask_larger_than_its_image_keeps_thin_strokes`; (16) `raster::tests::soft_mask_applies_to_a_shading_pattern_fill`; (17), (18) `raster::tests::reading_colors_keep_highlight_annotations_readable`; (19) `mesh_extent_tests::{a_mesh_paints_its_triangles_alone, a_mesh_keeps_the_graphics_state_alpha, a_mesh_off_the_canvas_paints_nothing}` (`cargo test -p hayro`); (20) `raster::tests::a_tiling_pattern_drawn_once_shows_its_cell`, `raster::tests::a_large_step_still_tiles_where_the_next_cell_shows`, `raster::tests::a_tiling_cell_too_thin_to_scale_keeps_the_old_path`; (21) `raster::tests::non_interpolated_minification_keeps_thin_strokes`, `raster::tests::non_integer_minification_averages_the_columns_each_pixel_covers`, `raster::tests::non_integer_minification_averages_colour_and_mask_together`, `raster::tests::non_interpolated_minification_does_not_bleed_transparent_colour`, `raster::tests::a_smaller_soft_mask_does_not_bleed_transparent_colour_when_minified`, `raster::tests::minified_image_edges_snap_to_whole_pixels`, `raster::tests::an_image_shrunk_one_way_and_stretched_the_other_is_still_averaged`, `raster::tests::wide_non_interpolated_sources_draw_after_shrinking`, `renderer::area_shrink_tests` (`cargo test -p hayro`); (22) `raster::tests::knockout_groups_replace_what_their_elements_cover`; (23) `raster::tests::axial_and_radial_shadings_draw_as_native_gradients`; (25) `tiling_cell_placement::{far_cells_of_a_fill_sit_at_whole_steps, far_cells_sit_at_whole_steps_under_a_turned_scaled_matrix, far_cells_of_a_stroke_sit_at_whole_steps, a_cell_pixmap_spans_exactly_one_step}` (`cargo test -p pdfcraft-render --test it`); (28) `raster::tests::large_non_interpolated_images_are_smoothed_when_enlarged`, `raster::tests::small_non_interpolated_images_keep_hard_pixels_when_enlarged`, `raster::tests::a_scan_shrunk_one_way_is_smoothed_the_way_it_is_stretched` | +| hayro | 0.7.1 | Apache-2.0 OR MIT | (1) `RenderSettings::{x_offset, y_offset}` to render one tile of a large page; (2) images over `MAX_IMAGE_PIXELS` (2^28) are skipped (a fuzzed `/W 4294967295` hung in resampling); (3) `tiling_cell_scale` keeps a tiling pattern's cell pixmap within 3000 px a side whatever its `/XStep`/`/YStep` (a fuzzed step far beyond `/BBox` allocated 2 GB); (4) stroke widths clamped to 4 × (canvas width + height) device pixels (a fuzzed `/LW 9223372036854775807` allocated 10 GB in stroke expansion); (5) `usable_dash_pattern`: a stroke whose dash pattern could cut it into more than `MAX_DASHES_PER_STROKE` (250,000) pieces, counting kurbo's restart of the pattern at every subpath and `ClosePath`, is drawn solid (a fuzzed `/D [[11] 0]` on a line from x = 92234775807 allocated over 5 GB, `[0.000001] 0 d` on a 20 pt line over 10 GB); (6) a dash pattern with a negative or non-finite entry, or a period that isn't positive, is drawn solid (`[-1 -1] 0 d` looped for ever in kurbo); (7) `render_size` and `render_into`: render into a buffer the caller allocates (PdfCraft renders into pixel-aligned memory that the GUI takes over as texture data); `render` is `render_into` on a new pixmap; (8) checked image source/target/RGBA, alpha-premultiplication scratch and intermediate sizes before resampling, finite transforms and exact source/mask buffers; sources wider than u16 may shrink before backend limits, with a 2^20 source-axis cap and conservative 128 MiB filter-planning estimate; an over-limit or failed resize keeps the original image resolution and geometry when the source fits the backend, otherwise skips it, with a logged warning; staging, resampling output/scratch, RGBA conversion and padding buffers use fallible reserves; (9) `may_paint_canvas`: a fill or stroke whose control-point box, grown by how far its paint reaches past the path (half the stroke width times the larger of the miter limit and √2) plus 0.5 user units and 2 pixels, misses the canvas is not drawn, so a tile no longer strokes and fills every path of its page (item 5 of #623); `set_skip_offscreen_paths` turns this off for the regression test; (10) `/Interpolate` governs the pre-resize too (#624): a minified image that does not request interpolation is pre-resized with nearest-neighbour (since (21), only inside Type 3 glyphs), planned without the Catmull-Rom filter limits (`image_resampling_size` takes the flag); a Type 3 glyph image is sampled bicubically only when interpolated; a minified image with a matching soft mask is nearest-sampled straight into premultiplied RGBA; (11) `may_paint_rect`: an image whose stream-size box, mapped to the device, misses the canvas is skipped before decoding (decoded data can only shrink against that box), so a tile of an image-heavy page no longer decodes, resamples and rasterizes every image on it; (12) preview renders (`InterpreterSettings::preview`: backdrops, thumbnails, grid cells, signature-drag backgrounds) skip images under `PREVIEW_MIN_IMAGE_AREA` device pixels, while final renders always draw them; (13) reading colours (#897): `RenderSettings::color_replacement` takes a `ColorReplacement` (text colour, background, `only_black`, `line_art`) that remaps solid text and line-art paint while rendering, glyph outlines counting as text and stencil-mask ink as line art; images, image alpha masks, soft masks and pattern cells keep their colours; `render_pdf` gains a `color_replacement: Option` argument (pass `None` for true colours); the upstream-only `[dev-dependencies.log]` entry is removed from `Cargo.toml` (log is a normal dependency since the logging patch), so `cargo test -p hayro` runs this crate's unit tests; (14) a shrunk image that is interpolated (or, since (21), area-averaged) and whose axes lie along the device's (upright, mirrored or a quarter turn, as on a page under `/Rotate`) is snapped to whole device pixels: shrunk once to its rounded device rectangle (halves round up, so a tile snaps an edge where the whole page does) and drawn 1:1 from the rounded corner, instead of a ceil-sized raster resampled again at a fractional scale and offset (`snap_minified_image`; scans looked soft next to Acrobat, and until #5530 a turned one still did, up to 200 grey levels off the same page drawn upright). An image counts as axis-aligned when the terms the snap drops move its far edges by at most a hundredth of a device pixel (`SNAP_SKEW`), not by a tolerance on the coefficients, which let a slight skew of a very wide image through; and the snapped 1:1 transform is used only when the shrink produced the snapped size, so a failed resize keeps the full transform instead of drawing the source raster at its source size; (15) a soft mask of another size than its image is filtered when shrunk, whatever its `/Interpolate` (#1649): such a mask carries the shape on its own (Word saves text as a tiny black image with a glyph-bitmap mask), and nearest-neighbour dropped every stroke thinner than the shrink factor; (16) under a soft mask or a blend mode, axial and radial shadings are sampled instead of drawn as native gradients (#2128): vello_cpu 0.0.8 copies an opaque gradient over fully covered tiles without the mask or blend mode, so PowerPoint's faded gradient fills came out solid; (17) the renderer keeps annotation paints out of the reading-colours mapping between `begin_annotations` and `end_annotations` (#1548); (18) the page renders as one transparency group; under reading colours its background is painted inside it as the initial backdrop, so a Multiply highlight blends with the dark background instead of staying an opaque block, and otherwise the background stays outside, so a blend mode over an empty part of the page keeps its colour instead of blending with the white paper (#2970); (19) `set_paint` samples a mesh shading's texture over its triangles alone (`EncodedShadingPattern::extent`, padded rather than repeated) and skips a mesh with none on the canvas: a BricsCAD plan's 310 one-triangle `sh` each covered the whole tile, 2 s per 1024 px tile at any zoom (the watchdog gave the page up on a busy machine), now about 70 ms; (20) a tiling pattern whose step is too large for a cell pixmap, where what is painted (a fill, or a stroke with its width) can only show the first cell (PowerPoint's `/XStep 2147480100` for a picture fill drawn once), has its `/BBox` drawn at full scale with a transparent border and padded instead of tiled (`shows_one_cell`; a cell whose scale isn't finite and positive, from a `/BBox` too thin to divide by, keeps the old path: `usable_cell_scale`); shrinking the cell to fit the 3000 px limit of (3) left the picture under a pixel, so network diagrams lost their pictures (#3349); (21) an image that does not request interpolation and is drawn below its resolution, outside a Type 3 glyph, is shrunk by area averaging (`area_shrink`: each device pixel the coverage-weighted mean of the samples it covers, in 16.16 fixed point whose weights sum to exactly one, premultiplied when the image has a mask; a soft mask no larger than the image is first sampled onto the image's grid by `mask_on_image_grid` (nearest-neighbour, or bilinearly when the mask asks for `/Interpolate`, so its edge stays as smooth as an enlarged mask's), so its colour and mask are averaged together instead of the averaged colour of transparent samples showing under a mask layer at the mask's edges) to its snapped device rectangle, instead of nearest-neighbour: `/Interpolate` asks for smoothing only when an image is enlarged (ISO 32000-2 §8.9.5.3), and nearest-neighbour kept one sample in N per axis, so the text of a 300 dpi scan at 100% lost most of its strokes (sampling them again at a fractional offset dropped more). PicScale's `Box` and `Area` filters were not used: shrinking, both average a window four target pixels wide. On a 13-page 300 dpi scan every rendered sample at 96, 120 and 192 dpi is within 1 of the exact area average (it was up to 255 off, 17-23% of samples by more than 1). A snapped image whose shrink failed keeps the full transform, so it is not drawn at its source size. The shrink never enlarges: a side drawn at or above its resolution (an image shrunk across and stretched down, as a 204 × 98 dpi fax is) keeps its samples for the sampler. Rows are averaged first (contiguous, cheap) and the fewer rows that leaves are then shrunk across with taps worked out once per image, which made the shrink about three times faster (a 300 dpi page at 96 dpi: 25 ms, nearest-neighbour 11 ms). A source with a side over `MAX_RESAMPLING_SOURCE_SIDE` (2^20) keeps (10)'s nearest-neighbour shrink, and `area_shrink` refuses it, so its accumulator row and tap table stay a few times one source row (a 2^27 × 2 image of zeros is a tiny file); (22) knockout groups: while one is innermost, fills, strokes, glyph outlines and shadings are drawn with `Compose::Copy` (weighted by coverage), so each element replaces what earlier elements of the group drew instead of compositing over it; an image or nested group, drawn into a layer of its own, first has what is under its shape cleared (`knockout_image` for an unmasked image's rectangle, only when `draw_image` will draw it, decoded at the size it is drawn; a knockout erase pass that draws a nested group's content with every paint clearing instead of painting, a shading fill's clear clipped to the shading's `/BBox` as its paint is) and then composites over it (#2971); (23) backport of hayro 0.8: shadings that `as_svg_gradient` can represent are painted as a `vello_cpu` linear or two-point radial gradient instead of a texture sampled pixel by pixel through the shading function (InDesign fades are luminosity soft masks of one full-page axial `sh`: 40 on a report page took 15 s at 144 dpi, now under 1 s); without upstream's half-pixel shift, which suits vello_cpu 0.3 but moves 0.0.8's gradients off pixel centres; (25) a tiling cell is drawn at the scale that makes one `/XStep` (`/YStep`) exactly the cell pixmap's whole number of pixels (`tiling_cell_pixels`, at least 1 and at most the cap), so the repeated pixmap puts cells at whole multiples of the step at any device scale; drawing at the unrounded scale repeated the cell up to half a pixel off at every step, and the error grew with distance from the pattern's origin (ISO 32000-2 8.7.3.1); (28) an image of page-scan size (`SMOOTHED_IMAGE_SIDE`: 512 samples or more on its shorter side) that does not request interpolation is drawn with `ImageQuality::Medium` (bilinear, as interpolated images are) when enlarged on either side, also when the other side is shrunk (the area shrink leaves that side at 1:1), outside a Type 3 glyph: scans rarely set `/Interpolate`, and drawn nearest-neighbour every sample of a zoomed scan became a block, so its text had staircase edges; icons, QR codes and pixel art keep their hard pixels. Bicubic (`High`) looked the same on a zoomed 300 dpi scan and drew four 1024 px tiles at 400% in 267 ms against 50 ms | (1) `raster::tests::tiles_match_full_render`; (2) `raster::tests::absurd_image_dimensions_are_skipped`; (3) `raster::tests::tiling_cells_stay_small_whatever_the_step`; (4) `raster::tests::absurd_predictor_columns_and_line_widths_render`; (5) `raster::tests::dash_patterns_that_cut_a_stroke_into_billions_of_pieces_render`; (6) `raster::tests::invalid_dash_patterns_terminate_and_are_drawn_solid`; (7) `raster::tests::pixels_are_a_plain_renders_bytes_in_a_buffer_a_gui_can_take_over`; (8) `raster::tests::image_resampling_rejects_anisotropic_target_growth`, `raster::tests::image_resampling_bounds_intermediate_before_planning`, `raster::tests::image_resampling_rejects_invalid_sources_and_scales`, `raster::tests::image_resampling_keeps_valid_sizes_and_exact_limits`, `raster::tests::resampling_fallback_preserves_original_geometry_and_pixels`, `raster::tests::image_resampling_wide_sources_shrink_before_backend_side_limits`, `raster::tests::image_resampling_wide_transparent_sources_still_shrink`, `raster::tests::image_resampling_padded_backend_dimensions_stay_checked`, `raster::tests::image_resampling_mismatched_alpha_mask_keeps_pixels_and_placement`; (9) `raster::tests::skipping_paths_off_the_canvas_changes_no_pixel`; (10) `raster::tests::non_interpolated_images_are_not_smoothed_when_minified`, `raster::tests::non_interpolated_gray_images_are_not_smoothed_when_minified`, `raster::tests::non_integer_minification_averages_the_columns_each_pixel_covers`, `raster::tests::non_integer_minification_averages_colour_and_mask_together`, `raster::tests::non_interpolated_alpha_images_keep_their_colours_and_mask_when_minified`, `raster::tests::non_interpolated_type3_glyph_images_are_not_smoothed_when_minified`, `raster::tests::image_resampling_nearest_accepts_wide_sources_the_filter_refuses`, `raster::tests::wide_non_interpolated_sources_draw_after_shrinking`; (11) `raster::tests::skipping_paths_off_the_canvas_changes_no_pixel`; (12) `raster::tests::preview_renders_skip_tiny_images_but_finals_draw_everything`; (13) `color_replacement_tests::{black_maps_to_text_colour, white_maps_to_background, midgrey_interpolates, alpha_copied_verbatim, only_black_threshold, line_art_false_skips_paths_but_not_text}`, `color_replacement_render_tests::{color_replacement_maps_text_and_line_art, color_replacement_maps_stencil_mask_ink}`; (14) `raster::tests::minified_image_edges_snap_to_whole_pixels`, `raster::tests::a_tile_snaps_half_pixel_image_edges_as_the_whole_page_does` (`cargo test -p hayro`), `raster::tests::a_turned_or_mirrored_minified_image_snaps_to_whole_pixels`, `rotated_image_snap::a_shrunk_image_on_a_turned_page_is_the_upright_render_turned` (`cargo test -p pdfcraft-render --test it`), `renderer::snap_tests::a_skew_that_moves_a_far_edge_is_not_snapped` (`cargo test -p hayro`); (15) `raster::tests::shrunk_soft_mask_larger_than_its_image_keeps_thin_strokes`; (16) `raster::tests::soft_mask_applies_to_a_shading_pattern_fill`; (17), (18) `raster::tests::reading_colors_keep_highlight_annotations_readable`; (19) `mesh_extent_tests::{a_mesh_paints_its_triangles_alone, a_mesh_keeps_the_graphics_state_alpha, a_mesh_off_the_canvas_paints_nothing}` (`cargo test -p hayro`); (20) `raster::tests::a_tiling_pattern_drawn_once_shows_its_cell`, `raster::tests::a_large_step_still_tiles_where_the_next_cell_shows`, `raster::tests::a_tiling_cell_too_thin_to_scale_keeps_the_old_path`; (21) `raster::tests::non_interpolated_minification_keeps_thin_strokes`, `raster::tests::non_integer_minification_averages_the_columns_each_pixel_covers`, `raster::tests::non_integer_minification_averages_colour_and_mask_together`, `raster::tests::non_interpolated_minification_does_not_bleed_transparent_colour`, `raster::tests::a_smaller_soft_mask_does_not_bleed_transparent_colour_when_minified`, `raster::tests::an_interpolated_smaller_soft_mask_stays_smooth_when_its_image_is_minified`, `raster::tests::minified_image_edges_snap_to_whole_pixels`, `raster::tests::an_image_shrunk_one_way_and_stretched_the_other_is_still_averaged`, `raster::tests::wide_non_interpolated_sources_draw_after_shrinking`, `renderer::area_shrink_tests` (`cargo test -p hayro`); (22) `raster::tests::knockout_groups_replace_what_their_elements_cover`; (23) `raster::tests::axial_and_radial_shadings_draw_as_native_gradients`; (25) `tiling_cell_placement::{far_cells_of_a_fill_sit_at_whole_steps, far_cells_sit_at_whole_steps_under_a_turned_scaled_matrix, far_cells_of_a_stroke_sit_at_whole_steps, a_cell_pixmap_spans_exactly_one_step}` (`cargo test -p pdfcraft-render --test it`); (28) `raster::tests::large_non_interpolated_images_are_smoothed_when_enlarged`, `raster::tests::small_non_interpolated_images_keep_hard_pixels_when_enlarged`, `raster::tests::a_scan_shrunk_one_way_is_smoothed_the_way_it_is_stretched` | | hayro-jbig2 | 0.3.0 | Apache-2.0 OR MIT | (1) at most 2^28 pixels per bitmap and page bitmap, and 2^25 cells per halftone grid and grayscale image (each side could be 65535: 4.3 billion pixels decoded one by one, and a 4-byte value per cell); (2) at most 2^26 symbol pixels per symbol dictionary (a fuzzed dictionary decoded for minutes); (3) a symbol dictionary's height classes are capped at one per new symbol and its export runs at `2 × symbols + 1`, each plus `EXTRA_EMPTY_STEPS` (1024): an empty height class or a zero-length export run doesn't advance, and past the end of its data the arithmetic decoder can decode them for ever (one changed byte hung the renderer); (4) a segment header's long-form referred-to segment count is capped at `u16::MAX` and a larger count is refused before its retention or referred-to bytes are read: the 29-bit count went straight to `Vec::with_capacity`, a reservation of about 2 GiB whose failure aborts the process (from storytold/craft-libs#176); (5) what a page's segments hold for later ones (intermediate regions, pattern and symbol dictionaries with their headers and retained contexts, and Huffman tables) is at most 64 MiB, two square bitmaps of 16,384 a side (`MAX_RETAINED_WORDS`), and a dictionary's new and exported symbols are counted against it as they are made: each was bounded on its own, but a few dozen bytes an intermediate region held 33 MB each, and 8 MiB symbols exported again and again killed the process, so a few kilobytes could exhaust memory; (6) a segment may refer to at most 2^20 symbols, checked before any are collected: references may repeat, and 65,535 of them to one dictionary, in a 78 KB stream, asked for 17 GB; (7) a pattern dictionary keeps at most 16 MiB of shifted copies, which only speed up drawing, and none that wouldn't fit what the page has left (`MAX_SHIFTED_WORDS`); (8) a custom Huffman table has at most 2^16 lines, and a stream with its globals at most 2^20 segments and 2^22 references, checked as they are read: both grew before anything counted them, and PDFs of a kilobyte or so reached 5 GB and more behind nested Flate; (9) a page past the bounds of (1) is refused when the stream is opened: a caller sizes its output from the header before the page bitmap is made, and an Indexed image in a 743-byte PDF asked for 137 GB, which aborts; (10) a stream may ask for at most 2^30 units of decoding work, four of the largest pages, about 6 s to 17 s (`MAX_WORK`): decoding a bitmap costs a pixel each and 4 a row, copying one a word each and 1 a row, and each integer an item reads costs 38, what one takes at worst, so one with no pixels pays too, before it is decoded: each bitmap was bounded, but nothing bounded how many there were, and a 795-byte PDF took 265 s; (11) `Image::decode_within` spends from what a run of images may still do, starting with its page's pixels, and `charge_reading` charges reading a stream and its globals before anything is parsed, so an operation over a document can bound what `MAX_WORK` bounds for one image, and images that fail or decode nothing add up too: a file of thousands of images made to reach that limit took as many times as long; (12) adding to a location, a reference offset, a symbol's corner or a Huffman range line near the end of an `i32`, and negating a decoded integer of 2^31, overflowed: a panic wherever overflow is checked. The refinement context, a text region's corners, the integer decoder and the Huffman table wrap, as a release build did, and `Bitmap::combine` saturates, which comes to the same, as a location that overflows is past the end of any page. What is drawn doesn't change | (1) `raster::tests::jbig2_regions_of_billions_of_pixels_are_refused`, `raster::tests::jbig2_halftone_grids_past_their_bound_are_refused`; (2) the nightly `cargo xtask fuzz` finding, which now finishes in about a second; (3) `raster::tests::jbig2_symbol_dictionaries_that_never_advance_terminate`; (4) `segment::tests::test_referred_to_count_is_capped` (`cargo test --manifest-path vendor/hayro-jbig2/Cargo.toml`); (5) `raster::tests::jbig2_regions_held_past_a_pages_bound_are_refused`, `jbig2_symbols_without_pixels_count_towards_a_pages_bound`, `jbig2_retained_contexts_count_towards_a_pages_bound`, `jbig2_huffman_tables_count_towards_a_pages_bound`, `jbig2_new_symbols_count_towards_a_pages_bound`, `jbig2_new_symbols_are_held_to_what_a_page_has_left`, `jbig2_pattern_dictionaries_count_towards_a_pages_bound`, and in the crate (`cargo test --manifest-path vendor/hayro-jbig2/Cargo.toml`) `decode::symbol::tests::exported_symbols_are_counted_before_they_are_copied`, `tests::the_room_left_shrinks_as_segments_are_stored`; (6) `raster::tests::jbig2_segments_referring_to_too_many_symbols_are_refused`; (7) `raster::tests::jbig2_pattern_dictionaries_too_large_to_shift_still_draw`, `jbig2_shifted_patterns_never_refuse_an_image`; (8) `raster::tests::jbig2_huffman_tables_of_too_many_lines_are_refused`, `jbig2_streams_of_too_many_segments_are_refused`, `jbig2_streams_of_too_many_references_are_refused`, and in the crate `tests::the_segments_of_a_stream_and_its_globals_count_together`; (9) `raster::tests::jbig2_pages_past_the_bounds_are_refused_before_anything_is_sized_from_them`, `tests::a_page_past_the_bounds_is_refused_when_opened`; (10) `raster::tests::jbig2_decoding_work_is_counted_item_by_item`; (11) `jbig2::decode::tests::a_run_of_images_spends_its_allowance_down_to_the_step` (pdfcraft-filters); (12) `raster::tests::jbig2_placements_at_the_end_of_an_i32_draw_nothing` | | lopdf | 0.45.0 | MIT | (1) PNG predictor rows longer than the data are refused before their buffers are allocated (a fuzzed `/Columns 4294967295` allocated two 4 GiB rows), and empty frames return without row buffers; (2) predictor row widths use saturating arithmetic or reject overflow; (3) sub-byte TIFF Predictor 2 scratch is bounded by the samples present in the input instead of the declared `/Colors`, aligned with [lopdf#589](https://github.com/J-F-Liu/lopdf/pull/589); (4) the object lookup keeps its cycle history through compressed objects, requires an ordinary in-file container, and caps the chain of indirect lookups at 16 (a chain of new objects, each a `/Length` compressed in a container whose own `/Length` names the next, or a stream whose `/Length` names the next stream, overflowed the stack: a dev build's 2 MiB thread within 800 hops); (5) object-stream members are parsed within their own extent (up to the next distinct offset), an entry whose offset repeats an earlier entry's is skipped (so a repeated object number can resolve to its earlier entry), and offsets are processed in index order; (6) `find_latest_trailer`: a rebuild takes a `/Type /XRef` stream dictionary as the trailer when the file has no `trailer` keyword (the latest in the file wins, as a later `trailer` would), so a concatenated or truncated file keeps its /Info and /ID instead of failing to load; a candidate whose /Root is the catalog beats one whose /Root is another dictionary, and an in-file /Root must match its generation (#2365) | (1) `inspect::tests::png_predictor_rows_longer_than_the_data_are_refused_before_allocating`, `inspect::tests::empty_png_predictor_frames_do_not_allocate_rows`; (2) `inspect::tests::tiff_subbyte_predictor_row_width_overflow_is_refused`; (3) `inspect::tests::tiff_subbyte_predictor_scratch_is_bounded_by_available_samples`, `inspect::tests::tiff_subbyte_predictors_keep_components_rows_and_padding`; (4) `inspect::tests::compressed_object_self_container_does_not_recurse`, `inspect::tests::compressed_object_mutual_containers_do_not_recurse`, `inspect::tests::object_stream_length_cycle_does_not_recurse`, `inspect::tests::compressed_object_container_chain_does_not_recurse`, `inspect::tests::compressed_length_chain_does_not_recurse`, `inspect::tests::plain_stream_length_chain_does_not_recurse`, `inspect::tests::compressed_objects_and_indirect_stream_length_still_load`, `inspect::tests::compressed_length_in_a_container_with_an_indirect_length_still_loads`; (5) `inspect::tests::object_stream_member_extents_bound_owned_strings`, `inspect::tests::object_stream_member_extents_preserve_index_order_and_whitespace`; (6) `inspect::tests::rebuilt_file_keeps_the_info_of_its_cross_reference_stream`, `inspect::tests::rebuilt_file_prefers_a_trailer_whose_root_is_the_catalog` | | hayro-syntax | 0.7.2 | Apache-2.0 OR MIT | (1) Page-tree cycle guard in `resolve_pages` (visited set + depth cap): a `/Kids` loop among compressed objects overflowed the stack and aborted the process; (2) `ccitt_size_ok`: CCITT images wider than 2^20 columns or over 2^28 pixels are not decoded (a fuzzed `/Columns 4294967295` allocated 4 GiB); (3) Flate/LZW predictor row widths are saturating `usize`, a zero width is refused and a row longer than the data is not allocated (a fuzzed `/Columns 9223372036854775807` wrapped to a 2^61-byte allocation and aborted); (4) the inline-image end-of-data check (`EI`) doesn't nest inside its own look-ahead and examines at most 256 candidates and steps (50 images with "EI" in their data ran for over ten minutes); an EI followed by a delimiter (EI%..., EI[...) ends the image like one followed by white space; only a regular character after EI skips the candidate; (5) a stream whose `/Length` is wrong is recovered only when `stream` is the next token — scanning ahead to a later object's `stream` made an indirect appearance-state dictionary (Quartz checkboxes: `/AP /N` → `<< /Yes … /Off … >>`) render as a non-form, so the widget was skipped; (6) JPX rescaling uses an exact integer sample range for 1–32-bit components (32 shifted past u32, and floating-point maxima at 25–31 bits wrapped to zero), and refuses zero-sized rows; (7) `decode_limit`: Flate (through flate2 and the fallback inflater), LZW, RunLength and ASCII85 never write past `MAX_DECODED_STREAM` (256 MiB), or, for the last filter of an image of known size, past twice its pixel bytes (a predictor tag byte per byte), at most 1 GiB; `Stream::decoded_within` for budgeted callers; a page's `/Contents` array shares one `MAX_DECODED_STREAM` (a 2 MB page whose content stream or 8 × 8 image inflated to 2 GB took the renderer past 3 GB in under a second; naming the stream twenty times in `/Contents`, 5 GB); (8) high-precision numbers (jsPDF MediaBox coordinates with over 19 digits) parse without overflowing the u64 mantissa: fractional digits past it are dropped, integer digits past it scale the value (a real, kept finite); (9) large dictionary indexes are built on first use and shared across repeated reads within each PDF (#307); the validated-byte cache is bounded, reader contexts remain separate, repairs and inline/content dictionaries bypass sharing, and small dictionaries stay eager; (10) decoded object-stream slots publish once under contention, cold loading rejects recursive dependencies, and xref repair serializes without holding recursive read guards; this permits one parser per native render pool with private worker caches; (11) `sync.rs`: a poisoned lock is recovered instead of unwrapped (a worker panic no longer cascades into every reader of the same PDF), and xref repair takes a blocking write lock instead of asserting a non-blocking one; (12) `SegmentList` (`util.rs`) returns `None` for an out-of-range index instead of panicking; (13) a page's `/UserUnit` scales `render_dimensions` and `initial_transform`, so the page and everything on it are sized in points, as Acrobat shows them; it is read as Acrobat reads it (the page's own entry, not inherited; a number from 1 up, at most 75,000; anything else counts as 1), and a page without it keeps exactly the transform it had; (14) revision 2–4 passwords are tried as given and then in PDFDocEncoding (`legacy_passwords` in `crypto/mod.rs`, which resolves the upstream `TODO: Convert to PDFDocEncoding`): passwords arrive as UTF-8 text, so a password with characters outside ASCII (`Çok güzel`, `ılık€`) written in PDFDocEncoding, as ISO 32000-2 §7.6.4.3.2 asks and qpdf and Acrobat write it, was always wrong; (15) page content is read a piece at a time (`page::ContentReader`, `Page::content_reader`): plain streams and Flate streams without a predictor are decoded as they are read, with a budget of 64 times their compressed size and never less than `MAX_DECODED_STREAM` (a stream that expands past it is cut and reported); other streams are decoded whole, as before, and share `MAX_DECODED_STREAM`. The content tokenizer reads a window at a time (`UntypedIter::new_window`, `TypedIter::needs_more`, `TypedIter::committed`): an instruction that runs past the window is not decided until more data follows, which `Reader` records by flagging reads and `at_end` past the end of the data, including through the clones the parser makes. The Flate decoder keeps its own 32 KiB history, so it pauses at any output size. `ContentReader::new` reuses raw/decrypted stream bytes for budget accounting and piece creation, and `decoded_raw_within` applies the existing Whole-stream filters and limits to those bytes without decrypting twice; (16) revision 5–6 (AES-256) passwords are tried in their SASLprep form (RFC 4013, then UTF-8, at most 127 bytes, as ISO 32000-2 §7.6.4.3.3 Algorithm 2.A asks and pdfcraft-crypt writes them) and then, if that differs, as given (`aes256_passwords` in `crypto/mod.rs`, which resolves the upstream `TODO: Do the above.`; `stringprep`, already in the tree through pdfcraft-crypt, is an optional dependency enabled by `std`, and without `std` passwords are used as given): the renderer used the password as typed, so one that SASLprep changes (a decomposed `café`, a no-break space, the fullwidth `ａｂ` a Japanese IME types in full-width mode) was wrong, although pdfcraft-cos accepted it: protecting a document with such an open password was rolled back ("the password is incorrect"), and a file with such a password did not open with it.; (17) CCITT `/Rows 0` means the height is not predetermined (ISO 32000-2 Table 11), not zero rows: scanned Group 4 pages written that way rendered blank; (18) The streaming Flate decoder (`lzw_flate::flate::fallback`) copies a back-reference in runs of up to its distance (`extend_from_within`), not byte by byte. The copy reads only bytes already in the output, and a copy whose source does not exist still writes nothing, so the bytes are those a byte-by-byte copy writes. Inflating the A1 sample's content stream (278.6 MB) takes about 14% less time (0.66 s to 0.56 s); the output is identical; (19) CCITT `/Rows` below the image's `/Height` decodes to the height: a plotter driver wrote `/Rows` equal to `/Columns`, so only the first rows were decoded and the rest drawn black (#3308) (20) `stream` followed by spaces or tabs and then the line break (`stream \r\n`, as Microsoft Reporting Services writes it) starts the data after the line break; the content stream was taken to start at the space and the page rendered blank.; (21) a JP2 colour specification naming an enumerated colour space the decoder refuses (CIEJab in a veraPDF fixture, YCbCr, PhotoYCC, YPbPr, bilevel…) no longer fails the image: it is decoded again from a copy whose `colr` names the space the decoder assumes without one, greyscale, sRGB or CMYK by component count, as MuPDF and Poppler draw it (`without_refused_enumerated_colour`; the page drew blank, #3306); (22) an object stream's index of `objnum offset` pairs is parsed once per stream and kept beside its decoded data (`ObjectStreamIndexes`, `XRef::with_object_stream_indexes`), and an offset past `usize` is malformed instead of wrapping: `XRef::get_with` parsed the whole index again for every object it fetched from the stream, so reading all of a stream of N objects took N² steps (a Firefox page whose 65,542 objects share one stream took 37 s to draw and 40 s to extract its text); (23) the inline-image end-of-data look-ahead no longer rejects a candidate `EI` because a byte above 0x7F lies between it and the next one: when the candidate is the end, those bytes are the content stream that follows the image, and its strings carry such bytes (WinAnsi `(Grüße) Tj`, the two-byte codes of any Type 0 font), so the operator check that follows decides alone, bounded as in (4) (a scanned page whose OCR text held an accented word and a later capital "EI" — EINGANG, BEI, KLEIN — drew and extracted nothing after the image: 40 of 275 pages across five real documents); (24) `xref::prev_offset`: a trailer `/Prev` written as a reference `N G R` is read as the offset that object holds when the section's own entries place it (iText's `indirectPrev.pdf`; each section's entries are read before its `Prev` is followed), and otherwise as offset N (tiff2pdf; JHOVE's PDF-HUL-157 files, #2366), as `pdfcraft-cos` and other readers take it, so the earlier sections' pages are seen instead of one; (25) direct arrays and dictionaries share a reader-local nesting cap of 128, counted while skipping and while parsing a dictionary (its lenient repair reads, rather than skips, an object found where a key belongs, so `<< << …` recursed through that read); deeper objects return `None` instead of overflowing the stack (including unused trailer values, trailer entries, content operands and marked-content properties); a chain of 65 direct stitching functions now stops at this cap, before the 64-function guard; (26) an xref stream subsection must be backed by its decoded data: all-zero `/W` widths (an entry reads no bytes) and a declared count longer than the data are refused before any entry is inserted, where a tiny stream with a large `/Size` or `/Index` count inserted that many entries (from storytold/craft-libs#239); a subsection running past object number `i32::MAX` is refused instead of overflowing or turning negative; (27) an xref stream's type field is read big-endian at its full `/W[0]` width, where only its first byte was read and wider fields shifted every later entry, and `/W[0]` over 8 bytes is refused like `/W[1]` (from storytold/craft-libs#241); (28) a content instruction keeps up to 256 operands (`MAX_OPERANDS`) instead of 10, and past that drops the oldest instead of ending the content stream: the eleventh operand of a twelve-ink DeviceN `scn` (ISO 32000-2 §8.6.6.5 allows any number of components) blanked the rest of the page (#4260); (29) a `/JBIG2Globals` stream that is itself JBIG2 isn't decoded: globals are raw segments, and a chain of 16,000 such streams overflowed the stack, which aborts the process; (31) `set_page_jbig2_work`: the JBIG2 images decoded on a thread spend one allowance, set per page, 2^30 steps (`MAX_PAGE_JBIG2_WORK`), so a page of many images made to reach their own limit stops at one, and what they cost to read counts, whether they decode or not; unbounded until a page sets it; (32) a group of array items read together (a tuple: an image's `/Decode`, `/W`, `/W2`, and a function's `/Domain`, `/Range`, `/Encode` and `/Decode`) can start with a reference: it was taken for the whole group and lost, which ended the array, so an image with `/Decode [7 0 R 0]` didn't draw and CID widths fell back to `/DW`; (33) an object that holds only a reference stands for what that names, stored on its own or in an object stream, up to 32 references: read as a number it was the reference's own object number, so a `/Decode` or `/W` entry naming such an object took a wrong value, and as anything else it was lost. Past 32, as any cycle goes, it reads as a missing object. An object stream isn't reached that way, as its checks are of the object named; reading its index counts against the 64 nested loads and reads its dictionary alone, as a chain of them, each with a member of the next for its `/N`, `/First` or `/Length`, ran out of stack; and a comment inside a reference counts as white space; (34) a table rebuilt by scanning the file (`fallback_xref_map`) finds an object by its number when its header names another generation, including one no generation may be (`3 18446744073709551616 obj` in `poppler-85140-0.pdf`, #5580): the header is read leniently only in that scan (`read_object_header`), the object is also filed under generation 0 unless a real `N 0 obj` holds that key (a later header for the number moves the stand-in), `XRef::get_with` matches by object number only in a repaired table, and `XRef::objects` yields each location once | (1) `cos/tests/writer.rs::page_tree_loops_in_object_streams_do_not_crash_readers`; (2) `raster::tests::absurd_mask_and_fax_sizes_are_refused_before_decoding`; (3) `raster::tests::absurd_predictor_columns_and_line_widths_render`; (4) `raster::tests::inline_images_with_ei_in_their_data_parse_in_linear_time`, `content::tests::inline_image_ei_may_be_followed_by_a_delimiter`; (5) `raster::tests::indirect_appearance_state_dict_is_not_another_stream`; (6) `raster::tests::jpx_stream_bit_depth_boundaries_do_not_panic`, `raster::tests::jpx_32_bit_image_does_not_prevent_other_page_content_from_rendering`, `raster::tests::jpx_gray_rgb_and_alpha_keep_eight_bit_samples`, `raster::tests::jpx_streams_reject_out_of_range_bit_depths`; (7) `raster::tests::decompression_bombs_stop_at_the_stream_limit`, `raster::tests::page_contents_and_nested_forms_share_one_budget`; (8) `inspect::tests::jspdf_high_precision_mediabox_does_not_overflow_dimensions`, `object::number::tests::high_precision_float_does_not_overflow_mantissa` (hayro-syntax); (9) `object::dict::lazy_index_tests`, `raster::tests::large_resource_indexes_preserve_pixels_and_text`; (10) `data::tests`, `xref::tests::concurrent_cold_object_stream_reads_all_succeed`, `raster::tests::shared_parser_is_lazy_and_initializes_once_under_contention`; (11) `data::tests::panicking_initializer_wakes_waiters_and_prevents_retries`, `xref::tests::concurrent_cold_object_stream_reads_all_succeed`; (12) `util::tests::segment_list_rejects_out_of_range_indices`; (13) `raster::tests::user_unit_scales_the_page_and_everything_on_it`; (14) `crypto::pdfcraft_password_tests`, `pdfcraft-engine` `tests::legacy_passwords_with_turkish_characters_open`; (15) `content::tests::windows_keep_every_instruction_at_every_split`, `content::tests::windows_keep_every_instruction_in_pieces_of_any_size`, `content::tests::windows_match_the_slice_on_random_content`, `content::tests::windows_drop_blanks_and_keep_a_comment_that_may_go_on`, `page::content_reader_tests::content_past_the_cap_is_read_to_the_end`, `page::content_reader_tests::a_stream_that_expands_past_the_budget_is_cut_there`, `page::content_reader_tests::a_small_expansion_under_the_cap_is_not_cut`, `raster::tests::page_content_past_the_old_cap_renders`, `raster::tests::content_reader_returns_the_same_bytes_in_any_pieces`, `raster::tests::content_bombs_stop_at_the_budget_and_say_so`, `page::content_reader_tests::plain_and_whole_pieces_keep_their_bytes`, `page::content_reader_tests::repeated_content_reference_counts_compressed_bytes_once`, `page::content_reader_tests::whole_cap_still_counts_later_unique_streams`; (16) `crypto::pdfcraft_password_tests::aes256_passwords_are_tried_in_their_saslprep_form_then_as_given`, `crypto::pdfcraft_password_tests::aes256_passwords_open_as_typed_and_in_their_saslprep_form`, `pdfcraft-engine` `tests::aes256_open_passwords_that_saslprep_changes_protect_and_reopen_as_typed`, `tests::aes256_files_whose_passwords_saslprep_changes_open_as_typed`; (17) `raster::tests::group4_fax_with_rows_zero_decodes_to_the_image_height`; (18) `filter::lzw_flate::tests::back_reference_longer_than_its_distance_repeats_its_source`, `back_reference_cut_by_the_limit_resumes_in_the_next_call` and `back_reference_reads_bytes_written_by_an_earlier_call`; each fails when the copy returns the wrong remaining count; (19) `raster::tests::group4_fax_with_rows_below_its_height_decodes_to_the_image_height`; (20) `raster::tests::space_before_the_line_break_after_stream_still_renders`; (21) `raster::tests::a_jp2_image_in_a_refused_enumerated_colour_space_is_drawn`, `filter::jpx::tests::refused_enumerated_colour_spaces_become_the_component_counts_default`, `filter::jpx::tests::what_the_decoder_handles_or_cant_be_read_is_left_alone` (`cargo test --manifest-path vendor/hayro-syntax/Cargo.toml`); (22) `raster::tests::objects_in_one_large_object_stream_are_read_in_linear_time`; (23) `text::tests::text_after_an_inline_image_survives_a_non_ascii_byte_before_a_later_ei`, `content::tests::an_inline_image_ends_at_its_own_ei_when_later_text_holds_a_non_ascii_byte` (hayro-syntax); (24) `xref::tests::prev_as_reference_reaches_the_earlier_section`, `xref::tests::prev_as_reference_to_an_object_holding_the_offset_reads_that_offset`, `pdfcraft-cos` `document::tests::a_prev_written_as_a_reference_still_reaches_the_earlier_sections`, `document::tests::a_prev_reference_to_an_object_holding_the_offset_reads_that_offset`; (25) `raster::tests::deeply_nested_direct_objects_do_not_overflow_the_stack`, `raster::tests::dictionaries_nested_without_keys_in_content_do_not_overflow_the_stack`, `raster::tests::dictionaries_nested_without_keys_in_the_trailer_do_not_overflow_the_stack`, `raster::tests::dictionaries_nested_without_keys_as_marked_content_properties_do_not_overflow_the_stack`, `raster::tests::dictionaries_nested_without_keys_are_read_up_to_the_nesting_cap`, `raster::tests::nesting_at_the_cap_fits_a_512_kib_stack` (native only), `raster::tests::function_limits_depth_guard_refuses_a_chain_of_references`; (26) `xref::tests::xref_stream_entries_need_backing_bytes`; (27) `xref::tests::xref_stream_type_field_uses_the_full_w0_width`, `xref::tests::xref_stream_loader_rejects_oversized_w0` (`cargo test --manifest-path vendor/hayro-syntax/Cargo.toml`); (28) `content::ops::tests::scn_with_twelve_components_keeps_them_all`, `content::ops::tests::operands_past_the_cap_keep_the_latest_in_order` (hayro-syntax), `raster::tests::devicen_inks::a_twelve_ink_devicen_colour_paints`, `raster::tests::devicen_inks::a_four_ink_devicen_colour_paints`, `raster::tests::devicen_inks::an_operator_with_hundreds_of_operands_does_not_end_the_content`; (29) `raster::tests::jbig2_globals_that_are_jbig2_themselves_are_not_decoded`, `raster::tests::jbig2_globals_are_read`, `raster::tests::the_jbig2_images_of_a_page_share_one_allowance`, `raster::tests::an_image_that_fails_spends_the_pages_allowance_too`; (32) `raster::tests::a_decode_array_with_references_in_it_is_read`, `raster::tests::a_functions_arrays_can_start_with_a_reference` (pdfcraft-render), `font::cid::width_tests::a_width_group_can_start_with_a_reference` (hayro-interpret); (33) `raster::tests::a_reference_to_a_reference_is_what_it_names`, `raster::tests::a_decode_array_with_references_in_it_is_read` (pdfcraft-render), `font::cid::width_tests::a_width_group_can_start_with_a_reference` (hayro-interpret), `xref::tests::an_object_stream_isnt_reached_through_a_reference`, `xref::tests::a_chain_of_object_streams_ends_at_the_limit_of_nested_loads`, `xref::tests::a_chain_of_object_streams_takes_no_more_work_for_two_ways_on`, `xref::tests::a_reference_held_in_either_storage_is_followed`, `xref::tests::a_page_a_reference_is_held_to_is_found_once` (hayro-syntax); (34) `raster::tests::a_rebuilt_table_finds_an_object_whose_header_names_another_generation`, `raster::tests::a_rebuilt_tables_generation_0_stand_in_is_not_a_second_object` (pdfcraft-render), `xref::tests::an_object_stream_and_its_first_member_are_both_enumerated` (hayro-syntax) | diff --git a/vendor/hayro/src/renderer.rs b/vendor/hayro/src/renderer.rs index 120a751281..c3d1d18520 100644 --- a/vendor/hayro/src/renderer.rs +++ b/vendor/hayro/src/renderer.rs @@ -193,27 +193,67 @@ fn image_buffer(len: usize) -> Option> { Some(out) } -/// PdfCraft patch: `mask` sampled onto the grid of a `width` × `height` image, each image sample -/// taking the mask sample under its centre. `None` for an empty side or a buffer over the limits. +/// PdfCraft patch: `mask` sampled onto the grid of a `width` × `height` image (no side smaller +/// than the mask's). Each image sample takes the mask sample under its centre or, for a mask that +/// asks for `/Interpolate`, the bilinear blend of the four around it, so the mask's edges stay as +/// smooth as when it was enlarged on its own. `None` for an empty side, an image side smaller than +/// the mask's, or a buffer over the limits. fn mask_on_image_grid(mask: &LumaData, width: u32, height: u32) -> Option> { if width == 0 || height == 0 || mask.width == 0 || mask.height == 0 { return None; } let mut out = image_buffer(source_byte_len(width, height, 1, MAX_IMAGE_PIXELS)?)?; - let mut columns = Vec::new(); - columns.try_reserve_exact(width as usize).ok()?; - columns.extend((0..width).map(|x| nearest_source_index(mask.width, width, x))); + let columns = grid_axis(mask.width, width, mask.interpolate)?; + let rows = grid_axis(mask.height, height, mask.interpolate)?; let mask_width = mask.width as usize; - for y in 0..height { - let start = nearest_source_index(mask.height, height, y).checked_mul(mask_width)?; - let row = mask.data.get(start..start.checked_add(mask_width)?)?; - for &x in &columns { - out.push(*row.get(x)?); + let row = |y: usize| { + let start = y.checked_mul(mask_width)?; + mask.data.get(start..start.checked_add(mask_width)?) + }; + let one = u64::from(GRID_ONE); + for &(y0, y1, fy) in &rows { + let (top, bottom, fy) = (row(y0)?, row(y1)?, u64::from(fy)); + for &(x0, x1, fx) in &columns { + let fx = u64::from(fx); + let blend = |r: &[u8]| Some(u64::from(*r.get(x0)?) * (one - fx) + u64::from(*r.get(x1)?) * fx); + // At most 255 × GRID_ONE², so the sum fits in a u64 and the result in a u8. + let v = (blend(top)? * (one - fy) + blend(bottom)? * fy + one * one / 2) / (one * one); + out.push(v.min(255) as u8); } } Some(out) } +/// PdfCraft patch: fixed-point unit of [`grid_axis`]'s blend weights. +const GRID_ONE: u32 = 1 << 16; + +/// PdfCraft patch: for each of `dst` samples (`dst >= src`) taken from `src` along one axis, the +/// two source samples it lies between and the weight of the second, in [`GRID_ONE`] units. +/// Centres line up: target `t` lies at `(t + ½)·src/dst − ½` source samples, held to the ends. +/// Without `smooth` both are the nearest sample, so the weight does not matter. +fn grid_axis(src: u32, dst: u32, smooth: bool) -> Option> { + if src == 0 || dst < src { + return None; + } + let mut out = Vec::new(); + out.try_reserve_exact(dst as usize).ok()?; + let last = u64::from(src - 1) << 16; + for t in 0..dst { + if !smooth { + let s = nearest_source_index(src, dst, t); + out.push((s, s, 0)); + continue; + } + // In u128, as `(2t + 1)·src·2^16` can pass 2^64 for a long side; the quotient is at + // most `src·2^16`, so it fits back in a u64. + let centre = ((2 * u128::from(t) + 1) * u128::from(src) << 16) / (2 * u128::from(dst)); + let at = u64::try_from(centre).ok()?.saturating_sub(u64::from(GRID_ONE / 2)).min(last); + let s = (at >> 16) as usize; + out.push((s, (s + 1).min(src as usize - 1), (at & 0xFFFF) as u32)); + } + Some(out) +} + /// PdfCraft patch (#624): the source index that PicScale's nearest-neighbour plan copies for /// destination `index` of `dst` samples taken from `src`. It is the same fixed-point centre /// sampling as `ResampleNearestPlan`, so these are the samples a nearest resize would copy. @@ -2908,7 +2948,18 @@ mod area_shrink_tests { assert_eq!(mask_on_image_grid(&mask(&[10, 20], 2, 1), 4, 2), Some(vec![10, 10, 20, 20, 10, 10, 20, 20])); // Centres of 7 samples over 3: 0.5/7·3 = 0.21 → 0, 1.5/7·3 = 0.64 → 0, 2.5/7·3 = 1.07 → 1, … assert_eq!(mask_on_image_grid(&mask(&[1, 2, 3], 3, 1), 7, 1), Some(vec![1, 1, 2, 2, 2, 3, 3])); - assert_eq!(mask_on_image_grid(&mask(&[1, 2, 3, 4], 1, 4), 1, 2), Some(vec![2, 4])); + } + + /// A mask that asks for `/Interpolate` is blended between the samples around each centre, + /// and held at its ends: the centres of 4 samples over 2 lie at -¼, ¼, ¾ and 1¼. + #[test] + fn an_interpolated_mask_is_blended_onto_the_image_grid() { + let smooth = |data: &[u8], width, height| LumaData { interpolate: true, ..mask(data, width, height) }; + assert_eq!(mask_on_image_grid(&smooth(&[0, 255], 2, 1), 4, 2), Some(vec![0, 64, 191, 255, 0, 64, 191, 255])); + // Centres of 7 samples over 3 lie 3/7 apart, from -2/7 to 2 2/7: a ramp, flat at the ends. + assert_eq!(mask_on_image_grid(&smooth(&[0, 70, 140], 3, 1), 7, 1), Some(vec![0, 10, 40, 70, 100, 130, 140])); + assert_eq!(mask_on_image_grid(&smooth(&[0, 255], 1, 2), 1, 4), Some(vec![0, 64, 191, 255])); + assert_eq!(mask_on_image_grid(&smooth(&[7, 9], 2, 1), 2, 1), Some(vec![7, 9]), "same size is unchanged"); } #[test] @@ -2916,6 +2967,7 @@ mod area_shrink_tests { assert_eq!(mask_on_image_grid(&mask(&[], 0, 1), 4, 1), None, "an empty mask"); assert_eq!(mask_on_image_grid(&mask(&[1, 2], 2, 1), 0, 1), None, "an empty image"); assert_eq!(mask_on_image_grid(&mask(&[1], 2, 1), 4, 1), None, "mask data shorter than its size"); + assert_eq!(mask_on_image_grid(&mask(&[1, 2, 3, 4], 1, 4), 1, 2), None, "a mask larger than the image"); } /// Every target sample's weights add up to exactly one, and together the targets take every From c5065554a1946e4ca8895048d70095f68a4ecec5 Mon Sep 17 00:00:00 2001 From: kronbii <145912020+Kronbii@users.noreply.github.com> Date: Fri, 30 Oct 2026 05:07:44 +0200 Subject: [PATCH 34/53] M2: shrink masked images a row at a time; keep near-1:1 scans sharp Review of #3858 (aiongg): - An area-shrunk image with a soft mask, a stencil's ink included, built its full-resolution RGBA before shrinking. A 10718 x 13701 stencil at page size peaked at 1.16 GB (587 MB of RGBA). It is now premultiplied a source row at a time as it is averaged, and the shrink works a target row at a time, so it peaks at 1.03 GB, as the nearest-neighbour path did. Same bytes; about 5% slower per shrink. - An image whose f32 scale passed 1 only by rounding was smoothed bilinearly and blurred by half a pixel at a half-pixel offset. It is now smoothed only above 1.01. - A shrunk /ImageMask stencil keeping its thin strokes (#4800) is now tested. Co-Authored-By: Claude Opus 5.5 --- crates/render/src/raster/tests.rs | 53 +++++++ parity/acrobat-features.toml | 4 +- vendor/README.md | 2 +- vendor/hayro/src/renderer.rs | 255 ++++++++++++++++++++++++------ 4 files changed, 267 insertions(+), 47 deletions(-) diff --git a/crates/render/src/raster/tests.rs b/crates/render/src/raster/tests.rs index f6c0530164..5692634c30 100644 --- a/crates/render/src/raster/tests.rs +++ b/crates/render/src/raster/tests.rs @@ -6814,6 +6814,59 @@ fn shrunk_soft_mask_larger_than_its_image_keeps_thin_strokes() { assert_eq!(grays[3], 255, "{grays:?}"); } +/// Issue #4800: scanners save the text of a page as a 1-bit `/ImageMask` stencil, painted in the +/// fill colour, and a stencil shrunk by nearest-neighbour lost the strokes thinner than the shrink +/// factor. Here a one-column red stroke of an 8-column stencil is drawn 4 pixels wide: nearest +/// reads columns 1, 3, 5 and 7 and misses it, while the area average makes it a half-covered red. +#[test] +fn a_shrunk_stencil_mask_keeps_thin_strokes() { + // Sample 0 paints (the default `/Decode [0 1]`), so 0xDF is a stroke in column 2. + let content = "q 1 0 0 rg 4 0 0 1 0 0 cm /Im0 Do Q"; + let pdf = format!( + "%PDF-1.4\n1 0 obj << /Type /Catalog /Pages 2 0 R >> endobj\n\ + 2 0 obj << /Type /Pages /Kids [3 0 R] /Count 1 >> endobj\n\ + 3 0 obj << /Type /Page /Parent 2 0 R /MediaBox [0 0 4 1] /Contents 4 0 R /Resources << /XObject << /Im0 5 0 R >> >> >> endobj\n\ + 4 0 obj << /Length {} >> stream\n{content}\nendstream endobj\n\ + 5 0 obj << /Type /XObject /Subtype /Image /Width 8 /Height 1 /ImageMask true /Filter /ASCIIHexDecode /Length 3 >> stream\nDF>\nendstream endobj\n\ + trailer << /Root 1 0 R >>\n%%EOF\n", + content.len() + ); + let mut renderer = PageRenderer::new(Arc::new(pdf.into_bytes()), RenderConfig::default()); + let page = renderer.render(RenderRequest { page: 0, kind: RequestKind::Pixels, scale: 1.0, ..Default::default() }); + assert!(page.error.is_none(), "{:?}", page.error); + assert_eq!((page.width, page.height), (4, 1)); + let px = page.rgba.as_chunks::<4>().0.to_vec(); + let [r, g, b, _] = px.get(1).copied().unwrap_or_default(); + assert!(r >= 254 && g.abs_diff(128) <= 1 && b == g, "half-covered red over white, got {px:?}"); + assert_eq!(px.get(3), Some(&[255, 255, 255, 255]), "{px:?}"); +} + +/// A scan drawn at its own resolution keeps its hard pixels when its scale passes 1 only by +/// rounding: `x_scale` is an `f32`, and a 192 dpi scan on a 2× display at 100% can come out a +/// few ulps above 1. Smoothed bilinearly at a half-pixel offset, one-pixel stripes became gray. +/// Here `600.0001` (an `f32` of 600.00012) draws 600 samples 2·10⁻⁷ wider than 1:1. +#[test] +fn a_scan_drawn_a_rounding_error_above_its_resolution_stays_sharp() { + let content = "q 600.0001 0 0 600 0.5 0 cm /Im0 Do Q"; + let image: String = (0..600 * 600).map(|i| if i % 2 == 0 { "00" } else { "FF" }).collect(); + let pdf = format!( + "%PDF-1.4\n1 0 obj << /Type /Catalog /Pages 2 0 R >> endobj\n\ + 2 0 obj << /Type /Pages /Kids [3 0 R] /Count 1 >> endobj\n\ + 3 0 obj << /Type /Page /Parent 2 0 R /MediaBox [0 0 601 600] /Contents 4 0 R /Resources << /XObject << /Im0 5 0 R >> >> >> endobj\n\ + 4 0 obj << /Length {} >> stream\n{content}\nendstream endobj\n\ + 5 0 obj << /Type /XObject /Subtype /Image /Width 600 /Height 600 /ColorSpace /DeviceGray /BitsPerComponent 8 /Filter /ASCIIHexDecode /Length {} >> stream\n{image}>\nendstream endobj\n\ + trailer << /Root 1 0 R >>\n%%EOF\n", + content.len(), + image.len() + 1 + ); + let mut renderer = PageRenderer::new(Arc::new(pdf.into_bytes()), RenderConfig::default()); + let page = renderer.render(RenderRequest { page: 0, kind: RequestKind::Pixels, scale: 1.0, ..Default::default() }); + assert!(page.error.is_none(), "{:?}", page.error); + let row: Vec = page.rgba.as_chunks::<4>().0.iter().skip(300 * page.width as usize).take(page.width as usize).map(|[r, ..]| *r).collect(); + let gray = row.iter().skip(10).take(580).filter(|v| (16..240).contains(*v)).count(); + assert_eq!(gray, 0, "one-pixel stripes were blurred: {:?