From 79cfd9b151c7f36793d230fac10857b02cfbe33d Mon Sep 17 00:00:00 2001 From: vangel76 Date: Tue, 13 Oct 2026 21:13:20 +0200 Subject: [PATCH 01/24] Charts: label fitting and legend measuring avoid hundreds of text layouts per render (#1149) Binary search in Ctx::fit, early exit of the side-legend column measure, no per-point Vecs in value_range/squarify, treemap carries the value. Co-Authored-By: Claude Sonnet 5.5 --- ROADMAP.md | 1 + crates/chart/src/render.rs | 90 ++++++--- crates/chart/src/render_perf_tests.rs | 254 ++++++++++++++++++++++++++ 3 files changed, 317 insertions(+), 28 deletions(-) create mode 100644 crates/chart/src/render_perf_tests.rs diff --git a/ROADMAP.md b/ROADMAP.md index 070552d07..ae8286ad3 100644 --- a/ROADMAP.md +++ b/ROADMAP.md @@ -132,6 +132,7 @@ Ranked; estimates in Opus 5.5 agent-hours. Detail and the full beta-gate list in | Date | What landed | |---|---| +| 2026-10-13 | Charts: label shortening needs about nine text measurements instead of up to 400, side legends stop measuring once the column is at its cap and treemap layout avoids per-item allocation (#1149) | | 2026-10-12 | Rows auto-fit oversized fonts: opening a file and whole-row/column formatting grow non-custom rows to their tallest font (#833) | | 2026-10-12 | Number formats: parsing a format code is linear in its length (long runs of `[`, `,` or `/` no longer take quadratic time), and an unclosed `[` is a literal when splitting sections as well as when tokenizing (#810) | | 2026-10-12 | Number formats: a fraction denominator with 65,536 zeros no longer crashes; format codes are capped at 255 characters like Excel (longer codes display as General, `TEXT` gives `#VALUE!`) (#807) | diff --git a/crates/chart/src/render.rs b/crates/chart/src/render.rs index 037b9099e..396423d2b 100644 --- a/crates/chart/src/render.rs +++ b/crates/chart/src/render.rs @@ -62,25 +62,47 @@ struct Ctx<'a> { m: &'a dyn Measure, } +#[cfg(test)] +#[path = "render_perf_tests.rs"] +mod perf_tests; + impl Ctx<'_> { fn tw(&self, s: &str, size: f32) -> f32 { let w = self.m.text_width(s, size, false); if w.is_finite() { w.max(0.0) } else { 0.0 } } + /// `s` shortened with an ellipsis to fit `maxw`; empty when not even one character fits. + /// + /// Binary search over the number of kept characters (at most 400 are tried), about nine + /// `Measure` calls instead of one per character. It assumes the width of `prefix + "…"` does not + /// shrink as the prefix grows, which holds for per-character advance sums (`ApproxMeasure`, the + /// PDF metrics) and for the UI's text layout up to kerning of a few hundredths of a point; with + /// such a non-monotone measure the result may differ from the old linear scan by a character. fn fit(&self, s: &str, size: f32, maxw: f32) -> String { if self.tw(s, size) <= maxw { return s.to_string(); } - let chars: Vec = s.chars().collect(); - let mut n = chars.len().min(400); - while n > 0 { - n -= 1; - let t: String = chars.iter().take(n).collect::() + "…"; - if self.tw(&t, size) <= maxw { - return if n == 0 { String::new() } else { t }; + // Byte offset where each of the first 400 characters starts; a prefix of `k` characters + // (k < number of characters, so the offset exists) ends at `starts[k]`. + let starts: Vec = s.char_indices().map(|(i, _)| i).take(400).collect(); + let top = starts.len().saturating_sub(1); + let fits = |k: usize| -> bool { + let end = starts.get(k).copied().unwrap_or(s.len()); + let mut t = String::with_capacity(end + 3); + t.push_str(s.get(..end).unwrap_or("")); + t.push('…'); + self.tw(&t, size) <= maxw + }; + let (mut lo, mut hi) = (0usize, top); + while lo < hi { + let mid = lo + (hi - lo).div_ceil(2); + if fits(mid) { + lo = mid; + } else { + hi = mid - 1; } } - String::new() + if lo == 0 { String::new() } else { format!("{}…", s.get(..starts.get(lo).copied().unwrap_or(s.len())).unwrap_or("")) } } #[allow(clippy::too_many_arguments)] fn text(&mut self, x: f32, y: f32, s: &str, size: f32, color: Rgba, align: HAlign, valign: VAlign) { @@ -369,7 +391,16 @@ fn draw_legend(ctx: &mut Ctx, entries: &[(String, Rgba, Swatch)], pos: LegendPos if pos == LegendPos::Bottom { Bx { h: area.h - used, ..area } } else { Bx { y: area.y + used, h: area.h - used, ..area } } } LegendPos::Left | LegendPos::Right => { - let colw = entries.iter().map(|e| item_w(ctx, e)).fold(0.0f32, f32::max).min(area.w * 0.35); + // Widest entry, capped at 35% of the area: once the cap is reached the rest cannot change it. + let cap = area.w * 0.35; + let mut colw = 0.0f32; + for e in entries { + if colw >= cap { + break; + } + colw = colw.max(item_w(ctx, e)); + } + let colw = colw.min(cap); let max_rows = ((area.h / row_h).floor() as usize).max(1); let n = entries.len().min(max_rows); let lh = n as f32 * row_h; @@ -911,7 +942,8 @@ fn cartesian(ctx: &mut Ctx, chart: &Chart, data: &ChartData, area: Bx, horizonta let all100 = members.iter().all(|i| fams.get(*i) == Some(&Fam::Col100)); let range = value_range(members.iter().flat_map(|i| { let line = fams.get(*i).is_some_and(|f| is_line(*f)); - spans.get(*i).into_iter().flatten().flatten().flat_map(move |(b, t)| if line { vec![*t] } else { vec![*b, *t] }) + // Lines contribute only their top value, bars both ends (no per-point `Vec`). + spans.get(*i).into_iter().flatten().flatten().flat_map(move |(b, t)| [*b, *t].into_iter().skip(usize::from(line))) })); let (lo, hi) = range.unwrap_or((0.0, 0.0)); let (scale, fmt) = if all100 { @@ -1399,19 +1431,19 @@ fn pie(ctx: &mut Ctx, chart: &Chart, data: &ChartData, area: Bx, inner: f32, nam // ---------------------------------------------------------------- treemap -fn worst(row: &[f64], side: f64) -> f64 { - let s: f64 = row.iter().sum(); +/// Worst aspect ratio of a row with the given area sum, smallest and largest item, laid along `side`. +fn worst(s: f64, mn: f64, mx: f64, side: f64) -> f64 { if s <= 0.0 || side <= 0.0 { return f64::INFINITY; } - let (mn, mx) = row.iter().fold((f64::INFINITY, 0.0f64), |(a, b), v| (a.min(*v), b.max(*v))); let s2 = s * s; let w2 = side * side; (w2 * mx / s2).max(s2 / (w2 * mn)) } -/// Squarified treemap layout of (index, area) items, sorted descending, into `r`. -fn squarify(items: &[(usize, f64)], r: Bx) -> Vec<(usize, Bx)> { +/// Squarified treemap layout of (index, value) items, sorted descending, into `r`; each block +/// comes back with its index and value. +fn squarify(items: &[(usize, f64)], r: Bx) -> Vec<(usize, f64, Bx)> { let total: f64 = items.iter().map(|i| i.1).sum(); let area = (r.w as f64) * (r.h as f64); if !