}", row.get(10..30)); +} + /// Type 3 glyph images used to be sampled bicubically (`ImageQuality::High`) whatever their /// `/Interpolate`, which blends the nearest pre-resize back in. The glyph is 4.6 × 2.3 pixels /// and its image pre-resizes to 5 × 3, so the final draw samples fractionally and bicubic diff --git a/parity/acrobat-features.toml b/parity/acrobat-features.toml index 367253b4a7..a0119d06e2 100644 --- a/parity/acrobat-features.toml +++ b/parity/acrobat-features.toml @@ -734,8 +734,8 @@ milestone = "M2" status = "partial" commands = [] tools = ["page_render"] -evidence = ["crates/render/src/raster/tests.rs::renders_a_blue_rectangle", "crates/automation/tests/it/automation.rs::open_inspect_render_and_find", "crates/render/src/raster/tests.rs::user_unit_scales_the_page_and_everything_on_it", "crates/render/src/raster/tests.rs::non_interpolated_minification_keeps_thin_strokes", "crates/render/src/raster/tests.rs::minified_image_edges_snap_to_whole_pixels", "crates/render/src/raster/tests.rs::large_non_interpolated_images_are_smoothed_when_enlarged", "crates/render/src/raster/tests.rs::small_non_interpolated_images_keep_hard_pixels_when_enlarged"] -notes = "hayro bootstrap renderer; own DisplayList devices in M2.6. Pages with /UserUnit are shown at their size in points, as Acrobat reads the value. An image drawn below its resolution is shrunk once to whole device pixels: by area averaging unless it asks for /Interpolate (then Catmull-Rom), so the text of a 300 dpi scan at 100% keeps its strokes (it was nearest-neighbour, which dropped them). Enlarged, an image of page-scan size (512 samples or more on its shorter side) is smoothed bilinearly whatever its /Interpolate, so a zoomed scan has no staircases; smaller images keep their hard pixels. Not measured against Acrobat" +evidence = ["crates/render/src/raster/tests.rs::renders_a_blue_rectangle", "crates/automation/tests/it/automation.rs::open_inspect_render_and_find", "crates/render/src/raster/tests.rs::user_unit_scales_the_page_and_everything_on_it", "crates/render/src/raster/tests.rs::non_interpolated_minification_keeps_thin_strokes", "crates/render/src/raster/tests.rs::a_shrunk_stencil_mask_keeps_thin_strokes", "crates/render/src/raster/tests.rs::minified_image_edges_snap_to_whole_pixels", "crates/render/src/raster/tests.rs::large_non_interpolated_images_are_smoothed_when_enlarged", "crates/render/src/raster/tests.rs::small_non_interpolated_images_keep_hard_pixels_when_enlarged"] +notes = "hayro bootstrap renderer; own DisplayList devices in M2.6. Pages with /UserUnit are shown at their size in points, as Acrobat reads the value. An image drawn below its resolution is shrunk once to whole device pixels: by area averaging unless it asks for /Interpolate (then Catmull-Rom), so the text of a 300 dpi scan at 100% keeps its strokes, a 1-bit /ImageMask text layer included (it was nearest-neighbour, which dropped them). Enlarged by more than 1%, an image of page-scan size (512 samples or more on its shorter side) is smoothed bilinearly whatever its /Interpolate, so a zoomed scan has no staircases; smaller images keep their hard pixels. Not measured against Acrobat" [[feature]] id = "view.render-shadings-patterns" diff --git a/vendor/README.md b/vendor/README.md index 0d90ad397e..28cc7ce9e7 100644 --- a/vendor/README.md +++ b/vendor/README.md @@ -40,7 +40,7 @@ Other resource categories retain their existing parent-chain lookup. | Crate | Version | Licence | Patches | Test | |---|---|---|---|---| | hayro-interpret | 0.7.0 | Apache-2.0 OR MIT | (1) select `/AP /N` appearance states via `/AS` (checkboxes, radios, stateful icons) and honour the NoView flag; (2) Type3 Unicode fallback to `/Encoding` glyph names, then printable ASCII; (3) `Type3Glyph::advance_width`; (4) `InterpreterSettings::ocg_overrides` for viewer layer toggles; (5) soft masks whose group lacks `/CS` were dropped (Chrome gradient text); (6) paint nesting cap `MAX_PAINT_NESTING` for tiling patterns and Type 3 glyphs (self-referencing content overflowed the stack); (7) CID width ranges in `/W` and `/W2` clamped to `MAX_CID` (a `0 4294967295 w` range hung); (8) `InterpreterSettings::hide_comments` skips markup annotations but keeps widgets and links (View ▸ Hide all comments); (9) `image_size_ok`: images and masks over 2^28 pixels are not decoded (a fuzzed `/W 4294967295` stencil mask allocated 4 GiB); (10) `MAX_NESTED_PAINTS`: nested paints (forms, Type 3 glyphs and tiling patterns inside one another) share a per-page budget of 50,000 (a glyph showing eight glyphs of its own font, sixteen deep, hung); (11) depends on skrifa 0.47 instead of 0.42 (upstream read-fonts fixed a Type 1 tokenizer that recursed through runs of numbers: a fuzzed font hung or overflowed the stack), with the one API change (`Charset::glyph_id` returns `Option`); (12) annotation appearances whose `/BBox` or `/Matrix` collapses to a line or a point are skipped (mapping them onto `/Rect` divided by zero and handed the device infinite or NaN matrices); (13) a Highlight annotation without a usable normal appearance is drawn from its `/QuadPoints`, `/C` and `/CA` with Multiply blending, as pdfcraft-annot draws highlights (it was not drawn at all); (14) simple-font `/Widths` (`read_widths`): entries only for codes 0–255 (one entry per code below `/FirstChar` was allocated: `4294967295` asked for 34 GB), an entry that isn't a number leaves its code missing instead of ending the array, and a `/LastChar` below `/FirstChar` gives no widths instead of discarding the font for a standard one; (15) the appearance box is mapped onto `/Rect` corner to corner: the translation now accounts for the scale (an appearance whose box didn't start at the origin and differed in size from `/Rect` was drawn shifted); (16) a clip set by `W`/`W*` takes effect after whichever path-painting operator ends the path (`S`, `f`, `B`, … as well as `n`; ISO 32000-2 §8.5.4): it was applied only on `n`, so `re W* S` clipped nothing and the clip stayed pending for a later, unrelated `n`; (17) `InterpreterSettings::cancelled`: once set, interpretation stops before the next content operator (a render pool sets it for every worker when dropped, after an edit or on closing, and for one worker when the watchdog gives up on its render, so renders whose answers nobody can receive stop instead of running to the end); (18) `MAX_PAGE_CONTENT`: the content a page holds at once (its contents, and the forms, tiling patterns and Type 3 glyph procedures being painted) stays within 512 MiB (`decode_content`; once it runs out, `InterpreterWarning::ContentTruncated` is reported (once per page by pdfcraft-render) so a truncated page is not a clean render; each level gives its bytes back when it finishes): forms nest fifty deep and each level kept its content while the next ran, so a form painting itself around a stream inflating to 256 MiB would have held over 12 GB; and Type 0 sampled functions take at most `MAX_SAMPLE_ENTRIES` (2^21) table points and `MAX_SAMPLE_VALUES` (2^23) samples, read no further than `/Size` asks (256 MiB of 1-bit samples asked for over 8 GB); (19) mesh shadings (types 4–7) are sampled only inside the area being drawn (`ShadingPattern::encode_within`, which the vendored hayro renderer calls with that area plus a pixel), within 8 visits per pixel of it plus 4 per triangle (`TriangleSampler`), and patches are triangulated one at a time: points far off the page were sampled over their whole bounding box, up to 65536 × 65536 entries (a fuzzed 147-byte patch stream; a synthetic page passed 6 GB in six seconds), ten thousand overlapping triangles ran for minutes, and every patch became 722 triangles before any limit; meshes hold at most `MAX_MESH_PATCHES` (2^16) patches and `MAX_MESH_TRIANGLES` (2^20) triangles, a lattice needs two to 2^19 vertices per row (`/VerticesPerRow 0` pushed empty rows for ever; one huge row read the whole stream), and a page samples at most `MAX_PAGE_MESH_TRIANGLES` (2^25) mesh triangles in all (a mesh painted a thousand times was triangulated a thousand times); each cut logs a warning; (20) PDF function construction is limited to 64 levels and 10,000 total nodes with cycle detection; calculator parsing and evaluation share procedure depth/work budgets, integer division/bitshift/rotation/index are panic-free, and fixed-size stack errors propagate; (21) a Type 0 font drawn with a substitute that isn't CID-keyed (it isn't embedded, or its program doesn't load) draws `.notdef` for a CID of an Adobe character collection (Japan1, GB1, CNS1, Korea1) that it can't find by Unicode: hayro took the CID for a glyph index of the substitute, which numbers its glyphs its own way, so the character was drawn as an unrelated one (𠮷 as "Ｑ" in BIZ UDPGothic); (22) a Type 0 font drawn with a substitute finds a character given with a variation selector (`/ToUnicode`, or Adobe-Japan1-UCS2, which maps the JIS X 0208 forms of the kanji JIS X 0213:2004 redrew that way: CID 1247 is 噂 U+E0100) through the substitute's variation sequences, else as the character alone: such a CID found no glyph and was drawn as an unrelated one; (23) backport of hayro 0.8's `gradient` module: `EncodedShadingPattern::as_svg_gradient` turns an axial or radial shading into stops of an sRGB gradient (within 0.01 per channel, a stop at every stitching-function bound via `Function::stitching_bounds`) when a native gradient draws it exactly (extended past the box drawn in; a radial one with no background colour); (24) `DecodedImageCache`: decoded image pixels shared across renders through the interpreter cache within a 64 MiB budget (an image-heavy page decoded its thousands of images again per tile and per scale), keyed by every decode input (`DecodedImageKey`: the stream's place and dictionary bytes, its object, mask/stencil/raster output, interpolation, decode target size, preview or final); images under a transfer function and inline images are not cached, an image over the budget is never kept, and a poisoned lock decodes uncached; (25) `Device::actual_text` hands the device a marked content sequence's `/ActualText` (inline or named in `/Properties`), so text extraction keeps the Thai, Indic and Arabic clusters Chrome marks this way (their shaped tone marks and vowels have no Unicode of their own); (26) non-embedded simple fonts drawn with a standard-font stand-in build the Latin Extended-A letters the stand-in lacks (`/Gbreve`, `/Idotaccent`, `/Scedilla`, `/aogonek`, `/ccaron`…) from the base letter and accent it has (`StandardKind::composite_glyph`, `ACCENTED_LETTERS`): PDFium's Foxit stand-ins hold only the Standard/WinAnsi/Mac Roman glyphs, so iText letters with Times New Roman unembedded showed blanks for Ğ, İ, Ş, ğ and ş; the comma-below letters (`/scommaaccent`, `/Tcommaaccent`, `/kcommaaccent`…, found by character, so `/uni0136` and the glyph list's `/Kcedilla` too) are the letter and the font's comma, smaller and under the baseline, since the stand-ins have no `commaaccent` (Romanian ș drew nothing and ț a cedilla in forms laid out from XFA; ģ, whose comma is turned and above, is not built); (27) an embedded TrueType font without a base encoding reads its codes as WinAnsi for text extraction (`EmbeddedKind::code_to_unicode_name`), matching pdftotext's output, instead of Standard: OCR text layers of scans write WinAnsi codes with no `/Encoding`, and Standard has no ü, ö or ß and reads é as Ø (glyph selection unchanged); (28) an image whose pixels hold at most 16 bits of samples (an 8-bit Separation or DeviceN image, a 1-, 2- or 4-bit gray or indexed one) is converted once per possible pixel value through a lookup table instead of once per pixel (`rgb_via_lookup`): a spot-colour tint transform is often a PostScript calculator program, run for every pixel, so a fillable character sheet's 3304 × 4105 DeviceN background took four seconds and 430 MB to draw a page; each entry is converted on its own (`convert_each`), so a tint transform giving the wrong number of components for a value the image never uses cannot shift the colours of the values it does; (29) `Device::begin_annotations`/`end_annotations` bracket the page's annotations, so a device can tell comment paints from document paints (reading colours recolour the document, never a comment, #1548); (30) the Annex D.2 alias codes read as the ink they draw: WinAnsiEncoding 0xA0 and 0xAD, and MacRomanEncoding 0xCA, carry the glyph names `space` and `hyphen` -- the names those tables also give 0x20 and 0x2D -- not `nbspace` and `sfthyphen`, whose glyph list values U+00A0 and U+00AD are invisible formatting characters. Extraction hid drawn ink behind them, so search and redaction matched neither (`XiaoBiaoSong.pdf` in the pdf.js corpus drew 104 soft hyphens; across the 989 files the soft hyphens we extract fall from 117 to 13 and the non-breaking spaces from 30 to 26, what remains coming from `/ToUnicode` maps, which say so explicitly). The base table only, in the text direction only: a `/Differences` name keeps its glyph list value and glyph lookup is untouched, both as pdftotext does; (31) a Type 0 sampled function keeps its samples in one flat array indexed by strides (`Type0::entry`) instead of a hash map keyed by each grid point: the map was built again, one hashed insert per sample, every time a shading using the function was painted (a 2,097,152-sample ramp painted 30 times took 38 s; `bug1721218` in the pdf.js corpus spent most of a 1.5 s page on it); the interpolation is unchanged, so results are bit for bit the same, a sample missing from a short stream still gives no value, and the maximum sample value is computed without overflowing at 32 bits per sample; (32) the decoded-image cache (24) keys a decode by the size the image is drawn at only when a JPX stream decodes it or its soft mask or mask (`decode_reads_target`): JPX is the only filter that decodes a reduced resolution for a size, so a JPEG or Flate scan was decoded again at every zoom step and kept once per zoom level; one decode now serves every size; `InterpreterCache::decoded_image_decodes` counts the cache's decodes; (33) Coons and tensor-product patches copy their vertices' colours less: each point of a patch's 20 × 20 triangle grid is interpolated once (`generate_patch_triangles`; each cell interpolated its four corners itself, 1,444 interpolations a patch for 400 points), a cell's second triangle takes the vertices its first copied, and the mesh sampler moves a patch's triangles into device space instead of copying each vertex (`TriangleSampler::sample_owned`); the same values, 10–16% faster on mesh-gradient pages; (34) `color::tint_transform_runs`: how many colours the calling thread has run through a Separation or DeviceN tint transform (counted once per batch in `convert_f32`), so a test counts patch (28)'s once-per-sample-value runs instead of timing them (its 10 s deadline failed on a busy FreeBSD VM in CI); (35) `EncodedShadingPattern::extent`: where a mesh shading (types 4–7) can paint at all, the pixels its triangles touched when it was encoded (`None` for the types that paint everywhere, and when the shading has a `/Background`), so the renderer samples a texture there alone and skips a mesh with nothing on the canvas; (36) a mesh shading's pixel whose corner no triangle holds strictly (on an edge two triangles share, or on the mesh's side) takes the triangle that covers its centre, edges included (`Triangle::covers`): testing only the corner left a white seam along every shared edge, and the colour is still taken at the corner; (38) the `/DefaultGray`, `/DefaultRGB` and `/DefaultCMYK` colour spaces of a resource dictionary (ISO 32000-2 §8.6.5.6) replace the device space content selects: with `g`/`G`, `rg`/`RG`, `k`/`K` and `cs`/`CS` (by name, or through a `/ColorSpace` resource that is a device space), and as the `/ColorSpace` of an image XObject, an inline image or a shading painted by `sh` or as a shading pattern (`DefaultColorSpaces`, looked up once per content window in the resources and those they inherit, so a form XObject without a default of its own draws in its page's, as Poppler and MuPDF draw it); a default that is a Lab, Indexed, Pattern or device space, has another number of components or is no colour space is ignored and device colour is drawn as before; device spaces inside other spaces (an Indexed base, a Separation or DeviceN alternate, a Pattern's underlying space), shading dictionaries, masks and transparency group spaces are unchanged, and the decoded-image cache (24) keys a decode by the default it was drawn in. hayro drew device colour as it was, so pages that calibrate it this way (common in PDF/A) came out in the wrong colours; (39) knockout groups (#2971): `Device::set_knockout` right after a `/K true` group is pushed; `Device::knockout_image` for an unmasked image before its group is pushed, when `Device::in_knockout_group`: the device clears its placement rectangle only if it will draw the image (on the canvas, not a preview's tiny image, and it decodes), decoded at its draw size so the draw reuses that decode; and, when `Device::knockout_erase_wanted`, a nested group's content drawn once between `begin_knockout_erase` and `end_knockout_erase` before the group itself, which then always gets a layer of its own; (40) an optional content membership dictionary's visibility expression (`/VE`: `/And`, `/Or`, `/Not`, nested up to 32 deep) decides its visibility ahead of `/OCGs` and `/P` (ISO 32000-2 §8.11.2.2); it was ignored, so content under it stayed visible with its layers hidden; a malformed expression, including one naming an object the document doesn't list in `/OCProperties /OCGs`, falls back to `/OCGs` and `/P` (#2986); (41) a transparency group that is neither isolated nor knockout, entered at full opacity with no soft mask and Normal blending, is drawn straight onto the page instead of into a layer of its own (§11.4.7: its elements composite with the backdrop; the layer made a `/BM /Multiply` highlight appearance an opaque block, #1968); (42) text in a Type 3 font is drawn filled whatever the text rendering mode, except mode 3 (invisible): only that mode has any effect on Type 3 fonts (ISO 32000-1 §9.3.6). Stroking ran the glyph procedure a second time, so `2 Tr` made bitmap fonts look bold, `1 Tr` painted an uncoloured glyph in the stroking colour, and `7 Tr` hid it (`7 Tr` since follows 64); (43) Lab colour spaces honour their own `/WhitePoint`: L\*a\*b\* goes to XYZ relative to that white, is Bradford-adapted to D65 and converted to sRGB (ISO 32000-2 §8.6.5.4) instead of through `LAB.icc` as if every white were D50 (a D65 blue came out 45 levels too red; the profile is removed); components normalised to [0, 1] (`manual_scale`, an `/Indexed` palette over Lab) map onto L\* 0..100 and the a\*/b\* `/Range`; (44) a Type 0 font whose `/ToUnicode` is the name `/Identity-H` or `/Identity-V` (not a CMap stream; WSCAD writes it) reads each code as its Unicode value, as Acrobat shows it (`Type0Font::identity_to_unicode`); the identity CMap the name resolved to maps codes to CIDs only, so such pages had no text to select or find; (46) an image XObject's group applies the fill opacity, so the fill alpha is reset to 1 before a stencil mask takes the fill colour as its paint, and a pattern paint drops the opacity it captured at `scn` (`Pattern::without_opacity`) (it was applied twice: a mask at `ca 0.5` was drawn at a quarter, #2975); (47) `path_as_rect`: a five-element path is a rectangle only when its four points are distinct, each step from one to the next runs along a single axis, and the last returns to the first. The length test admits one MoveTo and four LineTo as well as one MoveTo, three LineTo and a ClosePath, and asking only that every corner of the bounding box was visited says nothing about the order or the steps between, so a five-point polyline whose ends reached the corners was painted as that box -- a fourth side the path does not have -- and a clip of that shape was discarded; the coordinates are compared exactly, since the matrix is applied after this check and a `cm` of 10,000 turns `is_nearly_equal`'s 256th of a unit into tens of device units (#3327); (48) an `/ICCBased` colour space is cached by what its profile is read from (`IccProfileKey`): the stream object it names, looked up without reading the stream, and then the stream's dictionary and its own bytes (decrypted, not decoded), so a byte-for-byte copy of a profile in another object is read once, and a profile stream written directly in its array, which reports the object holding the array as every other stream written there does, is told apart by its bytes. hayro keyed it by the stream dictionary's bytes alone, so two different profiles with the same `/N` and `/Length` shared whichever was read first, and colours in the second were drawn through the first. `InterpreterCache::icc_profile_reads` counts the profiles read; (49) `OpenTypeFontBlob::new` also accepts a `post` table version 1.0, 2.0 or 3.0 written with its 16-bit halves swapped (`00 00 00 03`), as ScanSoft PDF Create! embeds every font; such fonts were refused and their text drawn with a substitute (Hebrew came out garbled or missing, #2946); `glyph_names` reads such a table from a copy with the version written correctly, so a 2.0 table's names serve `/Differences`. Any other unknown version is still refused (pdf.js 9462); (50) `CffFontBlob::glyph_index_by_name` looks a glyph name up in a name to glyph map built on first use (shared between clones) instead of searching the whole charset, with a string lookup per glyph, for every character code of a font; the first glyph with a name still wins; (51) at most 4096 graphics states saved at once, holding at most 64 MiB together (each `q` copies the whole state, dash array and clip list included, so short operators multiplied one large state); a `q` past either limit is counted but not copied, and its `Q` still removes the clips added since (from storytold/craft-libs#239); (52) a `/W2` run whose first CID is past `MAX_CID` stops there, as in `/W` (7), checked before the run's metrics are read: a first CID of 4294967295 inserted one entry and then overflowed, discarding the whole `/W2` table (from storytold/craft-libs#254); (55) a Separation or DeviceN tint transform that cannot be read keeps its colour space, and one that cannot be read or evaluated paints the alternate space's initial colour, as evaluation failures already did; that fallback, and a `cs`/`CS` colour space that cannot be read (drawn in DeviceGray as before), are now reported once per content stream as `InterpreterWarning::ColorFallback` (`RenderWarning::ColourFallback` in `pdfcraft-render`) instead of silently (#4260); (56) an annotation whose Invisible flag (bit 1) is set and whose `/Subtype` is none of the standard types of ISO 32000-2 Table 171 (`is_standard_annotation_type`; no `/Subtype` counts as unknown) is not drawn (Table 167), and a `/Popup` annotation's appearance is never painted on the page (§12.5.6.14: a popup has no appearance stream of its own; the viewer shows it as a window for its parent annotation). Both were drawn from their `/AP`; (57) a simple font drawn with a standard-font stand-in because the document names a font it doesn't embed (`StandardKind` with `fallback`) narrows a glyph to a narrower `/Widths` advance but no longer widens it to a wider one: the advance already places the next glyph (ISO 32000-2 §9.2.4; /Widths in Table 109), and widening drew a non-embedded `Arial,Bold` with 600-unit widths as a heavy fixed-pitch face whose `i` and `l` were solid blocks twice their width; an exact standard-14 font still matches its standard metrics both ways; (58) the fallback request for a Type 0 font that is not embedded (`FallbackFontQuery::with_descriptor`) takes its serif, italic and small-cap flags, `/FontWeight`, `/FontFamily` and `/FontStretch` from the descendant CIDFont's font descriptor (ISO 32000-2 §9.7.4.1, §9.7.6.1; §9.8): it read them from the Type 0 dictionary, which has none, so every such request asked for a regular sans-serif face and no resolver could tell a Song or Mincho font from a Hei or Gothic one, or see `/FontWeight 700`; (59) the NoRotate flag (`/F` bit 5) is honoured: the appearance, fitted to `/Rect`, is turned against the page's `/Rotate` about the rectangle's upper-left corner, which stays where it is, so it reads upright as the page is displayed (ISO 32000-2 §12.5.3). Sticky notes, which are saved NoRotate, were drawn turned with the page (#2583). Form fields and links stay in `/Rect`, where PdfCraft takes their click targets from it; (64) text in a Type 3 font follows ISO 32000-2 §9.3.6 in the clipping modes: under `7 Tr` it is not rendered (reported to the device as clipping text, `GlyphDrawMode::Clip`, so extraction keeps it), and in modes 4 to 7 a Type 3 glyph adds nothing to the clipping path (`clip_glyph`), so text shown only in a Type 3 font leaves no clip at `ET`. Since (42) `7 Tr` painted the glyphs in the fill colour (black by default), so a fill meant to follow showed them wherever it didn't cover them; 4 to 6 still render filled; (66) a Separation colour space whose colourant is the special name `All` paints every colourant at once (ISO 32000-2 §8.6.6.4): on screen and in an RGB print raster its tint t is the gray 1 − t, complemented because the device is additive, whatever its alternate space and tint transform say (hayro drew it through the tint transform, so a `/DeviceCMYK` transform giving cyan drew registration marks light cyan); `/None` still paints nothing, any other name still goes through its alternate, and a DeviceN space naming `All` keeps its alternate, since the clause gives the name that meaning in a Separation space; (63) a luminosity soft mask whose group colour space is `/DeviceCMYK` (by name, not an ICC-based or calibrated space) reads a `/DeviceCMYK` fill or stroke colour drawn in it, and its `/BC`, as their gray `1 - min(1, 0.3 c + 0.59 m + 0.11 y + k)` (ISO 32000-2 §10.4.2.3, §11.5.3: device colour converts to gray with no colour calibration; `Context::luminosity_in_device_cmyk`, `color::device_cmyk_to_gray`): it went through the CMYK display profile, where 100% K is a dark grey, so a 100% K mask let about an eighth of the masked paint through. CMYK images, shadings and tiling patterns inside such a mask, and CMYK on the page, still use the profile; (71) a lookup table in an indirect string reads: it was read as a stream first, which uses up a reference (ISO 32000-2 §8.6.6.3); (72) an uncoloured pattern's underlying space given as a name or a reference is used: it was skipped, so a CMYK cyan drew red (ISO 32000-2 §8.7.3.3); (73) `InterpreterSettings::jbig2_work`, set for each page in `interpret_page` and ended with it, and a JBIG2 image or an image with a JBIG2 mask past it left out with `ContentTruncated`; (74) colour spaces nest at most 32 levels, through every way one names another, so a chain of them ends instead of overflowing the stack; an ICC profile's `/Alternate` is read where it's met and isn't kept with the profile, or a page could build a chain past the limit on spaces already read; (75) a soft hyphen (U+00AD) that a CFF substitute for a CID font has no glyph for draws its hyphen, as Annex D.2 reads WinAnsi 0xAD in (30), instead of the CID taken as a glyph index (#5888) | (1) `raster::tests::widget_appearance_states`; (2, 3) `xtask text-oracle` on corpus/pdfjs issue918; (4) `raster::tests::layer_override_hides_content`; (5) `raster::tests::soft_mask_without_group_colour_space_is_applied`; (6) `raster::tests::self_referencing_tiling_pattern_terminates`, `raster::tests::self_referencing_type3_glyph_terminates`; (7) `raster::tests::huge_cid_width_ranges_terminate`; (8) `raster::tests::hiding_comments_keeps_fields`; (9) `raster::tests::absurd_mask_and_fax_sizes_are_refused_before_decoding`; (10) `raster::tests::type3_glyphs_that_fan_out_into_themselves_finish`; (11) `raster::tests::type1_font_programs_with_long_runs_of_numbers_load_quickly`; (12) `raster::tests::degenerate_annotation_appearances_are_skipped`; (13) `raster::tests::highlights_without_an_appearance_are_drawn`; (14) `raster::tests::malformed_simple_font_widths_keep_later_widths_and_the_font`, `raster::tests::huge_simple_font_first_char_terminates`; (15) `raster::tests::annotation_appearance_box_maps_onto_rect`; (16) `raster::tests::clipping_path_applies_after_any_painting_operator`; (17) `raster::tests::dropping_the_pool_cancels_its_renders`, `raster::tests::the_watchdog_stops_the_render_it_gives_up_on`; (18) `raster::tests::page_contents_and_nested_forms_share_one_budget`, `raster::tests::content_painted_again_and_again_keeps_drawing`, `raster::tests::sampled_functions_with_billions_of_samples_are_refused`; (19) `raster::tests::mesh_shadings_far_off_the_page_render`, `raster::tests::many_overlapping_mesh_triangles_finish`, `raster::tests::mesh_shadings_with_many_patches_or_no_rows_finish`; (20) `raster::tests::function_limits_calculator_arithmetic_and_depth`, `raster::tests::function_limits_public_construction_work_and_depth`, `raster::tests::function_limits_stitching_cycles_are_refused_and_shared_children_work`, `raster::tests::function_limits_invalid_calculator_keeps_the_page_renderable`; (21) `raster::tests::cids_a_substitute_lacks_are_not_drawn_as_other_glyphs`; (22) `raster::tests::variation_sequences_draw_their_kanji_in_a_substitute`; (23) `raster::tests::axial_and_radial_shadings_draw_as_native_gradients`; (24) `raster::tests::tiles_match_full_render`, `raster::tests::skipping_paths_off_the_canvas_changes_no_pixel` (both decode through the cache), `raster::tests::decoded_images_with_identical_dictionaries_keep_their_own_pixels`, `raster::tests::decoded_images_under_a_transfer_function_are_not_shared`; (25) `text::tests::actual_text_replaces_the_glyphs_it_marks_in_every_writing_direction`, `text::tests::actual_text_follows_the_pdf_rules_for_marked_content`; (26) `raster::tests::accented_letters_the_standard_stand_ins_lack_are_built_from_their_parts`, `raster::tests::comma_below_letters_in_standard_fonts_draw_a_comma_under_the_letter`; (27) `text::tests::truetype_without_an_encoding_reads_its_codes_as_win_ansi`; (28) `raster::tests::spot_colour_images_run_their_tint_transform_once_per_sample_value`, `raster::tests::images_converted_through_a_table_keep_every_samples_colour`, `raster::tests::a_malformed_tint_for_an_unused_sample_keeps_the_used_colours`; (29) `raster::tests::reading_colors_keep_highlight_annotations_readable`; (30) `text::tests::the_annex_d2_alias_codes_read_as_an_ordinary_hyphen_and_space`; (31) `raster::tests::shadings_with_large_sampled_functions_paint_quickly`, `raster::tests::sampled_functions_interpolate_their_grid_as_before`; (32) `raster::tests::decoded_images_serve_every_size_they_are_drawn_at`; (33) `raster::tests::patch_meshes_keep_their_corner_colours`; (34) `raster::tests::spot_colour_images_run_their_tint_transform_once_per_sample_value`; (35) `mesh_extent_tests::{a_mesh_paints_its_triangles_alone, a_mesh_keeps_the_graphics_state_alpha, a_mesh_off_the_canvas_paints_nothing}` (`cargo test -p hayro`); (36) `raster::tests::mesh_triangles_sharing_an_edge_leave_no_seam`; (38) `raster::tests::default_colour_spaces::device_rgb_operators_draw_in_default_rgb`, `raster::tests::default_colour_spaces::device_cmyk_operators_draw_in_default_cmyk`, `raster::tests::default_colour_spaces::device_gray_operators_draw_in_default_gray`, `raster::tests::default_colour_spaces::device_images_draw_in_the_default_colour_spaces`, `raster::tests::default_colour_spaces::device_shadings_draw_in_the_default_colour_spaces`, `raster::tests::default_colour_spaces::an_image_drawn_with_and_without_a_default_keeps_both_colours`, `raster::tests::default_colour_spaces::forms_inherit_the_default_colour_spaces_they_do_not_override`, `raster::tests::default_colour_spaces::unusable_defaults_leave_device_colour_alone`; (39) `raster::tests::knockout_groups_replace_what_their_elements_cover`; (40) `raster::tests::visibility_expressions_decide_what_a_membership_dictionary_shows`; (41) `raster::tests::a_multiply_highlight_inside_a_transparency_group_keeps_the_text`; (42) `raster::tests::rendering_modes_but_3_and_7_draw_type3_glyphs_as_filled`; (43) `raster::tests::lab_honours_the_space_white_point`, `raster::tests::indexed_palettes_over_lab_keep_their_colours`; (44) `text::tests::identity_to_unicode_names_give_text`; (46) `raster::tests::stencil_masks_are_drawn_at_the_fill_opacity_once`; (47) `raster::tests::a_five_point_polyline_is_not_painted_as_its_bounding_box`, `raster::tests::a_path_that_visits_every_corner_out_of_order_is_not_a_rectangle`, `raster::tests::a_rectangle_is_still_painted_as_a_rectangle`; (48) `raster::tests::icc_profile_cache::icc_profiles_with_identical_dictionaries_draw_their_own_colours`, `raster::tests::icc_profile_cache::icc_profiles_used_again_are_read_once`; (49) `raster::tests::a_font_whose_post_version_has_swapped_halves_is_still_used`; (50) `font::blob::tests::glyph_names_resolve_to_the_first_glyph_with_that_name`; (51) `context::saved_state_tests::repeated_saves_stay_bounded_and_balanced`; (52) `font::cid::tests::w2_run_past_the_largest_cid_keeps_the_rest_of_the_table` (hayro-interpret); (55) `raster::tests::devicen_inks::a_tint_transform_that_cannot_be_evaluated_draws_the_rest_of_the_page_and_warns`, `raster::tests::devicen_inks::a_tint_transform_that_cannot_be_read_falls_back_and_warns`; (56) `raster::tests::invisible_unknown_annotations_and_popups_paint_nothing`; (57) `substitute_widths::a_substitute_keeps_its_glyph_shapes_at_wider_advances`, `substitute_widths::a_substitute_still_narrows_glyphs_to_narrower_advances` (pdfcraft-render `tests/it`); (58) `cid_fallback_query::a_cid_font_fallback_request_carries_its_descendants_style`, `cid_fallback_query::a_cid_font_without_a_descriptor_asks_for_the_default_style` (pdfcraft-render integration tests); (59) `raster::tests::no_rotate_notes_stay_upright_at_their_corner_on_rotated_pages` (hayro-interpret); (64) `type3_text_clip::{clip_only_type3_text_is_not_rendered_and_leaves_no_clip, invisible_type3_text_is_not_rendered, other_modes_render_type3_glyphs_filled_and_never_clip, only_outline_glyphs_shape_a_clip_shared_with_type3_text, self_nesting_type3_text_in_a_clip_mode_stays_bounded}` (`cargo test -p pdfcraft-render --test it`), `raster::tests::rendering_modes_but_3_and_7_draw_type3_glyphs_as_filled`, `text::tests::clipping_mode_type3_text_is_kept_by_the_ocr_filter`; (66) `separation_all::{separation_all_fills_paint_the_complemented_tint_as_gray, separation_all_strokes_paint_the_complemented_tint_as_gray, separation_all_images_paint_each_sample_as_the_complemented_gray, separation_all_shadings_paint_the_complemented_gray, separation_none_paints_nothing, a_named_separation_still_paints_through_its_alternate, a_device_n_space_naming_all_still_paints_through_its_alternate}` (`cargo test -p pdfcraft-render --test it`); (63) `color::device_cmyk_gray_tests::{each_ink_darkens_by_its_iso_weight, ink_past_full_coverage_is_black_and_bad_components_read_as_zero}` (hayro-interpret), `it::luminosity_group_space::{black_ink_in_a_device_cmyk_group_masks_fully_and_no_ink_shows_fully, each_process_ink_in_a_device_cmyk_group_masks_by_its_weight, a_device_cmyk_backdrop_colour_masks_by_its_gray, gray_and_rgb_groups_mask_as_before, device_cmyk_on_the_page_keeps_its_display_colour}` (pdfcraft-render); (71) `raster::tests::an_indexed_lookup_table_is_read_in_every_legal_form` (pdfcraft-render); (72) `raster::tests::an_uncoloured_pattern_uses_its_underlying_space_in_every_form` (pdfcraft-render); (73) `raster::tests::the_jbig2_images_of_a_page_share_one_allowance`, `a_picture_whose_jbig2_mask_is_past_the_pages_allowance_is_not_drawn` and `a_pages_jbig2_allowance_ends_with_the_page` (pdfcraft-render); (74) `raster::tests::{a_chain_of_colour_spaces_ends, colour_spaces_are_read_to_the_limit_of_their_nesting, a_chain_built_on_spaces_already_read_ends, a_profile_read_at_the_limit_keeps_its_alternate_for_a_read_nearer_the_top}` (pdfcraft-render); (75) `raster::tests::a_soft_hyphen_a_substitute_lacks_draws_its_hyphen` (pdfcraft-render) | -| hayro | 0.7.1 | Apache-2.0 OR MIT | (1) `RenderSettings::{x_offset, y_offset}` to render one tile of a large page; (2) images over `MAX_IMAGE_PIXELS` (2^28) are skipped (a fuzzed `/W 4294967295` hung in resampling); (3) `tiling_cell_scale` keeps a tiling pattern's cell pixmap within 3000 px a side whatever its `/XStep`/`/YStep` (a fuzzed step far beyond `/BBox` allocated 2 GB); (4) stroke widths clamped to 4 × (canvas width + height) device pixels (a fuzzed `/LW 9223372036854775807` allocated 10 GB in stroke expansion); (5) `usable_dash_pattern`: a stroke whose dash pattern could cut it into more than `MAX_DASHES_PER_STROKE` (250,000) pieces, counting kurbo's restart of the pattern at every subpath and `ClosePath`, is drawn solid (a fuzzed `/D [[11] 0]` on a line from x = 92234775807 allocated over 5 GB, `[0.000001] 0 d` on a 20 pt line over 10 GB); (6) a dash pattern with a negative or non-finite entry, or a period that isn't positive, is drawn solid (`[-1 -1] 0 d` looped for ever in kurbo); (7) `render_size` and `render_into`: render into a buffer the caller allocates (PdfCraft renders into pixel-aligned memory that the GUI takes over as texture data); `render` is `render_into` on a new pixmap; (8) checked image source/target/RGBA, alpha-premultiplication scratch and intermediate sizes before resampling, finite transforms and exact source/mask buffers; sources wider than u16 may shrink before backend limits, with a 2^20 source-axis cap and conservative 128 MiB filter-planning estimate; an over-limit or failed resize keeps the original image resolution and geometry when the source fits the backend, otherwise skips it, with a logged warning; staging, resampling output/scratch, RGBA conversion and padding buffers use fallible reserves; (9) `may_paint_canvas`: a fill or stroke whose control-point box, grown by how far its paint reaches past the path (half the stroke width times the larger of the miter limit and √2) plus 0.5 user units and 2 pixels, misses the canvas is not drawn, so a tile no longer strokes and fills every path of its page (item 5 of #623); `set_skip_offscreen_paths` turns this off for the regression test; (10) `/Interpolate` governs the pre-resize too (#624): a minified image that does not request interpolation is pre-resized with nearest-neighbour (since (21), only inside Type 3 glyphs), planned without the Catmull-Rom filter limits (`image_resampling_size` takes the flag); a Type 3 glyph image is sampled bicubically only when interpolated; a minified image with a matching soft mask is nearest-sampled straight into premultiplied RGBA; (11) `may_paint_rect`: an image whose stream-size box, mapped to the device, misses the canvas is skipped before decoding (decoded data can only shrink against that box), so a tile of an image-heavy page no longer decodes, resamples and rasterizes every image on it; (12) preview renders (`InterpreterSettings::preview`: backdrops, thumbnails, grid cells, signature-drag backgrounds) skip images under `PREVIEW_MIN_IMAGE_AREA` device pixels, while final renders always draw them; (13) reading colours (#897): `RenderSettings::color_replacement` takes a `ColorReplacement` (text colour, background, `only_black`, `line_art`) that remaps solid text and line-art paint while rendering, glyph outlines counting as text and stencil-mask ink as line art; images, image alpha masks, soft masks and pattern cells keep their colours; `render_pdf` gains a `color_replacement: Option` argument (pass `None` for true colours); the upstream-only `[dev-dependencies.log]` entry is removed from `Cargo.toml` (log is a normal dependency since the logging patch), so `cargo test -p hayro` runs this crate's unit tests; (14) a shrunk image that is interpolated (or, since (21), area-averaged) and whose axes lie along the device's (upright, mirrored or a quarter turn, as on a page under `/Rotate`) is snapped to whole device pixels: shrunk once to its rounded device rectangle (halves round up, so a tile snaps an edge where the whole page does) and drawn 1:1 from the rounded corner, instead of a ceil-sized raster resampled again at a fractional scale and offset (`snap_minified_image`; scans looked soft next to Acrobat, and until #5530 a turned one still did, up to 200 grey levels off the same page drawn upright). An image counts as axis-aligned when the terms the snap drops move its far edges by at most a hundredth of a device pixel (`SNAP_SKEW`), not by a tolerance on the coefficients, which let a slight skew of a very wide image through; and the snapped 1:1 transform is used only when the shrink produced the snapped size, so a failed resize keeps the full transform instead of drawing the source raster at its source size; (15) a soft mask of another size than its image is filtered when shrunk, whatever its `/Interpolate` (#1649): such a mask carries the shape on its own (Word saves text as a tiny black image with a glyph-bitmap mask), and nearest-neighbour dropped every stroke thinner than the shrink factor; (16) under a soft mask or a blend mode, axial and radial shadings are sampled instead of drawn as native gradients (#2128): vello_cpu 0.0.8 copies an opaque gradient over fully covered tiles without the mask or blend mode, so PowerPoint's faded gradient fills came out solid; (17) the renderer keeps annotation paints out of the reading-colours mapping between `begin_annotations` and `end_annotations` (#1548); (18) the page renders as one transparency group; under reading colours its background is painted inside it as the initial backdrop, so a Multiply highlight blends with the dark background instead of staying an opaque block, and otherwise the background stays outside, so a blend mode over an empty part of the page keeps its colour instead of blending with the white paper (#2970); (19) `set_paint` samples a mesh shading's texture over its triangles alone (`EncodedShadingPattern::extent`, padded rather than repeated) and skips a mesh with none on the canvas: a BricsCAD plan's 310 one-triangle `sh` each covered the whole tile, 2 s per 1024 px tile at any zoom (the watchdog gave the page up on a busy machine), now about 70 ms; (20) a tiling pattern whose step is too large for a cell pixmap, where what is painted (a fill, or a stroke with its width) can only show the first cell (PowerPoint's `/XStep 2147480100` for a picture fill drawn once), has its `/BBox` drawn at full scale with a transparent border and padded instead of tiled (`shows_one_cell`; a cell whose scale isn't finite and positive, from a `/BBox` too thin to divide by, keeps the old path: `usable_cell_scale`); shrinking the cell to fit the 3000 px limit of (3) left the picture under a pixel, so network diagrams lost their pictures (#3349); (21) an image that does not request interpolation and is drawn below its resolution, outside a Type 3 glyph, is shrunk by area averaging (`area_shrink`: each device pixel the coverage-weighted mean of the samples it covers, in 16.16 fixed point whose weights sum to exactly one, premultiplied when the image has a mask; a soft mask no larger than the image is first sampled onto the image's grid by `mask_on_image_grid` (nearest-neighbour, or bilinearly when the mask asks for `/Interpolate`, so its edge stays as smooth as an enlarged mask's), so its colour and mask are averaged together instead of the averaged colour of transparent samples showing under a mask layer at the mask's edges) to its snapped device rectangle, instead of nearest-neighbour: `/Interpolate` asks for smoothing only when an image is enlarged (ISO 32000-2 §8.9.5.3), and nearest-neighbour kept one sample in N per axis, so the text of a 300 dpi scan at 100% lost most of its strokes (sampling them again at a fractional offset dropped more). PicScale's `Box` and `Area` filters were not used: shrinking, both average a window four target pixels wide. On a 13-page 300 dpi scan every rendered sample at 96, 120 and 192 dpi is within 1 of the exact area average (it was up to 255 off, 17-23% of samples by more than 1). A snapped image whose shrink failed keeps the full transform, so it is not drawn at its source size. The shrink never enlarges: a side drawn at or above its resolution (an image shrunk across and stretched down, as a 204 × 98 dpi fax is) keeps its samples for the sampler. Rows are averaged first (contiguous, cheap) and the fewer rows that leaves are then shrunk across with taps worked out once per image, which made the shrink about three times faster (a 300 dpi page at 96 dpi: 25 ms, nearest-neighbour 11 ms). A source with a side over `MAX_RESAMPLING_SOURCE_SIDE` (2^20) keeps (10)'s nearest-neighbour shrink, and `area_shrink` refuses it, so its accumulator row and tap table stay a few times one source row (a 2^27 × 2 image of zeros is a tiny file); (22) knockout groups: while one is innermost, fills, strokes, glyph outlines and shadings are drawn with `Compose::Copy` (weighted by coverage), so each element replaces what earlier elements of the group drew instead of compositing over it; an image or nested group, drawn into a layer of its own, first has what is under its shape cleared (`knockout_image` for an unmasked image's rectangle, only when `draw_image` will draw it, decoded at the size it is drawn; a knockout erase pass that draws a nested group's content with every paint clearing instead of painting, a shading fill's clear clipped to the shading's `/BBox` as its paint is) and then composites over it (#2971); (23) backport of hayro 0.8: shadings that `as_svg_gradient` can represent are painted as a `vello_cpu` linear or two-point radial gradient instead of a texture sampled pixel by pixel through the shading function (InDesign fades are luminosity soft masks of one full-page axial `sh`: 40 on a report page took 15 s at 144 dpi, now under 1 s); without upstream's half-pixel shift, which suits vello_cpu 0.3 but moves 0.0.8's gradients off pixel centres; (25) a tiling cell is drawn at the scale that makes one `/XStep` (`/YStep`) exactly the cell pixmap's whole number of pixels (`tiling_cell_pixels`, at least 1 and at most the cap), so the repeated pixmap puts cells at whole multiples of the step at any device scale; drawing at the unrounded scale repeated the cell up to half a pixel off at every step, and the error grew with distance from the pattern's origin (ISO 32000-2 8.7.3.1); (28) an image of page-scan size (`SMOOTHED_IMAGE_SIDE`: 512 samples or more on its shorter side) that does not request interpolation is drawn with `ImageQuality::Medium` (bilinear, as interpolated images are) when enlarged on either side, also when the other side is shrunk (the area shrink leaves that side at 1:1), outside a Type 3 glyph: scans rarely set `/Interpolate`, and drawn nearest-neighbour every sample of a zoomed scan became a block, so its text had staircase edges; icons, QR codes and pixel art keep their hard pixels. Bicubic (`High`) looked the same on a zoomed 300 dpi scan and drew four 1024 px tiles at 400% in 267 ms against 50 ms | (1) `raster::tests::tiles_match_full_render`; (2) `raster::tests::absurd_image_dimensions_are_skipped`; (3) `raster::tests::tiling_cells_stay_small_whatever_the_step`; (4) `raster::tests::absurd_predictor_columns_and_line_widths_render`; (5) `raster::tests::dash_patterns_that_cut_a_stroke_into_billions_of_pieces_render`; (6) `raster::tests::invalid_dash_patterns_terminate_and_are_drawn_solid`; (7) `raster::tests::pixels_are_a_plain_renders_bytes_in_a_buffer_a_gui_can_take_over`; (8) `raster::tests::image_resampling_rejects_anisotropic_target_growth`, `raster::tests::image_resampling_bounds_intermediate_before_planning`, `raster::tests::image_resampling_rejects_invalid_sources_and_scales`, `raster::tests::image_resampling_keeps_valid_sizes_and_exact_limits`, `raster::tests::resampling_fallback_preserves_original_geometry_and_pixels`, `raster::tests::image_resampling_wide_sources_shrink_before_backend_side_limits`, `raster::tests::image_resampling_wide_transparent_sources_still_shrink`, `raster::tests::image_resampling_padded_backend_dimensions_stay_checked`, `raster::tests::image_resampling_mismatched_alpha_mask_keeps_pixels_and_placement`; (9) `raster::tests::skipping_paths_off_the_canvas_changes_no_pixel`; (10) `raster::tests::non_interpolated_images_are_not_smoothed_when_minified`, `raster::tests::non_interpolated_gray_images_are_not_smoothed_when_minified`, `raster::tests::non_integer_minification_averages_the_columns_each_pixel_covers`, `raster::tests::non_integer_minification_averages_colour_and_mask_together`, `raster::tests::non_interpolated_alpha_images_keep_their_colours_and_mask_when_minified`, `raster::tests::non_interpolated_type3_glyph_images_are_not_smoothed_when_minified`, `raster::tests::image_resampling_nearest_accepts_wide_sources_the_filter_refuses`, `raster::tests::wide_non_interpolated_sources_draw_after_shrinking`; (11) `raster::tests::skipping_paths_off_the_canvas_changes_no_pixel`; (12) `raster::tests::preview_renders_skip_tiny_images_but_finals_draw_everything`; (13) `color_replacement_tests::{black_maps_to_text_colour, white_maps_to_background, midgrey_interpolates, alpha_copied_verbatim, only_black_threshold, line_art_false_skips_paths_but_not_text}`, `color_replacement_render_tests::{color_replacement_maps_text_and_line_art, color_replacement_maps_stencil_mask_ink}`; (14) `raster::tests::minified_image_edges_snap_to_whole_pixels`, `raster::tests::a_tile_snaps_half_pixel_image_edges_as_the_whole_page_does` (`cargo test -p hayro`), `raster::tests::a_turned_or_mirrored_minified_image_snaps_to_whole_pixels`, `rotated_image_snap::a_shrunk_image_on_a_turned_page_is_the_upright_render_turned` (`cargo test -p pdfcraft-render --test it`), `renderer::snap_tests::a_skew_that_moves_a_far_edge_is_not_snapped` (`cargo test -p hayro`); (15) `raster::tests::shrunk_soft_mask_larger_than_its_image_keeps_thin_strokes`; (16) `raster::tests::soft_mask_applies_to_a_shading_pattern_fill`; (17), (18) `raster::tests::reading_colors_keep_highlight_annotations_readable`; (19) `mesh_extent_tests::{a_mesh_paints_its_triangles_alone, a_mesh_keeps_the_graphics_state_alpha, a_mesh_off_the_canvas_paints_nothing}` (`cargo test -p hayro`); (20) `raster::tests::a_tiling_pattern_drawn_once_shows_its_cell`, `raster::tests::a_large_step_still_tiles_where_the_next_cell_shows`, `raster::tests::a_tiling_cell_too_thin_to_scale_keeps_the_old_path`; (21) `raster::tests::non_interpolated_minification_keeps_thin_strokes`, `raster::tests::non_integer_minification_averages_the_columns_each_pixel_covers`, `raster::tests::non_integer_minification_averages_colour_and_mask_together`, `raster::tests::non_interpolated_minification_does_not_bleed_transparent_colour`, `raster::tests::a_smaller_soft_mask_does_not_bleed_transparent_colour_when_minified`, `raster::tests::an_interpolated_smaller_soft_mask_stays_smooth_when_its_image_is_minified`, `raster::tests::minified_image_edges_snap_to_whole_pixels`, `raster::tests::an_image_shrunk_one_way_and_stretched_the_other_is_still_averaged`, `raster::tests::wide_non_interpolated_sources_draw_after_shrinking`, `renderer::area_shrink_tests` (`cargo test -p hayro`); (22) `raster::tests::knockout_groups_replace_what_their_elements_cover`; (23) `raster::tests::axial_and_radial_shadings_draw_as_native_gradients`; (25) `tiling_cell_placement::{far_cells_of_a_fill_sit_at_whole_steps, far_cells_sit_at_whole_steps_under_a_turned_scaled_matrix, far_cells_of_a_stroke_sit_at_whole_steps, a_cell_pixmap_spans_exactly_one_step}` (`cargo test -p pdfcraft-render --test it`); (28) `raster::tests::large_non_interpolated_images_are_smoothed_when_enlarged`, `raster::tests::small_non_interpolated_images_keep_hard_pixels_when_enlarged`, `raster::tests::a_scan_shrunk_one_way_is_smoothed_the_way_it_is_stretched` | +| hayro | 0.7.1 | Apache-2.0 OR MIT | (1) `RenderSettings::{x_offset, y_offset}` to render one tile of a large page; (2) images over `MAX_IMAGE_PIXELS` (2^28) are skipped (a fuzzed `/W 4294967295` hung in resampling); (3) `tiling_cell_scale` keeps a tiling pattern's cell pixmap within 3000 px a side whatever its `/XStep`/`/YStep` (a fuzzed step far beyond `/BBox` allocated 2 GB); (4) stroke widths clamped to 4 × (canvas width + height) device pixels (a fuzzed `/LW 9223372036854775807` allocated 10 GB in stroke expansion); (5) `usable_dash_pattern`: a stroke whose dash pattern could cut it into more than `MAX_DASHES_PER_STROKE` (250,000) pieces, counting kurbo's restart of the pattern at every subpath and `ClosePath`, is drawn solid (a fuzzed `/D [[11] 0]` on a line from x = 92234775807 allocated over 5 GB, `[0.000001] 0 d` on a 20 pt line over 10 GB); (6) a dash pattern with a negative or non-finite entry, or a period that isn't positive, is drawn solid (`[-1 -1] 0 d` looped for ever in kurbo); (7) `render_size` and `render_into`: render into a buffer the caller allocates (PdfCraft renders into pixel-aligned memory that the GUI takes over as texture data); `render` is `render_into` on a new pixmap; (8) checked image source/target/RGBA, alpha-premultiplication scratch and intermediate sizes before resampling, finite transforms and exact source/mask buffers; sources wider than u16 may shrink before backend limits, with a 2^20 source-axis cap and conservative 128 MiB filter-planning estimate; an over-limit or failed resize keeps the original image resolution and geometry when the source fits the backend, otherwise skips it, with a logged warning; staging, resampling output/scratch, RGBA conversion and padding buffers use fallible reserves; (9) `may_paint_canvas`: a fill or stroke whose control-point box, grown by how far its paint reaches past the path (half the stroke width times the larger of the miter limit and √2) plus 0.5 user units and 2 pixels, misses the canvas is not drawn, so a tile no longer strokes and fills every path of its page (item 5 of #623); `set_skip_offscreen_paths` turns this off for the regression test; (10) `/Interpolate` governs the pre-resize too (#624): a minified image that does not request interpolation is pre-resized with nearest-neighbour (since (21), only inside Type 3 glyphs), planned without the Catmull-Rom filter limits (`image_resampling_size` takes the flag); a Type 3 glyph image is sampled bicubically only when interpolated; a minified image with a matching soft mask is nearest-sampled straight into premultiplied RGBA; (11) `may_paint_rect`: an image whose stream-size box, mapped to the device, misses the canvas is skipped before decoding (decoded data can only shrink against that box), so a tile of an image-heavy page no longer decodes, resamples and rasterizes every image on it; (12) preview renders (`InterpreterSettings::preview`: backdrops, thumbnails, grid cells, signature-drag backgrounds) skip images under `PREVIEW_MIN_IMAGE_AREA` device pixels, while final renders always draw them; (13) reading colours (#897): `RenderSettings::color_replacement` takes a `ColorReplacement` (text colour, background, `only_black`, `line_art`) that remaps solid text and line-art paint while rendering, glyph outlines counting as text and stencil-mask ink as line art; images, image alpha masks, soft masks and pattern cells keep their colours; `render_pdf` gains a `color_replacement: Option` argument (pass `None` for true colours); the upstream-only `[dev-dependencies.log]` entry is removed from `Cargo.toml` (log is a normal dependency since the logging patch), so `cargo test -p hayro` runs this crate's unit tests; (14) a shrunk image that is interpolated (or, since (21), area-averaged) and whose axes lie along the device's (upright, mirrored or a quarter turn, as on a page under `/Rotate`) is snapped to whole