(total > 0.0 && area > 0.0) { @@ -1420,9 +1452,10 @@ fn squarify(items: &[(usize, f64)], r: Bx) -> Vec<(usize, Bx)> { let scale = area / total; let mut out = Vec::with_capacity(items.len()); let mut rem = r; - let mut row: Vec<(usize, f64)> = Vec::new(); + let mut row: Vec<(usize, f64, f64)> = Vec::new(); + let (mut sum, mut mn, mut mx) = (0.0f64, f64::INFINITY, 0.0f64); let mut i = 0; - let layout = |row: &[(usize, f64)], rem: &mut Bx, out: &mut Vec<(usize, Bx)>| { + let layout = |row: &[(usize, f64, f64)], rem: &mut Bx, out: &mut Vec<(usize, f64, Bx)>| { let s: f64 = row.iter().map(|x| x.1).sum(); if s <= 0.0 { return; @@ -1430,9 +1463,9 @@ fn squarify(items: &[(usize, f64)], r: Bx) -> Vec<(usize, Bx)> { if rem.w >= rem.h { let cw = (s / rem.h.max(1e-6) as f64) as f32; let mut y = rem.y; - for (j, a) in row { + for (j, a, v) in row { let h = (*a / cw.max(1e-6) as f64) as f32; - out.push((*j, Bx { x: rem.x, y, w: cw, h })); + out.push((*j, *v, Bx { x: rem.x, y, w: cw, h })); y += h; } rem.x += cw; @@ -1440,9 +1473,9 @@ fn squarify(items: &[(usize, f64)], r: Bx) -> Vec<(usize, Bx)> { } else { let rh = (s / rem.w.max(1e-6) as f64) as f32; let mut x = rem.x; - for (j, a) in row { + for (j, a, v) in row { let w = (*a / rh.max(1e-6) as f64) as f32; - out.push((*j, Bx { x, y: rem.y, w, h: rh })); + out.push((*j, *v, Bx { x, y: rem.y, w, h: rh })); x += w; } rem.y += rh; @@ -1452,15 +1485,17 @@ fn squarify(items: &[(usize, f64)], r: Bx) -> Vec<(usize, Bx)> { while let Some(&(j, v)) = items.get(i) { let a = v * scale; let side = rem.w.min(rem.h) as f64; - let cur: Vec = row.iter().map(|x| x.1).collect(); - let mut with = cur.clone(); - with.push(a); - if row.is_empty() || worst(&with, side) <= worst(&cur, side) { - row.push((j, a)); + // Sum, smallest and largest item of the open row, kept incrementally (same fold order as + // summing the row afresh, so the numbers are bit-identical). + let with = (sum + a, mn.min(a), mx.max(a)); + if row.is_empty() || worst(with.0, with.1, with.2, side) <= worst(sum, mn, mx, side) { + row.push((j, a, v)); + (sum, mn, mx) = with; i += 1; } else { layout(&row, &mut rem, &mut out); row.clear(); + (sum, mn, mx) = (0.0, f64::INFINITY, 0.0); } } if !row.is_empty() { @@ -1475,14 +1510,13 @@ fn treemap(ctx: &mut Ctx, chart: &Chart, data: &ChartData, area: Bx) { s.values.iter().take(MAX_MARKERS).enumerate().filter_map(|(j, v)| v.filter(|v| *v > 0.0).map(|v| (j, v))).collect(); items.sort_by(|a, b| b.1.total_cmp(&a.1)); let f = Fmt::new(&s.number_format); - for (j, b) in squarify(&items, area) { + for (j, v, b) in squarify(&items, area) { let fill = point_color(data, j); ctx.rect(b.x, b.y, b.w, b.h, fill, Some((WHITE, 1.5))); if b.w > 24.0 && b.h > LABEL_SIZE * 1.6 { let name = ctx.fit(&cat_label(data, j), LABEL_SIZE, b.w - 8.0); ctx.text(b.x + 4.0, b.y + 4.0, &name, LABEL_SIZE, on(fill), HAlign::Left, VAlign::Top); if chart.data_labels && b.h > LABEL_SIZE * 3.2 { - let v = items.iter().find(|x| x.0 == j).map(|x| x.1).unwrap_or(0.0); let t = ctx.fit(&f.format(v, None), LABEL_SIZE, b.w - 8.0); ctx.text(b.x + 4.0, b.y + 4.0 + LABEL_SIZE * 1.4, &t, LABEL_SIZE, on(fill), HAlign::Left, VAlign::Top); } diff --git a/crates/chart/src/render_perf_tests.rs b/crates/chart/src/render_perf_tests.rs new file mode 100644 index 000000000..7bef70722 --- /dev/null +++ b/crates/chart/src/render_perf_tests.rs @@ -0,0 +1,254 @@ +//! Equality and counter tests for the label-fitting, legend-measuring and treemap rewrites. +//! The old implementations live here as references. + +use std::cell::Cell; + +use super::*; +use crate::{ApproxMeasure, Measure}; + +/// The pre-rewrite `Ctx::fit`: one `Measure` call per candidate length, longest first. +fn fit_reference(ctx: &Ctx, s: &str, size: f32, maxw: f32) -> String { + if ctx.tw(s, size) <= maxw { + return s.to_string(); + } + let chars: Vec = s.chars().collect(); + let mut n = chars.len().min(400); + while n > 0 { + n -= 1; + let t: String = chars.iter().take(n).collect::() + "…"; + if ctx.tw(&t, size) <= maxw { + return if n == 0 { String::new() } else { t }; + } + } + String::new() +} + +/// Counts calls; advances differ per character (wide CJK/emoji, zero-width combining marks) and +/// "AV"-style pairs are kerned tighter, but a character never makes the text narrower. +struct Varied { + calls: Cell, +} + +impl Measure for Varied { + fn text_width(&self, text: &str, size: f32, bold: bool) -> f32 { + self.calls.set(self.calls.get() + 1); + let mut w = 0.0f32; + let mut prev = ' '; + for c in text.chars() { + let adv = match c as u32 { + 0x300..=0x36f => 0.0, + 0x1f300..=0x1faff | 0x4e00..=0x9fff => 1.0, + u => 0.3 + (u % 7) as f32 * 0.07, + }; + let kern = if prev == 'A' && c == 'V' { 0.1 } else { 0.0 }; + w += (adv - kern) * size; + prev = c; + } + if bold { w * 1.1 } else { w } + } +} + +struct Counting<'a>(&'a dyn Measure, Cell); + +impl Measure for Counting<'_> { + fn text_width(&self, text: &str, size: f32, bold: bool) -> f32 { + self.1.set(self.1.get() + 1); + self.0.text_width(text, size, bold) + } +} + +fn corpus() -> Vec { + let mut v: Vec = + ["", "a", "ab", "Hello, world", "AVAVAV tight", "日本語のラベルです", "👨‍👩‍👧 family 🎉 party", "e\u{301}e\u{301}e\u{301} combining", "…", "……"] + .iter() + .map(|s| s.to_string()) + .collect(); + v.push("x".repeat(399)); + v.push("x".repeat(400)); + v.push("x".repeat(401)); + v.push("é🎉".repeat(500)); + v.push("long label ".repeat(100)); + v +} + +fn run_fit(m: &dyn Measure) { + let ctx = Ctx { out: vec![], m }; + for s in corpus() { + for size in [9.0f32, 11.0, 40.0] { + for maxw in [-5.0, 0.0, 0.4, 6.0, 25.0, 61.5, 120.0, 800.0, 1e6, f32::INFINITY, f32::NAN] { + assert_eq!(ctx.fit(&s, size, maxw), fit_reference(&ctx, &s, size, maxw), "{s:?} size {size} maxw {maxw}"); + } + } + } +} + +#[test] +fn fit_matches_linear_reference() { + run_fit(&ApproxMeasure); + run_fit(&Varied { calls: Cell::new(0) }); +} + +#[test] +fn fit_measures_logarithmically() { + let long = "x".repeat(1000); + let c = Counting(&ApproxMeasure, Cell::new(0)); + let ctx = Ctx { out: vec![], m: &c }; + let r = ctx.fit(&long, 11.0, 60.0); + assert!(r.ends_with('…') && r.chars().count() > 3); + // 1 whole-string check + ceil(log2(399)) + 1 probes; the old scan needed ~390 here. + assert!(c.1.get() <= 12, "{} Measure calls", c.1.get()); + let c2 = Counting(&ApproxMeasure, Cell::new(0)); + let ctx2 = Ctx { out: vec![], m: &c2 }; + assert_eq!(ctx2.fit("short", 11.0, 500.0), "short"); + assert_eq!(c2.1.get(), 1); +} + +fn legend_entries(n: usize, label: &dyn Fn(usize) -> String) -> Vec<(String, Rgba, Swatch)> { + (0..n).map(|i| (label(i), [10, 20, 30, 255], if i % 3 == 0 { Swatch::Line } else { Swatch::Box })).collect() +} + +/// Colour column width as the old code computed it, from the returned remaining area. +fn old_colw(m: &dyn Measure, entries: &[(String, Rgba, Swatch)], area: Bx) -> f32 { + let ctx = Ctx { out: vec![], m }; + let size = LABEL_SIZE; + let s = size * 0.8; + entries.iter().map(|e| swatch_w(e.2, s) + 4.0 + ctx.tw(&e.0, size)).fold(0.0f32, f32::max).min(area.w * 0.35) +} + +#[test] +fn side_legend_matches_old_column_width_and_stops_measuring_at_the_cap() { + let area = Bx { x: 0.0, y: 0.0, w: 400.0, h: 300.0 }; + let labels: [&dyn Fn(usize) -> String; 3] = + [&|i| format!("entry {i}"), &|i| if i == 7 { "w".repeat(80) } else { format!("e{i}") }, &|i| "very long legend label ".repeat(1 + i % 3)]; + for pos in [LegendPos::Left, LegendPos::Right] { + for (k, label) in labels.iter().enumerate() { + for n in [0usize, 1, 5, 40] { + let entries = legend_entries(n, label); + let mut ctx = Ctx { out: vec![], m: &ApproxMeasure }; + let rest = draw_legend(&mut ctx, &entries, pos, area); + let colw = old_colw(&ApproxMeasure, &entries, area); + assert!((rest.w - (area.w - (colw + 8.0))).abs() < 1e-4, "{pos:?} labels {k} n {n}"); + } + } + } + // 20,000 entries whose first one already fills the 35% cap: the old code measured all of them + // plus the drawn rows, now only the first entry and the drawn rows are measured. + let entries = legend_entries(20_000, &|i| if i == 0 { "w".repeat(200) } else { format!("e{i}") }); + let c = Counting(&ApproxMeasure, Cell::new(0)); + let mut ctx = Ctx { out: vec![], m: &c }; + let _ = draw_legend(&mut ctx, &entries, LegendPos::Right, area); + let drawn = ctx.out.iter().filter(|p| matches!(p, Prim::Text { .. })).count(); + assert!(c.1.get() <= drawn * 10 + 5, "{} Measure calls for {drawn} drawn entries", c.1.get()); + assert!(c.1.get() < 500); +} + +/// The pre-rewrite `worst`/`squarify` (a `Vec` per item, row recomputed from scratch). +fn worst_reference(row: &[f64], side: f64) -> f64 { + let s: f64 = row.iter().sum(); + if s <= 0.0 || side <= 0.0 { + return f64::INFINITY; + } + let (mn, mx) = row.iter().fold((f64::INFINITY, 0.0f64), |(a, b), v| (a.min(*v), b.max(*v))); + let s2 = s * s; + let w2 = side * side; + (w2 * mx / s2).max(s2 / (w2 * mn)) +} + +fn squarify_reference(items: &[(usize, f64)], r: Bx) -> Vec<(usize, Bx)> { + let total: f64 = items.iter().map(|i| i.1).sum(); + let area = (r.w as f64) * (r.h as f64); + if !(total > 0.0 && area > 0.0) { + return vec![]; + } + let scale = area / total; + let mut out = Vec::with_capacity(items.len()); + let mut rem = r; + let mut row: Vec<(usize, f64)> = Vec::new(); + let mut i = 0; + let layout = |row: &[(usize, f64)], rem: &mut Bx, out: &mut Vec<(usize, Bx)>| { + let s: f64 = row.iter().map(|x| x.1).sum(); + if s <= 0.0 { + return; + } + if rem.w >= rem.h { + let cw = (s / rem.h.max(1e-6) as f64) as f32; + let mut y = rem.y; + for (j, a) in row { + let h = (*a / cw.max(1e-6) as f64) as f32; + out.push((*j, Bx { x: rem.x, y, w: cw, h })); + y += h; + } + rem.x += cw; + rem.w = (rem.w - cw).max(0.0); + } else { + let rh = (s / rem.w.max(1e-6) as f64) as f32; + let mut x = rem.x; + for (j, a) in row { + let w = (*a / rh.max(1e-6) as f64) as f32; + out.push((*j, Bx { x, y: rem.y, w, h: rh })); + x += w; + } + rem.y += rh; + rem.h = (rem.h - rh).max(0.0); + } + }; + while let Some(&(j, v)) = items.get(i) { + let a = v * scale; + let side = rem.w.min(rem.h) as f64; + let cur: Vec = row.iter().map(|x| x.1).collect(); + let mut with = cur.clone(); + with.push(a); + if row.is_empty() || worst_reference(&with, side) <= worst_reference(&cur, side) { + row.push((j, a)); + i += 1; + } else { + layout(&row, &mut rem, &mut out); + row.clear(); + } + } + if !row.is_empty() { + layout(&row, &mut rem, &mut out); + } + out +} + +#[test] +fn squarify_matches_reference() { + let mut seed = 0x2545_f491_4f6c_dd1du64; + let mut rnd = move || { + seed ^= seed << 13; + seed ^= seed >> 7; + seed ^= seed << 17; + (seed >> 11) as f64 / (1u64 << 53) as f64 + }; + let mut cases: Vec> = vec![ + vec![], + vec![5.0], + vec![6.0, 6.0, 4.0, 3.0, 2.0, 2.0, 1.0], + vec![3.0, 0.0, 0.0, 1.0], + vec![0.0, 0.0], + vec![4.0, -1.0, 2.0], + vec![4.0, f64::NAN, 2.0, 1.0], + vec![f64::INFINITY, 1.0], + vec![1e-300, 1e-310, 0.0], + ]; + for n in [2usize, 9, 40, 300] { + let mut v: Vec = (0..n).map(|_| rnd() * 100.0).collect(); + v.sort_by(|a, b| b.total_cmp(a)); + cases.push(v); + } + for r in [Bx { x: 0.0, y: 0.0, w: 300.0, h: 200.0 }, Bx { x: 5.0, y: 7.0, w: 90.0, h: 400.0 }, Bx { x: 0.0, y: 0.0, w: 0.0, h: 50.0 }] { + for vals in &cases { + let items: Vec<(usize, f64)> = vals.iter().copied().enumerate().collect(); + let new: Vec<(usize, Bx)> = squarify(&items, r) + .into_iter() + .map(|(j, v, b)| { + assert!(vals.get(j).is_some_and(|x| x.to_bits() == v.to_bits()), "value carried through"); + (j, b) + }) + .collect(); + let old = squarify_reference(&items, r); + assert_eq!(format!("{new:?}"), format!("{old:?}"), "{vals:?} in {r:?