device pixels: shrunk once to its rounded device rectangle (halves round up, so a tile snaps an edge where the whole page does) and drawn 1:1 from the rounded corner, instead of a ceil-sized raster resampled again at a fractional scale and offset (`snap_minified_image`; scans looked soft next to Acrobat, and until #5530 a turned one still did, up to 200 grey levels off the same page drawn upright). An image counts as axis-aligned when the terms the snap drops move its far edges by at most a hundredth of a device pixel (`SNAP_SKEW`), not by a tolerance on the coefficients, which let a slight skew of a very wide image through; and the snapped 1:1 transform is used only when the shrink produced the snapped size, so a failed resize keeps the full transform instead of drawing the source raster at its source size; (15) a soft mask of another size than its image is filtered when shrunk, whatever its `/Interpolate` (#1649): such a mask carries the shape on its own (Word saves text as a tiny black image with a glyph-bitmap mask), and nearest-neighbour dropped every stroke thinner than the shrink factor; (16) under a soft mask or a blend mode, axial and radial shadings are sampled instead of drawn as native gradients (#2128): vello_cpu 0.0.8 copies an opaque gradient over fully covered tiles without the mask or blend mode, so PowerPoint's faded gradient fills came out solid; (17) the renderer keeps annotation paints out of the reading-colours mapping between `begin_annotations` and `end_annotations` (#1548); (18) the page renders as one transparency group; under reading colours its background is painted inside it as the initial backdrop, so a Multiply highlight blends with the dark background instead of staying an opaque block, and otherwise the background stays outside, so a blend mode over an empty part of the page keeps its colour instead of blending with the white paper (#2970); (19) `set_paint` samples a mesh shading's texture over its triangles alone (`EncodedShadingPattern::extent`, padded rather than repeated) and skips a mesh with none on the canvas: a BricsCAD plan's 310 one-triangle `sh` each covered the whole tile, 2 s per 1024 px tile at any zoom (the watchdog gave the page up on a busy machine), now about 70 ms; (20) a tiling pattern whose step is too large for a cell pixmap, where what is painted (a fill, or a stroke with its width) can only show the first cell (PowerPoint's `/XStep 2147480100` for a picture fill drawn once), has its `/BBox` drawn at full scale with a transparent border and padded instead of tiled (`shows_one_cell`; a cell whose scale isn't finite and positive, from a `/BBox` too thin to divide by, keeps the old path: `usable_cell_scale`); shrinking the cell to fit the 3000 px limit of (3) left the picture under a pixel, so network diagrams lost their pictures (#3349); (21) an image that does not request interpolation and is drawn below its resolution, outside a Type 3 glyph, is shrunk by area averaging (`area_shrink`: each device pixel the coverage-weighted mean of the samples it covers, in 16.16 fixed point whose weights sum to exactly one, premultiplied when the image has a mask; a soft mask no larger than the image is first sampled onto the image's grid by `mask_on_image_grid` (nearest-neighbour, or bilinearly when the mask asks for `/Interpolate`, so its edge stays as smooth as an enlarged mask's), so its colour and mask are averaged together instead of the averaged colour of transparent samples showing under a mask layer at the mask's edges) to its snapped device rectangle, instead of nearest-neighbour: `/Interpolate` asks for smoothing only when an image is enlarged (ISO 32000-2 §8.9.5.3), and nearest-neighbour kept one sample in N per axis, so the text of a 300 dpi scan at 100% lost most of its strokes (sampling them again at a fractional offset dropped more). PicScale's `Box` and `Area` filters were not used: shrinking, both average a window four target pixels wide. On a 13-page 300 dpi scan every rendered sample at 96, 120 and 192 dpi is within 1 of the exact area average (it was up to 255 off, 17-23% of samples by more than 1). A snapped image whose shrink failed keeps the full transform, so it is not drawn at its source size. The shrink never enlarges: a side drawn at or above its resolution (an image shrunk across and stretched down, as a 204 × 98 dpi fax is) keeps its samples for the sampler. Each target row averages the whole source rows it covers (contiguous, cheap) and is then shrunk across with taps worked out once per image, which made the shrink about three times faster (a 300 dpi page at 96 dpi: 25 ms, nearest-neighbour 11 ms); working a row at a time (`area_shrink_rows`) keeps its work memory to a few rows. An image with a soft mask, a stencil's ink included, is premultiplied a source row at a time as it is averaged (`area_shrink_premultiplied`), never as a full-resolution RGBA copy: a 10718 × 13701 stencil shown at page size peaked at 1.16 GB with that copy and peaks at 1.03 GB without it, as with nearest-neighbour. A source with a side over `MAX_RESAMPLING_SOURCE_SIDE` (2^20) keeps (10)'s nearest-neighbour shrink, and `area_shrink` refuses it, so its accumulator row and tap table stay a few times one source row (a 2^27 × 2 image of zeros is a tiny file); (22) knockout groups: while one is innermost, fills, strokes, glyph outlines and shadings are drawn with `Compose::Copy` (weighted by coverage), so each element replaces what earlier elements of the group drew instead of compositing over it; an image or nested group, drawn into a layer of its own, first has what is under its shape cleared (`knockout_image` for an unmasked image's rectangle, only when `draw_image` will draw it, decoded at the size it is drawn; a knockout erase pass that draws a nested group's content with every paint clearing instead of painting, a shading fill's clear clipped to the shading's `/BBox` as its paint is) and then composites over it (#2971); (23) backport of hayro 0.8: shadings that `as_svg_gradient` can represent are painted as a `vello_cpu` linear or two-point radial gradient instead of a texture sampled pixel by pixel through the shading function (InDesign fades are luminosity soft masks of one full-page axial `sh`: 40 on a report page took 15 s at 144 dpi, now under 1 s); without upstream's half-pixel shift, which suits vello_cpu 0.3 but moves 0.0.8's gradients off pixel centres; (25) a tiling cell is drawn at the scale that makes one `/XStep` (`/YStep`) exactly the cell pixmap's whole number of pixels (`tiling_cell_pixels`, at least 1 and at most the cap), so the repeated pixmap puts cells at whole multiples of the step at any device scale; drawing at the unrounded scale repeated the cell up to half a pixel off at every step, and the error grew with distance from the pattern's origin (ISO 32000-2 8.7.3.1); (28) an image of page-scan size (`SMOOTHED_IMAGE_SIDE`: 512 samples or more on its shorter side) that does not request interpolation is drawn with `ImageQuality::Medium` (bilinear, as interpolated images are) when enlarged by more than 1% on either side (`SMOOTHED_IMAGE_SCALE`: a scan drawn at its resolution can come out a few ulps above 1, and blurred by half a pixel at a half-pixel offset), also when the other side is shrunk (the area shrink leaves that side at 1:1), outside a Type 3 glyph: scans rarely set `/Interpolate`, and drawn nearest-neighbour every sample of a zoomed scan became a block, so its text had staircase edges; icons, QR codes and pixel art keep their hard pixels. Bicubic (`High`) looked the same on a zoomed 300 dpi scan and drew four 1024 px tiles at 400% in 267 ms against 50 ms | (1) `raster::tests::tiles_match_full_render`; (2) `raster::tests::absurd_image_dimensions_are_skipped`; (3) `raster::tests::tiling_cells_stay_small_whatever_the_step`; (4) `raster::tests::absurd_predictor_columns_and_line_widths_render`; (5) `raster::tests::dash_patterns_that_cut_a_stroke_into_billions_of_pieces_render`; (6) `raster::tests::invalid_dash_patterns_terminate_and_are_drawn_solid`; (7) `raster::tests::pixels_are_a_plain_renders_bytes_in_a_buffer_a_gui_can_take_over`; (8) `raster::tests::image_resampling_rejects_anisotropic_target_growth`, `raster::tests::image_resampling_bounds_intermediate_before_planning`, `raster::tests::image_resampling_rejects_invalid_sources_and_scales`, `raster::tests::image_resampling_keeps_valid_sizes_and_exact_limits`, `raster::tests::resampling_fallback_preserves_original_geometry_and_pixels`, `raster::tests::image_resampling_wide_sources_shrink_before_backend_side_limits`, `raster::tests::image_resampling_wide_transparent_sources_still_shrink`, `raster::tests::image_resampling_padded_backend_dimensions_stay_checked`, `raster::tests::image_resampling_mismatched_alpha_mask_keeps_pixels_and_placement`; (9) `raster::tests::skipping_paths_off_the_canvas_changes_no_pixel`; (10) `raster::tests::non_interpolated_images_are_not_smoothed_when_minified`, `raster::tests::non_interpolated_gray_images_are_not_smoothed_when_minified`, `raster::tests::non_integer_minification_averages_the_columns_each_pixel_covers`, `raster::tests::non_integer_minification_averages_colour_and_mask_together`, `raster::tests::non_interpolated_alpha_images_keep_their_colours_and_mask_when_minified`, `raster::tests::non_interpolated_type3_glyph_images_are_not_smoothed_when_minified`, `raster::tests::image_resampling_nearest_accepts_wide_sources_the_filter_refuses`, `raster::tests::wide_non_interpolated_sources_draw_after_shrinking`; (11) `raster::tests::skipping_paths_off_the_canvas_changes_no_pixel`; (12) `raster::tests::preview_renders_skip_tiny_images_but_finals_draw_everything`; (13) `color_replacement_tests::{black_maps_to_text_colour, white_maps_to_background, midgrey_interpolates, alpha_copied_verbatim, only_black_threshold, line_art_false_skips_paths_but_not_text}`, `color_replacement_render_tests::{color_replacement_maps_text_and_line_art, color_replacement_maps_stencil_mask_ink}`; (14) `raster::tests::minified_image_edges_snap_to_whole_pixels`, `raster::tests::a_tile_snaps_half_pixel_image_edges_as_the_whole_page_does` (`cargo test -p hayro`), `raster::tests::a_turned_or_mirrored_minified_image_snaps_to_whole_pixels`, `rotated_image_snap::a_shrunk_image_on_a_turned_page_is_the_upright_render_turned` (`cargo test -p pdfcraft-render --test it`), `renderer::snap_tests::a_skew_that_moves_a_far_edge_is_not_snapped` (`cargo test -p hayro`); (15) `raster::tests::shrunk_soft_mask_larger_than_its_image_keeps_thin_strokes`; (16) `raster::tests::soft_mask_applies_to_a_shading_pattern_fill`; (17), (18) `raster::tests::reading_colors_keep_highlight_annotations_readable`; (19) `mesh_extent_tests::{a_mesh_paints_its_triangles_alone, a_mesh_keeps_the_graphics_state_alpha, a_mesh_off_the_canvas_paints_nothing}` (`cargo test -p hayro`); (20) `raster::tests::a_tiling_pattern_drawn_once_shows_its_cell`, `raster::tests::a_large_step_still_tiles_where_the_next_cell_shows`, `raster::tests::a_tiling_cell_too_thin_to_scale_keeps_the_old_path`; (21) `raster::tests::non_interpolated_minification_keeps_thin_strokes`, `raster::tests::non_integer_minification_averages_the_columns_each_pixel_covers`, `raster::tests::non_integer_minification_averages_colour_and_mask_together`, `raster::tests::non_interpolated_minification_does_not_bleed_transparent_colour`, `raster::tests::a_smaller_soft_mask_does_not_bleed_transparent_colour_when_minified`, `raster::tests::an_interpolated_smaller_soft_mask_stays_smooth_when_its_image_is_minified`, `raster::tests::a_shrunk_stencil_mask_keeps_thin_strokes`, `raster::tests::minified_image_edges_snap_to_whole_pixels`, `raster::tests::an_image_shrunk_one_way_and_stretched_the_other_is_still_averaged`, `raster::tests::wide_non_interpolated_sources_draw_after_shrinking`, `renderer::area_shrink_tests` (`cargo test -p hayro`); (22) `raster::tests::knockout_groups_replace_what_their_elements_cover`; (23) `raster::tests::axial_and_radial_shadings_draw_as_native_gradients`; (25) `tiling_cell_placement::{far_cells_of_a_fill_sit_at_whole_steps, far_cells_sit_at_whole_steps_under_a_turned_scaled_matrix, far_cells_of_a_stroke_sit_at_whole_steps, a_cell_pixmap_spans_exactly_one_step}` (`cargo test -p pdfcraft-render --test it`); (28) `raster::tests::large_non_interpolated_images_are_smoothed_when_enlarged`, `raster::tests::small_non_interpolated_images_keep_hard_pixels_when_enlarged`, `raster::tests::a_scan_shrunk_one_way_is_smoothed_the_way_it_is_stretched`, `raster::tests::a_scan_drawn_a_rounding_error_above_its_resolution_stays_sharp` | | hayro-jbig2 | 0.3.0 | Apache-2.0 OR MIT | (1) at most 2^28 pixels per bitmap and page bitmap, and 2^25 cells per halftone grid and grayscale image (each side could be 65535: 4.3 billion pixels decoded one by one, and a 4-byte value per cell); (2) at most 2^26 symbol pixels per symbol dictionary (a fuzzed dictionary decoded for minutes); (3) a symbol dictionary's height classes are capped at one per new symbol and its export runs at `2 × symbols + 1`, each plus `EXTRA_EMPTY_STEPS` (1024): an empty height class or a zero-length export run doesn't advance, and past the end of its data the arithmetic decoder can decode them for ever (one changed byte hung the renderer); (4) a segment header's long-form referred-to segment count is capped at `u16::MAX` and a larger count is refused before its retention or referred-to bytes are read: the 29-bit count went straight to `Vec::with_capacity`, a reservation of about 2 GiB whose failure aborts the process (from storytold/craft-libs#176); (5) what a page's segments hold for later ones (intermediate regions, pattern and symbol dictionaries with their headers and retained contexts, and Huffman tables) is at most 64 MiB, two square bitmaps of 16,384 a side (`MAX_RETAINED_WORDS`), and a dictionary's new and exported symbols are counted against it as they are made: each was bounded on its own, but a few dozen bytes an intermediate region held 33 MB each, and 8 MiB symbols exported again and again killed the process, so a few kilobytes could exhaust memory; (6) a segment may refer to at most 2^20 symbols, checked before any are collected: references may repeat, and 65,535 of them to one dictionary, in a 78 KB stream, asked for 17 GB; (7) a pattern dictionary keeps at most 16 MiB of shifted copies, which only speed up drawing, and none that wouldn't fit what the page has left (`MAX_SHIFTED_WORDS`); (8) a custom Huffman table has at most 2^16 lines, and a stream with its globals at most 2^20 segments and 2^22 references, checked as they are read: both grew before anything counted them, and PDFs of a kilobyte or so reached 5 GB and more behind nested Flate; (9) a page past the bounds of (1) is refused when the stream is opened: a caller sizes its output from the header before the page bitmap is made, and an Indexed image in a 743-byte PDF asked for 137 GB, which aborts; (10) a stream may ask for at most 2^30 units of decoding work, four of the largest pages, about 6 s to 17 s (`MAX_WORK`): decoding a bitmap costs a pixel each and 4 a row, copying one a word each and 1 a row, and each integer an item reads costs 38, what one takes at worst, so one with no pixels pays too, before it is decoded: each bitmap was bounded, but nothing bounded how many there were, and a 795-byte PDF took 265 s; (11) `Image::decode_within` spends from what a run of images may still do, starting with its page's pixels, and `charge_reading` charges reading a stream and its globals before anything is parsed, so an operation over a document can bound what `MAX_WORK` bounds for one image, and images that fail or decode nothing add up too: a file of thousands of images made to reach that limit took as many times as long; (12) adding to a location, a reference offset, a symbol's corner or a Huffman range line near the end of an `i32`, and negating a decoded integer of 2^31, overflowed: a panic wherever overflow is checked. The refinement context, a text region's corners, the integer decoder and the Huffman table wrap, as a release build did, and `Bitmap::combine` saturates, which comes to the same, as a location that overflows is past the end of any page. What is drawn doesn't change | (1) `raster::tests::jbig2_regions_of_billions_of_pixels_are_refused`, `raster::tests::jbig2_halftone_grids_past_their_bound_are_refused`; (2) the nightly `cargo xtask fuzz` finding, which now finishes in about a second; (3) `raster::tests::jbig2_symbol_dictionaries_that_never_advance_terminate`; (4) `segment::tests::test_referred_to_count_is_capped` (`cargo test --manifest-path vendor/hayro-jbig2/Cargo.toml`); (5) `raster::tests::jbig2_regions_held_past_a_pages_bound_are_refused`, `jbig2_symbols_without_pixels_count_towards_a_pages_bound`, `jbig2_retained_contexts_count_towards_a_pages_bound`, `jbig2_huffman_tables_count_towards_a_pages_bound`, `jbig2_new_symbols_count_towards_a_pages_bound`, `jbig2_new_symbols_are_held_to_what_a_page_has_left`, `jbig2_pattern_dictionaries_count_towards_a_pages_bound`, and in the crate (`cargo test --manifest-path vendor/hayro-jbig2/Cargo.toml`) `decode::symbol::tests::exported_symbols_are_counted_before_they_are_copied`, `tests::the_room_left_shrinks_as_segments_are_stored`; (6) `raster::tests::jbig2_segments_referring_to_too_many_symbols_are_refused`; (7) `raster::tests::jbig2_pattern_dictionaries_too_large_to_shift_still_draw`, `jbig2_shifted_patterns_never_refuse_an_image`; (8) `raster::tests::jbig2_huffman_tables_of_too_many_lines_are_refused`, `jbig2_streams_of_too_many_segments_are_refused`, `jbig2_streams_of_too_many_references_are_refused`, and in the crate `tests::the_segments_of_a_stream_and_its_globals_count_together`; (9) `raster::tests::jbig2_pages_past_the_bounds_are_refused_before_anything_is_sized_from_them`, `tests::a_page_past_the_bounds_is_refused_when_opened`; (10) `raster::tests::jbig2_decoding_work_is_counted_item_by_item`; (11) `jbig2::decode::tests::a_run_of_images_spends_its_allowance_down_to_the_step` (pdfcraft-filters); (12) `raster::tests::jbig2_placements_at_the_end_of_an_i32_draw_nothing` | | lopdf | 0.45.0 | MIT | (1) PNG predictor rows longer than the data are refused before their buffers are allocated (a fuzzed `/Columns 4294967295` allocated two 4 GiB rows), and empty frames return without row buffers; (2) predictor row widths use saturating arithmetic or reject overflow; (3) sub-byte TIFF Predictor 2 scratch is bounded by the samples present in the input instead of the declared `/Colors`, aligned with [lopdf#589](https://github.com/J-F-Liu/lopdf/pull/589); (4) the object lookup keeps its cycle history through compressed objects, requires an ordinary in-file container, and caps the chain of indirect lookups at 16 (a chain of new objects, each a `/Length` compressed in a container whose own `/Length` names the next, or a stream whose `/Length` names the next stream, overflowed the stack: a dev build's 2 MiB thread within 800 hops); (5) object-stream members are parsed within their own extent (up to the next distinct offset), an entry whose offset repeats an earlier entry's is skipped (so a repeated object number can resolve to its earlier entry), and offsets are processed in index order; (6) `find_latest_trailer`: a rebuild takes a `/Type /XRef` stream dictionary as the trailer when the file has no `trailer` keyword (the latest in the file wins, as a later `trailer` would), so a concatenated or truncated file keeps its /Info and /ID instead of failing to load; a candidate whose /Root is the catalog beats one whose /Root is another dictionary, and an in-file /Root must match its generation (#2365) | (1) `inspect::tests::png_predictor_rows_longer_than_the_data_are_refused_before_allocating`, `inspect::tests::empty_png_predictor_frames_do_not_allocate_rows`; (2) `inspect::tests::tiff_subbyte_predictor_row_width_overflow_is_refused`; (3) `inspect::tests::tiff_subbyte_predictor_scratch_is_bounded_by_available_samples`, `inspect::tests::tiff_subbyte_predictors_keep_components_rows_and_padding`; (4) `inspect::tests::compressed_object_self_container_does_not_recurse`, `inspect::tests::compressed_object_mutual_containers_do_not_recurse`, `inspect::tests::object_stream_length_cycle_does_not_recurse`, `inspect::tests::compressed_object_container_chain_does_not_recurse`, `inspect::tests::compressed_length_chain_does_not_recurse`, `inspect::tests::plain_stream_length_chain_does_not_recurse`, `inspect::tests::compressed_objects_and_indirect_stream_length_still_load`, `inspect::tests::compressed_length_in_a_container_with_an_indirect_length_still_loads`; (5) `inspect::tests::object_stream_member_extents_bound_owned_strings`, `inspect::tests::object_stream_member_extents_preserve_index_order_and_whitespace`; (6) `inspect::tests::rebuilt_file_keeps_the_info_of_its_cross_reference_stream`, `inspect::tests::rebuilt_file_prefers_a_trailer_whose_root_is_the_catalog` | | hayro-syntax | 0.7.2 | Apache-2.0 OR MIT | (1) Page-tree cycle guard in `resolve_pages` (visited set + depth cap): a `/Kids` loop among compressed objects overflowed the stack and aborted the process; (2) `ccitt_size_ok`: CCITT images wider than 2^20 columns or over 2^28 pixels are not decoded (a fuzzed `/Columns 4294967295` allocated 4 GiB); (3) Flate/LZW predictor row widths are saturating `usize`, a zero width is refused and a row longer than the data is not allocated (a fuzzed `/Columns 9223372036854775807` wrapped to a 2^61-byte allocation and aborted); (4) the inline-image end-of-data check (`EI`) doesn't nest inside its own look-ahead and examines at most 256 candidates and steps (50 images with "EI" in their data ran for over ten minutes); an EI followed by a delimiter (EI%..., EI[...) ends the image like one followed by white space; only a regular character after EI skips the candidate; (5) a stream whose `/Length` is wrong is recovered only when `stream` is the next token — scanning ahead to a later object's `stream` made an indirect appearance-state dictionary (Quartz checkboxes: `/AP /N` → `<< /Yes … /Off … >>`) render as a non-form, so the widget was skipped; (6) JPX rescaling uses an exact integer sample range for 1–32-bit components (32 shifted past u32, and floating-point maxima at 25–31 bits wrapped to zero), and refuses zero-sized rows; (7) `decode_limit`: Flate (through flate2 and the fallback inflater), LZW, RunLength and ASCII85 never write past `MAX_DECODED_STREAM` (256 MiB), or, for the last filter of an image of known size, past twice its pixel bytes (a predictor tag byte per byte), at most 1 GiB; `Stream::decoded_within` for budgeted callers; a page's `/Contents` array shares one `MAX_DECODED_STREAM` (a 2 MB page whose content stream or 8 × 8 image inflated to 2 GB took the renderer past 3 GB in under a second; naming the stream twenty times in `/Contents`, 5 GB); (8) high-precision numbers (jsPDF MediaBox coordinates with over 19 digits) parse without overflowing the u64 mantissa: fractional digits past it are dropped, integer digits past it scale the value (a real, kept finite); (9) large dictionary indexes are built on first use and shared across repeated reads within each PDF (#307); the validated-byte cache is bounded, reader contexts remain separate, repairs and inline/content dictionaries bypass sharing, and small dictionaries stay eager; (10) decoded object-stream slots publish once under contention, cold loading rejects recursive dependencies, and xref repair serializes without holding recursive read guards; this permits one parser per native render pool with private worker caches; (11) `sync.rs`: a poisoned lock is recovered instead of unwrapped (a worker panic no longer cascades into every reader of the same PDF), and xref repair takes a blocking write lock instead of asserting a non-blocking one; (12) `SegmentList` (`util.rs`) returns `None` for an out-of-range index instead of panicking; (13) a page's `/UserUnit` scales `render_dimensions` and `initial_transform`, so the page and everything on it are sized in points, as Acrobat shows them; it is read as Acrobat reads it (the page's own entry, not inherited; a number from 1 up, at most 75,000; anything else counts as 1), and a page without it keeps exactly the transform it had; (14) revision 2–4 passwords are tried as given and then in PDFDocEncoding (`legacy_passwords` in `crypto/mod.rs`, which resolves the upstream `TODO: Convert to PDFDocEncoding`): passwords arrive as UTF-8 text, so a password with characters outside ASCII (`Çok güzel`, `ılık€`) written in PDFDocEncoding, as ISO 32000-2 §7.6.4.3.2 asks and qpdf and Acrobat write it, was always wrong; (15) page content is read a piece at a time (`page::ContentReader`, `Page::content_reader`): plain streams and Flate streams without a predictor are decoded as they are read, with a budget of 64 times their compressed size and never less than `MAX_DECODED_STREAM` (a stream that expands past it is cut and reported); other streams are decoded whole, as before, and share `MAX_DECODED_STREAM`. The content tokenizer reads a window at a time (`UntypedIter::new_window`, `TypedIter::needs_more`, `TypedIter::committed`): an instruction that runs past the window is not decided until more data follows, which `Reader` records by flagging reads and `at_end` past the end of the data, including through the clones the parser makes. The Flate decoder keeps its own 32 KiB history, so it pauses at any output size. `ContentReader::new` reuses raw/decrypted stream bytes for budget accounting and piece creation, and `decoded_raw_within` applies the existing Whole-stream filters and limits to those bytes without decrypting twice; (16) revision 5–6 (AES-256) passwords are tried in their SASLprep form (RFC 4013, then UTF-8, at most 127 bytes, as ISO 32000-2 §7.6.4.3.3 Algorithm 2.A asks and pdfcraft-crypt writes them) and then, if that differs, as given (`aes256_passwords` in `crypto/mod.rs`, which resolves the upstream `TODO: Do the above.`; `stringprep`, already in the tree through pdfcraft-crypt, is an optional dependency enabled by `std`, and without `std` passwords are used as given): the renderer used the password as typed, so one that SASLprep changes (a decomposed `café`, a no-break space, the fullwidth `ａｂ` a Japanese IME types in full-width mode) was wrong, although pdfcraft-cos accepted it: protecting a document with such an open password was rolled back ("the password is incorrect"), and a file with such a password did not open with it.; (17) CCITT `/Rows 0` means the height is not predetermined (ISO 32000-2 Table 11), not zero rows: scanned Group 4 pages written that way rendered blank; (18) The streaming Flate decoder (`lzw_flate::flate::fallback`) copies a back-reference in runs of up to its distance (`extend_from_within`), not byte by byte. The copy reads only bytes already in the output, and a copy whose source does not exist still writes nothing, so the bytes are those a byte-by-byte copy writes. Inflating the A1 sample's content stream (278.6 MB) takes about 14% less time (0.66 s to 0.56 s); the output is identical; (19) CCITT `/Rows` below the image's `/Height` decodes to the height: a plotter driver wrote `/Rows` equal to `/Columns`, so only the first rows were decoded and the rest drawn black (#3308) (20) `stream` followed by spaces or tabs and then the line break (`stream \r\n`, as Microsoft Reporting Services writes it) starts the data after the line break; the content stream was taken to start at the space and the page rendered blank.; (21) a JP2 colour specification naming an enumerated colour space the decoder refuses (CIEJab in a veraPDF fixture, YCbCr, PhotoYCC, YPbPr, bilevel…) no longer fails the image: it is decoded again from a copy whose `colr` names the space the decoder assumes without one, greyscale, sRGB or CMYK by component count, as MuPDF and Poppler draw it (`without_refused_enumerated_colour`; the page drew blank, #3306); (22) an object stream's index of `objnum offset` pairs is parsed once per stream and kept beside its decoded data (`ObjectStreamIndexes`, `XRef::with_object_stream_indexes`), and an offset past `usize` is malformed instead of wrapping: `XRef::get_with` parsed the whole index again for every object it fetched from the stream, so reading all of a stream of N objects took N² steps (a Firefox page whose 65,542 objects share one stream took 37 s to draw and 40 s to extract its text); (23) the inline-image end-of-data look-ahead no longer rejects a candidate `EI` because a byte above 0x7F lies between it and the next one: when the candidate is the end, those bytes are the content stream that follows the image, and its strings carry such bytes (WinAnsi `(Grüße) Tj`, the two-byte codes of any Type 0 font), so the operator check that follows decides alone, bounded as in (4) (a scanned page whose OCR text held an accented word and a later capital "EI" — EINGANG, BEI, KLEIN — drew and extracted nothing after the image: 40 of 275 pages across five real documents); (24) `xref::prev_offset`: a trailer `/Prev` written as a reference `N G R` is read as the offset that object holds when the section's own entries place it (iText's `indirectPrev.pdf`; each section's entries are read before its `Prev` is followed), and otherwise as offset N (tiff2pdf; JHOVE's PDF-HUL-157 files, #2366), as `pdfcraft-cos` and other readers take it, so the earlier sections' pages are seen instead of one; (25) direct arrays and dictionaries share a reader-local nesting cap of 128, counted while skipping and while parsing a dictionary (its lenient repair reads, rather than skips, an object found where a key belongs, so `<< << …` recursed through that read); deeper objects return `None` instead of overflowing the stack (including unused trailer values, trailer entries, content operands and marked-content properties); a chain of 65 direct stitching functions now stops at this cap, before the 64-function guard; (26) an xref stream subsection must be backed by its decoded data: all-zero `/W` widths (an entry reads no bytes) and a declared count longer than the data are refused before any entry is inserted, where a tiny stream with a large `/Size` or `/Index` count inserted that many entries (from storytold/craft-libs#239); a subsection running past object number `i32::MAX` is refused instead of overflowing or turning negative; (27) an xref stream's type field is read big-endian at its full `/W[0]` width, where only its first byte was read and wider fields shifted every later entry, and `/W[0]` over 8 bytes is refused like `/W[1]` (from storytold/craft-libs#241); (28) a content instruction keeps up to 256 operands (`MAX_OPERANDS`) instead of 10, and past that drops the oldest instead of ending the content stream: the eleventh operand of a twelve-ink DeviceN `scn` (ISO 32000-2 §8.6.6.5 allows any number of components) blanked the rest of the page (#4260); (29) a `/JBIG2Globals` stream that is itself JBIG2 isn't decoded: globals are raw segments, and a chain of 16,000 such streams overflowed the stack, which aborts the process; (31) `set_page_jbig2_work`: the JBIG2 images decoded on a thread spend one allowance, set per page, 2^30 steps (`MAX_PAGE_JBIG2_WORK`), so a page of many images made to reach their own limit stops at one, and what they cost to read counts, whether they decode or not; unbounded until a page sets it; (32) a group of array items read together (a tuple: an image's `/Decode`, `/W`, `/W2`, and a function's `/Domain`, `/Range`, `/Encode` and `/Decode`) can start with a reference: it was taken for the whole group and lost, which ended the array, so an image with `/Decode [7 0 R 0]` didn't draw and CID widths fell back to `/DW`; (33) an object that holds only a reference stands for what that names, stored on its own or in an object stream, up to 32 references: read as a number it was the reference's own object number, so a `/Decode` or `/W` entry naming such an object took a wrong value, and as anything else it was lost. Past 32, as any cycle goes, it reads as a missing object. An object stream isn't reached that way, as its checks are of the object named; reading its index counts against the 64 nested loads and reads its dictionary alone, as a chain of them, each with a member of the next for its `/N`, `/First` or `/Length`, ran out of stack; and a comment inside a reference counts as white space; (34) a table rebuilt by scanning the file (`fallback_xref_map`) finds an object by its number when its header names another generation, including one no generation may be (`3 18446744073709551616 obj` in `poppler-85140-0.pdf`, #5580): the header is read leniently only in that scan (`read_object_header`), the object is also filed under generation 0 unless a real `N 0 obj` holds that key (a later header for the number moves the stand-in), `XRef::get_with` matches by object number only in a repaired table, and `XRef::objects` yields each location once | (1) `cos/tests/writer.rs::page_tree_loops_in_object_streams_do_not_crash_readers`; (2) `raster::tests::absurd_mask_and_fax_sizes_are_refused_before_decoding`; (3) `raster::tests::absurd_predictor_columns_and_line_widths_render`; (4) `raster::tests::inline_images_with_ei_in_their_data_parse_in_linear_time`, `content::tests::inline_image_ei_may_be_followed_by_a_delimiter`; (5) `raster::tests::indirect_appearance_state_dict_is_not_another_stream`; (6) `raster::tests::jpx_stream_bit_depth_boundaries_do_not_panic`, `raster::tests::jpx_32_bit_image_does_not_prevent_other_page_content_from_rendering`, `raster::tests::jpx_gray_rgb_and_alpha_keep_eight_bit_samples`, `raster::tests::jpx_streams_reject_out_of_range_bit_depths`; (7) `raster::tests::decompression_bombs_stop_at_the_stream_limit`, `raster::tests::page_contents_and_nested_forms_share_one_budget`; (8) `inspect::tests::jspdf_high_precision_mediabox_does_not_overflow_dimensions`, `object::number::tests::high_precision_float_does_not_overflow_mantissa` (hayro-syntax); (9) `object::dict::lazy_index_tests`, `raster::tests::large_resource_indexes_preserve_pixels_and_text`; (10) `data::tests`, `xref::tests::concurrent_cold_object_stream_reads_all_succeed`, `raster::tests::shared_parser_is_lazy_and_initializes_once_under_contention`; (11) `data::tests::panicking_initializer_wakes_waiters_and_prevents_retries`, `xref::tests::concurrent_cold_object_stream_reads_all_succeed`; (12) `util::tests::segment_list_rejects_out_of_range_indices`; (13) `raster::tests::user_unit_scales_the_page_and_everything_on_it`; (14) `crypto::pdfcraft_password_tests`, `pdfcraft-engine` `tests::legacy_passwords_with_turkish_characters_open`; (15) `content::tests::windows_keep_every_instruction_at_every_split`, `content::tests::windows_keep_every_instruction_in_pieces_of_any_size`, `content::tests::windows_match_the_slice_on_random_content`, `content::tests::windows_drop_blanks_and_keep_a_comment_that_may_go_on`, `page::content_reader_tests::content_past_the_cap_is_read_to_the_end`, `page::content_reader_tests::a_stream_that_expands_past_the_budget_is_cut_there`, `page::content_reader_tests::a_small_expansion_under_the_cap_is_not_cut`, `raster::tests::page_content_past_the_old_cap_renders`, `raster::tests::content_reader_returns_the_same_bytes_in_any_pieces`, `raster::tests::content_bombs_stop_at_the_budget_and_say_so`, `page::content_reader_tests::plain_and_whole_pieces_keep_their_bytes`, `page::content_reader_tests::repeated_content_reference_counts_compressed_bytes_once`, `page::content_reader_tests::whole_cap_still_counts_later_unique_streams`; (16) `crypto::pdfcraft_password_tests::aes256_passwords_are_tried_in_their_saslprep_form_then_as_given`, `crypto::pdfcraft_password_tests::aes256_passwords_open_as_typed_and_in_their_saslprep_form`, `pdfcraft-engine` `tests::aes256_open_passwords_that_saslprep_changes_protect_and_reopen_as_typed`, `tests::aes256_files_whose_passwords_saslprep_changes_open_as_typed`; (17) `raster::tests::group4_fax_with_rows_zero_decodes_to_the_image_height`; (18) `filter::lzw_flate::tests::back_reference_longer_than_its_distance_repeats_its_source`, `back_reference_cut_by_the_limit_resumes_in_the_next_call` and `back_reference_reads_bytes_written_by_an_earlier_call`; each fails when the copy returns the wrong remaining count; (19) `raster::tests::group4_fax_with_rows_below_its_height_decodes_to_the_image_height`; (20) `raster::tests::space_before_the_line_break_after_stream_still_renders`; (21) `raster::tests::a_jp2_image_in_a_refused_enumerated_colour_space_is_drawn`, `filter::jpx::tests::refused_enumerated_colour_spaces_become_the_component_counts_default`, `filter::jpx::tests::what_the_decoder_handles_or_cant_be_read_is_left_alone` (`cargo test --manifest-path vendor/hayro-syntax/Cargo.toml`); (22) `raster::tests::objects_in_one_large_object_stream_are_read_in_linear_time`; (23) `text::tests::text_after_an_inline_image_survives_a_non_ascii_byte_before_a_later_ei`, `content::tests::an_inline_image_ends_at_its_own_ei_when_later_text_holds_a_non_ascii_byte` (hayro-syntax); (24) `xref::tests::prev_as_reference_reaches_the_earlier_section`, `xref::tests::prev_as_reference_to_an_object_holding_the_offset_reads_that_offset`, `pdfcraft-cos` `document::tests::a_prev_written_as_a_reference_still_reaches_the_earlier_sections`, `document::tests::a_prev_reference_to_an_object_holding_the_offset_reads_that_offset`; (25) `raster::tests::deeply_nested_direct_objects_do_not_overflow_the_stack`, `raster::tests::dictionaries_nested_without_keys_in_content_do_not_overflow_the_stack`, `raster::tests::dictionaries_nested_without_keys_in_the_trailer_do_not_overflow_the_stack`, `raster::tests::dictionaries_nested_without_keys_as_marked_content_properties_do_not_overflow_the_stack`, `raster::tests::dictionaries_nested_without_keys_are_read_up_to_the_nesting_cap`, `raster::tests::nesting_at_the_cap_fits_a_512_kib_stack` (native only), `raster::tests::function_limits_depth_guard_refuses_a_chain_of_references`; (26) `xref::tests::xref_stream_entries_need_backing_bytes`; (27) `xref::tests::xref_stream_type_field_uses_the_full_w0_width`, `xref::tests::xref_stream_loader_rejects_oversized_w0` (`cargo test --manifest-path vendor/hayro-syntax/Cargo.toml`); (28) `content::ops::tests::scn_with_twelve_components_keeps_them_all`, `content::ops::tests::operands_past_the_cap_keep_the_latest_in_order` (hayro-syntax), `raster::tests::devicen_inks::a_twelve_ink_devicen_colour_paints`, `raster::tests::devicen_inks::a_four_ink_devicen_colour_paints`, `raster::tests::devicen_inks::an_operator_with_hundreds_of_operands_does_not_end_the_content`; (29) `raster::tests::jbig2_globals_that_are_jbig2_themselves_are_not_decoded`, `raster::tests::jbig2_globals_are_read`, `raster::tests::the_jbig2_images_of_a_page_share_one_allowance`, `raster::tests::an_image_that_fails_spends_the_pages_allowance_too`; (32) `raster::tests::a_decode_array_with_references_in_it_is_read`, `raster::tests::a_functions_arrays_can_start_with_a_reference` (pdfcraft-render), `font::cid::width_tests::a_width_group_can_start_with_a_reference` (hayro-interpret); (33) `raster::tests::a_reference_to_a_reference_is_what_it_names`, `raster::tests::a_decode_array_with_references_in_it_is_read` (pdfcraft-render), `font::cid::width_tests::a_width_group_can_start_with_a_reference` (hayro-interpret), `xref::tests::an_object_stream_isnt_reached_through_a_reference`, `xref::tests::a_chain_of_object_streams_ends_at_the_limit_of_nested_loads`, `xref::tests::a_chain_of_object_streams_takes_no_more_work_for_two_ways_on`, `xref::tests::a_reference_held_in_either_storage_is_followed`, `xref::tests::a_page_a_reference_is_held_to_is_found_once` (hayro-syntax); (34) `raster::tests::a_rebuilt_table_finds_an_object_whose_header_names_another_generation`, `raster::tests::a_rebuilt_tables_generation_0_stand_in_is_not_a_second_object` (pdfcraft-render), `xref::tests::an_object_stream_and_its_first_member_are_both_enumerated` (hayro-syntax) | diff --git a/vendor/hayro/src/renderer.rs b/vendor/hayro/src/renderer.rs index c3d1d18520..1e46e3b79e 100644 --- a/vendor/hayro/src/renderer.rs +++ b/vendor/hayro/src/renderer.rs @@ -302,6 +302,12 @@ fn nearest_premultiplied_rgba( /// icons, QR codes and pixel art are usually far smaller and keep their hard pixels. const SMOOTHED_IMAGE_SIDE: u32 = 512; +/// PdfCraft patch: the scale above which such an image counts as enlarged. A scan drawn at its own +/// resolution can come out a few ulps above 1 (the scale is an `f32`: a 192 dpi scan on a 2× +/// display at 100%), and sampled bilinearly at a half-pixel offset it blurred by half a pixel. +/// Below 1% nearest-neighbour repeats at most one sample in a hundred. +const SMOOTHED_IMAGE_SCALE: f32 = 1.01; + /// PdfCraft patch: fixed-point unit of [`area_shrink`]'s weights. const AREA_ONE: u64 = 1 << 16; @@ -365,60 +371,132 @@ impl AreaAxis { /// weighted by how much of it it covers. Nothing is invented between samples, which is what /// `/Interpolate` governs (ISO 32000-2 §8.9.5.3, for images drawn above their resolution), and /// nothing is dropped: nearest-neighbour keeps one sample in N per axis, so a scan shown at screen -/// size lost the strokes of its text. Premultiplied input stays premultiplied. Work memory is one -/// source row of `u32` accumulators, the horizontal taps (at most `sw + 2 × tw`) and a vertically -/// shrunk copy no larger than the source; a side over `MAX_RESAMPLING_SOURCE_SIDE` is refused, so -/// none of it exceeds a few times one source row. `None` if nothing shrinks, the sizes don't fit -/// or a buffer cannot be allocated. +/// size lost the strokes of its text. Premultiplied input stays premultiplied. `None` if nothing +/// shrinks, the sizes don't fit or a buffer cannot be allocated. fn area_shrink(data: &[u8], source: (u32, u32), target: (u32, u32)) -> Option> { + if data.len() != source_byte_len(source.0, source.1, N, MAX_IMAGE_PIXELS)? { + return None; + } + let row_len = (source.0 as usize).checked_mul(N)?; + area_shrink_rows::(source, target, |s, row| { + let start = s.checked_mul(row_len)?; + row.copy_from_slice(data.get(start..start.checked_add(row_len)?)?); + Some(()) + }) +} + +/// PdfCraft patch: [`area_shrink`] of a colour image under its same-size soft mask (a stencil's +/// ink under the stencil, or an image under a mask sampled onto its grid), premultiplied a source +/// row at a time as it is averaged. The result equals premultiplying the expanded RGBA and +/// shrinking that, without building it: a 10718 × 13701 stencil shown at page size peaked at +/// 1.16 GB with the full-resolution RGBA, which is 587 MB. +fn area_shrink_premultiplied( + color: &ImageData, + alpha: &[u8], + source: (u32, u32), + target: (u32, u32), +) -> Option> { + let (samples, channels) = match color { + ImageData::Luma(luma) => (luma.data.as_slice(), 1), + ImageData::Rgb(rgb) => (rgb.data.as_slice(), 3), + }; + if samples.len() != source_byte_len(source.0, source.1, channels, MAX_IMAGE_PIXELS)? + || alpha.len() != source_byte_len(source.0, source.1, 1, MAX_IMAGE_PIXELS)? + { + return None; + } + let width = source.0 as usize; + area_shrink_rows::<4>(source, target, |s, row| { + let start = s.checked_mul(width)?; + let end = start.checked_add(width)?; + let colors = samples.get(start.checked_mul(channels)?..end.checked_mul(channels)?)?; + let alphas = alpha.get(start..end)?; + for ((out, px), a) in row.chunks_exact_mut(4).zip(colors.chunks_exact(channels)).zip(alphas) { + let [r, g, b] = match px { + [v] => [*v; 3], + [r, g, b] => [*r, *g, *b], + _ => return None, + }; + out.copy_from_slice(&AlphaColor::from_rgba8(r, g, b, *a).premultiply().to_rgba8().to_u8_array()); + } + Some(()) + }) +} + +/// PdfCraft patch: the area shrink behind [`area_shrink`], reading source row `s` through +/// `read_row(s, row)` into a buffer of one row. Each target row averages the source rows it +/// covers, and that row is shrunk across at once with taps worked out once per image, so work +/// memory is one source row of samples, one of `u32` accumulators, one shrunk row and the +/// horizontal taps (at most `sw + 2 × tw`): never the image. A side over +/// `MAX_RESAMPLING_SOURCE_SIDE` is refused, so none of it exceeds a few times one source row. +fn area_shrink_rows( + source: (u32, u32), + target: (u32, u32), + mut read_row: impl FnMut(usize, &mut [u8]) -> Option<()>, +) -> Option> { let ((sw, sh), (tw, th)) = (source, target); // A side over MAX_RESAMPLING_SOURCE_SIDE would make the accumulator row and the tap table // several times larger than the samples (a 2^27 × 2 image of zeros is a tiny file). - if tw == 0 || th == 0 || tw > sw || th > sh || sw > MAX_RESAMPLING_SOURCE_SIDE || sh > MAX_RESAMPLING_SOURCE_SIDE { - return None; - } - let (sw_us, tw_us) = (sw as usize, tw as usize); - if data.len() != source_byte_len(sw, sh, N, MAX_IMAGE_PIXELS)? { + if tw == 0 + || th == 0 + || tw > sw + || th > sh + || (tw, th) == (sw, sh) + || sw > MAX_RESAMPLING_SOURCE_SIDE + || sh > MAX_RESAMPLING_SOURCE_SIDE + { return None; } + let row_len = (sw as usize).checked_mul(N)?; + let zeroed = |len: usize| { + let mut v = image_buffer(len)?; + v.resize(len, 0); + Some(v) + }; + let (mut source_row, mut shrunk_row) = (zeroed(row_len)?, image_buffer(row_len)?); + let mut acc: Vec = Vec::new(); + acc.try_reserve_exact(row_len).ok()?; + acc.resize(row_len, 0); + // The taps depend only on the widths, so they are worked out once, not again for each row. + let taps = if tw == sw { None } else { Some(AreaAxis::new(sw, tw)?) }; + let out_len = match taps { + None => source_byte_len(tw, th, N, MAX_IMAGE_PIXELS)?, + Some(_) => image_byte_len(tw, th, N, MAX_IMAGE_PIXELS)?, + }; + let mut out = image_buffer(out_len)?; // Weights sum to AREA_ONE (2^16) and samples are at most 255, so every sum fits in a u32. let round = |acc: u32| ((acc + (AREA_ONE as u32) / 2) >> 16).min(255) as u8; - // Vertical pass first: groups of whole source rows averaged into each target row. Rows are - // contiguous, so this pass is cheap, and it leaves the horizontal pass, which gathers samples - // per target, a third as many rows when a scan shrinks threefold. - let tall = if th == sh { - None - } else { - let row_len = sw_us * N; - let mut rows = image_buffer(source_byte_len(sw, th, N, MAX_IMAGE_PIXELS)?)?; - let mut acc: Vec = Vec::new(); - acc.try_reserve_exact(row_len).ok()?; - acc.resize(row_len, 0); - for t in 0..th { + // The last source row read: neighbouring target rows share the row on their boundary. + let mut loaded = None; + for t in 0..th { + // Vertical: the whole source rows target row `t` covers, averaged. + let row: &[u8] = if th == sh { + read_row(t as usize, &mut source_row)?; + &source_row + } else { acc.fill(0); for (s, w) in area_taps(sh, th, t) { - let row = data.get(s * row_len..(s + 1) * row_len)?; + if loaded != Some(s) { + read_row(s, &mut source_row)?; + loaded = Some(s); + } // A weight is at most AREA_ONE, so it fits in a u32. let w = w as u32; - for (a, v) in acc.iter_mut().zip(row) { + for (a, v) in acc.iter_mut().zip(&source_row) { *a += w * u32::from(*v); } } - rows.extend(acc.iter().map(|a| round(*a))); - } - Some(rows) - }; - if tw == sw { - return tall; - } - let rows = tall.as_deref().unwrap_or(data); - // Horizontal pass: each of those rows shrunk to `tw` samples. The taps depend only on the - // widths, so they are worked out once, not again for each row. - let taps = AreaAxis::new(sw, tw)?; - let mut out = image_buffer(image_byte_len(tw, th, N, MAX_IMAGE_PIXELS)?)?; - for row in rows.chunks_exact(sw_us * N) { - for t in 0..tw_us { - let (first, weights) = taps.get(t)?; + shrunk_row.clear(); + shrunk_row.extend(acc.iter().map(|a| round(*a))); + &shrunk_row + }; + // Horizontal: that row shrunk to `tw` samples. + let Some(taps) = &taps else { + out.extend_from_slice(row); + continue; + }; + for x in 0..tw as usize { + let (first, weights) = taps.get(x)?; let samples = row.get(first * N..(first + weights.len()) * N)?; let mut acc = [0_u32; N]; for (w, px) in weights.iter().zip(samples.chunks_exact(N)) { @@ -1263,7 +1341,7 @@ impl Renderer { // nothing, and the stretched side would otherwise be drawn nearest-neighbour. if !interpolate && !self.in_type3_glyph - && (x_scale > 1.0 || y_scale > 1.0) + && (x_scale > SMOOTHED_IMAGE_SCALE || y_scale > SMOOTHED_IMAGE_SCALE) && source_width.min(source_height) >= SMOOTHED_IMAGE_SIDE { quality = ImageQuality::Medium; @@ -1295,9 +1373,12 @@ impl Renderer { }; let needs_resize = (new_width, new_height) != (source_width, source_height); - // PdfCraft patch (#624): a minified Type 3 glyph image with a soft mask that is not - // interpolated is nearest-sampled straight into premultiplied target RGBA. That skips the - // full-resolution RGBA copy below and the premultiply pass after it. + // PdfCraft patch (#624): a minified image with a soft mask that is neither interpolated nor + // area-averaged (a Type 3 glyph image, or a source wider than MAX_RESAMPLING_SOURCE_SIDE) + // is nearest-sampled straight into premultiplied target RGBA. That skips the + // full-resolution RGBA copy below and the premultiply pass after it. PdfCraft patch: an + // area-averaged one (a stencil's ink, or an image under its mask) skips them too, and is + // premultiplied a source row at a time as it is averaged (`area_shrink_premultiplied`). let direct = match &alpha_data { Some(alpha) if needs_resize && !interpolate && !area => nearest_premultiplied_rgba( &image_data, @@ -1305,6 +1386,12 @@ impl Renderer { (source_width, source_height), (new_width, new_height), ), + Some(alpha) if needs_resize && area => area_shrink_premultiplied( + &image_data, + &alpha.data, + (source_width, source_height), + (new_width, new_height), + ), _ => None, }; let mut premultiplied = direct.is_some(); @@ -2935,9 +3022,89 @@ mod snap_tests { #[cfg(test)] mod area_shrink_tests { use super::{ - AREA_ONE, AreaAxis, LumaData, MAX_IMAGE_PIXELS, area_shrink, area_taps, mask_on_image_grid, + AREA_ONE, AreaAxis, ImageData, LumaData, MAX_IMAGE_PIXELS, RgbData, area_shrink, + area_shrink_premultiplied, area_taps, mask_on_image_grid, premultiply_rgba, }; + /// Deterministic samples that vary from one to the next. + fn noise(len: usize, seed: u32) -> Vec { + let mut x = seed.wrapping_mul(2_654_435_761) | 1; + (0..len) + .map(|_| { + x ^= x << 13; + x ^= x >> 17; + x ^= x << 5; + (x >> 24) as u8 + }) + .collect() + } + + /// The shrink as two whole passes, every row averaged down and then every row across, each + /// rounded to a byte: what `area_shrink` did before it worked a row at a time. + fn two_pass_shrink(data: &[u8], (sw, sh): (u32, u32), (tw, th): (u32, u32)) -> Vec { + let round = |acc: u64| ((acc + AREA_ONE / 2) >> 16).min(255) as u8; + let row_len = sw as usize * N; + let tall: Vec = (0..th) + .flat_map(|t| { + (0..row_len).map(move |i| { + round(area_taps(sh, th, t).map(|(s, w)| w * u64::from(data[s * row_len + i])).sum()) + }) + }) + .collect(); + tall.chunks_exact(row_len) + .flat_map(|row| { + (0..tw).flat_map(move |t| { + (0..N).map(move |c| round(area_taps(sw, tw, t).map(|(s, w)| w * u64::from(row[s * N + c])).sum())) + }) + }) + .collect() + } + + /// Working a row at a time gives the same bytes as two whole passes, in both directions, one + /// direction alone, and with source rows shared by neighbouring targets. + #[test] + fn a_row_at_a_time_shrink_equals_two_whole_passes() { + let sizes = [((7, 5), (3, 2)), ((10, 3), (6, 3)), ((4, 9), (4, 4)), ((13, 11), (1, 1)), ((31, 17), (12, 7))]; + for (i, (source, target)) in sizes.into_iter().enumerate() { + let gray = noise((source.0 * source.1) as usize, i as u32); + assert_eq!(area_shrink::<1>(&gray, source, target), Some(two_pass_shrink::<1>(&gray, source, target)), "{source:?} → {target:?}"); + let rgba = noise((source.0 * source.1 * 4) as usize, i as u32 + 99); + assert_eq!(area_shrink::<4>(&rgba, source, target), Some(two_pass_shrink::<4>(&rgba, source, target)), "{source:?} → {target:?}"); + } + } + + /// Premultiplying a row at a time as it is averaged gives the same bytes as premultiplying the + /// whole expanded RGBA first and shrinking that, for gray and RGB colour. + #[test] + fn premultiplying_row_by_row_equals_premultiplying_the_whole_image() { + let sizes = [((7, 5), (3, 2)), ((10, 3), (6, 3)), ((4, 9), (4, 4)), ((31, 17), (12, 7))]; + for (i, (source, target)) in sizes.into_iter().enumerate() { + let (w, h) = source; + let pixels = (w * h) as usize; + let alpha = noise(pixels, i as u32 + 7); + let gray = noise(pixels, i as u32 + 11); + let rgb = noise(pixels * 3, i as u32 + 13); + let colors = [ + ImageData::Luma(LumaData { data: gray.clone(), width: w, height: h, interpolate: false, scale_factors: (1.0, 1.0) }), + ImageData::Rgb(RgbData { data: rgb.clone(), width: w, height: h, interpolate: false, scale_factors: (1.0, 1.0) }), + ]; + for color in &colors { + let mut whole: Vec = match color { + ImageData::Luma(_) => gray.iter().zip(&alpha).flat_map(|(g, a)| [*g, *g, *g, *a]).collect(), + ImageData::Rgb(_) => rgb.chunks_exact(3).zip(&alpha).flat_map(|(c, a)| [c[0], c[1], c[2], *a]).collect(), + }; + premultiply_rgba(&mut whole); + assert_eq!( + area_shrink_premultiplied(color, &alpha, source, target), + area_shrink::<4>(&whole, source, target), + "{source:?} → {target:?}" + ); + } + let short = alpha.get(1..).unwrap_or_default(); + assert_eq!(area_shrink_premultiplied(&colors[0], short, source, target), None, "alpha shorter than the image"); + } + } + fn mask(data: &[u8], width: u32, height: u32) -> LumaData { LumaData { data: data.to_vec(), width, height, interpolate: false, scale_factors: (1.0, 1.0) } } From 51b68a8b356a30588eb1e3b733d3f3c7d8e0880c Mon Sep 17 00:00:00 2001 From: KenM76 Date: Wed, 4 Nov 2026 05:12:00 -0500 Subject: [PATCH 35/53] M12: prc-render: overflow-free projection and edges, shared near cuts, decode ceiling - View::project divides by z (or the orthographic half-height) before it multiplies, so an on-screen corner no longer turns NaN when x times the focal length, or the half-height times the aspect ratio, overflows. - View::clip cuts each edge from the same end whichever triangle meets it, so two triangles sharing an edge across the near plane get one cut point. - Scene::draw takes each edge with vec3::span, which halves the difference when it overflows, so a finite triangle wider than f64::MAX keeps its face normal. - assemble charges each file structure's meshes to the assembly ceilings as they are decoded, so decoding stops at the structure that passes one. Tests, each shown failing with its fix reverted: a_shared_edge_is_cut_at_the_same_point_from_either_triangle, an_on_screen_triangle_is_drawn_when_projecting_it_would_overflow, a_triangle_whose_edges_overflow_is_drawn, span_keeps_the_direction_of_a_difference_that_overflows, decoding_stops_at_the_structure_that_passes_the_ceiling. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SqmHYx8FiWM5PcsoiErgp9 --- crates/prc-render/src/assemble.rs | 22 ++++++++++++++++-- crates/prc-render/src/assemble/tests.rs | 13 +++++++++++ crates/prc-render/src/raster.rs | 30 ++++++++++++++++++++----- crates/prc-render/src/render.rs | 4 ++-- crates/prc-render/src/render/tests.rs | 21 +++++++++++++++++ crates/prc-render/src/vec3.rs | 12 ++++++++++ 6 files changed, 93 insertions(+), 9 deletions(-) diff --git a/crates/prc-render/src/assemble.rs b/crates/prc-render/src/assemble.rs index 7fbb9bd914..c9264d620e 100644 --- a/crates/prc-render/src/assemble.rs +++ b/crates/prc-render/src/assemble.rs @@ -49,9 +49,10 @@ pub fn place(mesh: &TriangleMesh, matrix: &Matrix, colour: Option<[u8; 4]>, tria /// # Errors /// [`RenderError::Model`] when a tessellation section or the tree cannot be read, or the tree /// places a tessellation its file structure does not have; [`RenderError::TooManyTriangles`] past -/// [`MAX_ASSEMBLED_TRIANGLES`] and [`RenderError::TooManyVertices`] past [`MAX_ASSEMBLED_VERTICES`]. +/// [`MAX_ASSEMBLED_TRIANGLES`] and [`RenderError::TooManyVertices`] past [`MAX_ASSEMBLED_VERTICES`], +/// counting both what is decoded and what is placed. pub fn assemble(prc: &PrcFile) -> Result, RenderError> { - let tessellations = prc.file_structures.iter().map(|fs| fs.tessellations()).collect::, _>>()?; + let tessellations = decode(prc.file_structures.iter().map(|fs| fs.tessellations()), Budget::default())?; let mut budget = Budget::default(); let mut out = Vec::new(); if !prc.stores_tree() { @@ -72,6 +73,23 @@ pub fn assemble(prc: &PrcFile) -> Result, RenderError> { Ok(out) } +/// Each file structure's tessellations, decoded one structure at a time and charged to `budget` +/// as they are, so decoding stops at the structure that passes a ceiling. +pub(crate) fn decode( + structures: impl IntoIterator, pdfcraft_prc::PrcError>>, + mut budget: Budget, +) -> Result>, RenderError> { + let mut out = Vec::new(); + for tessellations in structures { + let tessellations = tessellations?; + for mesh in tessellations.iter().filter_map(as_mesh) { + budget.spend(mesh)?; + } + out.push(tessellations); + } + Ok(out) +} + /// One placement as [`place_items`] reads it: a tessellation by file structure and index, where /// it goes and how it is coloured. pub(crate) struct Item<'a> { diff --git a/crates/prc-render/src/assemble/tests.rs b/crates/prc-render/src/assemble/tests.rs index 6947e78033..5018bdbe8c 100644 --- a/crates/prc-render/src/assemble/tests.rs +++ b/crates/prc-render/src/assemble/tests.rs @@ -112,3 +112,16 @@ fn a_placement_past_the_triangle_budget_is_an_error() { assert!(matches!(r, Err(RenderError::TooManyTriangles))); assert!(out.is_empty()); } + +#[test] +fn decoding_stops_at_the_structure_that_passes_the_ceiling() { + let pulled = std::cell::Cell::new(0); + let structures = std::iter::repeat_with(|| { + pulled.set(pulled.get() + 1); + Ok(vec![Tessellation::Mesh(triangle())]) + }) + .take(3); + let nearly_full = Budget { triangles: MAX_ASSEMBLED_TRIANGLES - 1, vertices: 0 }; + assert_eq!(decode(structures, nearly_full).err(), Some(RenderError::TooManyTriangles)); + assert_eq!(pulled.get(), 2, "the third structure is never decoded"); +} diff --git a/crates/prc-render/src/raster.rs b/crates/prc-render/src/raster.rs index e30105352c..4427df5973 100644 --- a/crates/prc-render/src/raster.rs +++ b/crates/prc-render/src/raster.rs @@ -70,9 +70,12 @@ impl View { out.push(p); } if p_in != q_in { - let t = (near - z(p)) / (z(q) - z(p)); - let mut cut = p; - for (c, d) in cut.iter_mut().zip(q) { + // Cut from the same end whichever way round the edge is met, so two triangles + // sharing it get the same cut point. + let (s, e) = if first(&p, &q) { (p, q) } else { (q, p) }; + let t = (near - z(s)) / (z(e) - z(s)); + let mut cut = s; + for (c, d) in cut.iter_mut().zip(e) { *c += t * (d - *c); } out.push(cut); @@ -85,13 +88,19 @@ impl View { /// in screen space and larger for nearer points. pub(crate) fn project(&self, [x, y, z, r, g, b]: ViewCorner, width: u32, height: u32) -> Corner { let (nx, ny, key) = match *self { - View::Perspective { focal, aspect, .. } => (x * focal / (z * aspect), y * focal / z, 1.0 / z), - View::Orthographic { half_height, aspect } => (x / (half_height * aspect), y / half_height, -z), + // Divided before multiplied, so an on-screen point's intermediate products stay finite. + View::Perspective { focal, aspect, .. } => (x / z * focal / aspect, y / z * focal, 1.0 / z), + View::Orthographic { half_height, aspect } => (x / half_height / aspect, y / half_height, -z), }; Corner { at: [(nx + 1.0) / 2.0 * f64::from(width), (1.0 - ny) / 2.0 * f64::from(height), key], shade: [r, g, b] } } } +/// Whether view corner `a`'s position comes before `b`'s, ordering x, then y, then z. +fn first(a: &ViewCorner, b: &ViewCorner) -> bool { + a.iter().zip(b).take(3).map(|(p, q)| p.total_cmp(q)).find(|o| o.is_ne()).is_none_or(std::cmp::Ordering::is_lt) +} + /// The image being drawn: `rgba` holds `width * height` straight-RGBA pixels and `depth` their /// depth keys, row by row from the top. pub(crate) struct Target<'b> { @@ -259,6 +268,17 @@ mod tests { assert_eq!(channel(127.5), 128); } + #[test] + fn a_shared_edge_is_cut_at_the_same_point_from_either_triangle() { + // Cutting this edge from P towards Q and from Q towards P rounds differently. + let view = View::Perspective { focal: 1.0, aspect: 1.0, near: 0.5 }; + let (p, q) = ([0.1, 0.1, 0.2, 1.0, 1.0, 1.0], [0.7, 0.2, 1.6, 1.0, 1.0, 1.0]); + let one = view.clip([p, q, [0.0, 3.0, 2.0, 1.0, 1.0, 1.0]]); + let other = view.clip([q, p, [3.0, 0.0, 2.0, 1.0, 1.0, 1.0]]); + let shared = one.iter().filter(|a| other.iter().any(|b| a[..3] == b[..3])).count(); + assert_eq!(shared, 2, "Q and the cut on PQ: {one:?} {other:?}"); + } + fn corner(x: f64, y: f64, key: f64, shade: f64) -> Corner { Corner { at: [x, y, key], shade: [shade; 3] } } diff --git a/crates/prc-render/src/render.rs b/crates/prc-render/src/render.rs index 9d7da84ecd..a27cfbe5fa 100644 --- a/crates/prc-render/src/render.rs +++ b/crates/prc-render/src/render.rs @@ -5,7 +5,7 @@ use crate::WorldMesh; use crate::raster::{Corner, Target, View, ViewCorner, at}; -use crate::vec3::{cross, dot, length, normalize, scale, sub, unit_cross}; +use crate::vec3::{cross, dot, length, normalize, scale, span, sub, unit_cross}; use pdfcraft_prc::TriangleMesh; /// The largest image [`render`] draws, in pixels (width × height). @@ -324,7 +324,7 @@ impl Scene { let Some(world) = mesh.triangles.get(t).and_then(|tri| corners(mesh, tri)) else { return Ok(()); }; - let Some(face) = unit_cross(sub(at3(&world, 1), at3(&world, 0)), sub(at3(&world, 2), at3(&world, 0))) else { + let Some(face) = unit_cross(span(at3(&world, 0), at3(&world, 1)), span(at3(&world, 0), at3(&world, 2))) else { return Ok(()); }; let stored = stored_normals(mesh, t); diff --git a/crates/prc-render/src/render/tests.rs b/crates/prc-render/src/render/tests.rs index 24a2d6b936..df089bfea5 100644 --- a/crates/prc-render/src/render/tests.rs +++ b/crates/prc-render/src/render/tests.rs @@ -308,3 +308,24 @@ fn a_small_model_far_from_the_origin_can_be_fitted() { let camera = Camera::fit_meshes(&[WorldMesh::new(m)], [1.0, 1.0, 1.0], [0.0, 1.0, 0.0], false, 1.0); assert!(camera.is_ok(), "{camera:?}"); } + +#[test] +fn a_triangle_whose_edges_overflow_is_drawn() { + // Its corners are finite but 2e308 apart, so the plain edge differences are infinite. + let m = mesh(vec![[-1e308, -1e308, 0.0], [1e308, -1e308, 0.0], [0.0, 1e308, 0.0]], vec![[0, 1, 2]]); + let image = render(&[WorldMesh::new(m)], &ortho(1e308), &small()).unwrap(); + assert_eq!(pixel(&image, 20, 20), [190, 192, 200, 255]); +} + +#[test] +fn an_on_screen_triangle_is_drawn_when_projecting_it_would_overflow() { + let wide = RenderOptions { width: 30, height: 10, ..RenderOptions::default() }; + let drawn = |image: &Image| image.rgba.chunks(4).filter(|px| px != &[255, 255, 255, 255]).count(); + // A tiny field of view: x times the focal length (about 1.1e300) overflows, though x over z does not. + let camera = Camera { eye: [0.0; 3], target: [0.0, 0.0, 1e308], up: [0.0, 1.0, 0.0], projection: Projection::Perspective { fov_y: 1e-298 } }; + let far = mesh(vec![[1.5e8, -3e7, 1e308], [2.5e8, -3e7, 1e308], [2e8, 3e7, 1e308]], vec![[0, 1, 2]]); + assert!(drawn(&render(&[WorldMesh::new(far)], &camera, &wide).unwrap()) > 0); + // Half the orthographic height times the aspect ratio overflows. + let big = mesh(vec![[-5e307, -5e307, 0.0], [5e307, -5e307, 0.0], [0.0, 5e307, 0.0]], vec![[0, 1, 2]]); + assert_eq!(pixel(&render(&[WorldMesh::new(big)], &ortho(1.5e308), &wide).unwrap(), 15, 5), [190, 192, 200, 255]); +} diff --git a/crates/prc-render/src/vec3.rs b/crates/prc-render/src/vec3.rs index 3bb438438a..bdbed70aed 100644 --- a/crates/prc-render/src/vec3.rs +++ b/crates/prc-render/src/vec3.rs @@ -8,6 +8,12 @@ pub(crate) fn sub([ax, ay, az]: Vec3, [bx, by, bz]: Vec3) -> Vec3 { [ax - bx, ay - by, az - bz] } +/// The direction from `from` to `to`: `to - from`, halved when the full difference overflows. +pub(crate) fn span(from: Vec3, to: Vec3) -> Vec3 { + let d = sub(to, from); + if d.iter().all(|c| c.is_finite()) { d } else { sub(scale(to, 0.5), scale(from, 0.5)) } +} + /// `a` times `s`. pub(crate) fn scale(a: Vec3, s: f64) -> Vec3 { a.map(|c| c * s) @@ -95,6 +101,12 @@ mod tests { assert_eq!(unit_cross([1e-170, 0.0, 0.0], [0.0, 1e160, 0.0]), Some([0.0, 0.0, 1.0])); } + #[test] + fn span_keeps_the_direction_of_a_difference_that_overflows() { + assert_eq!(span([1.0, 2.0, 3.0], [4.0, 0.0, 3.0]), [3.0, -2.0, 0.0]); + assert_eq!(span([-1e308, 0.0, 0.0], [1e308, 1e308, 0.0]), [1e308, 5e307, 0.0]); + } + #[test] fn length_survives_huge_and_tiny_vectors() { let near = |l: f64, want: f64| (l - want).abs() <= want * 1e-15; From c2d8c5f32782d716ac944572190b650a336a9d07 Mon Sep 17 00:00:00 2001 From: KenM76 Date: Wed, 4 Nov 2026 06:53:36 -0500 Subject: [PATCH 36/53] M12: prc-render: fully transparent triangles are not framed Review fix from #6087. Bounds::of and Camera::fit_meshes framed every finite triangle, but render skips those whose colour has zero alpha, so a clear, distant mesh could shrink the visible model in the image. The vertices they frame now skip fully transparent triangles, as render does. Test, shown failing with the fix reverted: a_fully_transparent_triangle_does_not_widen_the_box_or_the_fit. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SqmHYx8FiWM5PcsoiErgp9 --- crates/prc-render/src/render.rs | 17 +++++++++++++---- crates/prc-render/src/render/tests.rs | 16 ++++++++++++++++ 2 files changed, 29 insertions(+), 4 deletions(-) diff --git a/crates/prc-render/src/render.rs b/crates/prc-render/src/render.rs index a27cfbe5fa..bcaf8acced 100644 --- a/crates/prc-render/src/render.rs +++ b/crates/prc-render/src/render.rs @@ -56,7 +56,8 @@ pub struct Bounds { } impl Bounds { - /// The box around every finite vertex the meshes' triangles use, or `None` when there is none. + /// The box around every finite vertex the meshes' drawn triangles use, or `None` when there + /// is none. #[must_use] pub fn of(meshes: &[WorldMesh]) -> Option { let mut out: Option = None; @@ -83,9 +84,17 @@ impl Bounds { } } -/// Every corner of the triangles [`render`] draws: those whose three corners exist and are finite. +/// Every corner of the triangles [`render`] draws: those not fully transparent whose three +/// corners exist and are finite. fn vertices(meshes: &[WorldMesh]) -> impl Iterator + Clone + '_ { - meshes.iter().flat_map(|m| m.mesh.triangles.iter().filter_map(|tri| corners(&m.mesh, tri)).flatten()) + meshes + .iter() + .flat_map(|m| m.mesh.triangles.iter().enumerate().filter(|&(t, _)| !invisible(m, t)).filter_map(|(_, tri)| corners(&m.mesh, tri)).flatten()) +} + +/// Whether triangle `t` of `m` is fully transparent, so [`render`] does not draw it. +fn invisible(m: &WorldMesh, t: usize) -> bool { + m.colour_of(t).is_some_and(|c| c[3] == 0) } impl Camera { @@ -100,7 +109,7 @@ impl Camera { Camera::frame(bounds, corners, direction, up, (perspective, aspect)) } - /// As [`Camera::fit`], framing every vertex the meshes' triangles use rather than their box, + /// As [`Camera::fit`], framing every vertex the meshes' drawn triangles use rather than their box, /// so an oblique view is not padded by empty box corners. /// /// # Errors diff --git a/crates/prc-render/src/render/tests.rs b/crates/prc-render/src/render/tests.rs index df089bfea5..586292ef80 100644 --- a/crates/prc-render/src/render/tests.rs +++ b/crates/prc-render/src/render/tests.rs @@ -301,6 +301,22 @@ fn a_triangle_that_is_not_drawn_does_not_widen_the_box() { assert_eq!(b.max, [1.0, 1.0, 0.0]); } +#[test] +fn a_fully_transparent_triangle_does_not_widen_the_box_or_the_fit() { + // The far triangle is clear, so the camera frames the near one as if it were alone. + let m = mesh(vec![[0.0; 3], [1.0, 0.0, 0.0], [0.0, 1.0, 0.0], [1e6, 0.0, 0.0], [1e6, 1.0, 0.0]], vec![[0, 1, 2], [3, 4, 1]]); + let clear = WorldMesh { triangle_colours: Some(vec![None, Some([0, 0, 0, 0])]), ..WorldMesh::new(m.clone()) }; + assert_eq!(Bounds::of(std::slice::from_ref(&clear)).unwrap().max, [1.0, 1.0, 0.0]); + let alone = WorldMesh::new(mesh(m.positions, vec![[0, 1, 2]])); + let fit = |w: &WorldMesh| Camera::fit_meshes(std::slice::from_ref(w), [0.0, 0.0, -1.0], [0.0, 1.0, 0.0], true, 1.0).unwrap(); + assert_eq!(fit(&clear), fit(&alone)); + // A whole mesh painted clear is not framed either. + assert_eq!( + Bounds::of(&[painted(WorldMesh::new(mesh(vec![[0.0; 3], [9.0, 0.0, 0.0], [0.0, 9.0, 0.0]], vec![[0, 1, 2]])), Some([1, 2, 3, 0]))]), + None + ); +} + #[test] fn a_small_model_far_from_the_origin_can_be_fitted() { let (o, d) = (1e16, 4.0); From 52d68d7844243cd04e0b1deac38f1492e5212c20 Mon Sep 17 00:00:00 2001 From: FredApps <153173854+FredApps@users.noreply.github.com> Date: Wed, 4 Nov 2026 19:52:51 +0100 Subject: [PATCH 37/53] annot: share threed's MAX_MODEL_BYTES with model3d main's 3D-embed change (#5808) added crates/annot/src/threed.rs with its own pub MAX_MODEL_BYTES (same 256 MiB value), so both were re-exported from the crate root and pdfcraft-annot no longer compiled after merging main. model3d now uses threed's constant. --- crates/annot/src/lib.rs | 2 +- crates/annot/src/model3d.rs | 4 +--- 2 files changed, 2 insertions(+), 4 deletions(-) diff --git a/crates/annot/src/lib.rs b/crates/annot/src/lib.rs index ea7b0c227f..4f24d04b08 100644 --- a/crates/annot/src/lib.rs +++ b/crates/annot/src/lib.rs @@ -38,7 +38,7 @@ use lookup::{annots, page_ref}; pub use lookup::{ has_visible_fill_sign, is_fill_sign, isolate_annotation, isolate_annotations, page_refs, picture_rotation, set_annots, signature_image, }; -pub use model3d::{MAX_MODEL_BYTES, Model3d, Model3dKind, model_3d, model_3d_annotations}; +pub use model3d::{Model3d, Model3dKind, model_3d, model_3d_annotations}; pub(crate) use modify::n; pub use modify::{ add_reply, add_text_replacement, deletable, delete_annotation, erase_ink, fillable, goes_with, lift_annotation, move_annotation, set_contents, diff --git a/crates/annot/src/model3d.rs b/crates/annot/src/model3d.rs index 60d2df0f2a..b50342898e 100644 --- a/crates/annot/src/model3d.rs +++ b/crates/annot/src/model3d.rs @@ -2,11 +2,9 @@ //! 3D stream, or a 3D reference dictionary naming the stream another annotation shares. use super::lookup::{annots, page_ref}; +use super::threed::MAX_MODEL_BYTES; use super::{AnnotError, Dict, Document, Object}; -/// The largest decoded 3D stream [`model_3d`] reads, in bytes. -pub const MAX_MODEL_BYTES: usize = 256 << 20; - /// The format of a 3D stream, from its `/Subtype`. #[derive(Debug, Clone, Copy, PartialEq, Eq)] pub enum Model3dKind { From d22641a875b4066a17638c093e8a7db846c12e1a Mon Sep 17 00:00:00 2001 From: "Dylan M. Taylor" Date: Wed, 4 Nov 2026 17:44:52 -0500 Subject: [PATCH 38/53] Pass the Linux signing key to the release workflow (#6200) --- .github/workflows/release.yml | 1 + 1 file changed, 1 insertion(+) diff --git a/.github/workflows/release.yml b/.github/workflows/release.yml index d8549142f7..9403a1e044 100644 --- a/.github/workflows/release.yml +++ b/.github/workflows/release.yml @@ -42,3 +42,4 @@ jobs: AZURE_SIGNING_ENDPOINT: ${{ secrets.AZURE_SIGNING_ENDPOINT }} AZURE_SIGNING_ACCOUNT: ${{ secrets.AZURE_SIGNING_ACCOUNT }} AZURE_CERT_PROFILE: ${{ secrets.AZURE_CERT_PROFILE }} + LINUX_SIGNING_KEY: ${{ secrets.LINUX_SIGNING_KEY }} From 943718ea9854d217da9db7305a96087397a5ea7c Mon Sep 17 00:00:00 2001 From: FredApps <153173854+FredApps@users.noreply.github.com> Date: Thu, 5 Nov 2026 04:33:21 +0100 Subject: [PATCH 39/53] fix(ui): About credits tables fit without side-scrolling and their columns resize (#6052) * fix(ui): About credits tables fit without side-scrolling and their columns resize The Contributors table and the Models table were egui Grids inside a two-way scroll area: they opened wider than the dialog, so they scrolled sideways, and Company and Model sat almost touching. Both are now egui_extras tables. Each column opens at the width of its header and widest cell, the spare width goes to the text columns (Name; Company and Model), and every column edge can be dragged. Widening a column grows the table like a spreadsheet, and only then does it scroll sideways; the header stays put while the rows scroll. Counts and dates are right-aligned and every cell is vertically centred. The sorted column stays highlighted and its direction shows on the toolbar's sort-order button, so headers no longer reserve room for an arrow. The opening widths come from fit_widths, a pure function with unit tests that a fitted table fills its room exactly and that only the text columns give way when room is short. Co-Authored-By: Claude Opus 5.5 * Let every credits column be dragged, and refit changed tables The last column was egui's remainder column, which has no drag handle, so it could only be resized from the edge before it. It is now an ordinary column with its own handle. A table keeps its column widths, so switching to a longer name left the Name column at its old width and clipped the text. The table now goes back to the opening fit whenever that fit changes: a longer widest cell or a resized window. Dragged widths are dropped at that point. fit_widths could overflow the room when one column hit the minimum, because the other columns did not take up its share. Columns now give way in proportion to their width above the minimum. Co-Authored-By: Claude Opus 5.5 * Refit credits tables on what they measure, not the fitted widths Comparing the fitted widths missed the case it was meant for. Name is the only growing column, so a longer name takes its extra width from the spare room and the fit comes out the same: a dragged Name width stayed, clipping the longer name. It could also drop dragged widths by accident. The fit comes from the available width inside the scroll area, which scrolling can nudge, so scrolling a widened table could reset it. The table now compares the natural widths and the clip rect's width, which scrolling does not move. Co-Authored-By: Claude Opus 5.5 --------- Co-authored-by: Claude Opus 5.5 --- crates/ui-egui/src/dialogs.rs | 2 +- crates/ui-egui/src/dialogs/credits.rs | 380 ++++++++++++++++++++++---- 2 files changed, 327 insertions(+), 55 deletions(-) diff --git a/crates/ui-egui/src/dialogs.rs b/crates/ui-egui/src/dialogs.rs index d1c09dee05..0e7e2c0ed0 100644 --- a/crates/ui-egui/src/dialogs.rs +++ b/crates/ui-egui/src/dialogs.rs @@ -132,7 +132,7 @@ fn dialogs(app: &mut PdfCraftApp, ctx: &egui::Context) { Dialog::Print => 820.0, Dialog::FieldProps | Dialog::BulkFieldProps => 600.0, Dialog::Preferences => 610.0, - Dialog::About => 780.0, + Dialog::About => 820.0, Dialog::PinTool => 540.0, _ => 520.0, }); diff --git a/crates/ui-egui/src/dialogs/credits.rs b/crates/ui-egui/src/dialogs/credits.rs index ad45ec77b6..d69d545b3f 100644 --- a/crates/ui-egui/src/dialogs/credits.rs +++ b/crates/ui-egui/src/dialogs/credits.rs @@ -266,12 +266,17 @@ pub fn contributors_ui(ui: &mut egui::Ui) { let counts = crate::i18n::fmt(crate::i18n::t("{n} contributors · {c} commits"), &[("n", &list.len().to_string()), ("c", &group(TOTAL_COMMITS))]); ui.label(RichText::new(counts).small().color(t.text_muted)); ui.separator(); - egui::ScrollArea::both().id_salt("credits_list").max_height(LIST_HEIGHT).auto_shrink([false, true]).show(ui, |ui| { - if list.is_empty() { - ui.label(crate::i18n::t("No contributor data was built into this copy.")); - } else if v.table { - table(ui, &list, &mut v); - } else { + if list.is_empty() { + ui.label(crate::i18n::t("No contributor data was built into this copy.")); + } else if v.table { + // The table scrolls vertically itself, under a fixed header; it only scrolls sideways + // once its columns are dragged wider than the dialog. + ui.allocate_ui(egui::vec2(ui.available_width(), LIST_HEIGHT), |ui| { + ui.set_max_height(LIST_HEIGHT); + egui::ScrollArea::horizontal().id_salt("credits_list").auto_shrink([false, false]).show(ui, |ui| table(ui, &list, &mut v)); + }); + } else { + egui::ScrollArea::vertical().id_salt("credits_list").max_height(LIST_HEIGHT).auto_shrink([false, true]).show(ui, |ui| { ui.horizontal_wrapped(|ui| { for (i, c) in list.iter().enumerate() { if i > 0 { @@ -280,43 +285,154 @@ pub fn contributors_ui(ui: &mut egui::Ui) { ui.hyperlink_to(c.name(v.names), format!("https://github.com/{}", c.login)).on_hover_text(c.summary()); } }); - } - }); + }); + } ui.data_mut(|d| d.insert_temp(id, v)); } -fn table(ui: &mut egui::Ui, list: &[&Contributor], v: &mut View) { - egui::Grid::new("credits_table").striped(true).num_columns(SortKey::ALL.len()).show(ui, |ui| { - for k in SortKey::ALL { - let arrow = if v.key == k { if v.ascending { " ▲" } else { " ▼" } } else { "" }; - if ui - .button(RichText::new(format!("{}{arrow}", crate::i18n::t(k.label().1))).strong()) - .on_hover_text(crate::i18n::t(k.label().0)) - .clicked() - { - if v.key == k { - v.ascending = !v.ascending; - } else { - v.key = k; - v.ascending = k.default_ascending(); - } +/// How the contributor table's spare width is shared out: all of it to Name. +const CONTRIBUTOR_GROW: [f32; 10] = [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]; + +/// How the model table's spare width is shared out: to Company and Model, the text columns. +const MODEL_GROW: [f32; 6] = [1.0, 2.0, 0.0, 0.0, 0.0, 0.0]; + +/// The narrowest a column can be dragged. +const MIN_COLUMN: f32 = 28.0; + +/// The width a column needs for its header and its widest cell. +fn natural(ui: &egui::Ui, header: RichText, cells: impl Iterator) -> f32 { + let measure = |text: egui::WidgetText| text.into_galley(ui, Some(egui::TextWrapMode::Extend), f32::INFINITY, egui::TextStyle::Body).size().x; + let widest = cells.map(|c| measure(c.into())).fold(measure(header.into()), f32::max); + widest + 2.0 * ui.spacing().button_padding.x +} + +/// Opening column widths: each column's natural width, with the spare room shared out by `grow`, +/// so the columns fill `room` exactly. When there is too little room, only the growing columns +/// give way, each in proportion to its width above [`MIN_COLUMN`], and none goes below that. +fn fit_widths(natural: &[f32], grow: &[f32], room: f32) -> Vec { + let spare = room - natural.iter().sum::(); + let weight = grow.iter().sum::(); + let slack = |w: f32| (w - MIN_COLUMN).max(0.0); + let flexible: f32 = natural.iter().zip(grow).filter(|(_, g)| **g > 0.0).map(|(w, _)| slack(*w)).sum(); + natural + .iter() + .zip(grow) + .map(|(w, g)| { + if *g > 0.0 && spare >= 0.0 { + w + spare * g / weight + } else if *g > 0.0 && flexible > 0.0 { + w - (-spare).min(flexible) * slack(*w) / flexible + } else { + *w } - } - ui.end_row(); - for c in list { - ui.hyperlink_to(c.name(v.names), format!("https://github.com/{}", c.login)).on_hover_text(c.summary()); - ui.label(group(c.prs)); - ui.label(group(c.commits)); - ui.label(group(c.lines_added)); - ui.label(group(c.lines_deleted)); - ui.label(signed(c.lines_delta())); - ui.label(group(c.binary_added)); - ui.label(group(c.binary_deleted)); - ui.label(day(c.first_commit)); - ui.label(day(c.last_commit)); - ui.end_row(); - } - }); + }) + .collect() +} + +/// A striped table whose columns open at [`fit_widths`], so it fills the width and nothing scrolls +/// sideways. Every column, the last included, can be dragged wider or narrower at its right edge: +/// only a table dragged wider than the window scrolls sideways. Text too long for its column is +/// clipped. When a column's natural width or the window changes, the columns go back to a new fit, +/// dropping any dragged widths. +fn fitted_table<'a>(ui: &'a mut egui::Ui, id: &str, natural: &[f32], grow: &[f32]) -> egui_extras::TableBuilder<'a> { + // Every cell is followed by the item gap, and the vertical scrollbar keeps its margins. + let s = ui.spacing(); + let gaps = s.item_spacing.x * natural.len() as f32; + let bar = s.scroll.bar_width + s.scroll.bar_inner_margin + s.scroll.bar_outer_margin; + let widths = fit_widths(natural, grow, (ui.available_width() - gaps - bar).max(0.0)); + // The table keeps its own column widths, so a change in what they were fitted to has to clear + // them. A longer name can leave the fit itself unchanged, and scrolling can nudge the available + // width, so this compares the natural widths and the clip rect, which scrolling never moves. + let fit = egui::Id::unique(("credits_fit", id)); + let measured = (natural.to_vec(), ui.clip_rect().width()); + let refit = ui.data_mut(|d| d.get_temp::<(Vec, f32)>(fit)).as_ref() != Some(&measured); + ui.data_mut(|d| d.insert_temp(fit, measured)); + let mut t = + egui_extras::TableBuilder::new(ui).id_salt(id).striped(true).resizable(true).auto_shrink([false, false]).max_scroll_height(f32::INFINITY); + if refit { + t.reset(); + } + for w in widths { + t = t.column(egui_extras::Column::initial(w).at_least(MIN_COLUMN).clip(true)); + } + t +} + +/// Add a cell's widget right-aligned (counts, dates) or left-aligned (names), vertically centred. +fn aligned(ui: &mut egui::Ui, right: bool, add: impl FnOnce(&mut egui::Ui) -> R) -> R { + let layout = if right { egui::Layout::right_to_left(egui::Align::Center) } else { egui::Layout::left_to_right(egui::Align::Center) }; + ui.with_layout(layout, add).inner +} + +fn row_height(ui: &egui::Ui) -> f32 { + ui.text_style_height(&egui::TextStyle::Body) + 6.0 +} + +/// One contributor's table cells after the name, in `SortKey::ALL` order. +fn cells(c: &Contributor) -> [String; 9] { + [ + group(c.prs), + group(c.commits), + group(c.lines_added), + group(c.lines_deleted), + signed(c.lines_delta()), + group(c.binary_added), + group(c.binary_deleted), + day(c.first_commit).to_string(), + day(c.last_commit).to_string(), + ] +} + +/// A contributor-table header. The sorted column is highlighted; its direction shows on the +/// toolbar's sort-order button, so no arrow widens the column. +fn header_text(k: SortKey) -> RichText { + RichText::new(crate::i18n::t(k.label().1)).strong() +} + +fn table(ui: &mut egui::Ui, list: &[&Contributor], v: &mut View) { + let h = row_height(ui); + let rows: Vec<[String; 9]> = list.iter().map(|c| cells(c)).collect(); + let widths: Vec = SortKey::ALL + .iter() + .enumerate() + .map(|(i, &k)| match i.checked_sub(1) { + None => natural(ui, header_text(k), list.iter().map(|c| c.name(v.names))), + Some(col) => natural(ui, header_text(k), rows.iter().filter_map(|r| r.get(col).cloned())), + }) + .collect(); + fitted_table(ui, "credits_table", &widths, &CONTRIBUTOR_GROW) + .header(h + 4.0, |mut header| { + for k in SortKey::ALL { + header.col(|ui| { + let r = aligned(ui, k != SortKey::Name, |ui| ui.selectable_label(v.key == k, header_text(k))) + .on_hover_text(crate::i18n::t(k.label().0)); + + if r.clicked() { + if v.key == k { + v.ascending = !v.ascending; + } else { + v.key = k; + v.ascending = k.default_ascending(); + } + } + }); + } + }) + .body(|mut body| { + for (c, cells) in list.iter().zip(rows) { + body.row(h, |mut row| { + row.col(|ui| { + aligned(ui, false, |ui| ui.hyperlink_to(c.name(v.names), format!("https://github.com/{}", c.login))) + .on_hover_text(c.summary()); + }); + for text in cells { + row.col(|ui| { + aligned(ui, true, |ui| ui.label(text)); + }); + } + }); + } + }); } /// About ▸ Models: AI models credited in Co-Authored-By trailers. @@ -326,21 +442,50 @@ pub fn models_ui(ui: &mut egui::Ui) { return; } let assisted: u64 = MODELS.iter().map(|m| m.commits).max().unwrap_or(0).max(1); - egui::ScrollArea::both().id_salt("credits_models").max_height(LIST_HEIGHT).auto_shrink([false, true]).show(ui, |ui| { - egui::Grid::new("credits_models").striped(true).num_columns(6).show(ui, |ui| { - for h in MODEL_COLUMNS { - ui.label(RichText::new(crate::i18n::t(h)).strong()); - } - ui.end_row(); - for m in MODELS { - ui.label(m.company); - ui.label(m.model); - ui.label(m.version); - ui.label(group(m.commits)); - ui.label(format!("{:.0}%", 100.0 * m.commits as f64 / TOTAL_COMMITS.max(assisted) as f64)); - ui.label(format!("+{} / −{}", group(m.lines_added), group(m.lines_deleted))); - ui.end_row(); - } + let h = row_height(ui); + let titles = MODEL_COLUMNS; + // Company, Model and Version read as text; the counts line up on the right. + let right = [false, false, false, true, true, true]; + let rows: Vec<[String; 6]> = MODELS + .iter() + .map(|m| { + [ + m.company.to_string(), + m.model.to_string(), + m.version.to_string(), + group(m.commits), + format!("{:.0}%", 100.0 * m.commits as f64 / TOTAL_COMMITS.max(assisted) as f64), + format!("+{} / −{}", group(m.lines_added), group(m.lines_deleted)), + ] + }) + .collect(); + let widths: Vec = titles + .iter() + .enumerate() + .map(|(i, title)| natural(ui, RichText::new(crate::i18n::t(title)).strong(), rows.iter().filter_map(|r| r.get(i).cloned()))) + .collect(); + ui.allocate_ui(egui::vec2(ui.available_width(), LIST_HEIGHT), |ui| { + ui.set_max_height(LIST_HEIGHT); + egui::ScrollArea::horizontal().id_salt("credits_models").auto_shrink([false, false]).show(ui, |ui| { + fitted_table(ui, "credits_models", &widths, &MODEL_GROW) + .header(h + 4.0, |mut header| { + for (title, r) in titles.into_iter().zip(right) { + header.col(|ui| { + aligned(ui, r, |ui| ui.label(RichText::new(crate::i18n::t(title)).strong())); + }); + } + }) + .body(|mut body| { + for cells in rows { + body.row(h, |mut row| { + for (text, r) in cells.into_iter().zip(right) { + row.col(|ui| { + aligned(ui, r, |ui| ui.label(text)); + }); + } + }); + } + }); }); }); } @@ -421,4 +566,131 @@ mod tests { assert_eq!(day("2026-10-07T18:45:23Z"), "2026-10-07"); assert_eq!(day(""), "—"); } + + #[test] + fn columns_fill_the_room_and_spare_width_goes_to_growing_columns() { + assert_eq!(fit_widths(&[100.0, 40.0, 60.0], &[1.0, 0.0, 0.0], 300.0), [200.0, 40.0, 60.0]); + assert_eq!(fit_widths(&[100.0, 40.0, 60.0], &[1.0, 2.0, 0.0], 230.0), [110.0, 60.0, 60.0]); + let w = fit_widths(&[123.0, 45.5, 67.25], &[1.0, 0.0, 2.0], 512.0); + assert!((w.iter().sum::() - 512.0).abs() < 0.01, "a fitted table neither scrolls nor leaves a gap: {w:?}"); + } + + #[test] + fn too_little_room_shrinks_only_growing_columns() { + assert_eq!(fit_widths(&[100.0, 40.0, 60.0], &[1.0, 0.0, 0.0], 150.0), [50.0, 40.0, 60.0]); + assert_eq!(fit_widths(&[100.0, 40.0, 60.0], &[1.0, 0.0, 0.0], 0.0), [MIN_COLUMN, 40.0, 60.0]); + assert_eq!(fit_widths(&[100.0, 40.0], &[0.0, 0.0], 50.0), [100.0, 40.0]); + } + + #[test] + fn a_column_at_the_minimum_leaves_the_shortfall_to_the_others() { + let w = fit_widths(&[100.0, 300.0, 50.0], &[1.0, 1.0, 0.0], 140.0); + assert!((w.iter().sum::() - 140.0).abs() < 0.01, "the columns still fit the room: {w:?}"); + assert!(w.iter().all(|c| *c >= MIN_COLUMN), "no column goes below the minimum: {w:?}"); + assert_eq!(fit_widths(&[100.0, 300.0, 50.0], &[1.0, 1.0, 0.0], 0.0), [MIN_COLUMN, MIN_COLUMN, 50.0]); + } + + const WINDOW: f32 = 640.5; + const NATURAL: [f32; 3] = [120.0, 80.0, 80.0]; + const GROW: [f32; 3] = [1.0, 0.0, 0.0]; + + /// One frame of a table long enough to scroll; the rects of its header cells. + fn table_frame(ctx: &egui::Context, width: f32, natural: &[f32], events: Vec) -> Vec { + scrolled_frame(ctx, width, natural, 0.0, events) + } + + /// [`table_frame`], scrolled `offset` points sideways. + fn scrolled_frame(ctx: &egui::Context, width: f32, natural: &[f32], offset: f32, events: Vec) -> Vec { + let screen_rect = Some(egui::Rect::from_min_size(egui::Pos2::ZERO, egui::vec2(width, 300.0))); + let raw = egui::RawInput { screen_rect, events, ..Default::default() }; + let mut cells = Vec::new(); + let out = ctx.run_ui(raw, |ui| { + cells.clear(); + egui::ScrollArea::horizontal().auto_shrink([false, false]).horizontal_scroll_offset(offset).show(ui, |ui| { + fitted_table(ui, "test", natural, &GROW) + .header(20.0, |mut header| { + for _ in natural { + header.col(|ui| cells.push(ui.max_rect())); + } + }) + .body(|mut body| { + for _ in 0..40 { + body.row(20.0, |mut row| { + for _ in natural { + row.col(|ui| { + ui.label("x"); + }); + } + }); + } + }); + }); + }); + out.drop_without_applying_deltas(); + cells + } + + /// A new context with the table open and settled, and the rects of its header cells. + fn opened_table() -> (egui::Context, Vec) { + let ctx = egui::Context::default(); + table_frame(&ctx, WINDOW, &NATURAL, Vec::new()); + let cells = table_frame(&ctx, WINDOW, &NATURAL, Vec::new()); + (ctx, cells) + } + + /// Drags a column's right edge, which runs midway through the gap after `cell`, then lets go. + fn drag_edge(ctx: &egui::Context, cell: egui::Rect, by: f32) { + use egui::{Event, Modifiers, PointerButton}; + let start = egui::pos2(cell.right() + 4.0, cell.center().y); + let halfway = start + egui::vec2(by / 2.0, 0.0); + let end = start + egui::vec2(by, 0.0); + let moved = Event::PointerMoved; + let button = |pos, pressed| Event::PointerButton { pos, button: PointerButton::Primary, pressed, modifiers: Modifiers::NONE }; + for event in [moved(start), button(start, true), moved(halfway), moved(end), button(end, false)] { + table_frame(ctx, WINDOW, &NATURAL, vec![event]); + } + } + + #[test] + fn every_column_can_be_dragged_at_its_right_edge_and_stays_there() { + let (_, opened) = opened_table(); + for (dragged, cell) in opened.iter().enumerate() { + let (ctx, _) = opened_table(); + drag_edge(&ctx, *cell, -30.0); + table_frame(&ctx, WINDOW, &NATURAL, Vec::new()); + let after = table_frame(&ctx, WINDOW, &NATURAL, Vec::new()); + for (column, (was, now)) in opened.iter().zip(&after).enumerate() { + let want = if column == dragged { was.width() - 30.0 } else { was.width() }; + assert!((now.width() - want).abs() < 1.0, "dragging column {dragged}: column {column} is {now:?}, was {was:?}"); + } + } + } + + #[test] + fn a_changed_fit_replaces_the_column_widths() { + let (ctx, opened) = opened_table(); + drag_edge(&ctx, opened[0], -30.0); + let longer_name = table_frame(&ctx, WINDOW, &[200.0, 80.0, 80.0], Vec::new()); + assert!((longer_name[0].width() - opened[0].width()).abs() < 1.0, "a longer name drops a dragged width: {longer_name:?}"); + let wider = table_frame(&ctx, WINDOW, &[120.0, 200.0, 80.0], Vec::new()); + let widths: Vec = wider.iter().map(|c| c.width().round()).collect(); + assert!(widths.ends_with(&[200.0, 80.0]), "a longer cell widens its column: {widths:?}"); + assert!(wider[2].right() < WINDOW, "the table still fits the window: {wider:?}"); + let narrower = table_frame(&ctx, 400.0, &NATURAL, Vec::new()); + assert!(narrower[2].right() < 400.0, "a narrower window refits the table: {narrower:?}"); + } + + #[test] + fn scrolling_sideways_keeps_dragged_widths() { + let (ctx, opened) = opened_table(); + drag_edge(&ctx, opened[0], 100.0); + table_frame(&ctx, WINDOW, &NATURAL, Vec::new()); + let dragged = table_frame(&ctx, WINDOW, &NATURAL, Vec::new()); + assert!((dragged[0].width() - opened[0].width() - 100.0).abs() < 1.0, "the drag took: {dragged:?}"); + for offset in [13.3, 27.77, 41.1, 0.6] { + scrolled_frame(&ctx, WINDOW, &NATURAL, offset, Vec::new()); + let scrolled = scrolled_frame(&ctx, WINDOW, &NATURAL, offset, Vec::new()); + assert!((scrolled[0].width() - dragged[0].width()).abs() < 1.0, "scrolled {offset}: {scrolled:?}"); + } + } } From 863af864a70f18bcf780a298f1e9923ee792c2ac Mon Sep 17 00:00:00 2001 From: Mex <44814932+MMexicann@users.noreply.github.com> Date: Thu, 5 Nov 2026 09:28:43 +0100 Subject: [PATCH 40/53] M1: respect generations in PDF object references and full saves (#1143) * M1: respect generations in COS object references and full saves Resolve stale references to null instead of replacement objects with the same number. Validate complete identities in writer readers, indirect stream lengths and catalog checks, and keep generation-aware length caches. Cover cache reuse, edit overlays, snapshots, incremental and packed/classic full saves, and stream lengths with synthetic regressions and independent hayro checks. Co-Authored-By: Codex * M0: clarify refreshed CI roadmap note Co-authored-by: OpenAI Codex * M1: repair contradictory xref generation references Address maintainer feedback on the existing PR. Preserve upstream work and document the tested scope and remaining validation limits. * M1: cover repaired generations and freed-slot round trips Address maintainer feedback on the existing PR. Preserve upstream work and document tested scope and validation limits. * M1: document generation repair and bounded diagnostics Address maintainer feedback on the existing PR. Preserve upstream work and document tested scope and validation limits. * M1: record generation repair regression evidence Address maintainer feedback on the existing PR. Preserve upstream work and document tested scope and validation limits. * M1: document bounded COS generation repair Address maintainer feedback on the existing PR. Preserve upstream work and document tested scope and validation limits. * M1: record generation repair parity evidence Address maintainer feedback on the existing PR. Preserve upstream work and document tested scope and validation limits. * fix(cos): preserve repaired generations through preflight and saves Keep accepted xref/header aliases across repair-aware edits and snapshots, canonicalize full-save identities, and retain bounded generation diagnostics from older revisions. Add COS, writer/encryption, and real preflight regression coverage. * test(preflight): assert repaired aliases share saved identity Address the review nitpick by comparing reopened HeaderImage and RowImage references after damaged full saves, for classic/packed output and edited/restored snapshots. Focused preflight regression and rustfmt check pass. * M14: update certificate dependencies for RustSec advisories Bump rustls-webpki to0.103.16 and webpki-root-certs to1.0.10. Preserve all other locked packages and dependency lists, manifests, Git pins, and advisory policy. --------- Co-authored-by: MMexicann Co-authored-by: Codex Co-authored-by: FredApps --- README.md | 2 +- ROADMAP.md | 11 + crates/cos/README.md | 24 +- crates/cos/src/document/objects.rs | 160 ++++++- crates/cos/src/document/open.rs | 8 +- crates/cos/src/document/reader.rs | 31 +- crates/cos/src/document/state.rs | 8 +- crates/cos/src/document/tests.rs | 25 +- crates/cos/src/document/xref.rs | 34 +- crates/cos/src/writer.rs | 24 +- crates/cos/tests/it/generations.rs | 475 +++++++++++++++++++ crates/cos/tests/it/main.rs | 1 + crates/cos/tests/it/writer.rs | 31 +- crates/preflight/src/lib.rs | 10 +- crates/preflight/tests/generation_repairs.rs | 81 ++++ docs/roadmap.md | 10 +- parity/acrobat-features.toml | 16 +- 17 files changed, 861 insertions(+), 90 deletions(-) create mode 100644 crates/cos/tests/it/generations.rs create mode 100644 crates/preflight/tests/generation_repairs.rs diff --git a/README.md b/README.md index 3c1b67936a..f9a41cb3ca 100644 --- a/README.md +++ b/README.md @@ -94,7 +94,7 @@ Real-world typography: world scripts, vertical Japanese, colour emoji, gradients ### Fearless -Every save appends your changes and leaves the original bytes untouched. Writes are atomic, undo runs deep, and nothing is lost if you close by mistake. +Ordinary saves append your changes and leave the original bytes untouched. Edits to damaged objects whose xref and header generations disagree use a full rewrite, keeping both accepted references connected to the same repaired object. Writes are atomic, undo runs deep, and nothing is lost if you close by mistake. diff --git a/ROADMAP.md b/ROADMAP.md index a1432f1120..1f753c500f 100644 --- a/ROADMAP.md +++ b/ROADMAP.md @@ -21,6 +21,11 @@ > **Last reviewed:** 2026-10-23 · **Last updated:** 2026-10-23 · **Change:** minor (M2, PR #4172: extend on-screen CJK fallback to Combine thumbnails and figure previews; recognise FangSong and subset-prefixed Founder names; estimates unchanged) > **Last reviewed:** 2026-10-15 · **Last updated:** 2026-10-15 · **Change:** minor (typed small Edit font sizes and Fill & Sign text/date colour, #2444/#2496; estimates unchanged) +> **Last reviewed:** 2026-10-20 · **Last updated:** 2026-10-20 · **Change:** patch (M1, #1143: repaired aliases survive preflight, snapshots and normalized saves; retained materialized revisions report generation disagreements; estimates unchanged) + +> **Last reviewed:** 2026-10-18 · **Last updated:** 2026-10-18 · **Change:** minor (M1, #1143: explicit live xref/header generation repair policy and bounded diagnostics; estimates unchanged) + +> **Last reviewed:** 2026-10-15 · **Last updated:** 2026-10-15 · **Change:** minor (M1: refresh object-generation reference identity against current upstream; estimates unchanged) **Stage: alpha** · next: beta, ~25% (ready for real work 50% → 75%) and ~400–700 h away > **Last reviewed:** 2026-10-29 | **Last updated:** 2026-10-29 | **Change:** minor (M2: scans keep their strokes at screen size and are smoothed when zoomed; M7 #2517: a narrowed then widened paragraph keeps the typed spaces the wrap consumed, after save and reopen; estimates unchanged) @@ -339,6 +344,8 @@ Ranked; detail and the full milestone table in [docs/roadmap.md](docs/roadmap.md - **2026-10-20 (M5, arrow keys move comments):** With a comment selected, the arrow keys move it a screen pixel and ⌘/Ctrl or Shift and an arrow ten, instead of scrolling the view or turning the page; up is up on screen at any zoom and rotation, and each press is one undo step. Text markup and locked comments stay put. Estimates unchanged. +- **2026-10-20 (M1, PR #1143 review):** Repair-aware dictionary/preflight edits and snapshot restoration preserve both accepted generations of a damaged live slot. Such edits use the existing full-save fallback; classic and packed rewrites coalesce aliases so page-tree parent links retain one object identity. Explicit generation replacement remains single-generation. Materialized xref diagnostics inspect retained primary and hybrid rows before folding, retaining the 256-object cap plus summary and discarding abandoned-index notes. Synthetic red/green controls cover preflight, undo restoration, equal values, explicit narrowing, encryption packing and four save variants. Estimates unchanged. + - **2026-10-19 (M14, #3373):** The main menu and its long submenus scroll in short windows, keeping their commands reachable. Estimates unchanged. - **2026-10-19 (M3, #3387):** Save-prompt shortcuts yield to later consent and utility dialogs; a synthetic UI regression checks that an edited document stays open. Estimates unchanged. @@ -402,6 +409,8 @@ Ranked; detail and the full milestone table in [docs/roadmap.md](docs/roadmap.md - **2026-10-15 (community, M6, #2608):** In Prepare a form the selected field moves with the arrow keys: a point per press, ten with Shift, as the object nudge and dragging it do. Each press is one undo step ("Change field properties"), and the keys no longer turn the page while a field is selected. Estimates unchanged. - **2026-10-15 (M3, closable community card):** The ArtCraft community card on Home has a close button that hides it for the rest of the run; it returns on the next launch, and its links stay in the top bar's Discord button and the Help menu. A UI regression covers closing and the fresh-launch default. Estimates unchanged. - **2026-10-13 (community, M3, #1569 review):** Pixel-only Wayland wheel, wheel-tilt and continuous input use standalone moved events, so Ctrl zoom does not remain latched after the modifier is released (#991); vendored winit patch (13), with a policy test. Estimates unchanged. +- **2026-10-18 (M1, PR #1143 repair policy):** A live in-file reference resolves when its generation matches the xref row or a same-number header at that indexed offset. Neither-match and free slots remain null, including when old object bytes remain. Opening records bounded xref/header disagreements without parsing object bodies or expanding compact xref rows. Synthetic controls cover read order, clones, overlays, dictionary edits, full classic/packed saves and an independent reader of incremental output. **Correction to the 2026-10-15 note below:** its async-opening-test sentence was not evidenced by this COS change; no fresh async-opening or GUI validation is claimed. The full-reference stream-length recursion key is defensive hardening, not an independently isolated regression fix. Estimates unchanged. +- **2026-10-15 (M1, PR #1143 refresh):** COS reference lookups and full saves distinguish complete object references, so stale generations cannot expose replacement content. Snapshot restoration also keeps edits to objects with nonzero generations before incremental saves. Synthetic regressions cover caches, overlays, snapshots, stream lengths and classic/packed saves. Current upstream async-opening test corrections are included. Estimates unchanged. - **2026-10-15 (M14, #2383 candidate):** Device-limit requests now clamp unsupported maxima and minimum alignments before enabling the adapter's full 2D texture extent. This addresses the reported V3D request of 16 shader variables on an adapter supporting 15. Synthetic regressions fail before the clamp and pass afterward, along with the existing HiDPI case and Windows GPU-required application tests. Raspberry Pi startup remains unverified; milestone and overall estimates are unchanged. - **2026-10-12 (M6):** XFDF field import and data-file extraction join all text nodes in a value, preserving text after XML comments and processing instructions. Synthetic regressions cover Unicode, CDATA, empty values and PDF save/reopen. Estimates unchanged. - **2026-10-12 (M7, picture right-click menu):** The Edit text & images menu of a picture gains Cut Image, Redact Image (a redaction mark over the picture, applied from Redact a PDF), Create Link… and Recognize Text… (for that page), in all 21 catalogs. The menu was drawn only while the pointer was over the picture, so its items outside the picture (most of a logo's menu, and Delete below a page) could not be clicked; an open menu now stays drawn. Still missing from Acrobat's menu: Arrange (bring to front, send backward), Remove crop, Draw freehand and Edit image using… (opening an external editor). @@ -780,6 +789,7 @@ Newest first: the date, what moved, and the new overall estimate. | 2026-10-20 | minor | M6, #3502: form detection preserves comments across content streams; estimates unchanged. | | 2026-10-20 | minor | M10, #3532 partial: vector printing retains existing appearances with absent Subtype; estimates unchanged. | | 2026-10-20 | minor | M11 #3507: image merging preserves comments and state across content streams. Estimates unchanged. | +| 2026-10-18 | minor | M1, #1143: a live reference may match the xref generation or a same-number header at its indexed offset; neither-match and free slots remain null. Bounded repair diagnostics and synthetic cache, free-slot, edit and classic/packed save regressions; estimates unchanged. | | 2026-10-18 | minor | M5, PR #1971: nested source `/Page` and `/Pages` dictionaries export as null without traversing page content or the page tree. Synthetic saved/reopened three-page regression retains unrelated metadata and aliases. FDF import semantics, copied-graph limits and estimates unchanged. | | 2026-10-15 | minor | M5: budget retained FDF entries before cloning; discovery keeps lazy array owners and annotation positions, with alias/cache-eviction stress fixtures. Limits cover the copied graph; source parsing/page discovery retain existing COS/parser limits, with no total source-memory bound. FDF import semantics and estimates unchanged. | | 2026-10-15 | minor | M7: the Edit text caret is as tall as the row it is on in a region with several sizes; estimates unchanged. | @@ -790,6 +800,7 @@ Newest first: the date, what moved, and the new overall estimate. | 2026-10-14 | minor | M1: raw-deflate fallback preserves decoding-limit errors in strict and tolerant decoding; synthetic filter and COS stream regressions. Estimates unchanged. | | 2026-10-16 | minor | M14: Czech catalog completed (2,674 entries) and "tag" terminology unified. Estimates unchanged. | | 2026-10-13 | minor | M9: separate page validation overlays preserve signature artwork, and snapshot tabs reuse instead of multiplying; baked captions remain unchanged (#79 open); estimates unchanged | +| 2026-10-15 | minor | M1: refresh object-generation reference identity, retaining current upstream progress and regression tests. Estimates unchanged. | | 2026-10-16 | minor | M10 #1201 partial: lossless image export applies component Decode pairs and Indexed mapping. Estimates unchanged. | | 2026-10-15 | minor | M6: CSV export and spreadsheet merging quote tabs, and import distinguishes quoted tabs from TSV delimiters without changing field identities. Estimates unchanged. | | 2026-10-13 | minor | M4: rotation queries and edits resolve indirect page-tree values; inherited, negative and large rotations verified through both saves. Estimates unchanged. | diff --git a/crates/cos/README.md b/crates/cos/README.md index 7b576f293f..398b161625 100644 --- a/crates/cos/README.md +++ b/crates/cos/README.md @@ -53,7 +53,18 @@ another. Edits never touch the original bytes. - **Every cross-reference form:** classic tables with their quirky subsections, cross-reference streams, hybrid-reference files (`/XRefStm`), and object streams. - **Repair.** A file whose xref is wrong or missing is reconstructed by scanning for `obj` - headers. Repairs are *recorded*, never silent, so the UI can say the file was damaged. + headers. Repairs are *recorded*, never silent, so the UI can say the file was damaged. A live + in-file reference may match either its xref generation or a same-number object header at + that indexed offset; a missing, free, or neither-matching slot still resolves to null. + Header probes inspect at most 128 bytes at the indexed and header-relative offsets, without + parsing object bodies. On open, disagreements in retained xref revisions are recorded once + per object, up to 256 objects plus an omitted-details summary. This diagnostic scan does + not expand compact xref ranges; the winning xref entry remains authoritative for free + slots. Compressed objects require generation zero. Explicit `set` overlays accept only the + supplied generation; `set_existing` and `update_dict` retain both accepted generations of a + damaged live slot, including through snapshots, and require full rewrite. Headers beyond + this bounded probe are not additionally recovered. Materialized primary and hybrid rows are + diagnosed before folding, so a newer row does not hide a retained disagreement. - **Tolerant, never fatal.** The lexer works on a byte slice and a position; it allocates nothing for skipped data, bounds-checks every index, and caps nesting depth. Malformed input yields `CosError`, never a panic — this crate is the first thing a hostile file meets. Exhausted xref entries are skipped with a repair log; exhausted /Size hints allocate unused numbers without overflowing or replacing existing objects. @@ -66,14 +77,17 @@ another. Edits never touch the original bytes. ## Writing -`write_incremental` appends only the edited objects, a cross-reference section **in the same -style as the file's last one**, and a trailer with `/Prev`. The original bytes are never modified, -which is what keeps a signature over an earlier revision valid. +`write_incremental` ordinarily appends only the edited objects, a cross-reference section **in +the same style as the file's last one**, and a trailer with `/Prev`, preserving the original +bytes and signatures over earlier revisions. It delegates to `write_full` when reconstruction, +encryption changes or `full_save_required()` demand a rewrite. Repair-aware edits to a live +xref/header generation disagreement use this fallback rather than append a contradictory row. `write_full` writes only what the trailer can reach, renumbered densely — the Save As and garbage-collecting path. By default (`SaveOptions::object_streams`) it packs non-stream objects into compressed object streams with a cross-reference stream (PDF 1.5+); turning that off writes a -classic table any reader accepts. +classic table any reader accepts. Both forms map accepted xref/header aliases to one output +object, retaining shared identity such as page-tree parent links; unrelated generations stay null. Dictionaries keep their key order and streams keep their *encoded* bytes, so rewritten objects stay recognisable in a diff and byte-stable when re-serialized. diff --git a/crates/cos/src/document/objects.rs b/crates/cos/src/document/objects.rs index 62ef1e74cc..8d9917d749 100644 --- a/crates/cos/src/document/objects.rs +++ b/crates/cos/src/document/objects.rs @@ -1,6 +1,8 @@ +use std::collections::HashSet; + use super::reader::find; use super::state::{LENGTHS, LoadGuard, ObjStm, ObjectReader, ReadState, Slot, SourceIdentity, held_bytes, objstm_member}; -use super::xref::{Revision, XrefEntry}; +use super::xref::{Revision, XrefEntry, XrefIndex}; use super::{ Arc, CosError, Cow, Dict, Document, Lexer, MAX_DECODED, MAX_JBIG2_GLOBALS, ObjRef, Object, Stream, is_whitespace, parse_indirect_shared, }; @@ -129,8 +131,8 @@ impl Document { pub fn xref_entry_is_exact(&self, num: u32) -> bool { match self.xref_entry(num) { None | Some(XrefEntry::Free { .. }) => true, - Some(XrefEntry::InFile { offset, .. }) => { - let at = |off: usize| matches!(parse_indirect_shared(&self.data, off, &|_| None, &self.endstreams), Ok((id, _)) if id.num == num); + Some(XrefEntry::InFile { offset, generation }) => { + let at = |off: usize| matches!(parse_indirect_shared(&self.data, off, &|_| None, &self.endstreams), Ok((id, _)) if id == ObjRef::new(num, generation)); // Offsets may count from a header that does not start the file. usize::try_from(offset).is_ok_and(|off| at(off) || (self.header_offset != 0 && at(off.saturating_add(self.header_offset)))) } @@ -151,9 +153,93 @@ impl Document { // ── object access ─────────────────────────────────────────────────────────────────────── - /// Fetch an object by reference. Missing objects are `Null` (ISO 32000-2 §7.3.10). + /// Fetch an object by reference. Undefined references are `Null` (ISO 32000-2 §7.3.10). + /// A live file slot also accepts its actual header generation when a damaged xref row + /// disagrees; the disagreement is recorded in [`Self::repair_log`]. Free slots never do. pub fn get(&self, r: ObjRef) -> Arc { - self.try_get(r.num).unwrap_or_else(|_| Arc::new(Object::Null)) + self.try_get_reference(r).unwrap_or_else(|_| Arc::new(Object::Null)) + } + + pub(super) fn reference_matches(&self, r: ObjRef) -> bool { + match self.overlay.get(&r.num) { + Some(Slot::Set(generation, _)) => r.generation == *generation, + Some(Slot::RepairedSet(generation, alias, _)) => r.generation == *generation || r.generation == *alias, + Some(Slot::Freed(_)) => false, + None => match self.entries.lookup(r.num, &self.data) { + Some(XrefEntry::InFile { offset, generation }) => { + r.generation == generation || self.indexed_header_reference(r.num, offset) == Some(r) + } + Some(XrefEntry::InStream { .. }) => r.generation == 0, + None | Some(XrefEntry::Free { .. }) => false, + }, + } + } + + pub(super) fn try_get_reference(&self, r: ObjRef) -> Result, CosError> { + if !self.reference_matches(r) { + return Ok(Arc::new(Object::Null)); + } + self.try_get(r.num) + } + + /// A repair probe, not a whole-file search: only the indexed header (or the same offset + /// relative to a shifted PDF header) can establish this additional generation. A small + /// window bounds hostile whitespace/tokens; ordinary xref-matching reads are unchanged. + pub(super) fn indexed_header_reference(&self, num: u32, offset: u64) -> Option { + let at = |off: usize| { + let bytes = self.data.get(off..off.saturating_add(128).min(self.data.len()))?; + let mut lexer = Lexer::new(bytes, 0); + lexer.skip_ws(); + let number = std::str::from_utf8(lexer.token()).ok()?.parse::().ok()?; + lexer.skip_ws(); + let generation = std::str::from_utf8(lexer.token()).ok()?.parse::().ok()?; + (number == num && lexer.eat_keyword(b"obj")).then_some(ObjRef::new(number, generation)) + }; + let off = usize::try_from(offset).ok()?; + at(off).or_else(|| if self.header_offset == 0 { None } else { off.checked_add(self.header_offset).and_then(at) }) + } + + /// Add at most one bounded diagnostic per object. False stops diagnostic probing only; + /// parsers must still read the remaining xref rows needed to build their index. + pub(super) fn inspect_xref_generation_repair(&self, num: u32, entry: XrefEntry, recorded: &mut HashSet, log: &mut Vec) -> bool { + let XrefEntry::InFile { offset, generation } = entry else { return true }; + if recorded.contains(&num) { + return true; + } + let Some(header) = self.indexed_header_reference(num, offset) else { return true }; + if header.generation == generation { + return true; + } + if recorded.len() == 256 { + log.push("additional xref/header generation disagreements omitted after 256 objects".into()); + return false; + } + recorded.insert(num); + log.push(format!("object {num}: xref generation {generation} disagrees with object header generation {}", header.generation)); + true + } + + /// Diagnose once at open without expanding compact ranges or decoding an xref twice. + /// Materialized readers retain bounded notes while they read accepted sections, before + /// newer rows hide older ones. Discarded parse attempts/rebuilt indexes carry no notes. + pub(super) fn note_xref_generation_repairs(&self, log: &mut Vec) { + match &*self.entries { + XrefIndex::Materialized(_, notes) => log.extend(notes.iter().cloned()), + XrefIndex::Compact { revisions, .. } => { + let mut recorded = HashSet::new(); + for revision in revisions { + for section in std::iter::once(&revision.primary).chain(revision.hybrid.iter()) { + for num in section.numbers() { + if let Some(entry) = section.lookup(num, &self.data) + && !self.inspect_xref_generation_repair(num, entry, &mut recorded, log) + { + return; + } + } + } + } + } + } } /// Follow a reference if `o` is one; otherwise return `o` itself. An object can be nothing but @@ -388,7 +474,7 @@ impl Document { pub fn try_get(&self, num: u32) -> Result, CosError> { match self.overlay.get(&num) { - Some(Slot::Set(_, o)) => return Ok(o.clone()), + Some(Slot::Set(_, o) | Slot::RepairedSet(_, _, o)) => return Ok(o.clone()), Some(Slot::Freed(_)) => return Ok(Arc::new(Object::Null)), None => {} } @@ -519,11 +605,11 @@ impl Document { None | Some(XrefEntry::Free { .. }) => Ok(Object::Null), Some(XrefEntry::InFile { offset, .. }) => { let resolve = |r: ObjRef| { - if let Some(known) = LENGTHS.with(|m| m.borrow().get(&r.num).copied()) { + if let Some(known) = LENGTHS.with(|m| m.borrow().get(&r).copied()) { return known; } - let length = self.try_get(r.num).ok().and_then(|o| o.as_int()); - LENGTHS.with(|m| m.borrow_mut().insert(r.num, length)); + let length = self.try_get_reference(r).ok().and_then(|o| o.as_int()); + LENGTHS.with(|m| m.borrow_mut().insert(r, length)); length }; let off = offset as usize; @@ -640,6 +726,38 @@ impl Document { } } + /// Replace an accepted existing reference without losing a damaged source slot's aliases. + /// Unlike [`Self::set`], this preserves both xref/header generations when they disagree. + /// Such an edit forces a full rewrite so the output has consistent references, rather than + /// appending a new contradictory xref row. Missing, free and stale references are errors. + pub fn set_existing(&mut self, r: ObjRef, obj: impl Into) -> Result<(), CosError> { + if !self.reference_matches(r) { + return Err(CosError::MissingObject(r.num)); + } + let alias = self.alternate_generation(r); + if let Some(alias) = alias { + let obj = self.adopt(obj.into()); + self.overlay.insert(r.num, Slot::RepairedSet(r.generation, alias, Arc::new(obj))); + } else { + self.set(r, obj); + } + Ok(()) + } + + /// Other accepted generation for a known-live reference to a damaged slot. + pub(super) fn alternate_generation(&self, r: ObjRef) -> Option { + match self.overlay.get(&r.num) { + Some(Slot::RepairedSet(generation, alias, _)) => Some(if r.generation == *generation { *alias } else { *generation }), + Some(_) => None, + None => match self.entries.lookup(r.num, &self.data) { + Some(XrefEntry::InFile { offset, generation }) => self.indexed_header_reference(r.num, offset).and_then(|header| { + (header.generation != generation).then_some(if r.generation == generation { header.generation } else { generation }) + }), + _ => None, + }, + } + } + /// Add a new indirect object and return its reference. pub fn add(&mut self, obj: impl Into) -> ObjRef { let r = ObjRef::new(self.allocate_num(), 0); @@ -699,7 +817,8 @@ impl Document { let generation = current.generation(num); let same = saved_nums.binary_search(&num).is_ok() && self.generation(num) == generation - && saved_reader.try_get(ObjRef::new(num, 0))? == current_reader.try_get(ObjRef::new(num, 0))?; + && self.alternate_generation(ObjRef::new(num, generation)) == current.alternate_generation(ObjRef::new(num, generation)) + && saved_reader.try_get(ObjRef::new(num, generation))? == current_reader.try_get(ObjRef::new(num, generation))?; if !same { changed.push(num); } @@ -728,8 +847,14 @@ impl Document { let mut current_reader = current.object_reader(); for num in changed { let generation = current.generation(num); - let object = current_reader.try_get(ObjRef::new(num, 0))?; - self.set(ObjRef::new(num, generation), (*object).clone()); + let object = current_reader.try_get(ObjRef::new(num, generation))?; + if let Some(alias) = current.alternate_generation(ObjRef::new(num, generation)) { + let object = self.adopt((*object).clone()); + self.overlay.insert(num, Slot::RepairedSet(generation, alias, Arc::new(object))); + self.next_num = self.next_num.max(num.saturating_add(1)); + } else { + self.set(ObjRef::new(num, generation), (*object).clone()); + } } *self.trailer_mut() = current.trailer().clone(); Ok(()) @@ -740,14 +865,13 @@ impl Document { let mut obj = (*self.get(r)).clone(); let d = obj.as_dict_mut().ok_or(CosError::NotADictionary(r))?; f(d); - self.set(r, obj); - Ok(()) + self.set_existing(r, obj) } /// Current generation of an object number. pub fn generation(&self, num: u32) -> u16 { match self.overlay.get(&num) { - Some(Slot::Set(g, _)) => *g, + Some(Slot::Set(g, _) | Slot::RepairedSet(g, _, _)) => *g, Some(Slot::Freed(g)) => *g, None => match self.entries.lookup(num, &self.data) { Some(XrefEntry::InFile { generation, .. }) => generation, @@ -759,7 +883,7 @@ impl Document { pub(crate) fn overlay_entries(&self) -> impl Iterator>)> + '_ { self.overlay.iter().map(|(n, s)| match s { - Slot::Set(g, o) => (*n, *g, Some(o.clone())), + Slot::Set(g, o) | Slot::RepairedSet(g, _, o) => (*n, *g, Some(o.clone())), Slot::Freed(g) => (*n, *g, None), }) } @@ -782,7 +906,7 @@ impl Document { if source.peek().is_some_and(|n| *n == num) { source.next(); } - if matches!(slot, Slot::Set(..)) { + if matches!(slot, Slot::Set(..) | Slot::RepairedSet(..)) { numbers.push(num); } } @@ -800,7 +924,7 @@ impl Document { pub(super) fn root_is_catalog(&self) -> bool { let Some(r) = self.root() else { return false }; - matches!(self.try_get(r.num).as_deref(), Ok(Object::Dict(d)) if d.name(b"Type") == Some(b"Catalog") || d.contains(b"Pages")) + matches!(self.try_get_reference(r).as_deref(), Ok(Object::Dict(d)) if d.name(b"Type") == Some(b"Catalog") || d.contains(b"Pages")) } } diff --git a/crates/cos/src/document/open.rs b/crates/cos/src/document/open.rs index 70bbcc313e..7a4337c8da 100644 --- a/crates/cos/src/document/open.rs +++ b/crates/cos/src/document/open.rs @@ -1,5 +1,5 @@ use super::reader::find; -use super::state::Credentials; +use super::state::{Credentials, Slot}; use super::xref::XrefIndex; use super::{Arc, BTreeMap, CosError, Dict, Document, HashSet, ObjRef, Object}; @@ -156,6 +156,9 @@ impl Document { } doc.reconstruct(&mut log)?; } + // Header generations are plaintext, including in encrypted files. Inspect only the + // bounded headers, not object bodies, and retain the compact xref representation. + doc.note_xref_generation_repairs(&mut log); doc.repair_log = Arc::new(log); // Reconstruction may have found a different trailer: (re)authenticate against it. if (!ok || !authenticated) @@ -288,8 +291,9 @@ impl Document { } /// The next save must rewrite the whole file (see [`Document::require_full_save`]). + /// Repair-aware edits retaining two accepted generations cannot use one incremental slot. pub fn full_save_required(&self) -> bool { - self.full_save + self.full_save || self.overlay.values().any(|slot| matches!(slot, Slot::RepairedSet(..))) } /// Security was set or removed since the document was opened (the next save applies it). diff --git a/crates/cos/src/document/reader.rs b/crates/cos/src/document/reader.rs index d21bf9403f..e1691ace90 100644 --- a/crates/cos/src/document/reader.rs +++ b/crates/cos/src/document/reader.rs @@ -17,8 +17,8 @@ impl Document { return Ok((index, trailer, revisions, Vec::new())); } } - let (entries, trailer, revisions, oddities) = self.read_xref_chain_materialized(log)?; - Ok((XrefIndex::Materialized(Arc::new(entries)), trailer, revisions, oddities)) + let (entries, trailer, revisions, oddities, generation_repairs) = self.read_xref_chain_materialized(log)?; + Ok((XrefIndex::Materialized(Arc::new(entries), Arc::new(generation_repairs)), trailer, revisions, oddities)) } fn read_compact_xref_chain(&self, log: &mut Vec) -> Result)>, CosError> { @@ -323,7 +323,7 @@ impl Document { pub(super) fn read_xref_chain_materialized( &self, log: &mut Vec, - ) -> Result<(BTreeMap, Dict, Vec, Vec<(u64, u32)>), CosError> { + ) -> Result<(BTreeMap, Dict, Vec, Vec<(u64, u32)>, Vec), CosError> { let data = &self.data; let tail_start = data.len().saturating_sub(4096); let sx = rfind(data, b"startxref", tail_start).ok_or(CosError::Syntax { offset: data.len(), detail: "no startxref".into() })?; @@ -343,6 +343,9 @@ impl Document { // At most MAX_XREF_OBJECTS entries are kept in all; a stream past that is read again. let mut hybrids: HashMap>> = HashMap::new(); let (mut hybrid_rows, mut noted) = (0usize, HashSet::new()); + let mut generation_repairs = Vec::new(); + let mut recorded = HashSet::new(); + let mut diagnose = true; let mut next = Some(first); while let Some(off) = next.take() { if !visited.insert(off) || visited.len() > 10_000 { @@ -371,6 +374,12 @@ impl Document { } }; revisions.push(Revision { xref_offset: off, is_stream, header_relative }); + for (&num, &entry) in § { + if !diagnose { + break; + } + diagnose = self.inspect_xref_generation_repair(num, entry, &mut recorded, &mut generation_repairs); + } // Hybrid files (§7.5.8.4): the /XRefStm entries supplement the table and replace the // free entries it gives objects only a cross-reference stream can locate. let hybrid = sect_trailer.int(b"XRefStm").filter(|x| *x >= 0); @@ -390,6 +399,12 @@ impl Document { let mut stm = BTreeMap::new(); let read = match self.read_section(x as usize, &mut stm, &mut odd, log) { Ok(_) => { + for (&num, &entry) in &stm { + if !diagnose { + break; + } + diagnose = self.inspect_xref_generation_repair(num, entry, &mut recorded, &mut generation_repairs); + } for (&num, &e) in &stm { match sect.get(&num) { None | Some(XrefEntry::Free { .. }) => { @@ -427,7 +442,7 @@ impl Document { for k in [&b"Prev"[..], b"XRefStm", b"Type", b"W", b"Index", b"Filter", b"DecodeParms", b"Length"] { trailer.remove(k); } - Ok((entries, trailer, revisions, oddities)) + Ok((entries, trailer, revisions, oddities, generation_repairs)) } /// Read one section at `off`, adding entries not already known (newer sections win). `odd` @@ -606,7 +621,7 @@ impl Document { } if found > 0 { log.push(format!("the cross-reference data leaves out {found} objects the file contains; located them by their object headers")); - self.entries = Arc::new(XrefIndex::Materialized(Arc::new(filled))); + self.entries = Arc::new(XrefIndex::Materialized(Arc::new(filled), Arc::default())); } self.located = Arc::new(located); } @@ -631,7 +646,7 @@ impl Document { if entries.is_empty() { return Err(CosError::Syntax { offset: 0, detail: "no objects found".into() }); } - self.entries = Arc::new(XrefIndex::Materialized(Arc::new(entries.clone()))); + self.entries = Arc::new(XrefIndex::Materialized(Arc::new(entries.clone()), Arc::default())); self.cache.lock().map_err(|_| CosError::Poisoned)?.clear(); // Objects inside object streams. for num in objstms { @@ -648,7 +663,7 @@ impl Document { } } let object_count = entries.len(); - self.entries = Arc::new(XrefIndex::Materialized(Arc::new(entries))); + self.entries = Arc::new(XrefIndex::Materialized(Arc::new(entries), Arc::default())); self.cache.lock().map_err(|_| CosError::Poisoned)?.clear(); // Trailer: the last `trailer` dictionary or cross-reference stream dictionary with a // usable /Root (last in the file, whichever kind), else the catalog itself. One whose @@ -658,7 +673,7 @@ impl Document { // A file whose only trailers are cross-reference streams (and whose `startxref` is gone // or points nowhere) keeps its /Info, /ID and /Encrypt this way. let rank = |at: usize, d: &Dict| { - let root = self.try_get(d.reference(b"Root")?.num).ok()?; + let root = self.try_get_reference(d.reference(b"Root")?).ok()?; let Object::Dict(c) = &*root else { return None }; Some((c.name(b"Type") == Some(b"Catalog") || c.contains(b"Pages"), at)) }; diff --git a/crates/cos/src/document/state.rs b/crates/cos/src/document/state.rs index 96dde01c0f..f68a483290 100644 --- a/crates/cos/src/document/state.rs +++ b/crates/cos/src/document/state.rs @@ -7,7 +7,7 @@ thread_local! { /// Indirect `/Length` values resolved during the current outermost load: a damaged file /// retries a parse several ways, and each retry would resolve the same lengths again /// (exponential in the chain of streams whose lengths point at further streams). - pub(super) static LENGTHS: std::cell::RefCell>> = std::cell::RefCell::new(HashMap::new()); + pub(super) static LENGTHS: std::cell::RefCell>> = std::cell::RefCell::new(HashMap::new()); } /// Marks an object as being loaded on this thread. Loading can re-enter (indirect stream @@ -65,6 +65,9 @@ pub(super) fn objstm_member(num: i64, offset: i64, first: Option, len: us #[derive(Clone, Debug, PartialEq)] pub(super) enum Slot { Set(u16, Arc), + /// Both generations accepted from a damaged live source slot survive a repair-aware edit. + /// A full rewrite is required: a new xref slot cannot encode two generations. + RepairedSet(u16, u16, Arc), Freed(u16), } @@ -103,6 +106,7 @@ impl ReadState { a == b && match (x, y) { (Slot::Set(g, o), Slot::Set(h, p)) => g == h && Arc::ptr_eq(o, p), + (Slot::RepairedSet(g, a, o), Slot::RepairedSet(h, b, p)) => g == h && a == b && Arc::ptr_eq(o, p), (Slot::Freed(g), Slot::Freed(h)) => g == h, _ => false, } @@ -133,7 +137,7 @@ impl ObjectReader { } pub(crate) fn try_get(&mut self, reference: ObjRef) -> Result, CosError> { - let object = self.document.try_get(reference.num); + let object = self.document.try_get_reference(reference); // A missed eviction here only costs a re-parse, so never fail the read on a poisoned // lock. The editing-cache evictors propagate instead; they guard user edits. if let Ok(mut streams) = self.document.objstms.lock() { diff --git a/crates/cos/src/document/tests.rs b/crates/cos/src/document/tests.rs index f12437baaf..ad8f86fdea 100644 --- a/crates/cos/src/document/tests.rs +++ b/crates/cos/src/document/tests.rs @@ -599,7 +599,7 @@ pub(crate) fn build(bodies: &[&str], trailer: &str) -> Vec { fn assert_compact_matches_materialized(doc: &Document) { assert!(matches!(doc.entries.as_ref(), XrefIndex::Compact { .. }), "valid xrefs should not retain a per-entry map"); - let (entries, _, revisions, _) = doc.read_xref_chain_materialized(&mut Vec::new()).unwrap(); + let (entries, _, revisions, _, _) = doc.read_xref_chain_materialized(&mut Vec::new()).unwrap(); assert_eq!(revisions.as_slice(), doc.revisions()); let max = entries.keys().next_back().copied().unwrap_or(0); assert_eq!(doc.entries.max_num(), max); @@ -687,7 +687,7 @@ fn malformed_size_uses_the_materialized_compatibility_path() { let at = bytes.windows(marker.len()).position(|window| window == marker).unwrap(); bytes[at..at + marker.len()].copy_from_slice(b"/Size 2"); let doc = Document::open(Arc::new(bytes)).unwrap(); - assert!(matches!(doc.entries.as_ref(), XrefIndex::Materialized(_))); + assert!(matches!(doc.entries.as_ref(), XrefIndex::Materialized(..))); assert_eq!(doc.next_num, 4); assert_eq!(doc.object_numbers(), vec![1, 2, 3]); } @@ -1181,15 +1181,18 @@ fn reads_classic_revision_and_edits_are_copy_on_write() { fn object_numbers_merge_source_and_overlay_in_order() { let mut doc = Document::new_empty(); doc.overlay.clear(); - doc.entries = Arc::new(XrefIndex::Materialized(Arc::new(BTreeMap::from([ - (0, XrefEntry::Free { next_generation: 65535 }), - (2, XrefEntry::InFile { offset: 10, generation: 7 }), - (4, XrefEntry::Free { next_generation: 9 }), - (6, XrefEntry::InStream { stream: 80, index: 1 }), - (8, XrefEntry::InFile { offset: 20, generation: 5 }), - (10, XrefEntry::InFile { offset: 30, generation: 3 }), - (u32::MAX, XrefEntry::InFile { offset: 40, generation: 1 }), - ])))); + doc.entries = Arc::new(XrefIndex::Materialized( + Arc::new(BTreeMap::from([ + (0, XrefEntry::Free { next_generation: 65535 }), + (2, XrefEntry::InFile { offset: 10, generation: 7 }), + (4, XrefEntry::Free { next_generation: 9 }), + (6, XrefEntry::InStream { stream: 80, index: 1 }), + (8, XrefEntry::InFile { offset: 20, generation: 5 }), + (10, XrefEntry::InFile { offset: 30, generation: 3 }), + (u32::MAX, XrefEntry::InFile { offset: 40, generation: 1 }), + ])), + Arc::default(), + )); doc.set(ObjRef::new(1, 2), Object::Int(1)); // before the first live source slot doc.set(ObjRef::new(2, 8), Object::Int(2)); // override, without duplication doc.set(ObjRef::new(4, 9), Object::Int(4)); // reuse a source free slot diff --git a/crates/cos/src/document/xref.rs b/crates/cos/src/document/xref.rs index 02390ff7da..771df28b80 100644 --- a/crates/cos/src/document/xref.rs +++ b/crates/cos/src/document/xref.rs @@ -41,7 +41,7 @@ pub(super) struct XrefRevisionSection { #[derive(Clone, Debug)] pub(super) enum XrefIndex { Compact { revisions: Vec, max_num: u32, has_entries: bool }, - Materialized(Arc>), + Materialized(Arc>, Arc>), } pub(super) const MAX_COMPACT_XREF_ROWS: usize = 2_000_000; @@ -149,8 +149,8 @@ impl XrefSection { } } - pub(super) fn extend_numbers(&self, numbers: &mut Vec) { - for range in &self.ranges { + pub(super) fn numbers(&self) -> impl Iterator + '_ { + self.ranges.iter().flat_map(|range| { let available = match &self.data { XrefSectionData::Table => range.count as usize, XrefSectionData::Stream { bytes, widths } => { @@ -158,14 +158,12 @@ impl XrefSection { bytes.len().saturating_sub(range.offset).checked_div(row_width).unwrap_or(0) } }; - for i in 0..available.min(range.count as usize) { - if let Ok(i) = u32::try_from(i) - && let Some(num) = range.start.checked_add(i) - { - numbers.push(num); - } - } - } + (0..available.min(range.count as usize)).filter_map(move |i| u32::try_from(i).ok().and_then(|i| range.start.checked_add(i))) + }) + } + + pub(super) fn extend_numbers(&self, numbers: &mut Vec) { + numbers.extend(self.numbers()); } } @@ -189,20 +187,20 @@ impl XrefRevisionSection { impl XrefIndex { pub(super) fn empty() -> Self { - Self::Materialized(Arc::new(BTreeMap::new())) + Self::Materialized(Arc::new(BTreeMap::new()), Arc::default()) } pub(super) fn lookup(&self, num: u32, file: &[u8]) -> Option { match self { Self::Compact { revisions, .. } => revisions.iter().find_map(|revision| revision.lookup(num, file)), - Self::Materialized(entries) => entries.get(&num).copied(), + Self::Materialized(entries, _) => entries.get(&num).copied(), } } pub(super) fn is_empty(&self) -> bool { match self { Self::Compact { has_entries, .. } => !has_entries, - Self::Materialized(entries) => entries.is_empty(), + Self::Materialized(entries, _) => entries.is_empty(), } } @@ -211,14 +209,14 @@ impl XrefIndex { pub(super) fn remove_exhausted(&mut self) -> bool { match self { Self::Compact { .. } => false, - Self::Materialized(entries) => Arc::make_mut(entries).remove(&u32::MAX).is_some(), + Self::Materialized(entries, _) => Arc::make_mut(entries).remove(&u32::MAX).is_some(), } } /// Every entry, newest section first, as the materialized reader keeps them. pub(super) fn to_map(&self, file: &[u8]) -> BTreeMap { match self { - Self::Materialized(entries) => entries.as_ref().clone(), + Self::Materialized(entries, _) => entries.as_ref().clone(), Self::Compact { revisions, .. } => { let mut numbers = Vec::new(); for revision in revisions { @@ -232,13 +230,13 @@ impl XrefIndex { pub(super) fn max_num(&self) -> u32 { match self { Self::Compact { max_num, .. } => *max_num, - Self::Materialized(entries) => entries.keys().next_back().copied().unwrap_or(0), + Self::Materialized(entries, _) => entries.keys().next_back().copied().unwrap_or(0), } } pub(super) fn object_numbers(&self, file: &[u8]) -> Vec { match self { - Self::Materialized(entries) => { + Self::Materialized(entries, _) => { entries.iter().filter_map(|(&num, entry)| (!matches!(entry, XrefEntry::Free { .. })).then_some(num)).collect() } Self::Compact { revisions, .. } => { diff --git a/crates/cos/src/writer.rs b/crates/cos/src/writer.rs index dcfcb97997..6c4c71580a 100644 --- a/crates/cos/src/writer.rs +++ b/crates/cos/src/writer.rs @@ -307,8 +307,17 @@ pub fn write_full(doc: &Document, opts: &SaveOptions) -> Result, CosErro if matches!(*o, Object::Null) { continue; // dangling reference: written as null by renumbering below } - map.insert(r, output_number(order.len())?); - order.push((r, matches!(*o, Object::Stream(_)))); + // Accepted xref/header aliases denote one slot, including page-tree cycles. + // Check for null first: an unrelated or freed generation must remain dangling. + let canonical = ObjRef::new(r.num, reader.document.generation(r.num)); + if let Some(&num) = map.get(&canonical) { + map.insert(r, num); + continue; + } + let num = output_number(order.len())?; + map.insert(r, num); + map.insert(canonical, num); + order.push((canonical, matches!(*o, Object::Stream(_)))); visit_refs(&o, &mut queue); } let version = reader.document.version().chars().take(3).collect::(); @@ -319,13 +328,15 @@ pub fn write_full(doc: &Document, opts: &SaveOptions) -> Result, CosErro out.extend_from_slice(&[b'%', 0xE2, 0xE3, 0xCF, 0xD3, b'\n']); // The /Encrypt dictionary and everything it references (e.g. indirect /CF crypt-filter // dictionaries) are needed before anything can be decrypted: never inside object streams. - let mut encryption_objects: std::collections::HashSet = std::collections::HashSet::new(); + let mut encryption_objects: std::collections::HashSet = std::collections::HashSet::new(); let mut pending: Vec = Vec::new(); if let Some(encrypt) = trailer_in.get(b"Encrypt") { collect_refs(encrypt, &mut |r| pending.push(r)); } while let Some(r) = pending.pop() { - if encryption_objects.insert(r) { + if let Some(&num) = map.get(&r) + && encryption_objects.insert(num) + { collect_refs(&reader.get(r), &mut |x| pending.push(x)); } } @@ -333,14 +344,15 @@ pub fn write_full(doc: &Document, opts: &SaveOptions) -> Result, CosErro // damaged and could not be repaired", before it even asks for the password), while other // readers accept it. qpdf keeps the catalog out of object streams too. Verified with // Acrobat DC on macOS (#774). - let root_outside = if reader.document.output_security().0.is_some() { trailer_in.reference(b"Root") } else { None }; + let root_outside = + if reader.document.output_security().0.is_some() { trailer_in.reference(b"Root").and_then(|r| map.get(&r)).copied() } else { None }; let mut rows: BTreeMap = BTreeMap::new(); // Keep only references, not renamed copies of the entire object graph. Standalone // objects still precede object streams, preserving the established output order. let mut packed: Vec<(u32, ObjRef)> = Vec::new(); for (i, (r, is_stream)) in order.iter().enumerate() { let num = output_number(i)?; - let in_stream = opts.object_streams && !is_stream && !encryption_objects.contains(r) && root_outside != Some(*r); + let in_stream = opts.object_streams && !is_stream && !encryption_objects.contains(&num) && root_outside != Some(num); if in_stream { packed.push((num, *r)); } else { diff --git a/crates/cos/tests/it/generations.rs b/crates/cos/tests/it/generations.rs new file mode 100644 index 0000000000..3d56628c93 --- /dev/null +++ b/crates/cos/tests/it/generations.rs @@ -0,0 +1,475 @@ +//! An indirect reference identifies an object number AND its generation (ISO 32000-2 §7.3.10). + +use std::sync::Arc; + +use hayro_syntax::object::{Dict as HayroDict, ObjectIdentifier}; +use pdfcraft_cos::{Document, ObjRef, Object, SaveOptions, XrefEntry, write_full, write_incremental}; + +/// Object 3 was reused: only 3 1 R refers to the current value, never 3 0 R. +fn reused_object() -> Arc> { + object_file(&[ + (1, 0, "<< /Type /Catalog /Pages 2 0 R /Stale 3 0 R /Current 3 1 R >>"), + (2, 0, "<< /Type /Pages /Kids [] /Count 0 >>"), + (3, 1, "42"), + ]) +} + +fn object_file(objects: &[(u32, u16, &str)]) -> Arc> { + object_file_with_entry(objects, None) +} + +fn object_file_with_entry(objects: &[(u32, u16, &str)], changed: Option<(u32, u16, bool)>) -> Arc> { + let mut bytes = b"%PDF-1.7\n".to_vec(); + let mut offsets = Vec::new(); + for (num, generation, value) in objects { + offsets.push((*num, bytes.len(), *generation)); + bytes.extend_from_slice(format!("{num} {generation} obj\n{value}\nendobj\n").as_bytes()); + } + let xref = bytes.len(); + let size = objects.len() + 1; + bytes.extend_from_slice(format!("xref\n0 {size}\n0000000000 65535 f \n").as_bytes()); + for (num, offset, generation) in offsets { + let (generation, free) = changed.filter(|(n, _, _)| *n == num).map_or((generation, false), |(_, g, free)| (g, free)); + let (offset, kind) = if free { (0, 'f') } else { (offset, 'n') }; + bytes.extend_from_slice(format!("{offset:010} {generation:05} {kind} \n").as_bytes()); + } + bytes.extend_from_slice(format!("trailer\n<< /Size {size} /Root 1 0 R >>\nstartxref\n{xref}\n%%EOF\n").as_bytes()); + Arc::new(bytes) +} + +fn contradictory_entry(free: bool) -> Arc> { + object_file_with_entry( + &[(1, 0, "<< /Type /Catalog /Pages 2 0 R /Stale 3 0 R /Current 3 1 R >>"), (2, 0, "<< /Type /Pages /Kids [] /Count 0 >>"), (3, 1, "42")], + Some((3, if free { 2 } else { 0 }, free)), + ) +} + +#[test] +fn contradictory_xref_generation_recovers_the_matching_header_and_records_repair() { + for warm in [false, true] { + let doc = Document::open(contradictory_entry(false)).unwrap(); + if warm { + assert_eq!(*doc.try_get(3).unwrap(), Object::Int(42)); + } + assert_eq!(*doc.get(ObjRef::new(3, 1)), Object::Int(42)); + assert_eq!(*doc.resolve(&Object::Ref(ObjRef::new(3, 1))), Object::Int(42)); + // Preserve the existing xref-matching behavior, but never accept an unrelated generation. + assert_eq!(*doc.get(ObjRef::new(3, 0)), Object::Int(42)); + assert_eq!(*doc.get(ObjRef::new(3, 2)), Object::Null); + assert!(!doc.xref_entry_is_exact(3), "a contradictory generation is not an exact xref entry"); + assert!( + doc.repair_log() + .iter() + .any(|line| { line.contains("object 3") && line.contains("xref generation 0") && line.contains("header generation 1") }), + "{:?}", + doc.repair_log() + ); + let repairs = doc.repair_log().to_vec(); + for _ in 0..3 { + assert_eq!(*doc.get(ObjRef::new(3, 1)), Object::Int(42)); + } + assert_eq!(doc.repair_log(), repairs); + + let mut edited = doc.clone(); + edited.free(ObjRef::new(3, 1)); + for generation in [0, 1, 2] { + assert_eq!(*edited.get(ObjRef::new(3, generation)), Object::Null); + } + edited.set(ObjRef::new(3, 2), Object::Int(99)); + assert_eq!(*edited.get(ObjRef::new(3, 1)), Object::Null); + assert_eq!(*edited.get(ObjRef::new(3, 2)), Object::Int(99)); + assert_eq!(*doc.get(ObjRef::new(3, 1)), Object::Int(42)); + } +} + +/// A damaged row may name generation zero while the retained object and catalog use one. +fn contradictory_dictionary() -> Arc> { + object_file_with_entry( + &[(1, 0, "<< /Type /Catalog /Pages 2 0 R /Current 3 1 R >>"), (2, 0, "<< /Type /Pages /Kids [] /Count 0 >>"), (3, 1, "<< /Value 42 >>")], + Some((3, 0, false)), + ) +} + +fn assert_repaired_dictionary(bytes: Vec, value: i32) { + let bytes = Arc::new(bytes); + let oracle = hayro_syntax::Pdf::new(bytes.clone()).unwrap(); + let catalog = oracle.xref().get::>(ObjectIdentifier::new(1, 0)).unwrap(); + assert_eq!(catalog.get::>(b"Current").unwrap().get::(b"Value"), Some(value)); + let back = Document::open(bytes).unwrap(); + let catalog = back.get(back.root().unwrap()); + let current = back.resolve(catalog.as_dict().unwrap().get(b"Current").unwrap()); + assert_eq!(current.as_dict().unwrap().int(b"Value"), Some(i64::from(value))); +} + +#[test] +fn repaired_header_references_survive_full_saves_and_dictionary_edits() { + let original = contradictory_dictionary(); + let doc = Document::open(original.clone()).unwrap(); + for object_streams in [false, true] { + assert_repaired_dictionary(write_full(&doc, &SaveOptions { object_streams, ..SaveOptions::default() }).unwrap(), 42); + } + let mut edited = doc.clone(); + edited.update_dict(ObjRef::new(3, 1), |dict| dict.set(b"Value".to_vec(), Object::Int(77))).unwrap(); + assert_eq!(edited.generation(3), 1); + let incremental = write_incremental(&edited, &SaveOptions::default()).unwrap(); + assert!(edited.full_save_required(), "repair-aware dictionary edits normalize on full rewrite"); + assert_repaired_dictionary(incremental, 77); + for object_streams in [false, true] { + assert_repaired_dictionary(write_full(&edited, &SaveOptions { object_streams, ..SaveOptions::default() }).unwrap(), 77); + } + let mut restored = doc.clone(); + restored.overlay_changes_from(&edited).unwrap(); + assert_repaired_dictionary(write_incremental(&restored, &SaveOptions::default()).unwrap(), 77); +} + +#[test] +fn free_xref_slots_never_revive_retained_object_headers() { + for warm in [false, true] { + let doc = Document::open(contradictory_entry(true)).unwrap(); + assert!(matches!(doc.xref_entry(3), Some(XrefEntry::Free { next_generation: 2 }))); + if warm { + assert_eq!(*doc.try_get(3).unwrap(), Object::Null); + } + for generation in [0, 1, 2] { + assert_eq!(*doc.get(ObjRef::new(3, generation)), Object::Null); + } + assert_eq!(*doc.resolve(&Object::Ref(ObjRef::new(3, 1))), Object::Null); + assert!(doc.repair_log().is_empty(), "{:?}", doc.repair_log()); + } + // Control: the retained bytes really are a readable object when its slot is live. + let live = Document::open(contradictory_entry(false)).unwrap(); + assert_eq!(*live.try_get(3).unwrap(), Object::Int(42)); +} + +#[test] +fn incremental_generation_identity_is_visible_to_an_independent_reader() { + let original = reused_object(); + let mut doc = Document::open(original.clone()).unwrap(); + doc.set(ObjRef::new(3, 1), Object::Int(77)); + let bytes = write_incremental(&doc, &SaveOptions::default()).unwrap(); + assert!(bytes.starts_with(&original)); + let oracle = hayro_syntax::Pdf::new(Arc::new(bytes)).unwrap(); + let catalog = oracle.xref().get::>(ObjectIdentifier::new(1, 0)).unwrap(); + assert_eq!(catalog.get::(b"Current"), Some(77)); + assert_eq!(catalog.get::(b"Stale"), None); +} + +#[test] +fn generation_repair_diagnostics_are_bounded_without_disabling_recovery() { + let mut objects = vec![(1, 0, "<< /Type /Catalog /Pages 2 0 R >>"), (2, 0, "<< /Type /Pages /Kids [] /Count 0 >>")]; + objects.extend((3..303).map(|num| (num, 1, "42"))); + let bytes = object_file(&objects); + let bytes = String::from_utf8(bytes.as_ref().clone()).unwrap().replace("00001 n \n", "00000 n \n"); + for materialized in [false, true] { + let bytes = if materialized { bytes.replace("/Size 303", "/Size 302") } else { bytes.clone() }; + let doc = Document::open(Arc::new(bytes.into_bytes())).unwrap(); + assert_eq!(doc.repair_log().len(), 257); + assert!(doc.repair_log().last().unwrap().contains("omitted after 256 objects")); + assert_eq!(*doc.get(ObjRef::new(3, 1)), Object::Int(42)); + assert_eq!(*doc.get(ObjRef::new(302, 1)), Object::Int(42)); + assert_eq!(*doc.get(ObjRef::new(302, 2)), Object::Null); + assert_eq!(doc.repair_log().len(), 257); + } +} + +#[test] +fn references_require_the_current_generation_even_with_a_warm_cache() { + let bytes = reused_object(); + let oracle = hayro_syntax::Pdf::new(bytes.clone()).unwrap(); + assert_eq!(oracle.xref().get::(ObjectIdentifier::new(3, 0)), None); + assert_eq!(oracle.xref().get::(ObjectIdentifier::new(3, 1)), Some(42)); + + let doc = Document::open(bytes).unwrap(); + assert!(doc.repair_log().is_empty()); + assert_eq!(*doc.get(ObjRef::new(3, 0)), Object::Null); + assert_eq!(*doc.try_get(3).unwrap(), Object::Int(42)); + assert_eq!(*doc.get(ObjRef::new(3, 1)), Object::Int(42)); + assert_eq!(*doc.get(ObjRef::new(3, 0)), Object::Null); + assert_eq!(*doc.resolve(&Object::Ref(ObjRef::new(3, 0))), Object::Null); + assert_eq!(*doc.get(ObjRef::new(3, 2)), Object::Null); +} + +#[test] +fn overlay_generation_changes_do_not_retarget_old_references_or_clones() { + let mut doc = Document::open(reused_object()).unwrap(); + let snapshot = doc.clone(); + let old = ObjRef::new(3, 1); + let new = ObjRef::new(3, 2); + doc.free(old); + assert_eq!(*doc.get(old), Object::Null); + doc.set(new, Object::Int(99)); + assert_eq!(*doc.get(old), Object::Null); + assert_eq!(*doc.get(new), Object::Int(99)); + assert_eq!(*snapshot.get(old), Object::Int(42)); + assert_eq!(*snapshot.get(new), Object::Null); + + let saved = write_incremental(&doc, &SaveOptions::default()).unwrap(); + let back = Document::open(Arc::new(saved)).unwrap(); + assert!(back.repair_log().is_empty()); + assert_eq!(*back.get(old), Object::Null); + assert_eq!(*back.get(new), Object::Int(99)); +} + +#[test] +fn restoring_snapshot_preserves_changes_to_nonzero_generation_objects() { + for generation in [1, 2] { + let mut saved = Document::open(reused_object()).unwrap(); + let mut current = saved.clone(); + let reference = ObjRef::new(3, generation); + current.set(reference, Object::Int(99)); + current.update_dict(current.root().unwrap(), |catalog| catalog.set(b"Current".to_vec(), Object::Ref(reference))).unwrap(); + + saved.overlay_changes_from(¤t).unwrap(); + assert_eq!(*saved.get(reference), Object::Int(99), "generation {generation}"); + let back = Document::open(Arc::new(write_incremental(&saved, &SaveOptions::default()).unwrap())).unwrap(); + assert!(back.repair_log().is_empty()); + assert_eq!(*back.get(reference), Object::Int(99), "generation {generation}"); + let catalog = back.get(back.root().unwrap()); + assert_eq!(*back.resolve(catalog.as_dict().unwrap().get(b"Current").unwrap()), Object::Int(99)); + } +} + +#[test] +fn full_save_does_not_turn_stale_references_into_live_ones() { + let doc = Document::open(reused_object()).unwrap(); + for object_streams in [false, true] { + let saved = write_full(&doc, &SaveOptions { object_streams, ..SaveOptions::default() }).unwrap(); + let saved = Arc::new(saved); + let oracle = hayro_syntax::Pdf::new(saved.clone()).unwrap(); + let catalog = oracle.xref().get::>(ObjectIdentifier::new(1, 0)).unwrap(); + assert_eq!(catalog.get::(b"Current"), Some(42)); + assert_eq!(catalog.get::(b"Stale"), None); + let back = Document::open(saved).unwrap(); + assert!(back.repair_log().is_empty()); + let catalog = back.get(back.root().unwrap()); + let catalog = catalog.as_dict().unwrap(); + assert_eq!(catalog.get(b"Stale"), Some(&Object::Null)); + assert_eq!(*back.resolve(catalog.get(b"Current").unwrap()), Object::Int(42)); + } +} + +#[test] +fn compressed_objects_only_accept_generation_zero() { + let doc = Document::open(reused_object()).unwrap(); + let saved = write_full(&doc, &SaveOptions::default()).unwrap(); + let back = Document::open(Arc::new(saved)).unwrap(); + let catalog = back.get(back.root().unwrap()); + let current = catalog.as_dict().unwrap().reference(b"Current").unwrap(); + assert_eq!(current.generation, 0); + assert!(matches!(back.xref_entry(current.num), Some(XrefEntry::InStream { .. }))); + assert_eq!(*back.get(current), Object::Int(42)); + assert_eq!(*back.get(ObjRef::new(current.num, 1)), Object::Null); +} + +#[test] +fn indirect_stream_lengths_require_the_referenced_generation() { + for generation in [0, 1] { + // The embedded endstream is data when /Length resolves, but the existing missing- + // length recovery stops there when the reference is stale. + let stream = format!("<< /Length 4 {generation} R >>\nstream\nA\nendstream\nB\nendstream"); + let doc = Document::open(object_file(&[ + (1, 0, "<< /Type /Catalog /Pages 2 0 R >>"), + (2, 0, "<< /Type /Pages /Kids [] /Count 0 >>"), + (3, 0, &stream), + (4, 1, "13"), + ])) + .unwrap(); + let object = doc.get(ObjRef::new(3, 0)); + let Object::Stream(stream) = &*object else { panic!("expected stream") }; + assert_eq!(&stream.raw[..], if generation == 1 { b"A\nendstream\nB".as_slice() } else { b"A".as_slice() }); + } +} + +/// A newer valid row must not hide damage in a retained older revision. +fn shadowed_generation_disagreement(materialized: bool, shifted: bool) -> Arc> { + let original = contradictory_dictionary(); + let text = std::str::from_utf8(&original).unwrap(); + let previous: usize = text.rsplit("startxref\n").next().unwrap().lines().next().unwrap().parse().unwrap(); + let mut bytes = (*original).clone(); + let offset = bytes.len(); + bytes.extend_from_slice(b"3 1 obj\n<< /Value 77 >>\nendobj\n"); + let xref = bytes.len(); + let size = if materialized { 3 } else { 4 }; + bytes.extend_from_slice( + format!("xref\n3 1\n{offset:010} 00001 n \ntrailer\n<< /Size {size} /Root 1 0 R /Prev {previous} >>\nstartxref\n{xref}\n%%EOF\n").as_bytes(), + ); + if shifted { + bytes.splice(..0, b"junk before the header\n".iter().copied()); + } + Arc::new(bytes) +} + +#[test] +fn retained_generation_disagreements_are_reported_in_materialized_mode_too() { + for materialized in [false, true] { + for shifted in [false, true] { + let doc = Document::open(shadowed_generation_disagreement(materialized, shifted)).unwrap(); + assert_eq!(doc.revisions().len(), 2, "the xref chain is retained"); + assert!(!doc.reconstructed()); + assert_eq!(doc.get(ObjRef::new(3, 1)).as_dict().unwrap().int(b"Value"), Some(77)); + assert!( + doc.repair_log().iter().any(|line| line.contains("object 3: xref generation 0 disagrees with object header generation 1")), + "materialized={materialized}, shifted={shifted}: {:?}", + doc.repair_log() + ); + } + } +} + +#[test] +fn repair_aware_edits_keep_both_aliases_and_full_save_normalizes_them() { + let mut doc = Document::open(contradictory_entry(false)).unwrap(); + let snapshot = doc.clone(); + doc.set_existing(ObjRef::new(3, 0), Object::Int(77)).unwrap(); + for generation in [0, 1] { + assert_eq!(*doc.get(ObjRef::new(3, generation)), Object::Int(77)); + assert_eq!(*snapshot.get(ObjRef::new(3, generation)), Object::Int(42)); + } + let state = doc.read_state(); + assert!(doc.set_existing(ObjRef::new(3, 2), Object::Int(99)).is_err()); + assert!(state.matches(&doc)); + assert_eq!(*doc.get(ObjRef::new(3, 2)), Object::Null); + let mut restored = snapshot.clone(); + restored.overlay_changes_from(&doc).unwrap(); + for changed in [&doc, &restored] { + assert!(changed.full_save_required()); + for bytes in [ + write_incremental(changed, &SaveOptions::default()).unwrap(), + write_full(changed, &SaveOptions { object_streams: false, ..SaveOptions::default() }).unwrap(), + ] { + let bytes = Arc::new(bytes); + let oracle = hayro_syntax::Pdf::new(bytes.clone()).unwrap(); + let catalog = oracle.xref().get::>(ObjectIdentifier::new(1, 0)).unwrap(); + assert_eq!(catalog.get::(b"Current"), Some(77)); + assert_eq!(catalog.get::(b"Stale"), Some(77), "both source aliases stay live"); + let back = Document::open(bytes).unwrap(); + assert!(back.repair_log().is_empty(), "normalized output has no generation repairs"); + } + } + // Explicit replacement keeps its narrower, caller-supplied generation contract. + doc.set(ObjRef::new(3, 2), Object::Int(99)); + assert_eq!(*doc.get(ObjRef::new(3, 0)), Object::Null); + assert_eq!(*doc.get(ObjRef::new(3, 1)), Object::Null); + assert_eq!(*doc.get(ObjRef::new(3, 2)), Object::Int(99)); +} + +#[test] +fn hybrid_generation_diagnostics_observe_rows_before_the_table_hides_them() { + let mut bytes = b"%PDF-1.7\n".to_vec(); + let mut offsets = Vec::new(); + for (num, generation, body) in [(1, 0, "<< /Type /Catalog /Pages 2 0 R >>"), (2, 0, "<< /Type /Pages /Kids [] /Count 0 >>"), (3, 1, "42")] { + offsets.push(bytes.len()); + bytes.extend_from_slice(format!("{num} {generation} obj\n{body}\nendobj\n").as_bytes()); + } + let hybrid = bytes.len(); + offsets.push(hybrid); + bytes.extend_from_slice(b"4 0 obj\n<< /Type /XRef /Size 5 /W [1 4 2] /Index [3 1] /Length 7 >>\nstream\n"); + bytes.push(1); + bytes.extend_from_slice(&(offsets[2] as u32).to_be_bytes()); + bytes.extend_from_slice(&0u16.to_be_bytes()); + bytes.extend_from_slice(b"\nendstream\nendobj\n"); + let xref = bytes.len(); + bytes.extend_from_slice(b"xref\n0 5\n0000000000 65535 f \n"); + for (i, offset) in offsets.into_iter().enumerate() { + let generation = u16::from(i == 2); + bytes.extend_from_slice(format!("{offset:010} {generation:05} n \n").as_bytes()); + } + bytes.extend_from_slice(format!("trailer\n<< /Size 5 /Root 1 0 R /XRefStm {hybrid} >>\nstartxref\n{xref}\n%%EOF\n").as_bytes()); + let doc = Document::open(Arc::new(bytes)).unwrap(); + assert!(!doc.reconstructed()); + assert!(doc.xref_oddities().iter().any(|(_, num)| *num == 3), "the conflicting hybrid row is set aside"); + assert_eq!(*doc.get(ObjRef::new(3, 1)), Object::Int(42)); + assert_eq!(*doc.get(ObjRef::new(3, 0)), Object::Null, "the table still determines accepted references"); + assert!(doc.repair_log().iter().any(|line| line.contains("object 3: xref generation 0 disagrees with object header generation 1"))); +} + +#[test] +fn discarded_xref_repair_notes_do_not_survive_object_header_reconstruction() { + let bytes = + String::from_utf8((*contradictory_dictionary()).clone()).unwrap().replace("/Root 1 0 R", "/Root 99 0 R").replace("/Size 4", "/Size 3"); + let doc = Document::open(Arc::new(bytes.into_bytes())).unwrap(); + assert!(doc.reconstructed()); + assert!(doc.repair_log().iter().all(|line| !line.contains("xref generation")), "{:?}", doc.repair_log()); +} + +#[test] +fn restoring_equal_values_still_restores_the_accepted_generation_aliases() { + for edited in [false, true] { + for value in [42, 77] { + let mut current = Document::open(contradictory_entry(false)).unwrap(); + if edited { + current.set_existing(ObjRef::new(3, 0), Object::Int(42)).unwrap(); + } + let mut saved = Document::open(object_file(&[ + (1, 0, "<< /Type /Catalog /Pages 2 0 R /Stale 3 0 R /Current 3 1 R >>"), + (2, 0, "<< /Type /Pages /Kids [] /Count 0 >>"), + (3, 0, &value.to_string()), + ])) + .unwrap(); + assert_eq!(*saved.get(ObjRef::new(3, 1)), Object::Null); + saved.overlay_changes_from(¤t).unwrap(); + for generation in [0, 1] { + assert_eq!(*saved.get(ObjRef::new(3, generation)), Object::Int(42), "edited={edited}, value={value}"); + } + assert!(saved.full_save_required()); + } + } +} + +#[test] +fn repaired_page_tree_aliases_keep_one_identity_in_full_saves() { + let source = object_file_with_entry( + &[ + (1, 0, "<< /Type /Catalog /Pages 2 1 R /Stale 2 2 R >>"), + (2, 1, "<< /Type /Pages /Kids [3 0 R] /Count 1 >>"), + (3, 0, "<< /Type /Page /Parent 2 0 R /MediaBox [0 0 10 10] >>"), + ], + Some((2, 0, false)), + ); + for edited in [false, true] { + let mut doc = Document::open(source.clone()).unwrap(); + if edited { + doc.update_dict(ObjRef::new(2, 0), |dict| dict.set(b"Edited".to_vec(), Object::Bool(true))).unwrap(); + } + let mut restored = Document::open(source.clone()).unwrap(); + restored.overlay_changes_from(&doc).unwrap(); + for current in [&doc, &restored] { + for object_streams in [false, true] { + let opts = SaveOptions { object_streams, ..SaveOptions::default() }; + let mut saves = vec![write_full(current, &opts).unwrap()]; + if edited { + saves.push(write_incremental(current, &opts).unwrap()); + } + for bytes in saves { + let back = Document::open(Arc::new(bytes)).unwrap(); + let catalog = back.get(back.root().unwrap()); + let catalog = catalog.as_dict().unwrap(); + let pages_ref = catalog.reference(b"Pages").unwrap(); + let pages = back.get(pages_ref); + let child_ref = pages.as_dict().unwrap().get(b"Kids").unwrap().as_array().unwrap()[0].as_ref().unwrap(); + let child = back.get(child_ref); + assert_eq!(child.as_dict().unwrap().reference(b"Parent"), Some(pages_ref), "edited={edited}, packed={object_streams}"); + assert!(matches!(catalog.get(b"Stale"), Some(Object::Null)), "unrelated generations stay null"); + assert!(back.repair_log().is_empty()); + } + } + } + } +} + +#[test] +fn restoring_explicit_sets_removes_old_generation_aliases() { + for value in [42, 77] { + let mut current = Document::open(contradictory_entry(false)).unwrap(); + current.set(ObjRef::new(3, 0), Object::Int(42)); + for repaired_overlay in [false, true] { + let mut saved = Document::open(contradictory_entry(false)).unwrap(); + if repaired_overlay { + saved.set_existing(ObjRef::new(3, 0), Object::Int(value)).unwrap(); + } + saved.overlay_changes_from(¤t).unwrap(); + assert_eq!(*saved.get(ObjRef::new(3, 0)), Object::Int(42)); + assert_eq!(*saved.get(ObjRef::new(3, 1)), Object::Null, "value={value}, overlay={repaired_overlay}"); + } + } +} diff --git a/crates/cos/tests/it/main.rs b/crates/cos/tests/it/main.rs index 52cf05d6c2..1bd7ee1fd6 100644 --- a/crates/cos/tests/it/main.rs +++ b/crates/cos/tests/it/main.rs @@ -3,6 +3,7 @@ mod bounded_writer; mod corpus; mod encryption; +mod generations; mod proptests; mod shared_streams; mod writer; diff --git a/crates/cos/tests/it/writer.rs b/crates/cos/tests/it/writer.rs index eb03523189..2716b9242e 100644 --- a/crates/cos/tests/it/writer.rs +++ b/crates/cos/tests/it/writer.rs @@ -358,11 +358,32 @@ fn objects_referenced_by_the_encrypt_dictionary_stay_outside_object_streams() { let cf = doc.get(enc_ref).as_dict().unwrap().get(b"CF").cloned().unwrap(); let cf_ref = doc.add(cf); doc.update_dict(enc_ref, |d| d.set(b"CF".to_vec(), Object::Ref(cf_ref))).unwrap(); - let packed = write_full(&doc, &SaveOptions::default()).unwrap(); - let back = Document::open(Arc::new(packed.clone())).expect("reopens"); - assert!(back.repair_log().is_empty(), "{:?}", back.repair_log()); - assert_eq!(title(&back), b"Caf\xe9 (draft)"); - assert_eq!(hayro_pages(&packed), 4); + // Visit this alias from Root before the encryption walk reaches /CF's other alias. + doc.update_dict(doc.root().unwrap(), |d| d.set(b"CFProbe".to_vec(), Object::Ref(cf_ref))).unwrap(); + let classic = write_full(&doc, &SaveOptions { object_streams: false, ..SaveOptions::default() }).unwrap(); + let saved = Document::open(Arc::new(classic.clone())).unwrap(); + let cf_num = saved.get(saved.root().unwrap()).as_dict().unwrap().reference(b"CFProbe").unwrap().num; + for damaged in [false, true] { + let mut bytes = classic.clone(); + if damaged { + // Equal-length edits preserve the xref offsets and its generation-zero row. + for (from, to) in [(format!("\n{cf_num} 0 obj"), format!("\n{cf_num} 1 obj")), (format!("/CF {cf_num} 0 R"), format!("/CF {cf_num} 1 R"))] + { + let start = bytes.windows(from.len()).position(|w| w == from.as_bytes()).unwrap(); + bytes[start..start + from.len()].copy_from_slice(to.as_bytes()); + } + } + let doc = Document::open(Arc::new(bytes)).unwrap(); + let packed = write_full(&doc, &SaveOptions::default()).unwrap(); + let back = Document::open(Arc::new(packed.clone())).expect("reopens"); + assert!(back.repair_log().is_empty(), "{:?}", back.repair_log()); + assert_eq!(title(&back), b"Caf\xe9 (draft)"); + assert_eq!(hayro_pages(&packed), 4); + let probe = back.get(back.root().unwrap()).as_dict().unwrap().reference(b"CFProbe").unwrap(); + let encrypt = back.get(back.trailer().reference(b"Encrypt").unwrap()); + assert_eq!(encrypt.as_dict().unwrap().reference(b"CF"), Some(probe)); + assert!(matches!(back.xref_entry(probe.num), Some(pdfcraft_cos::XrefEntry::InFile { .. })), "crypt filters must stay outside object streams"); + } } #[test] diff --git a/crates/preflight/src/lib.rs b/crates/preflight/src/lib.rs index 1df39ebf05..1fb7272b33 100644 --- a/crates/preflight/src/lib.rs +++ b/crates/preflight/src/lib.rs @@ -894,7 +894,7 @@ pub fn convert(doc: &mut Document, level: Level) -> Result Result Result Result Arc> { + let mut bytes = b"%PDF-1.7\n".to_vec(); + let bodies = [ + "<< /Type /Catalog /Pages 2 0 R /HeaderImage 5 1 R /RowImage 5 0 R /StaleImage 5 2 R >>", + "<< /Type /Pages /Kids [3 0 R] /Count 1 >>", + "<< /Type /Page /Parent 2 0 R /MediaBox [0 0 10 10] /Resources << /XObject << /Im 5 1 R >> >> >>", + "null", + "<< /Type /XObject /Subtype /Image /Width 1 /Height 1 /BitsPerComponent 8 /ColorSpace /DeviceGray /Interpolate true /Length 1 >>\nstream\nX\nendstream", + ]; + let mut offsets = Vec::new(); + for (i, body) in bodies.iter().enumerate() { + offsets.push(bytes.len()); + let generation = u16::from(i == 4); + bytes.extend_from_slice(format!("{} {generation} obj\n{body}\nendobj\n", i + 1).as_bytes()); + } + let xref = bytes.len(); + bytes.extend_from_slice(b"xref\n0 6\n0000000000 65535 f \n"); + for (i, offset) in offsets.into_iter().enumerate() { + let generation = u16::from(i == 4 && !damaged); + bytes.extend_from_slice(format!("{offset:010} {generation:05} n \n").as_bytes()); + } + bytes.extend_from_slice(format!("trailer\n<< /Size 6 /Root 1 0 R >>\nstartxref\n{xref}\n%%EOF\n").as_bytes()); + Arc::new(bytes) +} + +fn assert_images(doc: &Document, converted: bool, dual: bool) { + let root = doc.get(doc.root().unwrap()); + let catalog = root.as_dict().unwrap(); + for key in [b"HeaderImage".as_slice(), b"RowImage".as_slice()] { + let value = doc.resolve(catalog.get(key).unwrap()); + if key == b"RowImage" && !dual { + assert_eq!(*value, Object::Null); + } else { + let Object::Stream(image) = &*value else { panic!("{key:?} became {value:?}") }; + assert_eq!(image.dict.get(b"Interpolate"), Some(&Object::Bool(!converted))); + assert_eq!(&image.raw[..], b"X"); + } + } + assert_eq!(*doc.resolve(catalog.get(b"StaleImage").unwrap()), Object::Null); +} + +#[test] +fn preflight_preserves_repaired_image_references_in_snapshots_and_saves() { + for damaged in [false, true] { + let original = image_file(damaged); + let mut doc = Document::open(original.clone()).unwrap(); + assert_images(&doc, false, damaged); + let snapshot = doc.clone(); + let report = convert(&mut doc, Level::A2b).unwrap(); + assert!(report.fixed.iter().any(|line| line.contains("cleared image interpolation"))); + assert_images(&doc, true, damaged); + assert_eq!(doc.full_save_required(), damaged); + assert_images(&snapshot, false, damaged); + let mut restored = snapshot.clone(); + restored.overlay_changes_from(&doc).unwrap(); + assert_images(&restored, true, damaged); + for changed in [&doc, &restored] { + for object_streams in [false, true] { + let opts = SaveOptions { object_streams, ..SaveOptions::default() }; + let incremental = write_incremental(changed, &opts).unwrap(); + if !damaged { + assert!(incremental.starts_with(&original)); + } + assert_images(&Document::open(Arc::new(incremental)).unwrap(), true, damaged); + let full = write_full(changed, &opts).unwrap(); + let reopened = Document::open(Arc::new(full)).unwrap(); + assert_images(&reopened, true, damaged); + if damaged { + let catalog = reopened.get(reopened.root().unwrap()); + let catalog = catalog.as_dict().unwrap(); + assert_eq!(catalog.reference(b"HeaderImage"), catalog.reference(b"RowImage"), "saved aliases must share one object"); + } + } + } + } +} diff --git a/docs/roadmap.md b/docs/roadmap.md index 8de636a500..25c0da7e00 100644 --- a/docs/roadmap.md +++ b/docs/roadmap.md @@ -63,6 +63,10 @@ > **Last reviewed:** 2026-10-17 · **Last updated:** 2026-10-17 · **Change:** minor (M8, gap 6 #2662: the redaction audit counts a cover over one word of a long line; estimates unchanged) · **Target:** Adobe Acrobat Pro (Acrobat DC, continuous track 26.002.21931, macOS) > **Last reviewed:** 2026-10-13 · **Last updated:** 2026-10-13 · **Change:** minor (M10: dotted-leader contents entries export as complete paragraphs; estimates unchanged) · **Target:** Adobe Acrobat Pro (Acrobat DC, continuous track 26.002.21931, macOS) > **Last reviewed:** 2026-10-22 · **Last updated:** 2026-10-22 · **Change:** minor (M6: calculation scripts read their notation only from a comment, and a regular expression after a condition as one; estimates unchanged) · **Target:** Adobe Acrobat Pro (Acrobat DC, continuous track 26.002.21931, macOS) + +> **Last reviewed:** 2026-10-20 · **Last updated:** 2026-10-20 · **Change:** patch (M1, #1143: repaired aliases survive preflight, snapshots and normalized saves; retained materialized revisions report generation disagreements; estimates unchanged) + +> **Last reviewed:** 2026-10-18 · **Last updated:** 2026-10-18 · **Change:** minor (M1, #1143: explicit live xref/header generation repair policy and bounded diagnostics; estimates unchanged) > **Last reviewed:** 2026-10-18 · **Last updated:** 2026-10-18 · **Change:** minor (M5, PR #1971: bounded nested FDF export excludes source page/page-tree dictionaries; refreshed against upstream 4db6ac4a; estimates unchanged) > **Last reviewed:** 2026-10-18 · **Last updated:** 2026-10-18 · **Change:** patch (M2 #3086: dotted TOC leaders no longer hide word spaces; estimates unchanged) > @@ -78,6 +82,7 @@ > **Last reviewed:** 2026-10-15 · **Last updated:** 2026-10-15 · **Change:** minor (M6: AcroForm XML text; estimates unchanged) > **Last reviewed:** 2026-10-15 · **Last updated:** 2026-10-15 · **Change:** minor (M6: preserve XML/XFDF attribute whitespace; refreshed upstream; estimates unchanged) · **Target:** Adobe Acrobat Pro (Acrobat DC, continuous track 26.002.21931, macOS) > **Last reviewed:** 2026-10-15 · **Last updated:** 2026-10-15 · **Change:** minor (M1: raw-deflate fallback preserves decoding-limit errors; refreshed upstream; estimates unchanged) · **Target:** Adobe Acrobat Pro (Acrobat DC, continuous track 26.002.21931, macOS) +> **Last reviewed:** 2026-10-15 · **Last updated:** 2026-10-15 · **Change:** minor (M1: refresh object-generation reference identity against current upstream; estimates unchanged) > **Last reviewed:** 2026-10-14 · **Last updated:** 2026-10-15 · **Change:** minor (M2: objects in one large object stream are read in linear time; estimates unchanged) · **Target:** Adobe Acrobat Pro (Acrobat DC, continuous track 26.002.21931, macOS) > **Last reviewed:** 2026-10-16; **Last updated:** 2026-10-16; **Change:** minor (M5 #2669, item 5: reply groups and page root counts per frame; estimates unchanged) > @@ -204,7 +209,7 @@ total matches the parity estimate. | M | Milestone | Plan size (h) | Done | Remaining (h, 2026-10-10) | Notes | |---|---|---|---|---|---| | M0 | Skeleton: workspace, xtask gates, CI | 15–30 | 98% | 1–2 | GitHub workflow, `deny.toml` (licence audit of every dependency), parity checklist (826 features, `xtask parity`) done. Missing: remaining crate stubs, testkit/oracle crates | -| M1 | COS: filters, crypt, parser, xref, writer | 120–200 | 80% | 20–40 | Done:
- filters and crypt: every standard-security revision R2–R6 (RC4, AES-128/256), SASLprep, permissions, creating encryption;
- cos: parse and repair (including invalid xref object-number ranges), decoded-stream limits including unfiltered data, decrypt on load, re-encrypt on save, incremental and full writing.