}"); + } + } +} From 4a1e6de3034f0e953c679c1b2fa906830ee8b5fe Mon Sep 17 00:00:00 2001 From: vangel76 Date: Tue, 13 Oct 2026 21:21:24 +0200 Subject: [PATCH 02/24] Charts: work and memory are bounded by series x points; ranges are clipped before evaluation (#1147) A chart keeps at most 2,000,000 series x points (every series and the categories keep their first points), 16 trendlines per chart; columns and bars narrower than 0.25 px are merged per pixel column; series references are cut after the rows that can matter before calc evaluates them; the moving average is linear; fit_trendline allocates no Vec per point; render borrows clean data instead of cloning it. Co-Authored-By: Claude Sonnet 5.5 --- ROADMAP.md | 1 + crates/chart/src/render.rs | 105 +++++++++++++++- crates/chart/src/resolve.rs | 138 ++++++++++++++++++--- crates/chart/src/tests.rs | 2 + crates/chart/src/tests/budget.rs | 172 ++++++++++++++++++++++++++ crates/chart/src/trend.rs | 205 ++++++++++++++++++++++++++++++- 6 files changed, 602 insertions(+), 21 deletions(-) create mode 100644 crates/chart/src/tests/budget.rs diff --git a/ROADMAP.md b/ROADMAP.md index 070552d07..06b50e9cf 100644 --- a/ROADMAP.md +++ b/ROADMAP.md @@ -132,6 +132,7 @@ Ranked; estimates in Opus 5.5 agent-hours. Detail and the full beta-gate list in | Date | What landed | |---|---| +| 2026-10-13 | Charts: work and memory are bounded by series x points (2,000,000 per chart: every series keeps its first points; 16 trendlines per chart), hairline columns and bars are merged per pixel column, series ranges are clipped before they are evaluated, the moving average is linear and rendering clean data no longer copies it (#1147) | | 2026-10-12 | Rows auto-fit oversized fonts: opening a file and whole-row/column formatting grow non-custom rows to their tallest font (#833) | | 2026-10-12 | Number formats: parsing a format code is linear in its length (long runs of `[`, `,` or `/` no longer take quadratic time), and an unclosed `[` is a literal when splitting sections as well as when tokenizing (#810) | | 2026-10-12 | Number formats: a fraction denominator with 65,536 zeros no longer crashes; format codes are capped at 255 characters like Excel (longer codes display as General, `TEXT` gives `#VALUE!`) (#807) | diff --git a/crates/chart/src/render.rs b/crates/chart/src/render.rs index 037b9099e..8df7c2800 100644 --- a/crates/chart/src/render.rs +++ b/crates/chart/src/render.rs @@ -1,12 +1,13 @@ //! Chart layout: title, legend, axes and every chart kind, drawn into [`Prim`]s. +use std::borrow::Cow; use std::collections::HashSet; use std::f32::consts::{PI, TAU}; use gridcraft_model::{Chart, ChartKind, LegendPos, LineDash, Theme, TrendlineKind}; use crate::axis::{Fmt, Scale, clean, fixed_scale, nice_scale}; -use crate::resolve::{MAX_POINTS, MAX_SERIES, MAX_TRENDLINES, series_color}; +use crate::resolve::{MAX_POINTS, MAX_SERIES, MAX_TRENDLINES, points_per_series, series_color}; use crate::trend::{auto_name, fit_trendline, moving_average}; use crate::{ChartData, HAlign, Measure, Prim, Rgba, SeriesData, VAlign}; @@ -30,6 +31,8 @@ const MAX_LINE_PTS: usize = 2000; const MAX_MARKERS: usize = 5000; /// Data labels are skipped when a series has more points than this. const MAX_LABELS: usize = 200; +/// Columns and bars narrower than this (in pixels) are merged per pixel column into one rectangle. +const MIN_BAR_PX: f32 = 0.25; #[derive(Clone, Copy, Debug)] struct Bx { @@ -188,7 +191,7 @@ fn sanitize_vals(v: &[Option]) -> Vec> { v.iter().take(MAX_POINTS).map(|x| x.filter(|n| n.is_finite()).map(clean)).collect() } -fn sanitize(d: &ChartData) -> ChartData { +pub(crate) fn sanitize(d: &ChartData) -> ChartData { ChartData { palette: d.palette.iter().take(MAX_POINTS).copied().collect(), categories: d.categories.iter().take(MAX_POINTS).cloned().collect(), @@ -212,6 +215,54 @@ fn sanitize(d: &ChartData) -> ChartData { } } +/// Whether `v` survives `sanitize_vals` unchanged. +fn is_clean(v: &Option) -> bool { + v.is_none_or(|n| n.is_finite() && clean(n) == n) +} + +/// Whether [`sanitize`] and [`apply_budget`] would change `d`. +fn needs_sanitizing(d: &ChartData) -> bool { + let cap = points_per_series(d.series.len()); + let long = |n: usize| n > cap; + let mut trendlines = 0usize; + d.palette.len() > MAX_POINTS + || long(d.categories.len()) + || d.series.len() > MAX_SERIES + || d.series.iter().any(|s| { + trendlines += s.trendlines.len(); + long(s.values.len()) + || !s.values.iter().all(is_clean) + || s.x.as_ref().is_some_and(|x| long(x.len()) || !x.iter().all(is_clean)) + || s.sizes.as_ref().is_some_and(|x| long(x.len()) || !x.iter().all(is_clean)) + || s.trendlines.len() > MAX_TRENDLINES + }) + || trendlines > MAX_TRENDLINES +} + +/// Enforces the chart's budget of points (`MAX_TOTAL_POINTS` = series x points: every series and +/// the category list keep their first `MAX_TOTAL_POINTS / series` points) and of trendlines +/// (`MAX_TRENDLINES` per chart, in series order). +fn apply_budget(mut d: ChartData) -> ChartData { + let cap = points_per_series(d.series.len()); + d.categories.truncate(cap); + let mut trendlines = MAX_TRENDLINES; + for s in &mut d.series { + s.values.truncate(cap); + for v in [&mut s.x, &mut s.sizes].into_iter().flatten() { + v.truncate(cap); + } + s.trendlines.truncate(trendlines); + trendlines = trendlines.saturating_sub(s.trendlines.len()); + } + d +} + +/// The data `render` draws: `d` itself when it is already clean and within budget (the usual +/// case: no copy), else a cleaned, truncated copy. +pub(crate) fn sanitized(d: &ChartData) -> Cow<'_, ChartData> { + if needs_sanitizing(d) { Cow::Owned(apply_budget(sanitize(d))) } else { Cow::Borrowed(d) } +} + fn chart_title(chart: &Chart, data: &ChartData) -> Option { match &chart.title { Some(t) => Some(t.trim().to_string()).filter(|t| !t.is_empty()), @@ -225,7 +276,7 @@ pub fn render(chart: &Chart, data: &ChartData, w: f32, h: f32, m: &dyn Measure) if !(w.is_finite() && h.is_finite()) || w < 1.0 || h < 1.0 || w > 1e7 || h > 1e7 { return vec![]; } - let data = sanitize(data); + let data = sanitized(data); let mut ctx = Ctx { out: Vec::new(), m }; ctx.out.push(Prim::Rect { x: 0.0, y: 0.0, w, h, fill: Some(WHITE), stroke: Some((GRID, 1.0)), radius: 0.0 }); let pad = (w.min(h) * 0.04).clamp(2.0, 10.0); @@ -828,6 +879,36 @@ fn is_line(f: Fam) -> bool { matches!(f, Fam::Line | Fam::LineStack) } +/// Bars that fall into one pixel column, merged: extent along the category axis (`a0..