Corpus: open/edit/save passes on 958 files, and all 7 password-protected files open.
Full saves now pack objects into compressed object streams. An unchanged incremental save of a classic table writes a zero-length cross-reference subsection instead of a section with no subsections (#1339). A full rewrite keeps unknown trailer entries, including objects reachable only from them (#1338). Fuzzing runs nightly (`xtask fuzz`). Raw-deflate fallback preserves decoding-limit errors, including through tolerant COS stream decoding. Done since: #2658: macOS atomic saves retain destination extended attributes, including Finder tags, and reject partial resource-fork reads after size changes. Metadata failures retain the original PDF and unsaved edit. Missing: ≥ 250 tests, own image codecs | +| M1 | COS: filters, crypt, parser, xref, writer | 120–200 | 80% | 20–40 | Done:
- filters and crypt: every standard-security revision R2–R6 (RC4, AES-128/256), SASLprep, permissions, creating encryption;
- cos: parse and repair (including invalid xref object-number ranges), decoded-stream limits including unfiltered data, decrypt on load, re-encrypt on save, incremental and full writing.
Corpus: open/edit/save passes on 958 files, and all 7 password-protected files open.
Full saves now pack objects into compressed object streams. An unchanged incremental save of a classic table writes a zero-length cross-reference subsection instead of a section with no subsections (#1339). A full rewrite keeps unknown trailer entries, including objects reachable only from them (#1338). Fuzzing runs nightly (`xtask fuzz`). Raw-deflate fallback preserves decoding-limit errors, including through tolerant COS stream decoding. Refreshed against upstream `99941daa` (2026-10-15). COS reference lookups and full saves distinguish complete object references, so stale generations cannot expose replacement content. Snapshot restoration preserves edited nonzero-generation objects through incremental save/reopen. Synthetic regressions cover caches, overlays, snapshots, stream lengths and classic/packed saves. For damaged live rows, either the xref generation or a same-number header at its indexed offset may identify the object; neither-match and free slots remain null. Header probes read at most 128 bytes per candidate offset, and opening records at most 256 object disagreements plus a summary without expanding compact xref rows. Synthetic dictionary-edit and independent save/reopen controls cover this repair policy. Repair-aware preflight edits and restored snapshots retain both accepted aliases, with full rewrite for repaired edits; classic/packed output coalesces them to one object identity. Materialized diagnostics inspect retained primary/hybrid rows before folding, with the same cap. Done since: #2658: macOS atomic saves retain destination extended attributes, including Finder tags, and reject partial resource-fork reads after size changes. Metadata failures retain the original PDF and unsaved edit. Missing: ≥ 250 tests, own image codecs | | M2 | Model, render, text | 200–350 | 15% | 130–230 | hayro bootstrap renderer (vendored patches); inspector predictor scratch is bounded by available input, and JPX sample rescaling handles 1–32 bits safely. Text extraction reaches word-F1 0.98 against pdftotext. Short same-line redraws split before placement rather than interleaving. Missing: model crate, fonts, DisplayList, renderer independent of hayro. Inspector page-label traversal and output are resource-bounded, with physical-number fallback warnings. Packed TIFF-predictor RGB rows retain later rows in decoding and rendering. Page marks, added content and print output retain small page-unit transform coefficients when serialized. Forms without Resources use page Properties for ActualText and layer visibility; own dictionaries stop fallback. Oversized missing inline-image samples, including header and terminator overhead, no longer pin the content window (#3107 / PR #3438); normal large streamed images retain deterministic sample boundaries. Radial luminosity fades have synthetic coverage for the existing #2132 mask guard, including per-pixel references, antialiased clips, previews, scales and tiles; no additional production fix was needed. Dotted TOC leaders no longer inflate letter tracking or hide word spaces (#3086); phrase search and selection are covered. Short tracked ellipses retain their gap samples and remain searchable. Missing spacing glyphs in non-embedded Japanese CID substitutes render blank without changing PDF advances or Unicode (#3601). The desktop viewer can use installed CJK fonts, including Windows GBK Song/Hei/Kai/FangSong and subset-prefixed Founder names; Combine thumbnails and accessibility figure previews use the same display-only fallback, without changing export or print font policy (#234, continuation of #1567, review #4172). Print previews exclude installed faces and isolate their font cache; GBK regressions cover view/print switching for inline and pooled whole-page/tile rendering. Installed-font helpers and their regression contracts have function documentation. GBK regression rasters match explicit-Unicode references with independently selected installed faces; wrong-glyph and wrong-family negative controls fail. Integration with main fa5f9f8c retains viewer fallback across the split engine session modules, renderer tests and Combine thumbnail workers. | | M2 | Model, render, text | 200–350 | 15% | 130–230 | hayro bootstrap renderer (vendored patches); inspector predictor scratch is bounded by available input, and JPX sample rescaling handles 1–32 bits safely. Text extraction reaches word-F1 0.98 against pdftotext. Short same-line redraws split before placement rather than interleaving. Missing: model crate, fonts, DisplayList, renderer independent of hayro. Inspector page-label traversal and output are resource-bounded, with physical-number fallback warnings. Packed TIFF-predictor RGB rows retain later rows in decoding and rendering. The renderer keeps its one-point crop-size floor in physical points after UserUnit scaling, with regression coverage for crop origins, rotations and f32 reciprocal rounding. Existing unit policy and nearly-zero fallback are retained. | | M2 | Model, render, text | 200–350 | 15% | 130–230 | hayro bootstrap renderer (vendored patches); inspector predictor scratch is bounded by available input, and JPX sample rescaling handles 1–32 bits safely. Text extraction reaches word-F1 0.98 against pdftotext. Short same-line redraws split before placement rather than interleaving. Missing: model crate, fonts, DisplayList, renderer independent of hayro. Inspector page-label traversal and output are resource-bounded, with physical-number fallback warnings | @@ -352,6 +357,8 @@ Estimates unchanged. | 2026-10-20 | minor | M10, #3532 partial: vector printing retains existing appearances with absent Subtype; estimates unchanged. | | 2026-10-19 | minor | M4 #3309: named-destination pruning keeps large documents usable; 60,000-name and deep-tree regressions verify targets through undo/redo and both save modes. Estimates unchanged. | | 2026-10-19 | minor | M6: Text field table (rows and columns of text fields as one edit, `form_add_field_table`); estimates unchanged | +| 2026-10-20 | patch | M1, #1143 review: preserve accepted aliases through preflight/snapshots, normalize them to one output identity with full-save fallback for repaired edits, and report retained materialized/hybrid generation disagreements before folding. Synthetic red/green and four-save controls; estimates unchanged. | +| 2026-10-18 | minor | M1, #1143: a live reference may match the xref generation or a same-number header at its indexed offset; neither-match and free slots remain null. Bounded repair diagnostics and synthetic cache, free-slot, edit and classic/packed save regressions; estimates unchanged. | | 2026-10-18 | minor | M5, PR #1971: nested source `/Page` and `/Pages` dictionaries export as null without traversing page content or the page tree. Synthetic saved/reopened three-page regression retains unrelated metadata and aliases. FDF import semantics, copied-graph limits and estimates unchanged. | | 2026-10-15 | minor | M5: budget retained FDF entries before cloning; discovery keeps lazy array owners and annotation positions, with alias/cache-eviction stress fixtures. Limits cover the copied graph; source parsing/page discovery retain existing COS/parser limits, with no total source-memory bound. FDF import semantics and estimates unchanged. | | 2026-10-15 | minor | M7: Edit text caret height follows its row, with a kittest regression; estimates unchanged | @@ -364,6 +371,7 @@ Estimates unchanged. | 2026-10-14 | minor | M1: raw-deflate fallback preserves decoding-limit errors in strict and tolerant decoding; synthetic filter and COS stream regressions. Estimates unchanged. | | 2026-10-16 | minor | M5 #2669, item 5: the Comments panel groups replies and counts all page roots once per frame. Estimates unchanged. | | 2026-10-15 | minor | M9 #1895: generic appearance rewriting deferred; separate validation overlays preserve original artwork, exports and printing. Baked captions remain unchanged (#79 open). Estimates unchanged. | +| 2026-10-15 | minor | M1: refresh object-generation reference identity, retaining current upstream progress and regression tests. Estimates unchanged. | | 2026-10-16 | minor | M10 #1201 partial: lossless image export applies component Decode pairs and Indexed mapping. Estimates unchanged. | | 2026-10-15 | minor | M6: CSV export and spreadsheet merging quote tabs, and import distinguishes quoted tabs from TSV delimiters without changing field identities. Estimates unchanged. | | 2026-10-13 | minor | M4: rotation queries and edits resolve indirect page-tree values; inherited, negative and large rotations verified through both saves. Estimates unchanged. | diff --git a/parity/acrobat-features.toml b/parity/acrobat-features.toml index 223fc66850..af283bed6e 100644 --- a/parity/acrobat-features.toml +++ b/parity/acrobat-features.toml @@ -70,8 +70,8 @@ milestone = "M1" status = "shipped" commands = [] tools = ["doc_info"] -evidence = ["crates/ui-egui/tests/it/viewer.rs::damaged_files_say_they_were_repaired"] -notes = "a notice says the file was repaired; Document Properties ▸ Advanced lists the repairs; doc_info reports them" +evidence = ["crates/ui-egui/tests/it/viewer.rs::damaged_files_say_they_were_repaired", "crates/cos/tests/it/generations.rs::contradictory_xref_generation_recovers_the_matching_header_and_records_repair", "crates/cos/tests/it/generations.rs::generation_repair_diagnostics_are_bounded_without_disabling_recovery", "crates/cos/tests/it/generations.rs::retained_generation_disagreements_are_reported_in_materialized_mode_too", "crates/cos/tests/it/generations.rs::hybrid_generation_diagnostics_observe_rows_before_the_table_hides_them", "crates/cos/tests/it/generations.rs::discarded_xref_repair_notes_do_not_survive_object_header_reconstruction"] +notes = "a notice says the file was repaired; Document Properties ▸ Advanced lists the repairs; doc_info reports them. COS records bounded live xref/header generation disagreements on open, including retained materialized and hybrid rows before folding; this follow-up validates the COS policy, without a fresh GUI claim." [[feature]] id = "core.xref-streams" @@ -82,8 +82,8 @@ milestone = "M1" status = "shipped" commands = [] tools = [] -evidence = ["crates/cos/tests/it/writer.rs::incremental_saves_stack_on_an_xref_stream_file", "crates/cos/tests/it/writer.rs::page_tree_loops_in_object_streams_do_not_crash_readers"] -notes = "" +evidence = ["crates/cos/tests/it/generations.rs::indirect_stream_lengths_require_the_referenced_generation", "crates/cos/tests/it/generations.rs::compressed_objects_only_accept_generation_zero", "crates/cos/tests/it/writer.rs::incremental_saves_stack_on_an_xref_stream_file", "crates/cos/tests/it/writer.rs::page_tree_loops_in_object_streams_do_not_crash_readers"] +notes = "Compressed objects accept generation zero only; indirect stream lengths are resolved by their complete reference. Synthetic regression coverage." [[feature]] id = "core.hybrid-xref" @@ -298,8 +298,8 @@ milestone = "M1" status = "shipped" commands = ["file.save_as"] tools = ["doc_save"] -evidence = ["crates/organize/src/tests.rs::full_save_drops_unreachable_objects", "crates/engine/src/tests.rs::full_save_is_readable_and_smaller_after_deletes"] -notes = "" +evidence = ["crates/cos/tests/it/generations.rs::full_save_does_not_turn_stale_references_into_live_ones", "crates/cos/tests/it/generations.rs::repaired_page_tree_aliases_keep_one_identity_in_full_saves", "crates/organize/src/tests.rs::full_save_drops_unreachable_objects", "crates/engine/src/tests.rs::full_save_is_readable_and_smaller_after_deletes"] +notes = "Full rewrites coalesce accepted repair aliases to one output object, preserving page-tree parent identity. They preserve reference-generation identity: stale references become null while live replacement objects remain reachable. Synthetic classic and object-stream save tests; this does not establish general rendering fidelity." [[feature]] id = "core.object-stream-compression" @@ -334,8 +334,8 @@ milestone = "M1" status = "shipped" commands = [] tools = [] -evidence = ["crates/cos/src/document/tests.rs::reads_classic_revision_and_edits_are_copy_on_write"] -notes = "" +evidence = ["crates/cos/tests/it/generations.rs::free_xref_slots_never_revive_retained_object_headers", "crates/cos/tests/it/generations.rs::restoring_equal_values_still_restores_the_accepted_generation_aliases", "crates/cos/tests/it/generations.rs::restoring_explicit_sets_removes_old_generation_aliases", "crates/preflight/tests/generation_repairs.rs::preflight_preserves_repaired_image_references_in_snapshots_and_saves", "crates/cos/tests/it/generations.rs::repaired_header_references_survive_full_saves_and_dictionary_edits", "crates/cos/tests/it/generations.rs::incremental_generation_identity_is_visible_to_an_independent_reader", "crates/cos/tests/it/generations.rs::restoring_snapshot_preserves_changes_to_nonzero_generation_objects", "crates/cos/tests/it/generations.rs::overlay_generation_changes_do_not_retarget_old_references_or_clones", "crates/cos/tests/it/generations.rs::references_require_the_current_generation_even_with_a_warm_cache", "crates/cos/src/document/tests.rs::reads_classic_revision_and_edits_are_copy_on_write"] +notes = "Generation-aware references stay distinct across warm caches, edit overlays and snapshots, including restoration and incremental save/reopen of edited nonzero-generation objects. A damaged live xref row may additionally resolve the matching same-number object header; unrelated generations and free slots remain null. Repair-aware edits preserve both aliases and require full rewrite; explicit replacement keeps only its supplied generation. Synthetic preflight/snapshot and classic/packed full/incremental controls include an independent reader. Broader real-file fidelity remains tracked separately." [[feature]] id = "core.write-validity" From 0f99608112988b3ecb231ae85d88009ccb7872ce Mon Sep 17 00:00:00 2001 From: Helen Kwok <107752876+helenkwok@users.noreply.github.com> Date: Mon, 26 Oct 2026 08:53:11 +1030 Subject: [PATCH 41/53] Contributors: credit helenkwok as Helen Kwok --- contributors/people.toml | 3 +++ 1 file changed, 3 insertions(+) diff --git a/contributors/people.toml b/contributors/people.toml index fecfb68c90..8ab196abc9 100644 --- a/contributors/people.toml +++ b/contributors/people.toml @@ -183,3 +183,6 @@ sync_github_name = true [people.Combiler01] sync_github_name = true + +[people.helenkwok] +real_name = "Helen Kwok" From d5b8a352487bbbaef15d3d586b96394b6300f74a Mon Sep 17 00:00:00 2001 From: Helen Kwok <107752876+helenkwok@users.noreply.github.com> Date: Mon, 26 Oct 2026 08:53:11 +1030 Subject: [PATCH 42/53] feat(web): page host mode, so a page can host PdfCraft in its own document (#4731) data-host="page" on the canvas opts a page that loads the module into its own document into the host bridge. The ?host=parent messages travel as non-bubbling CustomEvents on the canvas; host::Bridge holds a Transport (Parent or Page) with one send, and save, dirty, ready and failure reporting are shared. ?host=parent is checked first and unchanged. Hosting guide documents it. --- ROADMAP.md | 2 + apps/pdfcraft-web/Cargo.toml | 2 +- apps/pdfcraft-web/src/main.rs | 240 ++++++++++++++++++++++++---------- packaging/web/README.md | 48 +++++++ 4 files changed, 224 insertions(+), 68 deletions(-) diff --git a/ROADMAP.md b/ROADMAP.md index 1f753c500f..f82d8c13a6 100644 --- a/ROADMAP.md +++ b/ROADMAP.md @@ -30,6 +30,7 @@ > **Last reviewed:** 2026-10-29 | **Last updated:** 2026-10-29 | **Change:** minor (M2: scans keep their strokes at screen size and are smoothed when zoomed; M7 #2517: a narrowed then widened paragraph keeps the typed spaces the wrap consumed, after save and reopen; estimates unchanged) > **Last reviewed:** 2026-10-27 | **Last updated:** 2026-10-27 | **Change:** minor (M10: OCR model downloads use verified ONNX pairs with legacy discovery fallback; estimates unchanged) +> **Last reviewed:** 2026-10-25 | **Last updated:** 2026-10-25 | **Change:** minor (web: same-document host mode on the canvas, `data-host="page"`; estimates unchanged) > **Last reviewed:** 2026-10-25 | **Last updated:** 2026-10-25 | **Change:** minor (M14 localization current state: Arabic logical catalog, explicit RTL label direction and mirrored stock chrome; native/browser RTL workflows unqualified. Native release Hans face via LXGW WenKai Lite; estimates unchanged) > **Last reviewed:** 2026-10-23 · **Last updated:** 2026-10-23 · **Change:** minor (M7, #2636 partial: pasting added text boxes and images in the engine and through `content_copy`; estimates unchanged) > **Last reviewed:** 2026-10-23 · **Last updated:** 2026-10-23 · **Change:** minor (M11: image optimization accounts for page-local UserUnit; estimates unchanged) @@ -271,6 +272,7 @@ Ranked; detail and the full milestone table in [docs/roadmap.md](docs/roadmap.md - **2026-10-29 (M2):** Scanned pages look like the paper again. An image that doesn't ask for /Interpolate was shrunk by nearest-neighbour (since #635), which kept one sample in N per axis, so the text of a 300 dpi scan at 100% lost most of its strokes; it is now area-averaged to whole device pixels (on a 13-page 300 dpi scan every sample at 96–192 dpi is within 1 of the exact area average; it was up to 255 off). Enlarged, an image of page-scan size is smoothed bilinearly, so a zoomed scan has no staircases; icons, QR codes and pixel art keep hard pixels. Estimates unchanged. - **2026-10-26 (M11, JBIG2 symbols):** Lossless JBIG2 now stores each distinct mark of a black-and-white image once, as a symbol, and places its exact copies, when that is smaller than the generic region: on six pages of rendered 300 dpi text 28% of the generic region's size and 17% of Group 4's, and the same bits back through two independent decoders (hayro-jbig2 and Poppler). Scans, whose letters are seldom identical, gain little until refinement coding; a dictionary shared between pages (`/JBIG2Globals`) is next. Estimates unchanged. - **2026-10-25 (M5, comment replies):** One reply can be deleted from the Comments panel without its thread. On the selected card each reply has its own "…" menu (Edit, Copy text, Delete); Delete takes that reply with its pop-up and the replies to it, as `comment_delete` already did headlessly, and the comment and its other replies stay. The card stays selected, and the panel's editor stays open, on what they were on though the delete moved it in `/Annots`. Undo and Redo close the editor, since either can move what it was open on. A locked reply can't be deleted. Estimates unchanged. +- **2026-10-25 (web, same-document host mode):** A page that loads the PdfCraft web module into its own document (no iframe) can opt into the host bridge with `data-host="page"` on the app's canvas — the exact value `page`; anything else, including `?host=parent` unframed, keeps the normal app with downloads. It is the `?host=parent` bridge with the transport swapped: the same five messages travel as non-bubbling `CustomEvent`s on the canvas (`detail` holds the parent messages' fields without `type`; the host answers with a `pdfcraft:open` event), so saves reach the host instead of downloading and dirty and start-up-failure reporting are unchanged. Unit tests cover the opt-in rule; no parity or estimate changes. - **2026-10-24 (M10, #4158 partial):** Word, HTML, RTF and PowerPoint export keep a block that continues another on its line (an italic case name, a link, a superscript) as a run in its own style inside that paragraph, with the source's word space, instead of a paragraph of its own. Style changes that wrap onto the next line still split paragraphs (the reading blocks keep style boundaries). Estimates unchanged. - **2026-10-24 (M3, window move on Wayland, #3970 follow-up):** After a move or resize by PdfCraft's own title bar or edges, the button release given back to egui goes before the pointer leaving the window, so hover no longer sticks at the old place until the pointer moves. Found by CodeRabbit in review of #3970; a unit test covers it. Estimates unchanged. - **2026-10-24 (M14, #4253):** The annotation quick bar remembers Keep tool selected across restarts, including unpinning. Old and malformed settings keep safe defaults; the opt-in UI control channel reads and changes the same preference. Estimates unchanged. diff --git a/apps/pdfcraft-web/Cargo.toml b/apps/pdfcraft-web/Cargo.toml index ad8243d594..83ef1ca9f0 100644 --- a/apps/pdfcraft-web/Cargo.toml +++ b/apps/pdfcraft-web/Cargo.toml @@ -16,5 +16,5 @@ egui = { workspace = true, features = ["default_fonts", "monochrome_emoji_fonts" [target.'cfg(target_arch = "wasm32")'.dependencies] wasm-bindgen-futures = "0.4" -web-sys = { version = "0.3", features = ["Window", "Document", "Element", "HtmlCanvasElement", "HtmlElement", "Location", "Url", "UrlSearchParams", "Response", "Event", "MessageEvent", "EventTarget"] } +web-sys = { version = "0.3", features = ["Window", "Document", "Element", "HtmlCanvasElement", "HtmlElement", "Location", "Url", "UrlSearchParams", "Response", "Event", "MessageEvent", "EventTarget", "CustomEvent", "CustomEventInit"] } js-sys = "0.3" diff --git a/apps/pdfcraft-web/src/main.rs b/apps/pdfcraft-web/src/main.rs index 73d14f4a6c..91da171fba 100644 --- a/apps/pdfcraft-web/src/main.rs +++ b/apps/pdfcraft-web/src/main.rs @@ -5,18 +5,11 @@ #[cfg(target_arch = "wasm32")] fn main() { - use eframe::wasm_bindgen::JsCast; use pdfcraft_ui_egui::PdfCraftApp; let options = eframe::WebOptions { renderer: eframe::Renderer::Glow, ..Default::default() }; wasm_bindgen_futures::spawn_local(async move { - let Some(canvas) = web_sys::window() - .and_then(|w| w.document()) - .and_then(|d| d.get_element_by_id("pdfcraft")) - .and_then(|e| e.dyn_into::().ok()) - else { - return; - }; + let Some(canvas) = app_canvas() else { return }; let started = eframe::WebRunner::new() .start( canvas, @@ -54,9 +47,10 @@ fn main() { // eframe never raises a close request on a browser reload or tab close, so `guard_quit` never // prompts; ask the browser to confirm instead while a document has unsaved work (#812). warn_before_unload(app.unsaved_flag.clone()); - // `?host=parent`: a same-origin page embedding PdfCraft in an iframe (such - // as the Nextcloud app) passes files in and takes saves back (`host` below). - if query_param("host").as_deref() == Some("parent") + // Host modes (`host` below): `?host=parent`, a same-origin page embedding + // PdfCraft in an iframe (such as the Nextcloud app), or `data-host="page"` + // on the canvas, a host that runs the module in its own document. + if host_requested() && let Some(bridge) = host::Bridge::new() { // `?author=`: the host's name for the user signs their comments, @@ -81,7 +75,7 @@ fn main() { .await; if let Err(e) = started { eframe::web_sys::console::error_1(&e); - if query_param("host").as_deref() == Some("parent") + if host_requested() && let Some(bridge) = host::Bridge::new() { bridge.post_failed(&e.as_string().unwrap_or_else(|| format!("{e:?}"))); @@ -123,8 +117,37 @@ async fn fetch_bytes(url: &str) -> Result, String> { Ok(js_sys::Uint8Array::new(&buf).to_vec()) } -/// The `postMessage` bridge to a same-origin parent page. Messages are plain objects with a -/// `type`: +/// The app's canvas: the one element `index.html` and every embedding page must provide, with +/// the id the app looks up. +#[cfg(target_arch = "wasm32")] +fn app_canvas() -> Option { + use eframe::wasm_bindgen::JsCast; + web_sys::window()?.document()?.get_element_by_id("pdfcraft").and_then(|e| e.dyn_into::().ok()) +} + +/// Whether a host mode was asked for: `?host=parent` (the iframe bridge) or +/// `data-host="page"` on the canvas (the host document itself). Whether a bridge can actually +/// be built is [`host::Bridge::new`]'s answer. +#[cfg(target_arch = "wasm32")] +fn host_requested() -> bool { + query_param("host").as_deref() == Some("parent") || page_host_enabled(app_canvas().and_then(|c| c.get_attribute("data-host")).as_deref()) +} + +/// Host mode on the canvas is opt-in by attribute value: only the exact `page` turns it on +/// (`data-host="page"`). Everything else — `Page`, ` page`, `parent` (the URL opt-in), empty +/// or missing — keeps the normal app with downloads. +#[cfg(any(target_arch = "wasm32", test))] +fn page_host_enabled(data_host: Option<&str>) -> bool { + data_host == Some("page") +} + +/// Host modes: how the page embedding PdfCraft passes files in and takes saves back. Both are +/// opt-in and carry the same messages; only the transport differs, so only how an event goes +/// out and how `open` comes in change. +/// +/// `?host=parent` — the `postMessage` bridge for a same-origin page that embeds PdfCraft in an +/// iframe (such as the Nextcloud app). Messages are plain objects with a `type`, posted to the +/// parent window: /// /// - `pdfcraft:ready` (to the parent): the app is listening; send files now. /// - `pdfcraft:open` (from the parent): `{ name, bytes }`, where `bytes` is an `ArrayBuffer` or @@ -136,23 +159,52 @@ async fn fetch_bytes(url: &str) -> Result, String> { /// /// Only the parent window on the app's own origin is heard or answered. With a parent on /// another origin there is no bridge: the app runs as if unframed and saves download. +/// +/// `data-host="page"` on the app's canvas — for a host +/// that loads the wasm-bindgen module into its own document and renders on one of its own +/// canvases, with no iframe. The same messages travel as DOM `CustomEvent`s dispatched on that +/// canvas: the event name carries the `type`, and `detail` holds the same fields without it +/// (`pdfcraft:ready` carries `version`, `pdfcraft:save` carries `name`, `saveAs` and a `bytes` +/// `ArrayBuffer`). The host opens a file by dispatching a `pdfcraft:open` `CustomEvent` on the +/// canvas with `detail: { name, bytes }`. Events do not bubble, so nothing outside the canvas +/// hears them, and the canvas belongs to the document that loaded the module, so there is no +/// origin check. An unframed page that asks for `?host=parent` still gets no host mode: only +/// the exact attribute value `page` opts in. #[cfg(target_arch = "wasm32")] mod host { use eframe::wasm_bindgen::{JsCast, JsValue, closure::Closure}; + use super::page_host_enabled; + + #[derive(Clone)] + enum Transport { + /// A same-origin parent frame: send by `postMessage` with the origin, listen for + /// `message` events on the window. + Parent { window: web_sys::Window, parent: web_sys::Window, origin: String }, + /// The host document itself: send by dispatching a `CustomEvent` on the app's canvas, + /// and listen on it for `pdfcraft:open`. + Page { canvas: web_sys::HtmlCanvasElement }, + } + #[derive(Clone)] pub struct Bridge { - window: web_sys::Window, - parent: web_sys::Window, - origin: String, + transport: Transport, } impl Bridge { - /// `None` when the page isn't framed (there is no parent to talk to), or when the parent - /// is on another origin: messages only go to this origin, so the browser would drop - /// every save without an error while the document showed as saved (#2167). + /// Picks the transport: `?host=parent` needs a framing page with a same-origin parent; + /// otherwise `data-host="page"` on the app's canvas opts the host document itself in. `None` when the parent is on another origin (or when + /// `?host=parent` is used unframed): messages only go to this origin, so the browser + /// would drop every save without an error while the document showed as saved (#2167). pub fn new() -> Option { let window = web_sys::window()?; + let canvas = super::app_canvas(); + if super::query_param("host").as_deref() != Some("parent") { + if page_host_enabled(canvas.as_ref().and_then(|c| c.get_attribute("data-host")).as_deref()) { + return Some(Self { transport: Transport::Page { canvas: canvas? } }); + } + return None; + } let parent = window.parent().ok().flatten()?; if js_sys::Object::is(&parent, &window) { return None; @@ -162,59 +214,83 @@ mod host { if parent.location().origin().ok()? != origin { return None; } - Some(Self { window, parent, origin }) + Some(Self { transport: Transport::Parent { window, parent, origin } }) } - fn post(&self, message: &js_sys::Object, transfer: Option<&JsValue>) -> Result<(), String> { - let sent = match transfer { - Some(t) => self.parent.post_message_with_transfer(message, &self.origin, &js_sys::Array::of1(t)), - None => self.parent.post_message(message, &self.origin), - }; - sent.map_err(|e| e.as_string().unwrap_or_else(|| format!("{e:?}"))) + /// One outbound message, whatever the transport. `fields` is the payload: the parent + /// transport wraps it with a `type` in the posted object, the page transport puts it in + /// the event's `detail` as-is (the event name carries the type). `transfer` hands an + /// `ArrayBuffer` to the receiver instead of copying it (parent transport). + fn send(&self, name: &str, fields: &[(&str, JsValue)], transfer: Option<&JsValue>) -> Result<(), String> { + match &self.transport { + Transport::Parent { parent, origin, .. } => { + let mut all = Vec::with_capacity(fields.len().saturating_add(1)); + all.push(("type", JsValue::from_str(name))); + all.extend_from_slice(fields); + let Some(m) = message(&all) else { return Err("couldn't build the message".to_string()) }; + let sent = match transfer { + Some(t) => parent.post_message_with_transfer(&m, origin, &js_sys::Array::of1(t)), + None => parent.post_message(&m, origin), + }; + sent.map_err(|e| js_error(&e)) + } + Transport::Page { canvas } => { + let Some(detail) = message(fields) else { return Err("couldn't build the message".to_string()) }; + let init = web_sys::CustomEventInit::new(); + // The builder methods (`bubbles(..)`, `detail(..)`) are deprecated in + // web-sys 0.3.106 and fail clippy -D warnings; use the setters. + init.set_bubbles(false); + init.set_detail(&detail); + let event = web_sys::CustomEvent::new_with_event_init_dict(name, &init).map_err(|e| js_error(&e))?; + // The bool says whether a listener called `preventDefault`; not a send failure. + canvas.dispatch_event(&event).map(|_| ()).map_err(|e| js_error(&e))?; + Ok(()) + } + } } pub fn post_ready(&self) { - if let Some(m) = message(&[("type", "pdfcraft:ready".into()), ("version", env!("CARGO_PKG_VERSION").into())]) { - let _ = self.post(&m, None); - } + let _ = self.send("pdfcraft:ready", &[("version", env!("CARGO_PKG_VERSION").into())], None); } - /// The app couldn't start (no WebGL, for example): let the parent say so. + /// The app couldn't start (no WebGL, for example): let the host say so. pub fn post_failed(&self, error: &str) { - if let Some(m) = message(&[("type", "pdfcraft:failed".into()), ("error", error.into())]) { - let _ = self.post(&m, None); - } + let _ = self.send("pdfcraft:failed", &[("error", error.into())], None); } - /// Hear `pdfcraft:open` from the parent and queue the file like a `?file=` download. + /// Hear `pdfcraft:open` — from the same-origin parent (a `message` on the window) or + /// from the host document (a `pdfcraft:open` on the canvas) — and queue the file like a + /// `?file=` download. pub fn listen(&self, inbox: pdfcraft_ui_egui::Inbox, ctx: egui::Context) { - let parent = self.parent.clone(); - let origin = self.origin.clone(); - let on_message = Closure::::new(move |event: web_sys::MessageEvent| { - if event.origin() != origin || !event.source().is_some_and(|s| js_sys::Object::is(&s, &parent)) { - return; + match &self.transport { + Transport::Parent { window, parent, origin } => { + let parent = parent.clone(); + let origin = origin.clone(); + let on_message = Closure::::new(move |event: web_sys::MessageEvent| { + if event.origin() != origin || !event.source().is_some_and(|s| js_sys::Object::is(&s, &parent)) { + return; + } + let data = event.data(); + if string(&data, "type").as_deref() != Some("pdfcraft:open") { + return; + } + queue_open(&data, &inbox, &ctx); + }); + if window.add_event_listener_with_callback("message", on_message.as_ref().unchecked_ref()).is_ok() { + // The listener lives as long as the page. + on_message.forget(); + } } - let data = event.data(); - if string(&data, "type").as_deref() != Some("pdfcraft:open") { - return; + Transport::Page { canvas } => { + let on_open = Closure::::new(move |event: web_sys::Event| { + let Some(open) = event.dyn_ref::() else { return }; + queue_open(&open.detail(), &inbox, &ctx); + }); + if canvas.add_event_listener_with_callback("pdfcraft:open", on_open.as_ref().unchecked_ref()).is_ok() { + // The listener lives as long as the page. + on_open.forget(); + } } - let name = string(&data, "name").filter(|n| !n.trim().is_empty()).unwrap_or_else(|| "document.pdf".to_string()); - let Ok(bytes) = js_sys::Reflect::get(&data, &JsValue::from_str("bytes")) else { return }; - let bytes = if let Some(array) = bytes.dyn_ref::() { - array.to_vec() - } else if bytes.is_instance_of::() { - js_sys::Uint8Array::new(&bytes).to_vec() - } else { - return; - }; - if let Ok(mut q) = inbox.lock() { - q.push((name, bytes)); - } - ctx.request_repaint(); - }); - if self.window.add_event_listener_with_callback("message", on_message.as_ref().unchecked_ref()).is_ok() { - // The listener lives as long as the page. - on_message.forget(); } } @@ -222,10 +298,7 @@ mod host { let bridge = self.clone(); Box::new(move |name: &str, bytes: &[u8], save_as: bool| { let buffer = js_sys::Uint8Array::from(bytes).buffer(); - let m = - message(&[("type", "pdfcraft:save".into()), ("name", name.into()), ("saveAs", save_as.into()), ("bytes", buffer.clone().into())]) - .ok_or_else(|| "couldn't build the message".to_string())?; - bridge.post(&m, Some(&buffer)) + bridge.send("pdfcraft:save", &[("name", name.into()), ("saveAs", save_as.into()), ("bytes", buffer.clone().into())], Some(&buffer)) }) } @@ -235,14 +308,30 @@ mod host { Box::new(move |dirty: bool| { if dirty != last { last = dirty; - if let Some(m) = message(&[("type", "pdfcraft:dirty".into()), ("dirty", dirty.into())]) { - let _ = bridge.post(&m, None); - } + let _ = bridge.send("pdfcraft:dirty", &[("dirty", dirty.into())], None); } }) } } + /// The `pdfcraft:open` payload, `{ name, bytes }` where `bytes` is an `ArrayBuffer` or + /// `Uint8Array` holding a PDF: queue it like a `?file=` download and wake the app. + fn queue_open(data: &JsValue, inbox: &pdfcraft_ui_egui::Inbox, ctx: &egui::Context) { + let name = string(data, "name").filter(|n| !n.trim().is_empty()).unwrap_or_else(|| "document.pdf".to_string()); + let Ok(bytes) = js_sys::Reflect::get(data, &JsValue::from_str("bytes")) else { return }; + let bytes = if let Some(array) = bytes.dyn_ref::() { + array.to_vec() + } else if bytes.is_instance_of::() { + js_sys::Uint8Array::new(&bytes).to_vec() + } else { + return; + }; + if let Ok(mut q) = inbox.lock() { + q.push((name, bytes)); + } + ctx.request_repaint(); + } + fn message(fields: &[(&str, JsValue)]) -> Option { let m = js_sys::Object::new(); for (key, value) in fields { @@ -254,9 +343,26 @@ mod host { fn string(data: &JsValue, key: &str) -> Option { js_sys::Reflect::get(data, &JsValue::from_str(key)).ok()?.as_string() } + + fn js_error(value: &JsValue) -> String { + value.as_string().unwrap_or_else(|| format!("{value:?}")) + } } #[cfg(not(target_arch = "wasm32"))] fn main() { eprintln!("pdfcraft-web targets wasm32: run `trunk serve` or `trunk build --release` in apps/pdfcraft-web"); } + +#[cfg(test)] +mod page_host_tests { + use super::page_host_enabled; + + #[test] + fn only_the_exact_value_page_enables_the_page_host_mode() { + assert!(page_host_enabled(Some("page"))); + for off in [None, Some(""), Some("Page"), Some("PAGE"), Some(" page"), Some("page "), Some("parent")] { + assert!(!page_host_enabled(off), "{off:?} must keep host mode off"); + } + } +} diff --git a/packaging/web/README.md b/packaging/web/README.md index 7aaa49d6a5..9e32c3602e 100644 --- a/packaging/web/README.md +++ b/packaging/web/README.md @@ -78,6 +78,54 @@ Local test: `python3 -m http.server 8765` inside the folder, then open http://lo downloads) are blocked. - Don't send `X-Frame-Options: DENY` or a `frame-ancestors` CSP that excludes the embedding page. +## Same-document host mode (no iframe) + +A host application can also run PdfCraft from its own document: it loads the wasm-bindgen module +itself and gives PdfCraft one of its own canvases, so there is no iframe and no `?host=parent`. +It opts in with `data-host="page"` on that canvas — the exact value `page` — and the canvas +keeps the id the app looks up (`id="pdfcraft"`). + +The page and the app then talk with DOM `CustomEvent`s on the canvas instead of `postMessage`. +The app dispatches the same messages it would post to a parent — `pdfcraft:ready`, +`pdfcraft:save`, `pdfcraft:dirty`, `pdfcraft:failed` — with the message fields (without the +`type`) as the event's `detail`, and it opens a file when the page dispatches a `pdfcraft:open` +event with `detail: { name, bytes }`. Events do not bubble, so listen on the canvas itself. + +Add the listeners before `init`: the app announces itself with `pdfcraft:ready` while `init` is +still running, and a file sent in reply would otherwise be missed. + +```js +import init from "./pdfcraft-web-.js"; // the wasm-bindgen glue shipped with the app + +const canvas = document.getElementById("pdfcraft"); +canvas.dataset.host = "page"; // or + +canvas.addEventListener("pdfcraft:ready", () => { + // The app is listening; hand it a document now, if you have one. + canvas.dispatchEvent(new CustomEvent("pdfcraft:open", { + detail: { name: "handout.pdf", bytes: pdfBytes }, // ArrayBuffer or Uint8Array + })); +}); +canvas.addEventListener("pdfcraft:save", (event) => { + const { name, saveAs, bytes } = event.detail; // bytes is an ArrayBuffer + // Store the file under `name`; `saveAs` is true for Save As. Report your own errors. +}); +canvas.addEventListener("pdfcraft:dirty", (event) => { + unsaved = event.detail.dirty; // warn before leaving, as you like +}); +canvas.addEventListener("pdfcraft:failed", (event) => { + showError(event.detail.error); // the app couldn't start +}); + +await init({ module_or_path: "./pdfcraft-web-_bg.wasm" }); +``` + +- Saves are not downloads in this mode: the `bytes` in `pdfcraft:save` hold the saved PDF, and + the host decides where the file goes. +- `?author=` still signs the user's comments, as with `?host=parent`. +- An unframed page that uses `?host=parent` keeps the normal app with downloads; only + `data-host="page"` turns this mode on. + ## Renderer selection and fallback flags PdfCraft renders with wgpu. It uses **WebGPU** when the browser has it and falls back to From 5574a8bf4af089ce96d7188713a63885e4114738 Mon Sep 17 00:00:00 2001 From: Helen Kwok <107752876+helenkwok@users.noreply.github.com> Date: Sun, 1 Nov 2026 14:48:41 +0800 Subject: [PATCH 43/53] ROADMAP: a framed ?host=parent selects the parent bridge; only other requests keep downloads --- ROADMAP.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/ROADMAP.md b/ROADMAP.md index f82d8c13a6..f2d886c3b8 100644 --- a/ROADMAP.md +++ b/ROADMAP.md @@ -272,7 +272,7 @@ Ranked; detail and the full milestone table in [docs/roadmap.md](docs/roadmap.md - **2026-10-29 (M2):** Scanned pages look like the paper again. An image that doesn't ask for /Interpolate was shrunk by nearest-neighbour (since #635), which kept one sample in N per axis, so the text of a 300 dpi scan at 100% lost most of its strokes; it is now area-averaged to whole device pixels (on a 13-page 300 dpi scan every sample at 96–192 dpi is within 1 of the exact area average; it was up to 255 off). Enlarged, an image of page-scan size is smoothed bilinearly, so a zoomed scan has no staircases; icons, QR codes and pixel art keep hard pixels. Estimates unchanged. - **2026-10-26 (M11, JBIG2 symbols):** Lossless JBIG2 now stores each distinct mark of a black-and-white image once, as a symbol, and places its exact copies, when that is smaller than the generic region: on six pages of rendered 300 dpi text 28% of the generic region's size and 17% of Group 4's, and the same bits back through two independent decoders (hayro-jbig2 and Poppler). Scans, whose letters are seldom identical, gain little until refinement coding; a dictionary shared between pages (`/JBIG2Globals`) is next. Estimates unchanged. - **2026-10-25 (M5, comment replies):** One reply can be deleted from the Comments panel without its thread. On the selected card each reply has its own "…" menu (Edit, Copy text, Delete); Delete takes that reply with its pop-up and the replies to it, as `comment_delete` already did headlessly, and the comment and its other replies stay. The card stays selected, and the panel's editor stays open, on what they were on though the delete moved it in `/Annots`. Undo and Redo close the editor, since either can move what it was open on. A locked reply can't be deleted. Estimates unchanged. -- **2026-10-25 (web, same-document host mode):** A page that loads the PdfCraft web module into its own document (no iframe) can opt into the host bridge with `data-host="page"` on the app's canvas — the exact value `page`; anything else, including `?host=parent` unframed, keeps the normal app with downloads. It is the `?host=parent` bridge with the transport swapped: the same five messages travel as non-bubbling `CustomEvent`s on the canvas (`detail` holds the parent messages' fields without `type`; the host answers with a `pdfcraft:open` event), so saves reach the host instead of downloading and dirty and start-up-failure reporting are unchanged. Unit tests cover the opt-in rule; no parity or estimate changes. +- **2026-10-25 (web, same-document host mode):** A page that loads the PdfCraft web module into its own document (no iframe) can opt into the host bridge with `data-host="page"` on the app's canvas — the exact value `page`; any other value keeps the normal app with downloads, unless a same-origin framed `?host=parent` selects the parent bridge (unframed or cross-origin, `?host=parent` also keeps the normal app). It is the `?host=parent` bridge with the transport swapped: the same five messages travel as non-bubbling `CustomEvent`s on the canvas (`detail` holds the parent messages' fields without `type`; the host answers with a `pdfcraft:open` event), so saves reach the host instead of downloading and dirty and start-up-failure reporting are unchanged. Unit tests cover the opt-in rule; no parity or estimate changes. - **2026-10-24 (M10, #4158 partial):** Word, HTML, RTF and PowerPoint export keep a block that continues another on its line (an italic case name, a link, a superscript) as a run in its own style inside that paragraph, with the source's word space, instead of a paragraph of its own. Style changes that wrap onto the next line still split paragraphs (the reading blocks keep style boundaries). Estimates unchanged. - **2026-10-24 (M3, window move on Wayland, #3970 follow-up):** After a move or resize by PdfCraft's own title bar or edges, the button release given back to egui goes before the pointer leaving the window, so hover no longer sticks at the old place until the pointer moves. Found by CodeRabbit in review of #3970; a unit test covers it. Estimates unchanged. - **2026-10-24 (M14, #4253):** The annotation quick bar remembers Keep tool selected across restarts, including unpinning. Old and malformed settings keep safe defaults; the opt-in UI control channel reads and changes the same preference. Estimates unchanged. From 43efa9a9e39c7971cedbc95bfdc4f2c1a2c79b25 Mon Sep 17 00:00:00 2001 From: Helen Kwok <107752876+helenkwok@users.noreply.github.com> Date: Sun, 1 Nov 2026 15:04:08 +0800 Subject: [PATCH 44/53] packaging/web: say that ?host=parent takes precedence over data-host="page" --- packaging/web/README.md | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-) diff --git a/packaging/web/README.md b/packaging/web/README.md index 9e32c3602e..728c1a20ae 100644 --- a/packaging/web/README.md +++ b/packaging/web/README.md @@ -123,8 +123,10 @@ await init({ module_or_path: "./pdfcraft-web-_bg.wasm" }); - Saves are not downloads in this mode: the `bytes` in `pdfcraft:save` hold the saved PDF, and the host decides where the file goes. - `?author=` still signs the user's comments, as with `?host=parent`. -- An unframed page that uses `?host=parent` keeps the normal app with downloads; only - `data-host="page"` turns this mode on. +- An unframed page that uses `?host=parent` keeps the normal app with downloads. + `data-host="page"` turns this mode on only when `?host=parent` is absent: if `?host=parent` is + present it takes precedence, so a same-origin framed page uses the parent bridge, while an + unframed or cross-origin page gets no bridge and keeps normal downloads. ## Renderer selection and fallback flags From c79e48d6088e0ed1b78a53a7aee24a69cf2b4203 Mon Sep 17 00:00:00 2001 From: Helen Kwok <107752876+helenkwok@users.noreply.github.com> Date: Sun, 1 Nov 2026 17:30:54 +0800 Subject: [PATCH 45/53] web host mode: refuse an oversized host file before copying it, cap the name, and document trust and limits --- apps/pdfcraft-web/src/main.rs | 27 +++++++++++++++++++++++---- packaging/web/README.md | 6 ++++++ 2 files changed, 29 insertions(+), 4 deletions(-) diff --git a/apps/pdfcraft-web/src/main.rs b/apps/pdfcraft-web/src/main.rs index 91da171fba..2ff9874d5f 100644 --- a/apps/pdfcraft-web/src/main.rs +++ b/apps/pdfcraft-web/src/main.rs @@ -314,18 +314,37 @@ mod host { } } + /// The largest file the host may hand over to open, checked before it is copied (the app's own readers have their own limits). + const MAX_OPEN_BYTES: u64 = 512 << 20; + + /// The longest file name taken from the host. + const MAX_NAME_CHARS: usize = 255; + /// The `pdfcraft:open` payload, `{ name, bytes }` where `bytes` is an `ArrayBuffer` or /// `Uint8Array` holding a PDF: queue it like a `?file=` download and wake the app. fn queue_open(data: &JsValue, inbox: &pdfcraft_ui_egui::Inbox, ctx: &egui::Context) { - let name = string(data, "name").filter(|n| !n.trim().is_empty()).unwrap_or_else(|| "document.pdf".to_string()); + let name: String = string(data, "name") + .filter(|n| !n.trim().is_empty()) + .unwrap_or_else(|| "document.pdf".to_string()) + .chars() + .take(MAX_NAME_CHARS) + .collect(); let Ok(bytes) = js_sys::Reflect::get(data, &JsValue::from_str("bytes")) else { return }; - let bytes = if let Some(array) = bytes.dyn_ref::() { - array.to_vec() + let view = if let Some(array) = bytes.dyn_ref::() { + array.clone() } else if bytes.is_instance_of::() { - js_sys::Uint8Array::new(&bytes).to_vec() + js_sys::Uint8Array::new(&bytes) } else { return; }; + // A file from the host is untrusted input: refuse an oversized one before copying it into the app's memory. + if u64::from(view.length()) > MAX_OPEN_BYTES { + eframe::web_sys::console::warn_1( + &format!("PdfCraft: host file is {} bytes, over the {} MiB host open limit; not opened", view.length(), MAX_OPEN_BYTES >> 20).into(), + ); + return; + } + let bytes = view.to_vec(); if let Ok(mut q) = inbox.lock() { q.push((name, bytes)); } diff --git a/packaging/web/README.md b/packaging/web/README.md index 728c1a20ae..64552b1252 100644 --- a/packaging/web/README.md +++ b/packaging/web/README.md @@ -128,6 +128,12 @@ await init({ module_or_path: "./pdfcraft-web-_bg.wasm" }); present it takes precedence, so a same-origin framed page uses the parent bridge, while an unframed or cross-origin page gets no bridge and keeps normal downloads. +### Trust and limits + +- **Who is trusted.** In page host mode the page and PdfCraft share one document, so any script on the page can drive the app: open files in it and receive what it saves. Use this mode only for a page you control. The iframe mode (`?host=parent`) answers only the parent window, on the page's own origin. +- **Files from the host are untrusted input.** An opened file goes through the same reading as one picked with File ▸ Open or dropped on the window, with the same format limits; before it is copied into the app's memory, its size is checked against the host open limit (512 MiB) and a larger file is refused, and a name is cut to 255 characters. +- **The event and message names are the interface.** `pdfcraft:ready`, `pdfcraft:open`, `pdfcraft:save`, `pdfcraft:dirty` and `pdfcraft:failed`, and the fields they carry (the same as the `?host=parent` messages), stay stable; changing one is a breaking change. + ## Renderer selection and fallback flags PdfCraft renders with wgpu. It uses **WebGPU** when the browser has it and falls back to From 3df903d516d2dfb3e00103c4960d2cbf9e32752d Mon Sep 17 00:00:00 2001 From: Helen Kwok <107752876+helenkwok@users.noreply.github.com> Date: Thu, 5 Nov 2026 17:11:28 +0800 Subject: [PATCH 46/53] pdfcraft-web: speak the host messages through craft-web-host The transports, the message types and the origin checks come from the shared crate (storytold/craft-libs#420); PdfCraft keeps the save and dirty hooks, the open size limit and the author. --- Cargo.lock | 10 ++ apps/pdfcraft-web/Cargo.toml | 2 + apps/pdfcraft-web/src/main.rs | 328 +++++++++------------------------- packaging/web/README.md | 2 + 4 files changed, 101 insertions(+), 241 deletions(-) diff --git a/Cargo.lock b/Cargo.lock index 573b215b7c..1bf064f5d2 100644 --- a/Cargo.lock +++ b/Cargo.lock @@ -1303,6 +1303,15 @@ dependencies = [ "zip", ] +[[package]] +name = "craft-web-host" +version = "0.1.0" +source = "git+https://github.com/storytold/craft-libs?rev=2e2312ddaf21bc3d85cd3395d0955c7d239306f6#2e2312ddaf21bc3d85cd3395d0955c7d239306f6" +dependencies = [ + "wasm-bindgen", + "web-sys", +] + [[package]] name = "craft-winprint-gdi" version = "0.1.0" @@ -4729,6 +4738,7 @@ dependencies = [ name = "pdfcraft-web" version = "0.18.1" dependencies = [ + "craft-web-host", "eframe", "egui", "js-sys", diff --git a/apps/pdfcraft-web/Cargo.toml b/apps/pdfcraft-web/Cargo.toml index 83ef1ca9f0..56e10c0c29 100644 --- a/apps/pdfcraft-web/Cargo.toml +++ b/apps/pdfcraft-web/Cargo.toml @@ -11,6 +11,8 @@ workspace = true [dependencies] pdfcraft-ui-egui = { workspace = true } +# The host messages (storytold/craft-libs): one contract and one bridge for the web builds. +craft-web-host = { git = "https://github.com/storytold/craft-libs", rev = "2e2312ddaf21bc3d85cd3395d0955c7d239306f6", version = "0.1.0" } eframe = { workspace = true, features = ["default_fonts", "glow", "persistence", "accesskit"] } egui = { workspace = true, features = ["default_fonts", "monochrome_emoji_fonts"] } diff --git a/apps/pdfcraft-web/src/main.rs b/apps/pdfcraft-web/src/main.rs index 2ff9874d5f..ffa5a06068 100644 --- a/apps/pdfcraft-web/src/main.rs +++ b/apps/pdfcraft-web/src/main.rs @@ -47,12 +47,11 @@ fn main() { // eframe never raises a close request on a browser reload or tab close, so `guard_quit` never // prompts; ask the browser to confirm instead while a document has unsaved work (#812). warn_before_unload(app.unsaved_flag.clone()); - // Host modes (`host` below): `?host=parent`, a same-origin page embedding - // PdfCraft in an iframe (such as the Nextcloud app), or `data-host="page"` - // on the canvas, a host that runs the module in its own document. - if host_requested() - && let Some(bridge) = host::Bridge::new() - { + // Host modes (`host` below, over `craft-web-host`): `?host=parent`, a + // same-origin page embedding PdfCraft in an iframe (such as the Nextcloud + // app), or `data-host="page"` on the canvas, a host that runs the module in + // its own document. + if let Some(bridge) = app_canvas().as_ref().and_then(host::Bridge::connect) { // `?author=`: the host's name for the user signs their comments, // unless they chose a name in PdfCraft (the web default is "Guest"). if let Some(author) = query_param("author").map(|a| a.trim().to_string()).filter(|a| !a.is_empty()) @@ -75,9 +74,7 @@ fn main() { .await; if let Err(e) = started { eframe::web_sys::console::error_1(&e); - if host_requested() - && let Some(bridge) = host::Bridge::new() - { + if let Some(bridge) = app_canvas().as_ref().and_then(host::Bridge::connect) { bridge.post_failed(&e.as_string().unwrap_or_else(|| format!("{e:?