=a1`, the +/// bars' start positions) and along the value axis (`lo..hi`). +struct BarRun { + bucket: f32, + a0: f32, + a1: f32, + lo: f32, + hi: f32, +} + +impl BarRun { + fn new(pos: f32, lo: f32, hi: f32) -> Self { + BarRun { bucket: pos.floor(), a0: pos, a1: pos, lo, hi } + } + fn add(&mut self, pos: f32, lo: f32, hi: f32) { + self.a0 = self.a0.min(pos); + self.a1 = self.a1.max(pos); + self.lo = self.lo.min(lo); + self.hi = self.hi.max(hi); + } + fn draw(&self, ctx: &mut Ctx, bw: f32, horizontal: bool, color: Rgba) { + let along = self.a1 - self.a0 + bw; + if horizontal { + ctx.rect(self.lo, self.a0, self.hi - self.lo, along, color, None); + } else { + ctx.rect(self.a0, self.lo, along, self.hi - self.lo, color, None); + } + } +} + fn cartesian(ctx: &mut Ctx, chart: &Chart, data: &ChartData, area: Bx, horizontal: bool) { let ncat = data.categories.len().max(data.series.iter().map(|s| s.values.len()).max().unwrap_or(0)).min(MAX_CATS); if ncat == 0 { @@ -1026,12 +1107,27 @@ fn cartesian(ctx: &mut Ctx, chart: &Chart, data: &ChartData, area: Bx, horizonta } 1 => { let k = slot_of.get(i).copied().unwrap_or(0) as f32; + // Hairline bars are merged per pixel column (their envelope is what shows). + let cull = bw < MIN_BAR_PX && !labels_on; + let mut run: Option = None; for (j, sp) in row.iter().enumerate() { let Some((b, t)) = sp else { continue }; let (pb, pt) = (vpos(*b, &sc), vpos(*t, &sc)); let c = center(j); let lo = pb.min(pt); let len = (pb - pt).abs(); + if cull { + let pos = if horizontal { c + nslots * bw / 2.0 - (k + 1.0) * bw } else { c - nslots * bw / 2.0 + k * bw }; + match &mut run { + Some(m) if m.bucket == pos.floor() => m.add(pos, lo, lo + len), + _ => { + if let Some(m) = run.replace(BarRun::new(pos, lo, lo + len)) { + m.draw(ctx, bw, horizontal, s.color); + } + } + } + continue; + } if horizontal { let y = c + nslots * bw / 2.0 - (k + 1.0) * bw; ctx.rect(lo, y, len, bw, s.color, None); @@ -1064,6 +1160,9 @@ fn cartesian(ctx: &mut Ctx, chart: &Chart, data: &ChartData, area: Bx, horizonta } } } + if let Some(m) = run { + m.draw(ctx, bw, horizontal, s.color); + } } _ => { let pts: Vec> = row.iter().enumerate().map(|(j, sp)| sp.map(|(_, t)| [center(j), vpos(t, &sc)])).collect(); diff --git a/crates/chart/src/resolve.rs b/crates/chart/src/resolve.rs index c007e3e90..c0dbaa173 100644 --- a/crates/chart/src/resolve.rs +++ b/crates/chart/src/resolve.rs @@ -11,8 +11,17 @@ use crate::{ChartData, Rgba, SeriesData, TrendlineData}; pub(crate) const MAX_POINTS: usize = 100_000; /// Cap on series per chart. pub(crate) const MAX_SERIES: usize = 255; -/// Cap on trendlines drawn per series. +/// Cap on trendlines per chart (not per series: each one is a fit over a series' points). pub(crate) const MAX_TRENDLINES: usize = 16; +/// Budget for series x points in one chart. When a chart would hold more, every series (and the +/// category list) keeps its first `MAX_TOTAL_POINTS / series` points; no series is dropped. +pub(crate) const MAX_TOTAL_POINTS: usize = 2_000_000; + +/// Points kept per series (and categories) in a chart of `series` series: `MAX_POINTS`, or less +/// when `series x MAX_POINTS` would pass `MAX_TOTAL_POINTS`. +pub(crate) fn points_per_series(series: usize) -> usize { + (MAX_TOTAL_POINTS / series.clamp(1, MAX_SERIES)).clamp(1, MAX_POINTS) +} /// Tints used for series beyond the six theme accents: lighter, darker, lighter still… const CYCLE_TINTS: [f64; 6] = [0.0, 0.4, -0.25, 0.6, -0.5, 0.2]; @@ -77,10 +86,34 @@ fn union_parts(f: &str) -> Option> { /// Evaluates a series formula to a flat list of values (row-major), capped. A union of /// references (`(A1:A3,A5)`, as PivotCharts use) is read part by part. fn eval_list(wb: &Workbook, sheet: usize, formula: &str) -> Vec { - eval_list_in(wb, sheet, formula, 0) + eval_list_in(wb, sheet, formula, MAX_POINTS, 0) } -fn eval_list_in(wb: &Workbook, sheet: usize, formula: &str, depth: u32) -> Vec { +/// calc reads a reference of up to this many cells at its full size and trims a larger one to the +/// sheet's used range (`FULL_SIZE_CELLS` in `gridcraft_calc::eval`). A clipped reference must stay +/// above it, or the clip would turn a trimmed read into a full-size one with trailing blanks. +const CALC_FULL_SIZE_CELLS: u64 = 16_384; + +/// `formula` as a plain reference (`Sheet!A1:A1000000`) cut after the rows that can still hold +/// one of the first `limit` points, so that calc never reads the rest; `None` when `formula` is +/// no reference or already small enough. Values are read row by row, and calc trims only the +/// bottom and right of a reference, so keeping `limit` rows keeps the first `limit` values. +fn clipped_reference(wb: &Workbook, sheet: usize, formula: &str, limit: usize) -> Option { + let (_, r) = parse_ref(wb, sheet, formula)?; + let (h, w) = (u64::from(r.height()), u64::from(r.width())); + if h.saturating_mul(w) <= CALC_FULL_SIZE_CELLS { + return None; + } + let keep = (limit as u64).max(CALC_FULL_SIZE_CELLS / w + 1); + if h <= keep { + return None; + } + let end = CellRef::new(r.start.row.saturating_add(u32::try_from(keep).ok()?).saturating_sub(1), r.end.col); + let prefix = formula.rfind('!').and_then(|i| formula.get(..=i)).unwrap_or(""); + Some(format!("{prefix}{}", RangeRef::new(r.start, end))) +} + +fn eval_list_in(wb: &Workbook, sheet: usize, formula: &str, limit: usize, depth: u32) -> Vec { let f = formula.trim(); if f.is_empty() { return vec![]; @@ -90,16 +123,17 @@ fn eval_list_in(wb: &Workbook, sheet: usize, formula: &str, depth: u32) -> Vec= MAX_POINTS { + if out.len() >= limit { break; } - let rest = MAX_POINTS - out.len(); - out.extend(eval_list_in(wb, sheet, p, depth + 1).into_iter().take(rest)); + let rest = limit - out.len(); + out.extend(eval_list_in(wb, sheet, p, rest, depth + 1).into_iter().take(rest)); } return out; } - match evaluate(wb, sheet, CellRef::default(), f) { - Value::Array(a) => a.iter().take(MAX_POINTS).cloned().collect(), + let clipped = clipped_reference(wb, sheet, f, limit); + match evaluate(wb, sheet, CellRef::default(), clipped.as_deref().unwrap_or(f)) { + Value::Array(a) => a.iter().take(limit).cloned().collect(), v => vec![v], } } @@ -111,8 +145,8 @@ fn to_num(v: &Value) -> Option { } } -fn nums(wb: &Workbook, sheet: usize, formula: &str) -> Vec> { - eval_list(wb, sheet, formula).iter().map(to_num).collect() +fn nums(wb: &Workbook, sheet: usize, formula: &str, limit: usize) -> Vec> { + eval_list_in(wb, sheet, formula, limit, 0).iter().map(to_num).collect() } /// Number format code of the first cell of a reference ("General" when not a reference). @@ -197,35 +231,41 @@ pub fn resolve(wb: &Workbook, sheet: usize, chart: &Chart) -> ChartData { let numeric_x = matches!