}"))); } } @@ -125,263 +122,112 @@ fn app_canvas() -> Option { web_sys::window()?.document()?.get_element_by_id("pdfcraft").and_then(|e| e.dyn_into::().ok()) } -/// Whether a host mode was asked for: `?host=parent` (the iframe bridge) or -/// `data-host="page"` on the canvas (the host document itself). Whether a bridge can actually -/// be built is [`host::Bridge::new`]'s answer. -#[cfg(target_arch = "wasm32")] -fn host_requested() -> bool { - query_param("host").as_deref() == Some("parent") || page_host_enabled(app_canvas().and_then(|c| c.get_attribute("data-host")).as_deref()) -} - -/// Host mode on the canvas is opt-in by attribute value: only the exact `page` turns it on -/// (`data-host="page"`). Everything else — `Page`, ` page`, `parent` (the URL opt-in), empty -/// or missing — keeps the normal app with downloads. -#[cfg(any(target_arch = "wasm32", test))] -fn page_host_enabled(data_host: Option<&str>) -> bool { - data_host == Some("page") -} - -/// Host modes: how the page embedding PdfCraft passes files in and takes saves back. Both are -/// opt-in and carry the same messages; only the transport differs, so only how an event goes -/// out and how `open` comes in change. +/// Host modes: how the page embedding PdfCraft passes files in and takes saves back. The messages +/// and the two transports are `craft-web-host`'s (storytold/craft-libs), the same contract as the +/// other Craft apps' bridges, with the `pdfcraft:` prefix. This module is PdfCraft's part of it. /// -/// `?host=parent` — the `postMessage` bridge for a same-origin page that embeds PdfCraft in an -/// iframe (such as the Nextcloud app). Messages are plain objects with a `type`, posted to the -/// parent window: -/// -/// - `pdfcraft:ready` (to the parent): the app is listening; send files now. -/// - `pdfcraft:open` (from the parent): `{ name, bytes }`, where `bytes` is an `ArrayBuffer` or -/// `Uint8Array` holding a PDF. It opens in a tab, like a `?file=` URL. -/// - `pdfcraft:save` (to the parent): `{ name, saveAs, bytes }` when the user saves; `bytes` is -/// a transferred `ArrayBuffer`. The parent stores the file (and reports its own errors). -/// - `pdfcraft:dirty` (to the parent): `{ dirty }` whenever unsaved work appears or goes away. -/// - `pdfcraft:failed` (to the parent): `{ error }` when the app couldn't start. -/// -/// Only the parent window on the app's own origin is heard or answered. With a parent on -/// another origin there is no bridge: the app runs as if unframed and saves download. -/// -/// `data-host="page"` on the app's canvas — for a host -/// that loads the wasm-bindgen module into its own document and renders on one of its own -/// canvases, with no iframe. The same messages travel as DOM `CustomEvent`s dispatched on that -/// canvas: the event name carries the `type`, and `detail` holds the same fields without it -/// (`pdfcraft:ready` carries `version`, `pdfcraft:save` carries `name`, `saveAs` and a `bytes` -/// `ArrayBuffer`). The host opens a file by dispatching a `pdfcraft:open` `CustomEvent` on the -/// canvas with `detail: { name, bytes }`. Events do not bubble, so nothing outside the canvas -/// hears them, and the canvas belongs to the document that loaded the module, so there is no -/// origin check. An unframed page that asks for `?host=parent` still gets no host mode: only -/// the exact attribute value `page` opts in. +/// - `?host=parent` is the `postMessage` bridge for a same-origin page that embeds PdfCraft in an +/// iframe (such as the Nextcloud app). With a parent on another origin there is no bridge: the +/// app runs as if unframed and saves download (#2167). `data-host="page"` on the app's canvas +/// is for a host that loads the module into its own document, with no iframe: the same messages +/// travel as DOM `CustomEvent`s on that canvas. `?host=parent` wins when both are present. +/// - `pdfcraft:ready` once the app is listening; send files now. PdfCraft reads bytes, so it +/// says `file: false` and a host sends `bytes`. +/// - `pdfcraft:open` (from the host): `{ name, bytes }` holding a PDF. It opens in a tab, like a +/// `?file=` URL; a file over [`host::MAX_OPEN_BYTES`] is refused before it is copied. +/// - `pdfcraft:save` (to the host): `{ name, saveAs, bytes }` when the user saves. The host stores +/// the file (and reports its own errors). +/// - `pdfcraft:dirty` whenever unsaved work appears or goes away, `pdfcraft:failed` when the app +/// couldn't start. #[cfg(target_arch = "wasm32")] mod host { - use eframe::wasm_bindgen::{JsCast, JsValue, closure::Closure}; + use std::rc::Rc; - use super::page_host_enabled; + use craft_web_host::web::{FileData, Host, Message}; - #[derive(Clone)] - enum Transport { - /// A same-origin parent frame: send by `postMessage` with the origin, listen for - /// `message` events on the window. - Parent { window: web_sys::Window, parent: web_sys::Window, origin: String }, - /// The host document itself: send by dispatching a `CustomEvent` on the app's canvas, - /// and listen on it for `pdfcraft:open`. - Page { canvas: web_sys::HtmlCanvasElement }, - } + /// The app id in the host messages (`pdfcraft:ready`). + const APP: &str = "pdfcraft"; - #[derive(Clone)] - pub struct Bridge { - transport: Transport, - } - - impl Bridge { - /// Picks the transport: `?host=parent` needs a framing page with a same-origin parent; - /// otherwise `data-host="page"` on the app's canvas opts the host document itself in. `None` when the parent is on another origin (or when - /// `?host=parent` is used unframed): messages only go to this origin, so the browser - /// would drop every save without an error while the document showed as saved (#2167). - pub fn new() -> Option { - let window = web_sys::window()?; - let canvas = super::app_canvas(); - if super::query_param("host").as_deref() != Some("parent") { - if page_host_enabled(canvas.as_ref().and_then(|c| c.get_attribute("data-host")).as_deref()) { - return Some(Self { transport: Transport::Page { canvas: canvas? } }); - } - return None; - } - let parent = window.parent().ok().flatten()?; - if js_sys::Object::is(&parent, &window) { - return None; - } - let origin = window.location().origin().ok()?; - // Reading another origin's location throws, which is `None` too. - if parent.location().origin().ok()? != origin { - return None; - } - Some(Self { transport: Transport::Parent { window, parent, origin } }) - } - - /// One outbound message, whatever the transport. `fields` is the payload: the parent - /// transport wraps it with a `type` in the posted object, the page transport puts it in - /// the event's `detail` as-is (the event name carries the type). `transfer` hands an - /// `ArrayBuffer` to the receiver instead of copying it (parent transport). - fn send(&self, name: &str, fields: &[(&str, JsValue)], transfer: Option<&JsValue>) -> Result<(), String> { - match &self.transport { - Transport::Parent { parent, origin, .. } => { - let mut all = Vec::with_capacity(fields.len().saturating_add(1)); - all.push(("type", JsValue::from_str(name))); - all.extend_from_slice(fields); - let Some(m) = message(&all) else { return Err("couldn't build the message".to_string()) }; - let sent = match transfer { - Some(t) => parent.post_message_with_transfer(&m, origin, &js_sys::Array::of1(t)), - None => parent.post_message(&m, origin), - }; - sent.map_err(|e| js_error(&e)) - } - Transport::Page { canvas } => { - let Some(detail) = message(fields) else { return Err("couldn't build the message".to_string()) }; - let init = web_sys::CustomEventInit::new(); - // The builder methods (`bubbles(..)`, `detail(..)`) are deprecated in - // web-sys 0.3.106 and fail clippy -D warnings; use the setters. - init.set_bubbles(false); - init.set_detail(&detail); - let event = web_sys::CustomEvent::new_with_event_init_dict(name, &init).map_err(|e| js_error(&e))?; - // The bool says whether a listener called `preventDefault`; not a send failure. - canvas.dispatch_event(&event).map(|_| ()).map_err(|e| js_error(&e))?; - Ok(()) - } - } - } - - pub fn post_ready(&self) { - let _ = self.send("pdfcraft:ready", &[("version", env!("CARGO_PKG_VERSION").into())], None); - } - - /// The app couldn't start (no WebGL, for example): let the host say so. - pub fn post_failed(&self, error: &str) { - let _ = self.send("pdfcraft:failed", &[("error", error.into())], None); - } - - /// Hear `pdfcraft:open` — from the same-origin parent (a `message` on the window) or - /// from the host document (a `pdfcraft:open` on the canvas) — and queue the file like a - /// `?file=` download. - pub fn listen(&self, inbox: pdfcraft_ui_egui::Inbox, ctx: egui::Context) { - match &self.transport { - Transport::Parent { window, parent, origin } => { - let parent = parent.clone(); - let origin = origin.clone(); - let on_message = Closure::::new(move |event: web_sys::MessageEvent| { - if event.origin() != origin || !event.source().is_some_and(|s| js_sys::Object::is(&s, &parent)) { - return; - } - let data = event.data(); - if string(&data, "type").as_deref() != Some("pdfcraft:open") { - return; - } - queue_open(&data, &inbox, &ctx); - }); - if window.add_event_listener_with_callback("message", on_message.as_ref().unchecked_ref()).is_ok() { - // The listener lives as long as the page. - on_message.forget(); - } - } - Transport::Page { canvas } => { - let on_open = Closure::::new(move |event: web_sys::Event| { - let Some(open) = event.dyn_ref::() else { return }; - queue_open(&open.detail(), &inbox, &ctx); - }); - if canvas.add_event_listener_with_callback("pdfcraft:open", on_open.as_ref().unchecked_ref()).is_ok() { - // The listener lives as long as the page. - on_open.forget(); - } - } - } - } - - pub fn save_fn(&self) -> pdfcraft_ui_egui::HostSaveFn { - let bridge = self.clone(); - Box::new(move |name: &str, bytes: &[u8], save_as: bool| { - let buffer = js_sys::Uint8Array::from(bytes).buffer(); - bridge.send("pdfcraft:save", &[("name", name.into()), ("saveAs", save_as.into()), ("bytes", buffer.clone().into())], Some(&buffer)) - }) - } - - pub fn dirty_fn(&self) -> pdfcraft_ui_egui::HostDirtyFn { - let bridge = self.clone(); - let mut last = false; - Box::new(move |dirty: bool| { - if dirty != last { - last = dirty; - let _ = bridge.send("pdfcraft:dirty", &[("dirty", dirty.into())], None); - } - }) - } - } + /// The name a file gets when the host sent none (the crate says `document` for a blank name). + const DEFAULT_NAME: &str = "document.pdf"; /// The largest file the host may hand over to open, checked before it is copied (the app's own readers have their own limits). const MAX_OPEN_BYTES: u64 = 512 << 20; - /// The longest file name taken from the host. - const MAX_NAME_CHARS: usize = 255; + #[derive(Clone)] + pub struct Bridge { + host: Rc, + } - /// The `pdfcraft:open` payload, `{ name, bytes }` where `bytes` is an `ArrayBuffer` or - /// `Uint8Array` holding a PDF: queue it like a `?file=` download and wake the app. - fn queue_open(data: &JsValue, inbox: &pdfcraft_ui_egui::Inbox, ctx: &egui::Context) { - let name: String = string(data, "name") - .filter(|n| !n.trim().is_empty()) - .unwrap_or_else(|| "document.pdf".to_string()) - .chars() - .take(MAX_NAME_CHARS) - .collect(); - let Ok(bytes) = js_sys::Reflect::get(data, &JsValue::from_str("bytes")) else { return }; - let view = if let Some(array) = bytes.dyn_ref::() { - array.clone() - } else if bytes.is_instance_of::() { - js_sys::Uint8Array::new(&bytes) - } else { + impl Bridge { + /// The bridge for this page, or `None` when PdfCraft isn't hosted: no `?host=parent` and + /// no `data-host="page"` on the canvas, an unframed `?host=parent` page, or a parent on + /// another origin (messages only go to this origin, so the browser would drop every save + /// without an error while the document showed as saved, #2167). + pub fn connect(canvas: &web_sys::HtmlCanvasElement) -> Option { + Some(Self { host: Rc::new(Host::connect(APP, canvas)?) }) + } + + pub fn post_ready(&self) { + let _ = self.host.post_ready(env!("CARGO_PKG_VERSION"), false); + } + + /// The app couldn't start (no WebGL, for example): let the host say so. + pub fn post_failed(&self, error: &str) { + let _ = self.host.post_failed(error, None); + } + + /// Hear `pdfcraft:open` and queue the file like a `?file=` download. + pub fn listen(&self, inbox: pdfcraft_ui_egui::Inbox, ctx: egui::Context) { + self.host.listen(move |message| { + if let Message::Open { name, data } = message { + queue_open(&name, &data, &inbox, &ctx); + } + }); + } + + pub fn save_fn(&self) -> pdfcraft_ui_egui::HostSaveFn { + let host = self.host.clone(); + Box::new(move |name: &str, bytes: &[u8], save_as: bool| host.post_save(name, bytes, None, save_as)) + } + + pub fn dirty_fn(&self) -> pdfcraft_ui_egui::HostDirtyFn { + let host = self.host.clone(); + // A host starts out clean: the first report is a change only when it is dirty. + let mut last = false; + Box::new(move |dirty: bool| { + if dirty != last { + last = dirty; + host.post_dirty(dirty); + } + }) + } + } + + /// The `pdfcraft:open` payload, `{ name, bytes }` where `bytes` holds a PDF: queue it like a + /// `?file=` download and wake the app. The crate has cleaned the name (and cut it to 255 + /// characters); a blank one comes as `document` and becomes `document.pdf`. + fn queue_open(name: &str, data: &FileData, inbox: &pdfcraft_ui_egui::Inbox, ctx: &egui::Context) { + let name = if name == "document" { DEFAULT_NAME } else { name }.to_string(); + let FileData::Bytes(view) = data else { + eframe::web_sys::console::warn_1(&format!("PdfCraft: host file {name} came as a File; PdfCraft reads bytes only; not opened").into()); return; }; // A file from the host is untrusted input: refuse an oversized one before copying it into the app's memory. - if u64::from(view.length()) > MAX_OPEN_BYTES { + if data.size() > MAX_OPEN_BYTES as f64 { eframe::web_sys::console::warn_1( - &format!("PdfCraft: host file is {} bytes, over the {} MiB host open limit; not opened", view.length(), MAX_OPEN_BYTES >> 20).into(), + &format!("PdfCraft: host file is {} bytes, over the {} MiB host open limit; not opened", data.size(), MAX_OPEN_BYTES >> 20).into(), ); return; } - let bytes = view.to_vec(); if let Ok(mut q) = inbox.lock() { - q.push((name, bytes)); + q.push((name, view.to_vec())); } ctx.request_repaint(); } - - fn message(fields: &[(&str, JsValue)]) -> Option { - let m = js_sys::Object::new(); - for (key, value) in fields { - js_sys::Reflect::set(&m, &JsValue::from_str(key), value).ok()?; - } - Some(m) - } - - fn string(data: &JsValue, key: &str) -> Option { - js_sys::Reflect::get(data, &JsValue::from_str(key)).ok()?.as_string() - } - - fn js_error(value: &JsValue) -> String { - value.as_string().unwrap_or_else(|| format!("{value:?}")) - } } #[cfg(not(target_arch = "wasm32"))] fn main() { eprintln!("pdfcraft-web targets wasm32: run `trunk serve` or `trunk build --release` in apps/pdfcraft-web"); } - -#[cfg(test)] -mod page_host_tests { - use super::page_host_enabled; - - #[test] - fn only_the_exact_value_page_enables_the_page_host_mode() { - assert!(page_host_enabled(Some("page"))); - for off in [None, Some(""), Some("Page"), Some("PAGE"), Some(" page"), Some("page "), Some("parent")] { - assert!(!page_host_enabled(off), "{off:?} must keep host mode off"); - } - } -} diff --git a/packaging/web/README.md b/packaging/web/README.md index 64552b1252..3149dc75d5 100644 --- a/packaging/web/README.md +++ b/packaging/web/README.md @@ -128,6 +128,8 @@ await init({ module_or_path: "./pdfcraft-web-_bg.wasm" }); present it takes precedence, so a same-origin framed page uses the parent bridge, while an unframed or cross-origin page gets no bridge and keeps normal downloads. +The messages and both transports are the shared contract of [`craft-web-host`](https://github.com/storytold/craft-libs/tree/main/crates/web-host) (storytold/craft-libs#420), the same for every Craft app: its README lists every message, including `saved`, with the compatibility notes. In `pdfcraft:ready`, `version` is the contract's (`"1"`) and `appVersion` is this app's release. PdfCraft reads bytes, so a host sends `bytes`, not a `File`. + ### Trust and limits - **Who is trusted.** In page host mode the page and PdfCraft share one document, so any script on the page can drive the app: open files in it and receive what it saves. Use this mode only for a page you control. The iframe mode (`?host=parent`) answers only the parent window, on the page's own origin. From 9927a1bd12f11e3e63c13677443622ad54089bd9 Mon Sep 17 00:00:00 2001 From: FredApps Date: Thu, 5 Nov 2026 12:47:19 +0100 Subject: [PATCH 47/53] M10: page_enhance_scan as its own dispatch method Main split Automation::call into per-tool methods and its clippy cognitive-complexity limit (30) now fails with the inline arm. --- crates/automation/src/dispatch.rs | 62 ++++++++++++++++--------------- 1 file changed, 33 insertions(+), 29 deletions(-) diff --git a/crates/automation/src/dispatch.rs b/crates/automation/src/dispatch.rs index c3203f355e..45df846ebc 100644 --- a/crates/automation/src/dispatch.rs +++ b/crates/automation/src/dispatch.rs @@ -137,35 +137,7 @@ impl Automation { out["detected"] = json!(given.is_none()); out } - "page_enhance_scan" => { - let pages = match a.opt_ints("pages")? { - Some(_) => self.pages(&a, "pages")?, - None => Vec::new(), - }; - let level = match a.opt_int("level")? { - None => pdfcraft_engine::scan_clean::ScanCleanSettings::default().level, - Some(l @ 0..=100) => l as u8, - Some(l) => return Err(ToolError::InvalidArgs(format!("level must be between 0 and 100, not {l}"))), - }; - let mut settings = pdfcraft_engine::scan_clean::ScanCleanSettings::at(level); - if let Some(on) = a.opt_bool("edge_shadows")? { - settings.edge_shadows = on; - } - if let Some(on) = a.opt_bool("despeckle")? { - settings.despeckle = on; - } - let id = self.doc(&a)?.id; - let report = self.session.enhance_scanned_pages(id, &pages, settings).map_err(failed)?; - let numbered = |p: &[usize]| p.iter().map(|p| p + 1).collect::>(); - let skipped: Vec = - report.skipped.iter().map(|k| json!({ "pages": numbered(&k.pages), "name": k.name, "reason": k.reason })).collect(); - let mut out = summary(self.doc(&a)?); - out["pages_cleaned"] = json!(numbered(&report.pages)); - out["images"] = json!(report.images); - out["not_scanned"] = json!(numbered(&report.not_scanned)); - out["skipped"] = json!(skipped); - out - } + "page_enhance_scan" => self.page_enhance_scan(&a)?, "page_delete" => { let pages = self.pages(&a, "pages")?; self.apply(&a, Edit::DeletePages { pages })? @@ -948,3 +920,35 @@ impl Automation { Ok(out) } } + +impl Automation { + /// `page_enhance_scan`: Enhance ▸ Scanned document on the chosen pages (all by default). + fn page_enhance_scan(&mut self, a: &Args) -> Result { + let pages = match a.opt_ints("pages")? { + Some(_) => self.pages(a, "pages")?, + None => Vec::new(), + }; + let level = match a.opt_int("level")? { + None => pdfcraft_engine::scan_clean::ScanCleanSettings::default().level, + Some(l @ 0..=100) => l as u8, + Some(l) => return Err(ToolError::InvalidArgs(format!("level must be between 0 and 100, not {l}"))), + }; + let mut settings = pdfcraft_engine::scan_clean::ScanCleanSettings::at(level); + if let Some(on) = a.opt_bool("edge_shadows")? { + settings.edge_shadows = on; + } + if let Some(on) = a.opt_bool("despeckle")? { + settings.despeckle = on; + } + let id = self.doc(a)?.id; + let report = self.session.enhance_scanned_pages(id, &pages, settings).map_err(failed)?; + let numbered = |p: &[usize]| p.iter().map(|p| p + 1).collect::>(); + let skipped: Vec = report.skipped.iter().map(|k| json!({ "pages": numbered(&k.pages), "name": k.name, "reason": k.reason })).collect(); + let mut out = summary(self.doc(a)?); + out["pages_cleaned"] = json!(numbered(&report.pages)); + out["images"] = json!(report.images); + out["not_scanned"] = json!(numbered(&report.not_scanned)); + out["skipped"] = json!(skipped); + Ok(out) + } +} From 9cc1736dc92bc35351b416b93582f4c6cbf96dad Mon Sep 17 00:00:00 2001 From: FredApps Date: Thu, 5 Nov 2026 12:56:10 +0100 Subject: [PATCH 48/53] Enhance scanned file: decode bi-level images within the budget, DecodeParms resolved --- crates/optimize/src/images.rs | 2 +- crates/optimize/src/samples.rs | 43 +++++++++++++++++++++++++++++----- 2 files changed, 38 insertions(+), 7 deletions(-) diff --git a/crates/optimize/src/images.rs b/crates/optimize/src/images.rs index 9f2782ce1e..159ba04e8c 100644 --- a/crates/optimize/src/images.rs +++ b/crates/optimize/src/images.rs @@ -506,7 +506,7 @@ pub(crate) fn pixels(doc: &Document, s: &Stream, n: usize, budget: &mut ImageBud /// to three bytes for each of them and of a PNG predictor's bytes a row (ASCIIHex text under /// Flate, its two digits a byte broken into lines or spaced). More can only be a decompression /// bomb, and the image is left alone. -fn allowance(bytes: usize, rows: u32) -> Option { +pub(crate) fn allowance(bytes: usize, rows: u32) -> Option { bytes.checked_add(usize::try_from(rows).ok()?)?.checked_mul(3)?.checked_add(4096) } diff --git a/crates/optimize/src/samples.rs b/crates/optimize/src/samples.rs index d3d76c7f41..4800da6fd7 100644 --- a/crates/optimize/src/samples.rs +++ b/crates/optimize/src/samples.rs @@ -5,7 +5,7 @@ use pdfcraft_cos::{Document, ImageBudget, Object, Stream}; -use crate::images::{MAX_PIXELS, components, is_dct, is_identity_decode, jpeg, pixels, ycc_encoded}; +use crate::images::{MAX_PIXELS, allowance, components, is_dct, is_identity_decode, jpeg, pixels, ycc_encoded}; /// An image's samples, one byte each, row-major. #[derive(Clone, Debug, PartialEq)] @@ -83,7 +83,7 @@ pub fn decode_samples(doc: &Document, s: &Stream, budget: &mut ImageBudget) -> R Some(a) if is_inverted_decode(doc, a) => true, Some(_) => return Err(NotDecoded::Decode), }; - let data = unpack_bits(s, width, height, inverted).ok_or(NotDecoded::Corrupt)?; + let data = unpack_bits(doc, s, width, height, inverted, budget).ok_or(NotDecoded::Corrupt)?; return Ok(Samples { width, height, components: 1, data, jpeg: false, bilevel: true, inverted }); } if d.get(b"Decode").is_some_and(|a| !is_identity_decode(doc, a, n)) { @@ -107,12 +107,12 @@ fn packed_len(w: u32, h: u32) -> Option { usize::try_from(w).ok()?.div_ceil(8).checked_mul(usize::try_from(h).ok()?) } -/// A bi-level image's pixels, 0 (black) or 255 (white) as drawn. -fn unpack_bits(s: &Stream, w: u32, h: u32, inverted: bool) -> Option> { +/// A bi-level image's pixels, 0 (black) or 255 (white) as drawn, decoded within `budget` with +/// its `/DecodeParms` resolved, as the 8-bit images are. +fn unpack_bits(doc: &Document, s: &Stream, w: u32, h: u32, inverted: bool, budget: &mut ImageBudget) -> Option> { let stride = usize::try_from(w).ok()?.div_ceil(8); let need = packed_len(w, h)?; - // Some writers pad the data; anything far longer than the image needs is not this image. - let bits = s.decoded_within(need.saturating_add(need / 2).saturating_add(4096)).ok()?; + let bits = budget.decode_filters(doc, s, allowance(need, h)?).ok()?; let bits = bits.get(..need)?; let (w, h) = (usize::try_from(w).ok()?, usize::try_from(h).ok()?); let mut out = Vec::with_capacity(w.checked_mul(h)?); @@ -246,6 +246,37 @@ mod tests { assert_eq!(back.decoded().unwrap(), raw); } + #[test] + fn bilevel_decode_parms_by_reference_decode_like_direct_ones() { + // 10 × 2 with a TIFF predictor: each 1 bit flips the pixel from the one to its left. + let packed = vec![0b1000_0000, 0b0000_0000, 0b0100_0000, 0b0000_0000]; + let parms = || { + let mut p = pdfcraft_cos::Dict::new(); + for (k, v) in [("Predictor", 2), ("Colors", 1), ("BitsPerComponent", 1), ("Columns", 10)] { + p.set(k.as_bytes().to_vec(), Object::Int(v)); + } + Object::Dict(p) + }; + let dict = [ + ("Width", Object::Int(10)), + ("Height", Object::Int(2)), + ("ColorSpace", Object::name("DeviceGray")), + ("BitsPerComponent", Object::Int(1)), + ]; + let mut d = doc(); + let mut direct = Stream::flate(image(&dict, vec![]).dict, &packed); + direct.dict.set(b"DecodeParms".to_vec(), parms()); + let mut indirect = direct.clone(); + let r = d.add(parms()); + indirect.dict.set(b"DecodeParms".to_vec(), Object::Ref(r)); + let a = decode_samples(&d, &direct, &mut ImageBudget::with_work(u64::MAX)).unwrap(); + let b = decode_samples(&d, &indirect, &mut ImageBudget::with_work(u64::MAX)).unwrap(); + // Row 1: white from x = 0 on; row 2: black at x = 0, white from x = 1 on. + assert_eq!(&a.data[..10], &[255; 10]); + assert_eq!(&a.data[10..], &[0, 255, 255, 255, 255, 255, 255, 255, 255, 255]); + assert_eq!(b, a); + } + #[test] fn what_is_not_supported_says_why_and_never_panics() { let base = |extra: &[(&str, Object)], raw: Vec| { From 8d569066b8dd5ce97d71fa50dd121b577ed818ce Mon Sep 17 00:00:00 2001 From: ohagmann <268387159+ohagmann@users.noreply.github.com> Date: Thu, 5 Nov 2026 13:13:56 +0100 Subject: [PATCH 49/53] M5: Escape leaves a Fill & Sign tool, as it leaves a comment tool (#4539) * Escape leaves a Fill & Sign tool, as it leaves a comment tool The first Escape closes the open type box, the next one goes back to Select; Fill & Sign's Text tool otherwise opened a box at every click with no way out by the keyboard. Co-Authored-By: Claude Opus 5.5 * Fill & Sign: a new text box deselects, so two presses of Escape close it and leave the tool Co-Authored-By: Claude Opus 5.5 --------- Co-authored-by: Claude Opus 5.5 --- ROADMAP.md | 1 + crates/ui-egui/src/tools/comments/input.rs | 6 ++- crates/ui-egui/src/tools/fill_sign.rs | 8 +++- crates/ui-egui/tests/it/fill_sign.rs | 51 ++++++++++++++++++++++ parity/acrobat-features.toml | 4 +- 5 files changed, 64 insertions(+), 6 deletions(-) diff --git a/ROADMAP.md b/ROADMAP.md index 1f753c500f..9c1ecb136d 100644 --- a/ROADMAP.md +++ b/ROADMAP.md @@ -299,6 +299,7 @@ Ranked; detail and the full milestone table in [docs/roadmap.md](docs/roadmap.md Estimates unchanged. - **2026-10-21 (M3, window resize and move on Wayland):** With PdfCraft's own title bar on GNOME/Wayland, a second resize by a window edge, or a second move by the empty tab strip, works straight after the first; it waited for a click inside the window, because the compositor keeps the button's release from egui. A unit test covers the release given back; checked by hand in a test build. Estimates unchanged. - **2026-10-24 (M5, move preview):** A comment being moved is drawn where it goes while it is dragged (a rectangle or oval showed only its blue frame): the comment is rendered alone once as the drag starts and laid over the page, which shows it where it was until the move is applied. Engine `Document::comment_layer`; synthetic engine and UI regressions. Estimates unchanged. +- **2026-10-24 (M5, Escape in Fill & Sign):** Escape leaves a Fill & Sign tool as it leaves a comment tool: the first press closes the open type box, the next goes back to Select, so the Text tool no longer opens a box at every click with no way out by the keyboard. A synthetic UI regression covers the Text and Checkmark tools. Estimates unchanged. - **2026-10-22 (M4, #3916 partial):** Document edits and Undo/Redo cancel an Organize page drag. Pointer-release regressions check exact page order and save/reopen. - **2026-10-22 (M10, #3752 paper):** CUPS jobs ask for the chosen paper in every size. The 13 sizes the Print dialog sent no paper for (A6, A0–A2, ARCH, ANSI C–E) use PWG 5101.1 names, checked against a synthetic PPD with `cupsfilter`, and `doc_print` to a printer sends the paper too. The macOS tray half (Auto Select instead of the PPD's default tray) is a craft-print change that needs a pin bump here. Estimates unchanged. diff --git a/crates/ui-egui/src/tools/comments/input.rs b/crates/ui-egui/src/tools/comments/input.rs index ac3c2a6878..23a5ab3f06 100644 --- a/crates/ui-egui/src/tools/comments/input.rs +++ b/crates/ui-egui/src/tools/comments/input.rs @@ -606,9 +606,11 @@ pub(crate) fn keys(ctx: &egui::Context, view: &mut DocView, tool: &mut QuickTool } let cv = &mut view.comments; if ctx.input(|i| i.key_pressed(egui::Key::Escape)) { - // One step back per press: cancel the gesture, then deselect, then drop the tool. + // One step back per press: cancel the gesture, then deselect, then drop the tool (a + // comment tool, or a Fill & Sign one: its Text tool would otherwise open a box at every + // click with no way out by the keyboard). let cancelled = cv.gesture.take().is_some() || cv.selected.take().is_some(); - if !cancelled && matches!(tool, QuickTool::Comment(_)) { + if !cancelled && matches!(tool, QuickTool::Comment(_) | QuickTool::Fill(_)) { *tool = QuickTool::Select; } } diff --git a/crates/ui-egui/src/tools/fill_sign.rs b/crates/ui-egui/src/tools/fill_sign.rs index 224ba8b49d..a156dfe3d2 100644 --- a/crates/ui-egui/src/tools/fill_sign.rs +++ b/crates/ui-egui/src/tools/fill_sign.rs @@ -491,7 +491,10 @@ pub(crate) fn page_input(cx: PageToolContext<'_>, input: PageInput<'_>, free: bo match tool { FillTool::Text => { // Clicking away from an open box commits it: the page sees this click before the box - // sees its lost focus, so replacing the box first would drop what was typed. + // sees its lost focus, so replacing the box first would drop what was typed. A new box + // deselects, as a comment tool's note or text box does: Escape then closes the box and + // leaves the tool in two presses. + view.comments.selected = None; let open = view.fill_text.replace(TypeBox { page, at: top, text: String::new(), size: text_size, color: text_color, focus: true }); open.filter(|tb| !tb.text.trim().is_empty()).map(|tb| { FillAction::Edit(Box::new(typed_with_color( @@ -587,7 +590,8 @@ pub(crate) fn type_box(ctx: &egui::Context, view: &mut DocView, info: &DocInfo, r.request_focus(); t.focus = false; } - if ui.input(|i| i.key_pressed(egui::Key::Escape)) { + // Taken here: the same press must not also leave the tool (one step back per press). + if ui.input_mut(|i| i.consume_key(egui::Modifiers::NONE, egui::Key::Escape)) { cancel = true; } else if r.lost_focus() { commit = true; diff --git a/crates/ui-egui/tests/it/fill_sign.rs b/crates/ui-egui/tests/it/fill_sign.rs index 9f565d834a..b5ad9f0522 100644 --- a/crates/ui-egui/tests/it/fill_sign.rs +++ b/crates/ui-egui/tests/it/fill_sign.rs @@ -159,6 +159,57 @@ fn clicking_elsewhere_with_add_text_keeps_the_typed_text() { assert!(h.state().views[0].fill_text.as_ref().is_some_and(|t| t.text.is_empty()), "a new, empty box is open"); } +/// Escape steps back out of a Fill & Sign tool as out of a comment tool: the first press closes +/// the open type box, the next one leaves the Text tool for Select, so a click opens no new box. +#[test] +fn escape_leaves_the_fill_and_sign_tools() { + let mut h = harness(); + h.state_mut().execute("sign.fill.text"); + click(&mut h, 50.0, 250.0); + assert!(h.state().views[0].fill_text.is_some(), "a box is open"); + h.key_press(egui::Key::Escape); + h.run_steps(2); + assert!(h.state().views[0].fill_text.is_none(), "the first Escape closes the box"); + assert_eq!(h.state().quick_tool, QuickTool::Fill(FillTool::Text), "the tool stays for the next box"); + h.key_press(egui::Key::Escape); + h.run_steps(2); + assert_eq!(h.state().quick_tool, QuickTool::Select, "the next Escape leaves the tool"); + click(&mut h, 50.0, 150.0); + assert!(h.state().views[0].fill_text.is_none(), "no new box"); + assert!(items(&h).is_empty(), "nothing was added: {:?}", items(&h)); + // A mark tool too. + h.state_mut().execute("sign.fill.check"); + h.run_steps(2); + assert_eq!(h.state().quick_tool, QuickTool::Fill(FillTool::Check)); + h.key_press(egui::Key::Escape); + h.run_steps(2); + assert_eq!(h.state().quick_tool, QuickTool::Select); +} + +/// A box opened while a comment is selected deselects it: two presses of Escape still close the +/// box and leave the tool. +#[test] +fn escape_leaves_the_text_tool_in_two_presses_after_a_selection() { + let mut h = harness(); + h.state_mut().execute("sign.fill.text"); + click(&mut h, 50.0, 250.0); + h.event(egui::Event::Text("Ada".into())); + h.run_steps(1); + commit_typed_text(&mut h); + assert_eq!(items(&h).len(), 1, "{:?}", items(&h)); + h.state_mut().views[0].fill_text = None; + h.state_mut().views[0].comments.selected = Some((0, 0)); + click(&mut h, 50.0, 150.0); + assert!(h.state().views[0].fill_text.is_some(), "a box is open"); + assert_eq!(h.state().views[0].comments.selected, None, "the new box deselected the comment"); + h.key_press(egui::Key::Escape); + h.run_steps(2); + assert!(h.state().views[0].fill_text.is_none(), "the first Escape closes the box"); + h.key_press(egui::Key::Escape); + h.run_steps(2); + assert_eq!(h.state().quick_tool, QuickTool::Select, "the second Escape leaves the tool"); +} + #[test] fn fill_text_size_controls_new_text_and_survives_settings_restore() { let mut h = harness(); diff --git a/parity/acrobat-features.toml b/parity/acrobat-features.toml index af283bed6e..6ff5eca28e 100644 --- a/parity/acrobat-features.toml +++ b/parity/acrobat-features.toml @@ -5423,8 +5423,8 @@ milestone = "M5" status = "shipped" commands = ["sign.fill.text"] tools = ["fill_sign_add"] -evidence = ["crates/annot/src/tests.rs::indirect_typewriter_names_keep_outer_whitespace_in_summaries", "crates/ui-egui/tests/it/comments.rs::typewriter_edit_composer_keeps_trailing_blank_lines", "crates/ui-egui/tests/it/comments.rs::other_edit_composers_keep_trimming_trailing_whitespace", "crates/ui-egui/tests/it/comments.rs::moved_fill_and_sign_text_is_drawn_where_it_goes_before_release", "crates/engine/src/tests.rs::fill_sign_multiline_text_survives_engine_save_reopen_and_render", "crates/ui-egui/tests/it/fill_sign.rs::fill_sign_enter_adds_lines_outside_click_commits_and_escape_cancels", "crates/automation/tests/it/automation.rs::multiline_fill_sign_engine_round_trip_preserves_lines_and_appearance", "crates/ui-egui/tests/it/fill_sign.rs::text_marks_and_date", "crates/annot/src/tests.rs::fill_and_sign_items_are_drawn", "crates/automation/tests/it/automation.rs::fill_and_sign_through_tools", "crates/engine/src/tests.rs::typed_text_reads_across_as_displayed_on_rotated_pages", "crates/ui-egui/tests/it/fill_sign.rs::fill_text_colour_survives_preview_commit_settings_and_save", "crates/automation/tests/it/automation.rs::fill_sign_text_and_date_colours_are_saved_and_validated"] -notes = "Indirect IT is resolved when recognizing typewriter contents and Fill & Sign marks; comment listing and properties retain direct Subtype as before; composer edits preserve the typewriter text as written, while other comments retain their existing trimming. Enter inserts a newline; the editor grows with its lines, outside clicks commit and Escape cancels. Nonempty text preserves leading/internal/trailing line breaks, including after save/reopen; fill_sign_add sizes all wrapped rows for the saved appearance (#5850). typewriter FreeText, 10 pt Helvetica. Reads across as displayed on pages with /Rotate (sideways-stored scans): the box follows the displayed axes and the appearance is turned back (PdfCraft key /PCTextRotate); other apps that regenerate the appearance from /DA draw it unturned. A later page rotation does not re-turn it. Field detection does not work on scanned (image-only) forms yet; typed fractional font sizes commit once; Fill & Sign text/date colour is remembered and preserved in preview and saved appearances (#2444)" +evidence = ["crates/annot/src/tests.rs::indirect_typewriter_names_keep_outer_whitespace_in_summaries", "crates/ui-egui/tests/it/comments.rs::typewriter_edit_composer_keeps_trailing_blank_lines", "crates/ui-egui/tests/it/comments.rs::other_edit_composers_keep_trimming_trailing_whitespace", "crates/ui-egui/tests/it/comments.rs::moved_fill_and_sign_text_is_drawn_where_it_goes_before_release", "crates/engine/src/tests.rs::fill_sign_multiline_text_survives_engine_save_reopen_and_render", "crates/ui-egui/tests/it/fill_sign.rs::fill_sign_enter_adds_lines_outside_click_commits_and_escape_cancels", "crates/automation/tests/it/automation.rs::multiline_fill_sign_engine_round_trip_preserves_lines_and_appearance", "crates/ui-egui/tests/it/fill_sign.rs::text_marks_and_date", "crates/annot/src/tests.rs::fill_and_sign_items_are_drawn", "crates/automation/tests/it/automation.rs::fill_and_sign_through_tools", "crates/engine/src/tests.rs::typed_text_reads_across_as_displayed_on_rotated_pages", "crates/ui-egui/tests/it/fill_sign.rs::fill_text_colour_survives_preview_commit_settings_and_save", "crates/automation/tests/it/automation.rs::fill_sign_text_and_date_colours_are_saved_and_validated", "crates/ui-egui/tests/it/fill_sign.rs::escape_leaves_the_fill_and_sign_tools"] +notes = "Indirect IT is resolved when recognizing typewriter contents and Fill & Sign marks; comment listing and properties retain direct Subtype as before; composer edits preserve the typewriter text as written, while other comments retain their existing trimming. Enter inserts a newline; the editor grows with its lines, outside clicks commit and Escape cancels. Nonempty text preserves leading/internal/trailing line breaks, including after save/reopen; fill_sign_add sizes all wrapped rows for the saved appearance (#5850). typewriter FreeText, 10 pt Helvetica. Escape closes the open box, and a second Escape leaves the Fill & Sign tool for Select, as it leaves a comment tool. Reads across as displayed on pages with /Rotate (sideways-stored scans): the box follows the displayed axes and the appearance is turned back (PdfCraft key /PCTextRotate); other apps that regenerate the appearance from /DA draw it unturned. A later page rotation does not re-turn it. Field detection does not work on scanned (image-only) forms yet; typed fractional font sizes commit once; Fill & Sign text/date colour is remembered and preserved in preview and saved appearances (#2444)" [[feature]] id = "form.fill-sign-marks" From 65aa495c982c105619f1d14a04517b749df51366 Mon Sep 17 00:00:00 2001 From: FredApps Date: Thu, 5 Nov 2026 13:19:49 +0100 Subject: [PATCH 50/53] Enhance scanned file: re-render the page shown after a successful run --- crates/ui-egui/src/canvas/tests.rs | 37 +++++++++++++++++++ crates/ui-egui/src/workflows/scan_clean_ui.rs | 4 ++ 2 files changed, 41 insertions(+) diff --git a/crates/ui-egui/src/canvas/tests.rs b/crates/ui-egui/src/canvas/tests.rs index d5b8312d6c..5260ccc4ed 100644 --- a/crates/ui-egui/src/canvas/tests.rs +++ b/crates/ui-egui/src/canvas/tests.rs @@ -2380,3 +2380,40 @@ fn a_camera_corner_at_the_page_edge_is_hit_on_the_page_only() { assert!(crate::camera_ui::hit(&xf, 0, &v, pos2(14.0, 24.0)).is_some(), "on the page"); assert!(crate::camera_ui::hit(&xf, 0, &v, pos2(6.0, 16.0)).is_none(), "just off it"); } + +/// #5125: after Enhance scanned file, the page shown is re-rendered, not the raster of the +/// scan before it was cleaned (same scale, so nothing else would replace it). +#[test] +fn enhancing_a_scan_re_renders_the_page_shown() { + let (w, h) = (64usize, 64usize); + let px: Vec = (0..w * h).map(|i| if i % w < 4 { 30 } else { 198 }).collect(); + let mut pdf = format!( + "%PDF-1.7\n1 0 obj\n<< /Type /Catalog /Pages 2 0 R >>\nendobj\n2 0 obj\n<< /Type /Pages /Kids [3 0 R] /Count 1 >>\nendobj\n3 0 obj\n<< /Type /Page /Parent 2 0 R /MediaBox [0 0 {w} {h}] /Resources << /XObject << /Im0 4 0 R >> >> /Contents 5 0 R >>\nendobj\n4 0 obj\n<< /Type /XObject /Subtype /Image /Width {w} /Height {h} /ColorSpace /DeviceGray /BitsPerComponent 8 /Length {} >>\nstream\n", + px.len() + ) + .into_bytes(); + pdf.extend_from_slice(&px); + let content = format!("q {w} 0 0 {h} 0 0 cm /Im0 Do Q"); + pdf.extend_from_slice( + format!( + "\nendstream\nendobj\n5 0 obj\n<< /Length {} >>\nstream\n{content}\nendstream\nendobj\ntrailer\n<< /Root 1 0 R >>\n%%EOF\n", + content.len() + ) + .as_bytes(), + ); + let mut app = PdfCraftApp::new(); + app.open_bytes("scan.pdf", None, pdf).unwrap(); + let (i, doc) = app.active_ids().unwrap(); + let ctx = egui::Context::default(); + let tex = ctx.load_texture("before", egui::ColorImage::filled([1, 1], Color32::WHITE), TextureOptions::LINEAR); + app.views[i].pages.insert(0, PageTex { tag: 1000, tex }); + let job = app.session.scan_clean_job(doc, &[], pdfcraft_engine::scan_clean::ScanCleanSettings::default()).unwrap(); + let cleaned = job.run(|_, _| true).unwrap(); + assert_eq!(cleaned.report.pages, vec![0], "the page is a scan and gets cleaned"); + let progress = + std::sync::Arc::new(std::sync::Mutex::new(crate::scan_clean_ui::ScanCleanProgress { result: Some(Ok(cleaned)), ..Default::default() })); + app.scan_clean_run = Some(crate::scan_clean_ui::ScanCleanRun { doc, progress }); + app.poll_scan_clean(); + assert!(app.scan_clean_run.is_none()); + assert_eq!(app.views[i].pages.get(&0).unwrap().tag, STALE_TAG, "the old raster is re-rendered"); +} diff --git a/crates/ui-egui/src/workflows/scan_clean_ui.rs b/crates/ui-egui/src/workflows/scan_clean_ui.rs index 64f8ce48eb..81d6764af9 100644 --- a/crates/ui-egui/src/workflows/scan_clean_ui.rs +++ b/crates/ui-egui/src/workflows/scan_clean_ui.rs @@ -230,6 +230,10 @@ impl PdfCraftApp { self.progress_notice = None; match result.and_then(|cleaned| self.session.apply_scan_clean(doc, cleaned)) { Ok(report) => { + // The page rasters, tiles and thumbnails show the old pixels at the same scale. + for v in self.views.iter_mut().filter(|v| v.id == doc) { + v.invalidate_content(); + } let msg = summary(&report); self.notify(msg); } From 04cee34bfe1f6c64fb581a625b145c5a04391d6f Mon Sep 17 00:00:00 2001 From: ohagmann <268387159+ohagmann@users.noreply.github.com> Date: Thu, 5 Nov 2026 13:21:29 +0100 Subject: [PATCH 51/53] M5: a double-click on a comment card's text edits it (#4544) * A double-click on a comment card's text edits it On the selected card in the Comments panel only the header took the double-click: the text is a selectable label, so a double-click there selected a word and the editor stayed closed (a tester's report). It now opens the editor too; dragging over the text still selects it for copying. Co-Authored-By: Claude Opus 5.5 * Tests: the double-clicked card's editor is drawn too Co-Authored-By: Claude Opus 5.5 * Review: the double-click test also starts from a card that isn't selected It selected the card before any input, so it never covered what the change is for in the Comments panel as one uses it: a first click on the text of an unselected card selects it, the second edits it. The test now runs both ways. Co-Authored-By: Claude Opus 5.5 --------- Co-authored-by: Claude Opus 5.5 --- crates/ui-egui/src/panels/comments_panel.rs | 12 ++++-- crates/ui-egui/tests/it/comments.rs | 47 +++++++++++++++++++++ parity/acrobat-features.toml | 4 +- 3 files changed, 57 insertions(+), 6 deletions(-) diff --git a/crates/ui-egui/src/panels/comments_panel.rs b/crates/ui-egui/src/panels/comments_panel.rs index d54daf5717..591a667159 100644 --- a/crates/ui-egui/src/panels/comments_panel.rs +++ b/crates/ui-egui/src/panels/comments_panel.rs @@ -433,7 +433,7 @@ fn card(ui: &mut egui::Ui, view: &mut DocView, context: CardContext<'_>) -> Opti buttons_left = ui.min_rect().left(); }); }); - egui::Frame::NONE.inner_margin(egui::Margin { left: 34, right: 0, top: 4, bottom: 0 }).show(ui, |ui| { + let body = egui::Frame::NONE.inner_margin(egui::Margin { left: 34, right: 0, top: 4, bottom: 0 }).show(ui, |ui| { ui.set_width(ui.available_width()); if editing { if let Some(e) = inline_editor(ui, t, view, key, 2) { @@ -531,7 +531,7 @@ fn card(ui: &mut egui::Ui, view: &mut DocView, context: CardContext<'_>) -> Opti // The header selects the card, except where its buttons are. let mut r = head.response.rect; r.max.x = r.max.x.min(buttons_left - 2.0); - r + (r, body.response.rect) }); let rect = frame.response.rect; if selected { @@ -545,7 +545,8 @@ fn card(ui: &mut egui::Ui, view: &mut DocView, context: CardContext<'_>) -> Opti ui.add_space(5.0); // The header row selects the card (and shows the comment on the page); unselected cards are // clickable everywhere. - let hit = if selected { frame.inner } else { rect }; + let (head, body) = frame.inner; + let hit = if selected { head } else { rect }; let click = ui .interact(hit, ui.scope_id().with(("card", key)), Sense::click()) .on_hover_cursor(egui::CursorIcon::PointingHand) @@ -555,7 +556,10 @@ fn card(ui: &mut egui::Ui, view: &mut DocView, context: CardContext<'_>) -> Opti view.comments.reply.clear(); *nav = Some(Nav::Flash(a.page, a.shown_rect(rotation))); } - if click.double_clicked() && allowed { + // On the selected card a double-click on the text edits it too (dragging over it selects + // words of it, for copying). + let body = (selected && !editing).then(|| ui.interact(body, ui.scope_id().with(("card-body", key)), Sense::click())); + if (click.double_clicked() || body.is_some_and(|b| b.double_clicked())) && allowed { view.comments.open_editor(a.page, a.index, a.contents.clone().unwrap_or_default()); } let menu_id = ui.scope_id().with(("card-menu", key)); diff --git a/crates/ui-egui/tests/it/comments.rs b/crates/ui-egui/tests/it/comments.rs index 129bcbf3e4..29396830f7 100644 --- a/crates/ui-egui/tests/it/comments.rs +++ b/crates/ui-egui/tests/it/comments.rs @@ -143,6 +143,53 @@ fn sidebar_comment_centres_the_annotation_at_high_zoom() { } } +/// A double-click on the text of a card in the Comments panel edits it, as one on its header does +/// (a click there would select a word): on a selected card, and on one the first click selects. +#[test] +fn double_clicking_a_cards_text_edits_it() { + for selected in [true, false] { + double_click_a_cards_text(selected); + } +} + +fn double_click_a_cards_text(selected: bool) { + // Frames 1/60 s apart, so two clicks make a double-click. + let mut h = Harness::builder().with_size(egui::vec2(1400.0, 900.0)).with_step_dt(1.0 / 60.0).build_eframe(move |_cc| { + let mut app = PdfCraftApp::new(); + app.set_option("language", "en").unwrap(); + app.open_bytes("text.pdf", None, TEXT_FIXTURE.to_vec()).expect("opens"); + app.set_option("left", "closed").unwrap(); + app.set_option("panel", "comments").unwrap(); + let shape = pdfcraft_engine::Shape::Note { at: [150.0, 90.0], icon: pdfcraft_engine::NoteIcon::Comment }; + assert!(app.apply_edit(pdfcraft_engine::Edit::AddAnnotation(pdfcraft_engine::NewAnnotation { + page: 0, + style: pdfcraft_engine::Style::default_for(&shape), + shape, + contents: "Edit this note".into(), + author: "Tester".into(), + }))); + app.views[0].comments.selected = selected.then_some((0, 0)); + app + }); + h.run_steps(4); + crate::support::settle(&mut h); + assert_eq!(h.state().views[0].comments.selected.is_some(), selected); + let p = h.query_all_by_label("Edit this note").next().expect("the card's text").rect().center(); + h.hover_at(p); + h.run_steps(1); + for _ in 0..2 { + for pressed in [true, false] { + h.event(egui::Event::PointerButton { pos: p, button: egui::PointerButton::Primary, pressed, modifiers: egui::Modifiers::NONE }); + } + h.run_steps(1); + } + h.run_steps(2); + assert_eq!(h.state().views[0].comments.selected, Some((0, 0)), "selected: {selected}"); + assert_eq!(h.state().views[0].comments.editing, Some((0, 0, "Edit this note".into())), "selected: {selected}"); + // And the editor is drawn. + note_editor(&h); +} + #[test] fn dragging_over_text_with_the_highlighter_highlights_it() { let mut h = harness(|app| app.set_option("quick", "highlight").unwrap()); diff --git a/parity/acrobat-features.toml b/parity/acrobat-features.toml index 6ff5eca28e..f543284d54 100644 --- a/parity/acrobat-features.toml +++ b/parity/acrobat-features.toml @@ -4118,8 +4118,8 @@ milestone = "M5" status = "shipped" commands = ["comment.list"] tools = ["comment_edit"] -evidence = ["crates/annot/src/tests.rs::text_box_text_and_colour_changes_are_drawn", "crates/annot/src/tests.rs::editing_a_text_box_refits_its_rectangle", "crates/automation/tests/it/automation.rs::comments_through_tools", "crates/ui-egui/tests/it/comments.rs::ctrl_enter_posts_a_highlight_comment_and_shift_enter_adds_a_line", "crates/ui-egui/tests/it/comments.rs::one_reply_is_deleted_from_its_menu_and_the_thread_stays", "crates/ui-egui/tests/it/comments.rs::undo_and_redo_close_a_reply_editor_left_open"] -notes = "double-click a comment or card, or Edit in the menus (a reply from its own \"…\" menu); Ctrl+Enter (also Command+Enter) saves text in the focused comment editor, Shift+Enter adds a newline; a text box's rectangle follows the new text (wrap width and top edge stay, the height fits the wrapped lines). Undo or Redo closes the panel's editor without saving it, since either can move the comment it was open on" +evidence = ["crates/annot/src/tests.rs::text_box_text_and_colour_changes_are_drawn", "crates/annot/src/tests.rs::editing_a_text_box_refits_its_rectangle", "crates/automation/tests/it/automation.rs::comments_through_tools", "crates/ui-egui/tests/it/comments.rs::ctrl_enter_posts_a_highlight_comment_and_shift_enter_adds_a_line", "crates/ui-egui/tests/it/comments.rs::one_reply_is_deleted_from_its_menu_and_the_thread_stays", "crates/ui-egui/tests/it/comments.rs::undo_and_redo_close_a_reply_editor_left_open", "crates/ui-egui/tests/it/comments.rs::double_clicking_a_cards_text_edits_it"] +notes = "double-click a comment or card (its header or its text), or Edit in the menus; double-click a comment or card, or Edit in the menus (a reply from its own \"…\" menu); Ctrl+Enter (also Command+Enter) saves text in the focused comment editor, Shift+Enter adds a newline; a text box's rectangle follows the new text (wrap width and top edge stay, the height fits the wrapped lines). Undo or Redo closes the panel's editor without saving it, since either can move the comment it was open on" [[feature]] id = "comment.copy-comment-text" From 96fe832e12c6de3ba88005496ba3d57cc3039fa9 Mon Sep 17 00:00:00 2001 From: ohagmann <268387159+ohagmann@users.noreply.github.com> Date: Thu, 5 Nov 2026 14:01:18 +0100 Subject: [PATCH 52/53] M5: Shift holds lines level or plumb while drawing (ortho) (#4285) MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit * M5: Shift holds lines level or plumb while drawing (ortho) With Shift held, a new line or arrow runs level or plumb from where it starts, whichever is nearer to the pointer (a CAD drawing's ortho mode), and each new side of connected lines, a polygon or a cloud is held the same way from the corner before, rubber band included. Dragging an end or corner of a drawn one with Shift is held the same way; it turned to 45° steps before. The measuring tools keep their 45° steps. measure::snap::ortho next to constrain; comment_add takes `ortho` for line, arrow, polygon, cloud and polyline. Parity comment.draw-ortho. Co-Authored-By: Claude Opus 5.5 * Ortho: over the first corner the rubber band shows the closing side, Shift or not From review: a click there closes the polygon or cloud to its first corner, while the preview with Shift held showed a level or plumb side. The click and the preview now share one rule (closes_at). Co-Authored-By: Claude Opus 5.5 * Rebase: Shift is read where main's split put the drawing input, and the ortho helpers are imported there Co-Authored-By: Claude Opus 5.5 * Rebase: Shift holds a drawn line, not a freehand stroke (a hunk had landed in the wrong branch) Co-Authored-By: Claude Opus 5.5 * Ortho: the parity note names the closing side's exception; the test's doc comment is back above its test Co-Authored-By: Claude Opus 5.5 --------- Co-authored-by: Claude Opus 5.5 Co-authored-by: FredApps <153173854+FredApps@users.noreply.github.com> --- ROADMAP.md | 1 + crates/automation/src/comments.rs | 16 +++- crates/automation/src/tools.rs | 1 + crates/automation/tests/it/automation.rs | 38 +++++++++ crates/measure/src/snap.rs | 6 ++ crates/measure/src/tests.rs | 10 +++ crates/ui-egui/src/tools/comments/geometry.rs | 18 ++-- crates/ui-egui/src/tools/comments/input.rs | 18 ++-- crates/ui-egui/src/tools/comments/paint.rs | 14 +++- crates/ui-egui/tests/it/comments.rs | 83 +++++++++++++++++-- docs/ui-parity.md | 1 + parity/acrobat-features.toml | 14 +++- 12 files changed, 197 insertions(+), 23 deletions(-) diff --git a/ROADMAP.md b/ROADMAP.md index 49f7d4101d..5eaf0f5035 100644 --- a/ROADMAP.md +++ b/ROADMAP.md @@ -302,6 +302,7 @@ Ranked; detail and the full milestone table in [docs/roadmap.md](docs/roadmap.md - **2026-10-21 (M3, window resize and move on Wayland):** With PdfCraft's own title bar on GNOME/Wayland, a second resize by a window edge, or a second move by the empty tab strip, works straight after the first; it waited for a click inside the window, because the compositor keeps the button's release from egui. A unit test covers the release given back; checked by hand in a test build. Estimates unchanged. - **2026-10-24 (M5, move preview):** A comment being moved is drawn where it goes while it is dragged (a rectangle or oval showed only its blue frame): the comment is rendered alone once as the drag starts and laid over the page, which shows it where it was until the move is applied. Engine `Document::comment_layer`; synthetic engine and UI regressions. Estimates unchanged. - **2026-10-24 (M5, Escape in Fill & Sign):** Escape leaves a Fill & Sign tool as it leaves a comment tool: the first press closes the open type box, the next goes back to Select, so the Text tool no longer opens a box at every click with no way out by the keyboard. A synthetic UI regression covers the Text and Checkmark tools. Estimates unchanged. +- **2026-10-23 (M5, ortho):** Shift while drawing holds a line or arrow level or plumb, whichever is nearer (a CAD drawing's ortho mode), and each new side of connected lines, a polygon or a cloud from the corner before; dragging an end or corner with Shift is held the same way (it turned to 45° steps). `measure::snap::ortho`; `comment_add ortho`. Estimates unchanged. - **2026-10-22 (M4, #3916 partial):** Document edits and Undo/Redo cancel an Organize page drag. Pointer-release regressions check exact page order and save/reopen. - **2026-10-22 (M10, #3752 paper):** CUPS jobs ask for the chosen paper in every size. The 13 sizes the Print dialog sent no paper for (A6, A0–A2, ARCH, ANSI C–E) use PWG 5101.1 names, checked against a synthetic PPD with `cupsfilter`, and `doc_print` to a printer sends the paper too. The macOS tray half (Auto Select instead of the PPD's default tray) is a craft-print change that needs a pin bump here. Estimates unchanged. diff --git a/crates/automation/src/comments.rs b/crates/automation/src/comments.rs index 135d0868ee..a1e5e05263 100644 --- a/crates/automation/src/comments.rs +++ b/crates/automation/src/comments.rs @@ -253,6 +253,12 @@ impl Automation { if endings.is_some() && !matches!