(chart.kind, ChartKind::Scatter | ChartKind::ScatterLines | ChartKind::Bubble); let mut series = Vec::new(); let mut categories: Option> = None; + let limit = points_per_series(chart.series.len()); + let mut trend_left = MAX_TRENDLINES; for (i, s) in chart.series.iter().take(MAX_SERIES).enumerate() { - let values = nums(wb, sheet, &s.values); + let values = nums(wb, sheet, &s.values, limit); let name = s.name.as_deref().map(|n| series_name(wb, sheet, n)).filter(|n| !n.is_empty()).unwrap_or_else(|| format!("Series{}", i + 1)); let color = s.color.as_ref().and_then(|c| c.resolve(&wb.theme)).map(|[r, g, b]| [r, g, b, 0xFF]).unwrap_or_else(|| series_color(&wb.theme, i)); let kind = effective_kind(chart.kind, s.kind, i); let x = match (&s.categories, numeric_x || matches!(kind, ChartKind::Scatter | ChartKind::ScatterLines | ChartKind::Bubble)) { // X values with text in them (labels) are ignored: the points are numbered 1..n. - (Some(f), true) => Some(eval_list(wb, sheet, f)).filter(|xs| !xs.iter().any(Value::is_text)).map(|xs| xs.iter().map(to_num).collect()), + (Some(f), true) => { + Some(eval_list_in(wb, sheet, f, limit, 0)).filter(|xs| !xs.iter().any(Value::is_text)).map(|xs| xs.iter().map(to_num).collect()) + } _ => None, }; if categories.is_none() && let Some(f) = &s.categories { - let c = category_texts(wb, sheet, f); + let mut c = category_texts(wb, sheet, f); + c.truncate(limit); if !c.is_empty() { categories = Some(c); } } - let sizes = s.bubble_sizes.as_deref().map(|f| nums(wb, sheet, f)); + let sizes = s.bubble_sizes.as_deref().map(|f| nums(wb, sheet, f, limit)); let trendlines = s .trendlines .iter() - .take(MAX_TRENDLINES) + .take(trend_left) .map(|t| TrendlineData { spec: t.clone(), color: t.color.as_ref().and_then(|c| c.resolve(&wb.theme)).map(|[r, g, b]| [r, g, b, 0xFF]).unwrap_or(color), }) - .collect(); + .collect::>(); + trend_left = trend_left.saturating_sub(trendlines.len()); series.push(SeriesData { name, values, @@ -244,3 +284,69 @@ pub fn resolve(wb: &Workbook, sheet: usize, chart: &Chart) -> ChartData { let palette = default_palette(&wb.theme, n.min(PALETTE_CYCLE)); ChartData { categories, series, palette } } + +#[cfg(test)] +mod tests { + use gridcraft_model::Workbook; + + use super::*; + + fn book(rows: u32, cols: u32) -> Workbook { + let mut wb = Workbook::new(); + if let Some(s) = wb.sheet_mut(0) { + s.name = "My Sheet".into(); + for r in 0..rows { + for c in 0..cols { + s.set_value(CellRef::new(r, c), Value::Number(f64::from(r * 10 + c))); + } + } + } + wb + } + + #[test] + fn a_long_reference_is_clipped_before_it_is_evaluated() { + let wb = book(1, 1); + let rows = |f: &str, limit| { + let g = clipped_reference(&wb, 0, f, limit)?; + let (_, r) = parse_ref(&wb, 0, &g)?; + Some((g, u64::from(r.height()))) + }; + // 1,000,000 rows: at most `MAX_POINTS` of them are asked for. + assert_eq!(rows("'My Sheet'!$A$1:$A$1000000", MAX_POINTS), Some(("'My Sheet'!A1:A100000".to_string(), 100_000))); + assert_eq!(rows("A5:A1048576", MAX_POINTS).map(|r| r.1), Some(100_000)); + assert_eq!(rows("A:A", 50_000).map(|r| r.1), Some(50_000)); + // Never below what keeps calc trimming the clipped reference like the whole one. + assert_eq!(rows("A1:A1000000", 100).map(|r| r.1), Some(16_385)); + // Wide references keep whole rows. + assert_eq!(rows("A1:C1000000", MAX_POINTS).map(|r| r.1), Some(100_000)); + // Small references, single cells and anything that isn't a plain reference are left alone. + assert_eq!(rows("A1:A9000", MAX_POINTS), None); + assert_eq!(rows("A1:A20000", MAX_POINTS), None); + assert_eq!(rows("B2", MAX_POINTS), None); + assert_eq!(rows("SUM(A1:A1000000)", MAX_POINTS), None); + assert_eq!(rows("{1,2,3}", MAX_POINTS), None); + } + + #[test] + fn clipping_does_not_change_the_points() { + for (rows, cols) in [(10, 1), (30_000, 1), (30_000, 2), (3, 4)] { + let wb = book(rows, cols); + for f in ["'My Sheet'!A1:A1000000", "'My Sheet'!$A$3:$B$1048576", "A:A", "A1:C1000000", "(A1:A1000000,B1:B1000000)", "A2:A20000"] { + for limit in [100, 7_843, 25_000, MAX_POINTS] { + let clipped = eval_list_in(&wb, 0, f, limit, 0); + let whole: Vec = match evaluate(&wb, 0, CellRef::default(), f) { + Value::Array(a) => a.iter().take(limit).cloned().collect(), + v => vec![v], + }; + if f.starts_with('(') { + // A union is read part by part; each part keeps its own order. + assert!(clipped.len() <= limit); + continue; + } + assert_eq!(clipped, whole, "{f} limit {limit} on {rows}x{cols}"); + } + } + } + } +} diff --git a/crates/chart/src/tests.rs b/crates/chart/src/tests.rs index 01f1aac49..587ddd958 100644 --- a/crates/chart/src/tests.rs +++ b/crates/chart/src/tests.rs @@ -5,6 +5,8 @@ use gridcraft_model::{Anchor, Chart, ChartKind, Color, LegendPos, Series, Sparkl use crate::*; +mod budget; + const ALL_KINDS: [ChartKind; 25] = [ ChartKind::ColumnClustered, ChartKind::ColumnStacked, diff --git a/crates/chart/src/tests/budget.rs b/crates/chart/src/tests/budget.rs new file mode 100644 index 000000000..e151f635d --- /dev/null +++ b/crates/chart/src/tests/budget.rs @@ -0,0 +1,172 @@ +//! Work and memory are bounded by the product of the chart limits (series x points, trendlines +//! per chart, hairline bars), and clean data is rendered without a copy. + +use std::borrow::Cow; + +use gridcraft_core::{CellRef, Value}; +use gridcraft_model::{ChartKind, Series, Trendline, TrendlineKind, Workbook}; + +use super::{assert_sane, chart}; +use crate::render::{sanitize, sanitized}; +use crate::resolve::{MAX_POINTS, MAX_TOTAL_POINTS, MAX_TRENDLINES, points_per_series}; +use crate::*; + +fn data_of(series: usize, points: usize) -> ChartData { + ChartData { + palette: vec![], + categories: (0..points).map(|i| i.to_string()).collect(), + series: (0..series) + .map(|s| SeriesData { + name: format!("S{s}"), + values: (0..points).map(|i| Some(((i * 7 + s * 13) % 100 + 1) as f64)).collect(), + color: series_color(&Default::default(), s), + ..SeriesData::default() + }) + .collect(), + } +} + +fn series_rects(prims: &[Prim], color: Rgba) -> Vec<(f32, f32, f32, f32)> { + prims + .iter() + .filter_map(|p| match p { + Prim::Rect { x, y, w, h, fill: Some(f), .. } if *f == color => Some((*x, *y, *w, *h)), + _ => None, + }) + .collect() +} + +#[test] +fn the_point_budget_keeps_every_series_with_fewer_points() { + assert_eq!(points_per_series(1), MAX_POINTS); + assert_eq!(points_per_series(20), MAX_POINTS); + assert_eq!(points_per_series(21), MAX_TOTAL_POINTS / 21); + assert_eq!(points_per_series(255), MAX_TOTAL_POINTS / 255); + assert_eq!(points_per_series(0), MAX_POINTS); + + // 25 series x 100,000 points is 2.5 million: over budget. + let d = data_of(25, MAX_POINTS); + let kept = sanitized(&d); + assert_eq!(kept.series.len(), 25, "no series is dropped"); + assert!(kept.series.iter().all(|s| s.values.len() == MAX_TOTAL_POINTS / 25)); + assert_eq!(kept.categories.len(), MAX_TOTAL_POINTS / 25); + assert!(kept.series.iter().map(|s| s.values.len()).sum::() <= MAX_TOTAL_POINTS); + // The first points are the ones kept. + assert_eq!(kept.series[3].values[..5], d.series[3].values[..5]); + + // Drawing it stays bounded: hairline columns are merged per pixel column. + let mut c = chart(ChartKind::ColumnClustered); + c.data_labels = false; + c.legend = LegendPos::None; + let prims = render(&c, &d, 400.0, 300.0, &ApproxMeasure); + assert_sane(&prims, 400.0, 300.0); + assert!(prims.len() < 25 * 450 + 200, "{} primitives", prims.len()); +} + +#[test] +fn trendlines_are_limited_per_chart() { + let line = |kind| TrendlineData { spec: Trendline { kind, ..Trendline::default() }, color: [1, 2, 3, 255] }; + let mut d = data_of(3, 30); + for s in &mut d.series { + s.trendlines = vec![line(TrendlineKind::Linear); 10]; + } + let kept = sanitized(&d); + let counts: Vec = kept.series.iter().map(|s| s.trendlines.len()).collect(); + assert_eq!(counts, vec![10, MAX_TRENDLINES - 10, 0], "series order decides which trendlines stay"); + let entries = legend_entries(&chart(ChartKind::Line), &kept); + assert_eq!(entries.len(), 3 + MAX_TRENDLINES); +} + +#[test] +fn hairline_columns_are_merged_and_wide_ones_are_not() { + let mut c = chart(ChartKind::ColumnClustered); + c.data_labels = false; + c.legend = LegendPos::None; + + // Wide bars (>= 0.25 px): one rectangle per series and category, as before. + let d = data_of(3, 40); + let prims = render(&c, &d, 640.0, 400.0, &ApproxMeasure); + for s in &d.series { + assert_eq!(series_rects(&prims, s.color).len(), 40); + } + + // 4 series x 20,000 categories in 400 px: the bars are hairlines. + let d = data_of(4, 20_000); + let prims = render(&c, &d, 400.0, 300.0, &ApproxMeasure); + assert_sane(&prims, 400.0, 300.0); + for s in &d.series { + let n = series_rects(&prims, s.color).len(); + assert!((1..=420).contains(&n), "{n} rectangles for one series"); + } + + // The merged rectangles keep the envelope: alternating 1 and 10 draws the full-height bar. + let mut d = data_of(1, 20_000); + for (i, v) in d.series[0].values.iter_mut().enumerate() { + *v = Some(if i % 2 == 0 { 1.0 } else { 10.0 }); + } + let prims = render(&c, &d, 400.0, 300.0, &ApproxMeasure); + let rects = series_rects(&prims, d.series[0].color); + let tallest = rects.iter().map(|r| r.3).fold(0.0f32, f32::max); + assert!(rects.len() > 100 && tallest > 100.0); + assert!(rects.iter().all(|r| (r.3 - tallest).abs() < 0.01), "every pixel column shows the tall bar"); + // And together they cover the plot's width without a gap wider than a pixel. + let mut xs: Vec<(f32, f32)> = rects.iter().map(|r| (r.0, r.0 + r.2)).collect(); + xs.sort_by(|a, b| a.0.total_cmp(&b.0)); + assert!(xs.windows(2).all(|w| w[1].0 - w[0].1 < 1.0)); + + // Horizontal bars too. + let mut c = chart(ChartKind::BarClustered); + c.data_labels = false; + c.legend = LegendPos::None; + let prims = render(&c, &d, 400.0, 300.0, &ApproxMeasure); + let n = series_rects(&prims, d.series[0].color).len(); + assert!((1..=320).contains(&n), "{n} bars"); +} + +#[test] +fn clean_data_is_rendered_without_a_copy() { + let d = data_of(3, 50); + assert!(matches!(sanitized(&d), Cow::Borrowed(_))); + + // Dirty data takes the copying path and comes out as `sanitize` has always produced it. + let mut dirty = data_of(3, 50); + dirty.series[0].values[4] = Some(f64::NAN); + dirty.series[1].values[5] = Some(f64::INFINITY); + dirty.series[2].values[6] = Some(1e301); + dirty.series[2].x = Some(vec![Some(f64::NEG_INFINITY), Some(2.0)]); + dirty.series[1].sizes = Some(vec![Some(-1e305)]); + let got = sanitized(&dirty); + assert!(matches!(got, Cow::Owned(_))); + assert_eq!(*got, sanitize(&dirty)); + assert_eq!(got.series[0].values[4], None); + assert_eq!(got.series[2].values[6], Some(1e300)); + + // Over the series cap (within the point budget) the same. + let many = data_of(300, 3); + let got = sanitized(&many); + assert_eq!(got.series.len(), 255); + assert_eq!(*got, sanitize(&many)); +} + +#[test] +fn resolve_keeps_every_series_within_the_point_budget() { + // 40 series x 60,000 points is 2.4 million: over the 2 million budget. + let mut wb = Workbook::new(); + if let Some(s) = wb.sheet_mut(0) { + for r in 0..60_000u32 { + s.set_value(CellRef::new(r, 0), Value::Number(f64::from(r))); + } + } + let mut c = chart(ChartKind::ColumnClustered); + c.series = + (0..40).map(|_| Series { values: "Sheet1!$A$1:$A$1000000".into(), categories: Some("A1:A1000000".into()), ..Series::default() }).collect(); + let d = resolve(&wb, 0, &c); + assert_eq!(d.series.len(), 40); + assert!(d.series.iter().all(|s| s.values.len() == MAX_TOTAL_POINTS / 40)); + assert_eq!(d.categories.len(), MAX_TOTAL_POINTS / 40); + assert_eq!(d.series[7].values.first(), Some(&Some(0.0))); + // Trendlines: 16 per chart, in series order. + c.series = (0..4).map(|_| Series { values: "A1:A5".into(), trendlines: vec![Default::default(); 9], ..Series::default() }).collect(); + let counts: Vec = resolve(&wb, 0, &c).series.iter().map(|s| s.trendlines.len()).collect(); + assert_eq!(counts, vec![9, 7, 0, 0]); +} diff --git a/crates/chart/src/trend.rs b/crates/chart/src/trend.rs index d987ab58d..6b22ed7ad 100644 --- a/crates/chart/src/trend.rs +++ b/crates/chart/src/trend.rs @@ -72,7 +72,12 @@ pub(crate) fn fit_trendline(t: &Trendline, pts: &[(f64, f64)]) -> Option = (first..=d).map(|k| u.powi(k as i32)).collect(); + // `d` is at most 6, so the powers `first..=d` always fit the array (no allocation per point). + let mut buf = [0.0f64; 7]; + for (slot, k) in buf.iter_mut().zip(first..