(kind, "line" | "arrow" | "polyline" | "callout") { return Err(ToolError::InvalidArgs("endings apply to a line, arrow, polyline or callout".into())); } + // Shift while drawing: each end or corner held level or plumb with the one before. + let ortho = a.opt_bool("ortho")?.unwrap_or(false); + if ortho && !matches!(kind, "line" | "arrow" | "polygon" | "cloud" | "polyline") { + return Err(ToolError::InvalidArgs("ortho applies to a line, arrow, polygon, cloud or polyline".into())); + } + let held = |last: [f64; 2], p: [f64; 2]| if ortho { pdfcraft_engine::measure::snap::ortho(last, p) } else { p }; let markup = match kind { "highlight" => Some(Markup::Highlight), "underline" => Some(Markup::Underline), @@ -325,7 +331,8 @@ impl Automation { Shape::TextBox { rect: rect_to_user(&info, a.need::<4>("rect", "a text box")?), font_size } } "line" | "arrow" => { - let (f, t) = (a.need::<2>("from", "a line")?, a.need::<2>("to", "a line")?); + let f = a.need::<2>("from", "a line")?; + let t = held(f, a.need::<2>("to", "a line")?); let [start, end] = ending_pair(kind, endings.as_deref())?; Shape::Line { from: to_user(&info, f[0], f[1]), to: to_user(&info, t[0], t[1]), start, end } } @@ -352,10 +359,15 @@ impl Automation { "polygon" | "cloud" | "polyline" => { let wrong = || ToolError::InvalidArgs(format!("{kind} needs `points`: an array of [x, y] points")); let pts = a.get("points").and_then(|p| p.as_array()).ok_or_else(wrong)?; + let mut last = None; let vertices = pts .iter() .map(|p| match p.as_array().map(|p| p.iter().map(|v| v.as_f64()).collect::>()).as_deref() { - Some([Some(x), Some(y)]) => Ok(to_user(&info, *x, *y)), + Some([Some(x), Some(y)]) => { + let p = last.map_or([*x, *y], |l| held(l, [*x, *y])); + last = Some(p); + Ok(to_user(&info, p[0], p[1])) + } _ => Err(wrong()), }) .collect::>>()?; diff --git a/crates/automation/src/tools.rs b/crates/automation/src/tools.rs index 14b55361c5..0728679c8e 100644 --- a/crates/automation/src/tools.rs +++ b/crates/automation/src/tools.rs @@ -861,6 +861,7 @@ pub fn tools() -> Vec { "opacity": { "type": "number", "minimum": 0, "maximum": 1 }, "width": { "type": "number", "minimum": 0, "description": "Line width in points." }, "endings": { "type": "array", "items": { "type": "string", "enum": ["None", "Square", "Circle", "Diamond", "OpenArrow", "ClosedArrow", "Butt", "ROpenArrow", "RClosedArrow", "Slash"] }, "minItems": 1, "maxItems": 2, "description": "Line endings. Two names (start, end) for a line, arrow or polyline; one name for a callout." }, + "ortho": { "type": "boolean", "description": "line/arrow/polygon/cloud/polyline: hold each end or corner level or plumb with the one before, whichever is nearer (Shift while drawing)." }, "font_size": { "type": "number", "exclusiveMinimum": 0 }, }), &["doc", "page", "type"], diff --git a/crates/automation/tests/it/automation.rs b/crates/automation/tests/it/automation.rs index 35c4db79c2..5bde9d27f9 100644 --- a/crates/automation/tests/it/automation.rs +++ b/crates/automation/tests/it/automation.rs @@ -6493,6 +6493,44 @@ fn drawing_comments_through_tools() { assert!(matches!(png[0], Content::Png { .. })); } +/// comment_add `ortho`, Shift's hold while drawing: a line's end is held level or plumb with its +/// start, and each corner of connected lines or a polygon with the corner before. +#[test] +fn lines_and_corners_are_held_level_or_plumb_through_the_tool() { + let dir = workdir("ortho"); + let mut a = auto(&dir); + let doc = ok(&mut a, "doc_open", json!({ "path": "a.pdf" }))["doc"].as_u64().unwrap(); + for args in [ + json!({ "type": "line", "from": [20, 40], "to": [120, 52], "ortho": true }), + json!({ "type": "arrow", "from": [180, 40], "to": [170, 140], "ortho": true }), + json!({ "type": "polyline", "points": [[20, 200], [90, 210], [95, 260], [130, 250]], "ortho": true }), + json!({ "type": "line", "from": [20, 280], "to": [120, 290] }), + ] { + let mut args = args; + args.as_object_mut().unwrap().extend([("doc".to_string(), json!(doc)), ("page".to_string(), json!(1))]); + ok(&mut a, "comment_add", args); + } + let list = ok(&mut a, "comment_list", json!({ "doc": doc, "page": 1 })); + // In the order added: the list isn't sorted by index. + let mut listed = list["comments"].as_array().unwrap().clone(); + listed.sort_by_key(|c| c["index"].as_u64()); + let points: Vec> = + listed.iter().map(|c| c["points"].as_array().unwrap().iter().map(|p| [p[0].as_f64().unwrap(), p[1].as_f64().unwrap()]).collect()).collect(); + let want: [&[[f64; 2]]; 4] = [ + &[[20.0, 40.0], [120.0, 40.0]], + &[[180.0, 40.0], [180.0, 140.0]], + &[[20.0, 200.0], [90.0, 200.0], [90.0, 260.0], [130.0, 260.0]], + &[[20.0, 280.0], [120.0, 290.0]], + ]; + assert_eq!(points.len(), want.len(), "{points:?}"); + for (got, want) in points.iter().zip(want) { + assert_eq!(got.len(), want.len(), "{got:?}"); + assert!(got.iter().zip(want).all(|(g, w)| (g[0] - w[0]).abs() < 0.01 && (g[1] - w[1]).abs() < 0.01), "{got:?} for {want:?}"); + } + let refused = a.call("comment_add", &json!({ "doc": doc, "page": 1, "type": "rectangle", "rect": [10, 10, 50, 50], "ortho": true })); + assert!(matches!(refused, Err(ToolError::InvalidArgs(_))), "ortho is for lines and corners: {refused:?}"); +} + #[test] fn signing_with_an_image_typed_or_drawn_graphic_through_tools() { let dir = workdir("signing-graphic"); diff --git a/crates/measure/src/snap.rs b/crates/measure/src/snap.rs index 930ffe3053..b26c389e96 100644 --- a/crates/measure/src/snap.rs +++ b/crates/measure/src/snap.rs @@ -189,6 +189,12 @@ pub fn constrain(last: Point, point: Point) -> Point { let angle = (v[1].atan2(v[0]) / std::f64::consts::FRAC_PI_4).round() * std::f64::consts::FRAC_PI_4; [last[0] + length * angle.cos(), last[1] + length * angle.sin()] } +/// Hold a proposed vertex level or plumb with the last vertex, whichever is nearer (a CAD +/// drawing's ortho mode): the vertex slides onto that axis rather than turning onto it. +pub fn ortho(last: Point, point: Point) -> Point { + let v = sub(point, last); + if v[0].abs() >= v[1].abs() { [point[0], last[1]] } else { [last[0], point[1]] } +} struct Walker<'a> { doc: &'a Document, geometry: Geometry, diff --git a/crates/measure/src/tests.rs b/crates/measure/src/tests.rs index 5b277e5546..88906910c7 100644 --- a/crates/measure/src/tests.rs +++ b/crates/measure/src/tests.rs @@ -449,6 +449,16 @@ fn listing_skips_repeated_refs_and_caps_what_it_keeps() { assert!(!listing.measurements.is_empty()); } #[test] +fn ortho_holds_a_vertex_level_or_plumb_with_the_last() { + let last = [10.0, 20.0]; + assert_eq!(snap::ortho(last, [50.0, 27.0]), [50.0, 20.0], "mostly across: level"); + assert_eq!(snap::ortho(last, [-30.0, 13.0]), [-30.0, 20.0], "to the left too"); + assert_eq!(snap::ortho(last, [14.0, 80.0]), [10.0, 80.0], "mostly up: plumb"); + assert_eq!(snap::ortho(last, [8.0, -40.0]), [10.0, -40.0], "and down"); + assert_eq!(snap::ortho(last, [30.0, 40.0]), [30.0, 20.0], "a tie stays level"); + assert_eq!(snap::ortho(last, last), last); +} +#[test] fn a_stream_over_the_byte_budget_uses_it_up_once() { // 50 references to one Flate stream that inflates past the 16 MiB budget decode once. let big = pdfcraft_cos::Stream::flate(Dict::new(), &vec![b' '; 17 << 20]); diff --git a/crates/ui-egui/src/tools/comments/geometry.rs b/crates/ui-egui/src/tools/comments/geometry.rs index ee67b92bb0..4f3b338fe6 100644 --- a/crates/ui-egui/src/tools/comments/geometry.rs +++ b/crates/ui-egui/src/tools/comments/geometry.rs @@ -32,15 +32,21 @@ pub(super) fn point_handle(cx: &PageCx<'_>, a: &Annotation, p: Pos2) -> Option, to: [f64; 2], shift: bool) -> [f64; 2] { + match last { + Some(l) if shift => pdfcraft_engine::measure::snap::ortho(*l, to), + _ => to, + } +} + +/// `points` with point `i` moved to `to`; with Shift, held level or plumb with the point before +/// it (a line's other end; the first corner is held to the second). pub(crate) fn reshaped(points: &[[f64; 2]], i: usize, to: [f64; 2], shift: bool) -> Vec<[f64; 2]> { let mut out = points.to_vec(); let anchor = if i == 0 { points.get(1) } else { points.get(i - 1) }; - let to = match anchor { - Some(a) if shift => pdfcraft_engine::measure::snap::constrain(*a, to), - _ => to, - }; + let to = held(anchor, to, shift); if let Some(p) = out.get_mut(i) { *p = to; } diff --git a/crates/ui-egui/src/tools/comments/input.rs b/crates/ui-egui/src/tools/comments/input.rs index 23a5ab3f06..caddd3e837 100644 --- a/crates/ui-egui/src/tools/comments/input.rs +++ b/crates/ui-egui/src/tools/comments/input.rs @@ -1,8 +1,8 @@ use super::geometry::{ - CARET_SIZE, HANDLES, clamp_to, corner_ratio, drawn_box, handle_pos, is_markup, point_handle, reaches_points, reshapable, reshaped, resizable, - resized, snapped_caret, + CARET_SIZE, HANDLES, clamp_to, corner_ratio, drawn_box, handle_pos, held, is_markup, point_handle, reaches_points, reshapable, reshaped, + resizable, resized, snapped_caret, }; -use super::paint::{drawn_shape, edit_anchor, erase, new_comment}; +use super::paint::{closes_at, drawn_shape, edit_anchor, erase, new_comment}; use super::state::{Composer, ComposerKind, Gesture, PageCx}; use super::{Annotation, CommentPrefs, CommentTool, DocInfo, DocView, Edit, Pos2, QuickTool, Rect, Shape, vec2}; use crate::page_cursor::{Layer, Subject}; @@ -346,6 +346,8 @@ impl Gestures<'_> { } fn draw_shape(&self, view: &mut DocView, tool: CommentTool) -> bool { + // Shift holds a line level or plumb (ortho). + let shift = self.resp.ctx.input(|i| i.modifiers.shift); let Self { ui, resp, cx, pointer, origin, page_rect, pressed_here, over_page, .. } = *self; let cv = &mut view.comments; if !cx.allowed { @@ -374,7 +376,8 @@ impl Gestures<'_> { } } else { points.truncate(1); - points.push(u); + let line = matches!(*tool, CommentTool::Line | CommentTool::Arrow); + points.push(held(points.first(), u, line && shift)); } } if resp.drag_stopped() @@ -408,6 +411,8 @@ impl Gestures<'_> { } fn click_points(&self, view: &mut DocView, tool: CommentTool) -> bool { + // Shift holds the next side of connected lines, a polygon or a cloud level or plumb. + let shift = self.resp.ctx.input(|i| i.modifiers.shift); let Self { ui, resp, cx, pointer, over_page, .. } = *self; let cv = &mut view.comments; if !cx.allowed { @@ -434,8 +439,9 @@ impl Gestures<'_> { match cv.gesture.as_mut() { Some(Gesture::Draw { page, tool: t, points }) if *page == cx.page && *t == tool => { // Clicking the first point again closes a polygon. - let closes = points.len() >= 3 && tool != CommentTool::PolyLine && cx.to_screen(points[0]).distance(p) <= 6.0; - let repeat = points.last().is_some_and(|l| cx.to_screen(*l).distance(p) < 2.0); + let closes = closes_at(cx, tool, points, p); + let u = held(points.last(), u, shift); + let repeat = points.last().is_some_and(|l| cx.to_screen(*l).distance(cx.to_screen(u)) < 2.0); if closes { let pts = std::mem::take(points); cv.gesture = None; diff --git a/crates/ui-egui/src/tools/comments/paint.rs b/crates/ui-egui/src/tools/comments/paint.rs index 8faae3d02e..f993ecf85b 100644 --- a/crates/ui-egui/src/tools/comments/paint.rs +++ b/crates/ui-egui/src/tools/comments/paint.rs @@ -1,4 +1,4 @@ -use super::geometry::{HANDLES, handle_pos, is_markup, reaches_points, reshapable, reshaped, resizable}; +use super::geometry::{HANDLES, handle_pos, held, is_markup, reaches_points, reshapable, reshaped, resizable}; use super::state::{CommentView, Gesture, PageCx, SETTLE_SECS}; use super::{ Annotation, Color32, CommentTool, CornerRadius, DocInfo, DocView, Edit, LineEnding, NewAnnotation, Pos2, QuickTool, Rect, SELECT_BLUE, Shape, @@ -165,10 +165,12 @@ fn paint_draw(painter: &egui::Painter, cx: &PageCx<'_>, tool: &CommentTool, poin painter.add(egui::Shape::ellipse_stroke(r.center(), r.size() / 2.0, stroke)); } CommentTool::Polygon | CommentTool::PolyLine | CommentTool::Cloud => { - // The vertices so far, and a rubber band to the pointer. + // The vertices so far, and a rubber band to where a click would put the next one: the + // first, where the click closes the shape, Shift or not. let mut line = pts.clone(); if let Some(p) = cx.ptr.pos { - line.push(p); + let shift = painter.ctx().input(|i| i.modifiers.shift); + line.push(if closes_at(cx, *tool, points, p) { pts[0] } else { cx.to_screen(held(points.last(), cx.to_user(p), shift)) }); } painter.add(egui::Shape::line(line, stroke)); if let Some(first) = pts.first().filter(|_| *tool != CommentTool::PolyLine && pts.len() >= 3) { @@ -187,6 +189,12 @@ fn paint_draw(painter: &egui::Painter, cx: &PageCx<'_>, tool: &CommentTool, poin } } +/// Whether a click at `p` (on screen) closes the polygon or cloud drawn so far (`points`, user +/// space): on its first point again, once it has three. +pub(super) fn closes_at(cx: &PageCx<'_>, tool: CommentTool, points: &[[f64; 2]], p: Pos2) -> bool { + points.len() >= 3 && tool != CommentTool::PolyLine && cx.to_screen(points[0]).distance(p) <= 6.0 +} + /// The shape a finished drawing gesture makes (`None` if it is too small to mean anything). pub(super) fn drawn_shape(tool: CommentTool, points: &[[f64; 2]]) -> Option { let (first, last) = (*points.first()?, *points.last()?); diff --git a/crates/ui-egui/tests/it/comments.rs b/crates/ui-egui/tests/it/comments.rs index 29396830f7..674cd90d26 100644 --- a/crates/ui-egui/tests/it/comments.rs +++ b/crates/ui-egui/tests/it/comments.rs @@ -956,8 +956,8 @@ fn comments_are_selected_moved_and_deleted_with_the_select_tool() { } /// A line's ends have handles: dragging one with the Select tool changes the line's length and -/// angle (Shift holds it to 45°), as one undo step. The line tool keeps drawing from an end: a -/// new line starts there rather than dragging the one just drawn. +/// angle (Shift holds it level or plumb with the other end), as one undo step. The line tool keeps +/// drawing from an end: a new line starts there rather than dragging the one just drawn. #[test] fn line_ends_are_dragged_to_change_length_and_angle() { let mut h = harness(|app| app.set_option("quick", "line").unwrap()); @@ -981,7 +981,7 @@ fn line_ends_are_dragged_to_change_length_and_angle() { assert!((p[1][0] - 200.0).abs() < 1.0 && (p[1][1] - 160.0).abs() < 1.0, "the dragged end follows: {p:?}"); let id = h.state().views[0].id; assert_eq!(h.state().session.get(id).unwrap().can_undo(), Some("Reshape comment")); - // Shift: held to the nearest 45° from the other end (39° here). + // Shift: held level or plumb with the other end, here level (110 pt across, 90 up). let a = at(&h, 200.0, 160.0); let b = at(&h, 150.0, 190.0); h.hover_at(a); @@ -998,8 +998,7 @@ fn line_ends_are_dragged_to_change_length_and_angle() { h.event(egui::Event::ModifiersChanged(egui::Modifiers::NONE)); h.run_steps(1); let p = ends(&h); - let (dx, dy) = (p[1][0] - p[0][0], p[1][1] - p[0][1]); - assert!(dx > 50.0 && (dx - dy).abs() < 0.5, "45°: {p:?}"); + assert!((p[1][1] - p[0][1]).abs() < 0.01 && (p[1][0] - 150.0).abs() < 1.0, "level: {p:?}"); // Undo puts the line back as it was before the drag. h.state_mut().undo(); h.run_steps(2); @@ -1083,6 +1082,80 @@ fn alt_arrows_navigate_instead_of_nudging_the_selected_comment() { assert!(comments(&h)[0].rect[0] > before[0], "→ nudged it"); } +/// Over the first corner, where a click closes the polygon, the rubber band runs to that corner +/// with Shift held too: the closing side the click will make, not a level or plumb one. +#[test] +fn the_rubber_band_shows_the_closing_side_with_shift_held() { + let _gpu = crate::gpu(); + let mut h = harness(|app| app.set_option("quick", "polygon").unwrap()); + for p in [(40.0, 40.0), (140.0, 40.0), (140.0, 140.0)] { + click_pt(&mut h, p); + } + h.event(egui::Event::ModifiersChanged(egui::Modifiers::SHIFT)); + h.hover_at(at(&h, 40.0, 40.0) + egui::vec2(2.0, 2.0)); + h.run_steps(3); + let pixels = h.render().unwrap(); + // Halfway along the diagonal closing side, (90, 90). + let m = at(&h, 90.0, 90.0); + let drawn = (-2..=2).any(|d: i32| { + let v = pixels.get_pixel((m.x as i32 + d) as u32, (m.y as i32 + d) as u32).0; + v[0] > 150 && v[1] < 100 && v[2] < 100 + }); + assert!(drawn, "the closing side is drawn"); + h.event(egui::Event::ModifiersChanged(egui::Modifiers::NONE)); + h.run_steps(1); +} + +/// Shift holds a new line or arrow level or plumb with where it starts, and each new side of +/// connected lines or a polygon with the corner before (a CAD drawing's ortho mode). Without +/// Shift they go where the pointer is. +#[test] +fn shift_holds_new_lines_and_sides_level_or_plumb() { + let mut h = harness(|app| app.set_option("quick", "line").unwrap()); + let shift = |h: &mut Harness<'static, PdfCraftApp>, on: bool| { + h.event(egui::Event::ModifiersChanged(if on { egui::Modifiers::SHIFT } else { egui::Modifiers::NONE })); + h.run_steps(1); + }; + let points = |h: &Harness<'static, PdfCraftApp>, subtype: &str| { + // In the order drawn: the list isn't sorted by index. + let mut drawn: Vec<_> = comments(h).into_iter().filter(|a| a.subtype == subtype).collect(); + drawn.sort_by_key(|a| a.index); + drawn.into_iter().map(|a| a.points).collect::>() + }; + let close = |p: [f32; 2], q: [f32; 2]| (p[0] - q[0]).abs() < 1.0 && (p[1] - q[1]).abs() < 1.0; + shift(&mut h, true); + drag_pt(&mut h, (40.0, 100.0), (140.0, 120.0)); + drag_pt(&mut h, (200.0, 40.0), (215.0, 150.0)); + shift(&mut h, false); + drag_pt(&mut h, (40.0, 160.0), (140.0, 190.0)); + let lines = points(&h, "Line"); + assert_eq!(lines.len(), 3, "{lines:?}"); + assert!(close(lines[0][0], [40.0, 100.0]) && close(lines[0][1], [140.0, 100.0]), "mostly across: level: {lines:?}"); + assert!(close(lines[1][0], [200.0, 40.0]) && close(lines[1][1], [200.0, 150.0]), "mostly up: plumb: {lines:?}"); + assert!(close(lines[2][1], [140.0, 190.0]), "no Shift, no hold: {lines:?}"); + + h.state_mut().set_option("quick", "arrow").unwrap(); + shift(&mut h, true); + drag_pt(&mut h, (260.0, 180.0), (150.0, 170.0)); + shift(&mut h, false); + let arrow = points(&h, "Line").pop().expect("an arrow"); + assert!(close(arrow[1], [150.0, 180.0]), "an arrow too, leftwards: {arrow:?}"); + + // Connected lines: each click with Shift is held to the corner before. + h.state_mut().set_option("quick", "polyline").unwrap(); + click_pt(&mut h, (20.0, 20.0)); + shift(&mut h, true); + click_pt(&mut h, (90.0, 30.0)); + click_pt(&mut h, (95.0, 80.0)); + shift(&mut h, false); + click_pt(&mut h, (130.0, 60.0)); + h.key_press(egui::Key::Enter); + h.run_steps(3); + let poly = points(&h, "PolyLine").pop().expect("connected lines"); + assert_eq!(poly.len(), 4, "{poly:?}"); + assert!(close(poly[1], [90.0, 20.0]) && close(poly[2], [90.0, 80.0]) && close(poly[3], [130.0, 60.0]), "{poly:?}"); +} + #[test] fn a_shape_just_drawn_resizes_from_its_handles_without_the_select_tool() { // #2338: the new shape showed its handles, but only the Select tool answered them. diff --git a/docs/ui-parity.md b/docs/ui-parity.md index f950e36a94..dbb38ce730 100644 --- a/docs/ui-parity.md +++ b/docs/ui-parity.md @@ -49,6 +49,7 @@ No dump of Acrobat's menu tree has been diffed against the registry yet (gap 24) | Select and move several objects at once | yes | on one page: Shift/Ctrl/⌘-click and marquee, one undo step (#844); not across pages, no group resize | partial | | Moved comments drawn where they go while dragged | yes | yes: the comment's own pixels follow the pointer, the page shows it where it was until release; stamps and Fill & Sign marks also hide the original | shipped | | Object handles: corners keep proportion, edges free | yes | yes for added images and signatures; Shift on a corner keeps the proportions of comments, stamps and form fields (#2199); no Alt resize from the centre | shipped | +| Shift while drawing a line | holds it to 45° steps | holds it level or plumb (a CAD drawing's ortho): new lines and arrows, each side of connected lines, polygons and clouds, and a dragged end or corner | shipped | | Arrange, align and distribute page objects | yes | fields only | partial | | Rulers, grids, snap to grid, guides | yes | no | planned | | Cursor coordinates, line weights | yes | no | planned | diff --git a/parity/acrobat-features.toml b/parity/acrobat-features.toml index c62ad80a2d..d4a8c37392 100644 --- a/parity/acrobat-features.toml +++ b/parity/acrobat-features.toml @@ -4095,7 +4095,19 @@ status = "shipped" commands = [] tools = ["comment_edit"] evidence = ["crates/annot/src/tests.rs::lines_and_polygons_are_reshaped_by_their_points", "crates/ui-egui/tests/it/comments.rs::line_ends_are_dragged_to_change_length_and_angle", "crates/automation/tests/it/automation.rs::comments_are_reshaped_through_the_tool"] -notes = "A selected line, arrow, connected lines, polygon or cloud has a handle on each end or corner (instead of the frame); dragging one changes the length and angle (Shift holds it to 45° from the point before), with the new shape drawn while dragging, as one undo step (Edit::ReshapeAnnotation). The drawing tools keep drawing from an end (a new line starts there); the Select tool reshapes. /Rect follows the points with room for the width and arrowheads, and the appearance is drawn anew. Headless: comment_edit `points`; comment_list reports them. Measurements keep the points they measured; locked comments keep their shape. Not yet: adding or removing a corner." +notes = "A selected line, arrow, connected lines, polygon or cloud has a handle on each end or corner (instead of the frame); dragging one changes the length and angle (Shift holds it level or plumb with the point before, comment.draw-ortho), with the new shape drawn while dragging, as one undo step (Edit::ReshapeAnnotation). The drawing tools keep drawing from an end (a new line starts there); the Select tool reshapes. /Rect follows the points with room for the width and arrowheads, and the appearance is drawn anew. Headless: comment_edit `points`; comment_list reports them. Measurements keep the points they measured; locked comments keep their shape. Not yet: adding or removing a corner." + +[[feature]] +id = "comment.draw-ortho" +title = "Shift holds lines level or plumb while drawing (ortho)" +area = "E" +tier = "P1" +milestone = "M5" +status = "shipped" +commands = [] +tools = ["comment_add"] +evidence = ["crates/measure/src/tests.rs::ortho_holds_a_vertex_level_or_plumb_with_the_last", "crates/ui-egui/tests/it/comments.rs::shift_holds_new_lines_and_sides_level_or_plumb", "crates/ui-egui/tests/it/comments.rs::the_rubber_band_shows_the_closing_side_with_shift_held", "crates/ui-egui/tests/it/comments.rs::line_ends_are_dragged_to_change_length_and_angle", "crates/automation/tests/it/automation.rs::lines_and_corners_are_held_level_or_plumb_through_the_tool"] +notes = "With Shift held, a new line or arrow runs level or plumb from where it starts, whichever is nearer to the pointer (a CAD drawing's ortho mode), and each new side of connected lines, a polygon or a cloud from the corner before, the rubber band included. Over the first corner, where a click closes a polygon or cloud, the closing side runs straight to that corner, Shift or not. A dragged end or corner of a drawn one is held the same way (comment.reshape). Acrobat holds lines to 45° steps instead; on drawings and plans the level and plumb lines are the ones wanted. Headless: comment_add `ortho`. Measuring tools keep their 45° steps." [[feature]] id = "comment.move-preview" From 983e2f2ad6bfa4b7377163050f9b3fe16af980d3 Mon Sep 17 00:00:00 2001 From: ohagmann <268387159+ohagmann@users.noreply.github.com> Date: Thu, 5 Nov 2026 14:25:56 +0100 Subject: [PATCH 53/53] =?UTF-8?q?M3:=20=E2=86=92=20/=20=E2=86=90=20scroll?= =?UTF-8?q?=20a=20page=20wider=20than=20the=20window=20before=20turning=20?= =?UTF-8?q?it=20(#4540)?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit * → / ← scroll a page wider than the window sideways before turning it On a plan zoomed in, the arrow keys turned the page (#185) and never showed its other side: the view moved sideways only by the scroll bar, the wheel or the Hand. Now → / ← scroll a line that way while there is more of the page, and turn the page at its edge, as ↓ / ↑ do in single-page view. ⌘→ / ⌘← turn it at once; a page that fits across turns as before. Co-Authored-By: Claude Opus 5.5 * Review: → / ← read the current page's or spread's edges, not the widest page's; ui-parity's change record In continuous view the content is as wide as the widest page, so on a narrower current page that shows whole → scrolled toward the wide page's edge instead of turning. The edges are now the current page's (or spread's) as drawn, or the content's. Co-Authored-By: Claude Opus 5.5 * Review: a spread's second page off screen still counts for → / ← The row's edges came from the pages' screen transforms, and a page gets one only while it meets the window. With a zoomed two-page spread scrolled so its second page was still wholly to the right of the window, the row ended with the first page, so the next → turned the spread instead of scrolling to the second page. The edges are now taken from the layout rectangles of every page in the row, placed as the drawn pages are. A TwoUp case in the viewer tests zooms a spread until its page gap sits at the window's right edge and checks that → scrolls on rather than turning. Co-Authored-By: Claude Opus 5.5 * Review: → / ← read the row after continuous view has moved the current page In continuous view the current page is updated while the pages are laid out, after the row's extent was taken, so the edges could describe the previous row and the next → or ← turn the page instead of scrolling the page that had just become current. The row is now read after that update. Co-Authored-By: Claude Opus 5.5 * Review: a page turned by → / ← at its edge starts at the edge the reader comes from Turning at a wide page's right edge with →, the next page kept the horizontal scroll (the same column, as other jumps keep it), so an equally wide page opened at its right edge and the next → turned again without showing its left side; ← had the same in reverse. When plain → or ← turns the page, a page or spread wider than the window now starts at its left (after →) or right (after ←). ⌘→ / ⌘←, Page Down / Page Up and every other jump keep the column. The sideways-arrow test now goes across two wide pages both ways: → scrolls the next page from its left edge, ← turns back to the previous page shown from its right edge. Co-Authored-By: Claude Opus 5.5 * Rebase: the sideways arrow keys through main's PageEnv, PageOut and GlideCx, and shown_row with its layout Co-Authored-By: Claude Opus 5.5 --------- Co-authored-by: Claude Opus 5.5 Co-authored-by: FredApps Co-authored-by: FredApps <153173854+FredApps@users.noreply.github.com> --- ROADMAP.md | 1 + crates/ui-egui/src/canvas.rs | 69 +++++++++++-- crates/ui-egui/src/canvas/navigation.rs | 32 +++++- crates/ui-egui/tests/it/viewer.rs | 132 ++++++++++++++++++++++++ docs/ui-parity.md | 5 +- parity/acrobat-features.toml | 4 +- 6 files changed, 229 insertions(+), 14 deletions(-) diff --git a/ROADMAP.md b/ROADMAP.md index b3b0691e16..17ef6e4d75 100644 --- a/ROADMAP.md +++ b/ROADMAP.md @@ -305,6 +305,7 @@ Ranked; detail and the full milestone table in [docs/roadmap.md](docs/roadmap.md - **2026-10-24 (M5, move preview):** A comment being moved is drawn where it goes while it is dragged (a rectangle or oval showed only its blue frame): the comment is rendered alone once as the drag starts and laid over the page, which shows it where it was until the move is applied. Engine `Document::comment_layer`; synthetic engine and UI regressions. Estimates unchanged. - **2026-10-24 (M5, Escape in Fill & Sign):** Escape leaves a Fill & Sign tool as it leaves a comment tool: the first press closes the open type box, the next goes back to Select, so the Text tool no longer opens a box at every click with no way out by the keyboard. A synthetic UI regression covers the Text and Checkmark tools. Estimates unchanged. - **2026-10-23 (M5, ortho):** Shift while drawing holds a line or arrow level or plumb, whichever is nearer (a CAD drawing's ortho mode), and each new side of connected lines, a polygon or a cloud from the corner before; dragging an end or corner with Shift is held the same way (it turned to 45° steps). `measure::snap::ortho`; `comment_add ortho`. Estimates unchanged. +- **2026-10-24 (M3, arrow keys on a wide page):** → / ← scroll a page wider than the window (a plan zoomed in) sideways a line at a time and turn the page at its edge, as ↓ / ↑ do in single-page view; ⌘→ / ⌘← turn it at once, and a page that fits across turns as before (#185). A synthetic viewer regression covers both. Estimates unchanged. - **2026-10-22 (M4, #3916 partial):** Document edits and Undo/Redo cancel an Organize page drag. Pointer-release regressions check exact page order and save/reopen. - **2026-10-22 (M10, #3752 paper):** CUPS jobs ask for the chosen paper in every size. The 13 sizes the Print dialog sent no paper for (A6, A0–A2, ARCH, ANSI C–E) use PWG 5101.1 names, checked against a synthetic PPD with `cupsfilter`, and `doc_print` to a printer sends the paper too. The macOS tray half (Auto Select instead of the PPD's default tray) is a craft-print change that needs a pin bump here. Estimates unchanged. diff --git a/crates/ui-egui/src/canvas.rs b/crates/ui-egui/src/canvas.rs index be2dd97d53..96210fb2fd 100644 --- a/crates/ui-egui/src/canvas.rs +++ b/crates/ui-egui/src/canvas.rs @@ -492,6 +492,10 @@ pub struct DocView { pub forward: Vec, /// Pending navigation: page and fraction down the page to align with the viewport top. pub goto: Option<(usize, f32)>, + /// With `goto`: → / ← turned the page at an edge, so a page or spread wider than the window + /// starts at the edge the reader comes from (`false`: its left, after →; `true`: its right, + /// after ←) instead of keeping the column. + pub(crate) goto_side: Option, /// Pending position within that page from a destination ([`Self::go_to_dest`]): the point /// (fractions of the displayed page; `None` leaves that axis to `goto`) to put `rel` points /// from the viewport's top-left corner. @@ -499,6 +503,12 @@ pub struct DocView { /// Pending keyboard scrolling, in points down (negative: up): ↓ / ↑, and Page Down / /// Page Up where the pages scroll. pub key_scroll: f32, + /// Pending keyboard scrolling sideways, in points right (negative: left): → / ← on a page + /// wider than the window. + pub key_side: f32, + /// As last drawn: whether the view was at its left and at its right edge (both when the + /// pages fit across), so → / ← turn the page there and scroll sideways elsewhere. + side_edges: Option<(bool, bool)>, /// Single-page view, as last drawn: the page shown and whether the view was at its top and /// at its bottom, so ↑ / ↓ turn the page there instead of doing nothing (#273). single_edges: Option<(usize, bool, bool)>, @@ -928,8 +938,11 @@ impl DocView { back: Vec::new(), forward: Vec::new(), goto: None, + goto_side: None, goto_point: None, key_scroll: 0.0, + key_side: 0.0, + side_edges: None, single_edges: None, flash: None, compare_marks: Vec::new(), @@ -1162,6 +1175,7 @@ struct PageEnv<'a> { info: &'a DocInfo, initials: &'a Option, key_scroll: f32, + key_side: f32, middle_gesture: bool, modal: bool, navigation_enabled: bool, @@ -1211,6 +1225,9 @@ struct PageOut { page_action: Option, primary_down: bool, refused: Option, + /// The current page or spread where it is drawn, every page of it: one off screen (the second + /// page of a zoomed spread) has no screen transform, but still decides where the row ends. + row_on_screen: Option, stamp_placed: bool, } @@ -1597,6 +1614,7 @@ fn draw_pages( hand, info, key_scroll, + key_side, middle_gesture, modal, navigation_enabled, @@ -1626,7 +1644,7 @@ fn draw_pages( // ScrollArea ignores wheel input during a widget drag. Defer the wheel contribution // until after the text widgets when a selection owns that drag. let wheel_delta = if selection_wheel { Vec2::ZERO } else { precise_scroll + coarse_scroll }; - let delta = auto_delta + edge_delta - vec2(0.0, key_scroll) + wheel_delta + glide; + let delta = auto_delta + edge_delta - vec2(key_side, key_scroll) + wheel_delta + glide; if delta != Vec2::ZERO && delta.is_finite() { ui.scroll_with_delta_animation(delta, egui::style::ScrollAnimation::none()); } @@ -1779,6 +1797,7 @@ fn draw_pages( page_action: _, primary_down: _, refused, + row_on_screen: _, stamp_placed, } = page_out; let mut page_action: Option = None; @@ -1831,6 +1850,9 @@ fn draw_pages( if view.continuous { view.current = current; } + // The row as it is now: in continuous view the current page may have changed above. + let row_on_screen = + shown_row(view, env.rects).iter().filter_map(|&i| env.rects.get(i)).map(|r| r.translate(origin.to_vec2())).reduce(Rect::union); resp.context_menu(|ui| { let is_crop_over_image = tool == QuickTool::EditText && view.image_selection.clone().is_some_and(|sel| { @@ -2006,6 +2028,7 @@ fn draw_pages( page_action, primary_down: ptr.primary_down, refused, + row_on_screen, stamp_placed, }; (wanted, visible_now) @@ -2468,6 +2491,7 @@ fn page_scroll_area(ui: &egui::Ui, view: &mut DocView, geometry: &ScrollGeometry let point = pos2(r.left() + fx * r.width(), r.top() - y_shift + fy * r.height()); scroll = scroll.scroll_offset(anchored_scroll(point, rel, vec2(content_w, content_h), avail.size())); } else if let Some((page, frac)) = view.goto.take() { + let side = view.goto_side.take(); let page = page.min(rects.len() - 1); let r = rects[page]; // Clamped here: egui clamps only after drawing, so a page that fits was drawn MARGIN - page_gap() @@ -2483,7 +2507,27 @@ fn page_scroll_area(ui: &egui::Ui, view: &mut DocView, geometry: &ScrollGeometry // quick-action bar as the layout does. let (x, w) = (view.last_scroll.x, avail.width()); let (left, right) = (r.left() - SIDE, r.right() + SIDE); - let target = if right - left <= w { if left >= x && right <= x + w { x } else { r.center().x - w / 2.0 } } else { x.clamp(left, right - w) }; + // Turned by → / ← at an edge (`goto_side`): a page or spread wider than the window starts at + // the edge the reader comes from. + let row = shown_row(view, rects).iter().filter_map(|&i| rects.get(i)).copied().reduce(Rect::union).unwrap_or(r); + let (row_left, row_right) = (row.left() - SIDE, row.right() + SIDE); + let target = match side { + Some(from_right) if row_right - row_left > w => { + if from_right { + row_right - w + } else { + row_left + } + } + _ if right - left <= w => { + if left >= x && right <= x + w { + x + } else { + r.center().x - w / 2.0 + } + } + _ => x.clamp(left, right - w), + }; if (target - x).abs() > 0.5 { scroll = scroll.horizontal_scroll_offset(target.clamp(0.0, (content_w - w).max(0.0))); } @@ -2733,6 +2777,7 @@ struct GlideCx { now: f64, may_glide: bool, key_scroll: f32, + key_side: f32, auto_delta: Vec2, zoom_delta: f32, edge_delta: Vec2, @@ -2746,11 +2791,11 @@ struct GlideCx { /// automatic flash uses `goto`, not these deltas, so its programmatic scroll is never mistaken for /// a reader action. Returns this frame's glide. fn step_glide(ui: &egui::Ui, view: &mut DocView, avail: Rect, cx: &GlideCx) -> Vec2 { - let GlideCx { now, may_glide, key_scroll, auto_delta, zoom_delta, edge_delta, wheel_scrolled } = *cx; + let GlideCx { now, may_glide, key_scroll, key_side, auto_delta, zoom_delta, edge_delta, wheel_scrolled } = *cx; let over_page = ui.rect_contains_pointer(avail); ui.input(|i| view.fling.read(&i.events, now, may_glide && over_page)); let interrupted = glide_interrupted(ui); - if !may_glide || interrupted || key_scroll != 0.0 || auto_delta != Vec2::ZERO || zoom_delta != 1.0 { + if !may_glide || interrupted || key_scroll != 0.0 || key_side != 0.0 || auto_delta != Vec2::ZERO || zoom_delta != 1.0 { view.fling.stop(); } let glide = view.fling.step(now); @@ -2832,7 +2877,8 @@ pub fn document_area(app: &mut PdfCraftApp, index: usize, ui: &mut egui::Ui) { }; // Keyboard scrolling applies to this frame's page view only, never later. let key_scroll = std::mem::take(&mut app.views[index].key_scroll); - if key_scroll != 0.0 { + let key_side = std::mem::take(&mut app.views[index].key_side); + if key_scroll != 0.0 || key_side != 0.0 { app.views[index].end_find_following(); } // The Search panel closed: its search moves to the find bar. @@ -2932,7 +2978,7 @@ pub fn document_area(app: &mut PdfCraftApp, index: usize, ui: &mut egui::Ui) { read_wheel(ui, view, avail, single_paging, edge_turned, wheel_enabled, animates); let may_glide = !paging_owns_vertical && !middle_gesture && navigation_enabled && unobstructed && animates; let wheel_scrolled = precise_scroll != Vec2::ZERO || line_base != Vec2::ZERO || coarse_scroll != Vec2::ZERO || has_page; - let motion = GlideCx { now, may_glide, key_scroll, auto_delta, zoom_delta, edge_delta, wheel_scrolled }; + let motion = GlideCx { now, may_glide, key_scroll, key_side, auto_delta, zoom_delta, edge_delta, wheel_scrolled }; let glide = step_glide(ui, view, avail, &motion); // The wheel may have turned the page above: draw the row it turned to. Measured from the old // row, the jump to the new page scrolled both out of view, a blank frame on every turn forward. @@ -3039,6 +3085,7 @@ pub fn document_area(app: &mut PdfCraftApp, index: usize, ui: &mut egui::Ui) { info, initials: &initials, key_scroll, + key_side, middle_gesture, modal, navigation_enabled, @@ -3089,6 +3136,7 @@ pub fn document_area(app: &mut PdfCraftApp, index: usize, ui: &mut egui::Ui) { page_action, primary_down, mut refused, + row_on_screen, stamp_placed, } = page_out; view.single_edges = visible_pages.first().copied().filter(|_| !view.continuous).map(|page| { @@ -3098,6 +3146,15 @@ pub fn document_area(app: &mut PdfCraftApp, index: usize, ui: &mut egui::Ui) { let max_y = (out.content_size.y - out.inner_rect.height()).max(0.0); (page, out.state.clamped_offset().y <= MARGIN - view.page_gap() + 1.0, out.state.clamped_offset().y >= max_y - 1.0) }); + // Within a point, as above: at an edge of the scrolled content, or of the current page or + // spread shown whole that way (in continuous view the content is as wide as the widest page, + // and a narrower current page shows whole inside it). + let max_x = (out.content_size.x - out.inner_rect.width()).max(0.0); + let inner = out.inner_rect; + view.side_edges = Some(( + out.state.clamped_offset().x <= 1.0 || row_on_screen.is_some_and(|r| r.left() >= inner.left() - 1.0), + out.state.clamped_offset().x >= max_x - 1.0 || row_on_screen.is_some_and(|r| r.right() <= inner.right() + 1.0), + )); view.viewport_screen = out.inner_rect; view.auto_scroll.paint(ui, avail); diff --git a/crates/ui-egui/src/canvas/navigation.rs b/crates/ui-egui/src/canvas/navigation.rs index bc98a243fa..1a2a0343a6 100644 --- a/crates/ui-egui/src/canvas/navigation.rs +++ b/crates/ui-egui/src/canvas/navigation.rs @@ -424,14 +424,39 @@ pub fn shortcuts(view: &mut DocView, ctx: &egui::Context, speed: crate::scroll_s } } // As in Acrobat: → / ← go to the next / previous page in every layout (#185), and so do - // ⌘Page Down / ⌘Page Up, or plain Page Down / Page Up in single-page view. + // ⌘Page Down / ⌘Page Up, or plain Page Down / Page Up in single-page view. On a page wider + // than the window (a plan zoomed in) → / ← scroll sideways a line first, and turn the page + // once there is nothing left to scroll that way, as ↓ / ↑ do in single-page view; ⌘→ / ⌘← + // turn it at once. let command = ctx.input(|i| i.modifiers.command); let scrolls = view.continuous; - if key(Key::ArrowRight) || (key(Key::PageDown) && (command || !scrolls)) { + let line = KEY_SCROLL_LINE * speed.factor(); + // Only an up-to-date view counts: no jump pending. + let sideways = |view: &DocView, right: bool| { + !command && view.goto.is_none() && view.side_edges.is_some_and(|(left_edge, right_edge)| !if right { right_edge } else { left_edge }) + }; + // Turned at an edge, the next page starts at its left and the previous one at its right, as + // reading on across: on two wide pages → doesn't turn twice without showing the next one's left. + let at_edge = |view: &mut DocView, right: bool| { + if !command && view.goto.is_some() { + view.goto_side = Some(!right); + } + }; + if key(Key::ArrowRight) && sideways(view, true) { + view.key_side += line; + } else if key(Key::ArrowRight) || (key(Key::PageDown) && (command || !scrolls)) { view.step_page(true); + if key(Key::ArrowRight) { + at_edge(view, true); + } } - if key(Key::ArrowLeft) || (key(Key::PageUp) && (command || !scrolls)) { + if key(Key::ArrowLeft) && sideways(view, false) { + view.key_side -= line; + } else if key(Key::ArrowLeft) || (key(Key::PageUp) && (command || !scrolls)) { view.step_page(false); + if key(Key::ArrowLeft) { + at_edge(view, false); + } } // ↓ / ↑ scroll a line, and Page Down / Page Up a screen where the pages scroll. In // single-page view, ↓ / ↑ turn the page once there is nothing left to scroll that way, as @@ -439,7 +464,6 @@ pub fn shortcuts(view: &mut DocView, ctx: &egui::Context, speed: crate::scroll_s // The Scroll speed preference sets how far ↓ / ↑ go as well as the wheel (#985); Page Down / // Page Up keep a screen. if !command { - let line = KEY_SCROLL_LINE * speed.factor(); let screen = (view.viewport_h - KEY_SCROLL_LINE).max(KEY_SCROLL_LINE); let steps = [(Key::ArrowDown, line), (Key::ArrowUp, -line), (Key::PageDown, screen), (Key::PageUp, -screen)]; for (k, by) in steps { diff --git a/crates/ui-egui/tests/it/viewer.rs b/crates/ui-egui/tests/it/viewer.rs index 6f7d3d3cc8..d383582915 100644 --- a/crates/ui-egui/tests/it/viewer.rs +++ b/crates/ui-egui/tests/it/viewer.rs @@ -1484,6 +1484,138 @@ fn up_and_down_turn_pages_in_single_page_view() { assert_eq!(press(&mut h, Key::ArrowDown, 1), 0, "scrolled up from the bottom of page 1"); } +#[test] +fn left_and_right_scroll_a_page_wider_than_the_window_before_turning_it() { + // A plan zoomed in: → / ← turned the page (#185) and so never showed its other side; the + // view moved sideways only by the scroll bar, the wheel or the Hand. + use egui::{Key, Modifiers}; + let mut h = harness(); + h.state_mut().active = Some(0); + h.state_mut().views[0].set_zoom(12.0); + h.run_steps(6); + let press = |h: &mut Harness<'static, PdfCraftApp>, modifiers, key| { + h.key_press_modifiers(modifiers, key); + h.run_steps(4); + let v = &h.state().views[0]; + (v.current, v.page_screen_rect(v.current).map(|r| r.left())) + }; + let start = h.state().views[0].page_screen_rect(0).map(|r| r.left()).expect("on screen"); + let (page, left) = press(&mut h, Modifiers::NONE, Key::ArrowRight); + assert_eq!(page, 0, "→ scrolls the wide page rather than turning it"); + assert!(left.is_some_and(|l| l < start - 10.0), "the page moved left: {left:?} from {start}"); + // At its right edge the next → turns the page. + let mut n = 1; + while press(&mut h, Modifiers::NONE, Key::ArrowRight).0 == 0 { + n += 1; + assert!(n < 500, "→ never reached the right edge"); + } + assert!(n > 3, "→ scrolled across page 1 before turning ({n} presses)"); + assert_eq!(h.state().views[0].current, 1); + // Page 2 starts at its left edge, as reading on across: → scrolls it rather than turning + // again at once. + let shown = |h: &Harness<'static, PdfCraftApp>, page: usize| { + let v = &h.state().views[0]; + (v.page_screen_rect(page).expect("on screen"), v.viewport_rect()) + }; + let (r, view) = shown(&h, 1); + assert!(r.left() >= view.left() && r.right() > view.right(), "page 2 from its left edge: {r:?} in {view:?}"); + assert_eq!(press(&mut h, Modifiers::NONE, Key::ArrowRight).0, 1, "→ scrolls page 2 from its left edge"); + // ← goes back to that edge and turns back: page 1 comes at its right edge, so the next ← scrolls it. + let mut n = 1; + while press(&mut h, Modifiers::NONE, Key::ArrowLeft).0 == 1 { + n += 1; + assert!(n < 50, "← never turned back from page 2"); + } + assert_eq!(h.state().views[0].current, 0); + let (r, view) = shown(&h, 0); + assert!(r.right() <= view.right() && r.left() < view.left(), "page 1 from its right edge: {r:?} in {view:?}"); + assert_eq!(press(&mut h, Modifiers::NONE, Key::ArrowLeft).0, 0, "← scrolls page 1 from its right edge"); + // ⌘→ turns at once. + assert_eq!(press(&mut h, Modifiers::COMMAND, Key::ArrowRight).0, 1, "⌘→ turns the page at once"); + assert_eq!(press(&mut h, Modifiers::COMMAND, Key::ArrowRight).0, 2); + // A page that fits across: → turns pages as before. + h.state_mut().views[0].fit = pdfcraft_ui_egui::canvas::Fit::Width; + h.run_steps(6); + assert_eq!(press(&mut h, Modifiers::NONE, Key::ArrowRight).0, 3); +} + +/// Continuous view with pages of different widths: the content is as wide as the widest page, but +/// on a narrow page that shows whole → turns the page; on the wide one it scrolls first. +#[test] +fn left_and_right_follow_the_current_pages_width_in_continuous_view() { + use egui::Key; + let mut h = Harness::builder().with_size(egui::vec2(1400.0, 900.0)).build_eframe(|_cc| { + let mut app = PdfCraftApp::new(); + app.set_option("language", "en").unwrap(); + app.open_bytes("mixed.pdf", None, sized_fixture(&[(200, 1200), (3000, 1200), (200, 1200)])).unwrap(); + app + }); + crate::support::settle(&mut h); + h.state_mut().active = Some(0); + assert!(h.state().views[0].continuous); + // Each page taller than the window, so one is the current page. + h.state_mut().views[0].set_zoom(1.0); + h.run_steps(6); + h.state_mut().views[0].go_to_page(0); + h.run_steps(6); + let press = |h: &mut Harness<'static, PdfCraftApp>| { + h.key_press(Key::ArrowRight); + h.run_steps(4); + let v = &h.state().views[0]; + (v.current, v.page_screen_rect(1).map(|r| r.left())) + }; + assert_eq!(h.state().views[0].current, 0); + let (page, start) = press(&mut h); + assert_eq!(page, 1, "→ on the narrow page shown whole turns it"); + let (page, left) = press(&mut h); + assert_eq!(page, 1, "→ on the wide page scrolls it"); + assert!(left.zip(start).is_some_and(|(l, s)| l < s - 10.0), "the wide page moved left: {left:?} from {start:?}"); +} + +/// A zoomed two-page spread scrolled so its first page ends inside the window and the second is +/// still wholly to the right of it: → scrolls on to the second page rather than turning the +/// spread. (Read from the pages on screen alone, the spread seemed to end with its first page.) +#[test] +fn right_scrolls_to_a_spreads_second_page_off_screen_before_turning() { + use egui::Key; + // The page gap (points on screen). + const GAP: f32 = 14.0; + let build = |zoom: f32| { + let mut h = Harness::builder().with_size(egui::vec2(1400.0, 900.0)).build_eframe(move |_cc| { + let mut app = PdfCraftApp::new(); + app.set_option("language", "en").unwrap(); + app.open_bytes("spread.pdf", None, sized_fixture(&[(600, 800); 4])).unwrap(); + app.set_option("layout", "two-page").unwrap(); + // Before the first layout: zoomed about the content's top left, scrolled to it. + app.views[0].set_zoom(zoom); + app + }); + crate::support::settle(&mut h); + h.state_mut().active = Some(0); + h.run_steps(4); + h + }; + let shown = + |h: &Harness<'static, PdfCraftApp>| (h.state().views[0].viewport_rect(), h.state().views[0].page_screen_rect(0).expect("page 1 on screen")); + // Zoom until page 1's right edge sits half a gap inside the window's. + let mut zoom = 2.0; + let mut h = build(zoom); + for _ in 0..3 { + let (inner, first) = shown(&h); + zoom *= (inner.right() - GAP / 2.0 - first.left()) / first.width(); + h = build(zoom); + } + let (inner, first) = shown(&h); + assert!(first.right() <= inner.right() && first.right() + GAP > inner.right(), "the spread's gap at the window's edge: {first:?} in {inner:?}"); + assert!(h.state().views[0].page_screen_rect(1).is_none(), "its second page off screen"); + assert!(h.state().views[0].layout == pdfcraft_ui_egui::canvas::PageLayout::TwoUp && !h.state().views[0].continuous); + h.key_press(Key::ArrowRight); + h.run_steps(4); + let v = &h.state().views[0]; + assert!(v.current < 2, "→ doesn't turn the spread: page {}", v.current + 1); + assert!(v.page_screen_rect(1).is_some(), "it scrolls the second page into view"); +} + #[test] fn v_h_and_space_pick_the_quick_tools() { // The toolbar's tooltips promise "Select (V)" and "Hand (H)", but the keys did nothing. diff --git a/docs/ui-parity.md b/docs/ui-parity.md index dbb38ce730..69072e259c 100644 --- a/docs/ui-parity.md +++ b/docs/ui-parity.md @@ -1,6 +1,6 @@ # UI and interaction parity -> **Last reviewed:** 2026-10-21 · **Last updated:** 2026-10-21 · **Change:** minor (Pages inspector thumbnail size slider and buttons #3808; full screen presents one page at a time #1459; the selected field nudges with the arrow keys in Prepare a form #2608) · **Target:** Adobe Acrobat Pro (Acrobat DC, continuous track 26.002.21931, macOS) +> **Last reviewed:** 2026-10-21 · **Last updated:** 2026-10-24 · **Change:** minor (→ / ← scroll a page wider than the window before turning it) · **Target:** Adobe Acrobat Pro (Acrobat DC, continuous track 26.002.21931, macOS) How PdfCraft's tools, panels, shortcuts and feel compare with Acrobat Pro's current interface (the 2023+ "new Acrobat": All tools pane on the left, quick-action toolbar, Home view). Behaviour is @@ -44,7 +44,7 @@ No dump of Acrobat's menu tree has been diffed against the registry yet (gap 24) | Zoom: wheel, pinch, keys, marquee, fit page/width/height/visible | yes | yes | shipped | | Fixed zoom percentages match Acrobat's physical size | yes | differ (#739, documented in #736) | gap | | Dynamic zoom, loupe, pan-and-zoom window | yes | no | planned | -| Hand tool, middle-button pan, autoscroll | yes | hand tool; Linux autoscroll | partial | +| Hand tool, middle-button pan, autoscroll | yes | hand tool; Linux autoscroll; → / ← scroll a page wider than the window before turning it | partial | | Text selection, column select | yes, Alt-drag columns | reading-order selection; Alt/Option-drag or Edit ▸ Column select (#740) | shipped | | Select and move several objects at once | yes | on one page: Shift/Ctrl/⌘-click and marquee, one undo step (#844); not across pages, no group resize | partial | | Moved comments drawn where they go while dragged | yes | yes: the comment's own pixels follow the pointer, the page shows it where it was until release; stamps and Fill & Sign marks also hide the original | shipped | @@ -83,6 +83,7 @@ No dump of Acrobat's menu tree has been diffed against the registry yet (gap 24) | Date | Change | Summary | |---|---|---| +| 2026-10-24 | minor | → / ← scroll a page wider than the window sideways before turning it, by the current page's or spread's edges in continuous view; ⌘→ / ⌘← turn at once | | 2026-10-21 | minor | Pages inspector thumbnail sizing (#3808), separate from page and Organize zoom; view setting retained on reload and exposed through UI control. Not persisted between sessions. | | 2026-10-15 | minor | A selected field nudges with the arrow keys in Prepare a form: a point, Shift ten points, one undo step per press (#2608) | | 2026-10-14 | minor | Full screen shows one fitted page (or spread) at a time on black with click and key paging, as Acrobat's Full Screen Mode (#379, #1459, #1353) | diff --git a/parity/acrobat-features.toml b/parity/acrobat-features.toml index d4a8c37392..1a3edc4c36 100644 --- a/parity/acrobat-features.toml +++ b/parity/acrobat-features.toml @@ -1154,8 +1154,8 @@ milestone = "M3" status = "shipped" commands = [] tools = [] -evidence = ["crates/ui-egui/tests/it/control.rs::state_and_view_options", "crates/ui-egui/tests/it/ui.rs::theme_and_view_options_apply", "crates/ui-egui/tests/it/viewer.rs::two_page_view_steps_a_spread_at_a_time", "crates/ui-egui/tests/it/viewer.rs::arrow_and_page_keys_move_through_a_scrolling_document"] -notes = "in two-page view, next and previous page move a whole spread (with or without a cover page); → / ← turn pages in every layout, ↓ / ↑ scroll a line and Page Down / Page Up a screen (the page in single-page view), as in Acrobat; in single-page view ↓ / ↑ turn the page once there is nothing left to scroll that way, like the wheel" +evidence = ["crates/ui-egui/tests/it/control.rs::state_and_view_options", "crates/ui-egui/tests/it/ui.rs::theme_and_view_options_apply", "crates/ui-egui/tests/it/viewer.rs::two_page_view_steps_a_spread_at_a_time", "crates/ui-egui/tests/it/viewer.rs::arrow_and_page_keys_move_through_a_scrolling_document", "crates/ui-egui/tests/it/viewer.rs::left_and_right_scroll_a_page_wider_than_the_window_before_turning_it"] +notes = "in two-page view, next and previous page move a whole spread (with or without a cover page); → / ← turn pages in every layout (on a page wider than the window, a plan zoomed in, they scroll sideways a line first and turn the page at its edge; ⌘→ / ⌘← turn at once), ↓ / ↑ scroll a line and Page Down / Page Up a screen (the page in single-page view), as in Acrobat; in single-page view ↓ / ↑ turn the page once there is nothing left to scroll that way, like the wheel" [[feature]] id = "view.goto-page-label"