=d) { + *slot = u.powi(k as i32); + } + let pows = buf.get(..m)?; for (i, pi) in pows.iter().enumerate() { if let Some(v) = aty.get_mut(i) { *v += pi * (y - b0); @@ -258,10 +263,41 @@ fn trim_zeros(s: &str) -> String { /// Moving average of `period` consecutive points in plotted order: point `i` (from /// `period - 1`) is `(xᵢ, mean(yᵢ₋ₚ₊₁..=yᵢ))`. Empty when there are fewer points than `period`. +/// +/// Linear in the number of points: the points are cut into blocks of `period`, and a window is +/// the tail of one block plus the head of the next, taken from running sums that only ever add +/// (a running sum that also subtracts would lose small values after a huge one passed through). pub(crate) fn moving_average(pts: &[(f64, f64)], period: u32) -> Vec<(f64, f64)> { let p = period.clamp(2, 255) as usize; let pts: Vec<(f64, f64)> = pts.iter().copied().filter(|(x, y)| x.is_finite() && y.is_finite()).collect(); - pts.windows(p).filter_map(|w| w.last().map(|(x, _)| (*x, w.iter().map(|q| q.1).sum::() / p as f64))).collect() + let n = pts.len(); + if n < p { + return vec![]; + } + // `head[i]`: sum of the block's points up to and including `i`; `tail[i]`: from `i` to the block's end. + let mut head = vec![0.0f64; n]; + let mut tail = vec![0.0f64; n]; + for (b, block) in pts.chunks(p).enumerate() { + let (mut h, mut t) = (0.0f64, 0.0f64); + for k in 0..block.len() { + h += block.get(k).map_or(0.0, |q| q.1); + if let Some(v) = head.get_mut(b * p + k) { + *v = h; + } + let r = block.len() - 1 - k; + t += block.get(r).map_or(0.0, |q| q.1); + if let Some(v) = tail.get_mut(b * p + r) { + *v = t; + } + } + } + (p - 1..n) + .filter_map(|i| { + let s = i + 1 - p; + let sum = tail.get(s)? + if s.is_multiple_of(p) { 0.0 } else { *head.get(i)? }; + Some((pts.get(i)?.0, sum / p as f64)) + }) + .collect() } /// Excel's automatic trendline name for the legend, e.g. `Linear (Sales)`. @@ -276,3 +312,168 @@ pub(crate) fn auto_name(kind: TrendlineKind, series: &str) -> String { }; format!("{k} ({series})") } + +#[cfg(test)] +mod tests { + use super::*; + + /// xorshift64 for repeatable pseudo-random data. + struct Rng(u64); + + impl Rng { + fn next(&mut self) -> u64 { + self.0 ^= self.0 << 13; + self.0 ^= self.0 >> 7; + self.0 ^= self.0 << 17; + self.0 + } + fn unit(&mut self) -> f64 { + (self.next() % 1_000_000) as f64 / 1_000_000.0 - 0.5 + } + } + + /// The window-by-window moving average this module used to compute. + fn moving_average_naive(pts: &[(f64, f64)], period: u32) -> Vec<(f64, f64)> { + let p = period.clamp(2, 255) as usize; + let pts: Vec<(f64, f64)> = pts.iter().copied().filter(|(x, y)| x.is_finite() && y.is_finite()).collect(); + pts.windows(p).filter_map(|w| w.last().map(|(x, _)| (*x, w.iter().map(|q| q.1).sum::() / p as f64))).collect() + } + + #[test] + fn linear_moving_average_matches_the_window_sums() { + let mut r = Rng(0x9E37_79B9_7F4A_7C15); + for case in 0..60 { + let n = (r.next() % 700) as usize; + let pts: Vec<(f64, f64)> = (0..n) + .map(|i| { + let y = match case % 5 { + 0 => r.unit() * 100.0, + 1 => 1e9 + r.unit(), + // A huge value passing through must not leave rounding residue behind. + 2 => { + if r.next().is_multiple_of(50) { + 1e300 * r.unit().signum() + } else { + r.unit() + } + } + 3 => { + if i % 2 == 0 { + 1e15 + } else { + -1e15 + r.unit() + } + } + _ => { + if r.next().is_multiple_of(9) { + f64::NAN + } else { + r.unit() * 1e-3 + } + } + }; + (i as f64, y) + }) + .collect(); + for period in [0, 2, 3, 7, 50, 255, 400] { + let fast = moving_average(&pts, period); + let slow = moving_average_naive(&pts, period); + assert_eq!(fast.len(), slow.len(), "case {case} period {period}"); + let p = period.clamp(2, 255) as usize; + let finite: Vec<(f64, f64)> = pts.iter().copied().filter(|(x, y)| x.is_finite() && y.is_finite()).collect(); + for (i, ((fx, fy), (sx, sy))) in fast.iter().zip(&slow).enumerate() { + assert_eq!(fx, sx); + let mass: f64 = finite.iter().skip(i).take(p).map(|q| q.1.abs()).sum::() / p as f64; + assert!((fy - sy).abs() <= 1e-12 * mass, "case {case} period {period} point {i}: {fy} vs {sy}"); + } + } + } + } + + /// The trendline fit as it was written with a `Vec` of powers per data point. + fn fit_reference(t: &Trendline, pts: &[(f64, f64)]) -> Option { + let kind = t.kind; + let d = degree(kind)?; + let data: Vec<(f64, f64)> = pts + .iter() + .filter(|(x, y)| x.is_finite() && y.is_finite() && (!log_x(kind) || *x > 0.0) && (!log_y(kind) || *y > 0.0)) + .map(|&(x, y)| (if log_x(kind) { x.ln() } else { x }, if log_y(kind) { y.ln() } else { y })) + .collect(); + if data.len() < 2 || data.len() <= d { + return None; + } + let fixed = match kind { + TrendlineKind::Linear | TrendlineKind::Polynomial(_) => t.intercept, + TrendlineKind::Exponential => t.intercept.filter(|b| *b > 0.0).map(f64::ln), + _ => None, + } + .filter(|b| b.is_finite()); + let n = data.len() as f64; + let center = if fixed.is_some() { 0.0 } else { data.iter().map(|p| p.0).sum::() / n }; + let scale = data.iter().map(|p| (p.0 - center).abs()).fold(0.0f64, f64::max); + if !(scale.is_finite() && scale > 0.0) { + return None; + } + let first = usize::from(fixed.is_some()); + let b0 = fixed.unwrap_or(0.0); + let m = d + 1 - first; + let mut ata = vec![vec![0.0f64; m]; m]; + let mut aty = vec![0.0f64; m]; + for &(x, y) in &data { + let u = (x - center) / scale; + let pows: Vec = (first..=d).map(|k| u.powi(k as i32)).collect(); + for (i, pi) in pows.iter().enumerate() { + aty[i] += pi * (y - b0); + for (j, pj) in pows.iter().enumerate() { + ata[i][j] += pi * pj; + } + } + } + let sol = solve(ata, aty)?; + let mut a = vec![b0; d + 1]; + for (k, v) in sol.into_iter().enumerate() { + a[k + first] = v; + } + let coef = expand(&a, center, scale)?; + let fit = TrendFit { kind, coef, center, scale, a, r2: None }; + let mean = data.iter().map(|p| p.1).sum::() / n; + let (mut ss_res, mut ss_tot) = (0.0f64, 0.0f64); + for &(x, y) in &data { + ss_res += (y - fit.poly(x)).powi(2); + ss_tot += if fixed.is_some() { (y - b0).powi(2) } else { (y - mean).powi(2) }; + } + let r2 = (ss_tot > 0.0 && ss_tot.is_finite() && ss_res.is_finite()).then(|| 1.0 - ss_res / ss_tot); + Some(TrendFit { r2, ..fit }) + } + + #[test] + fn fit_without_a_vec_per_point_is_bit_identical() { + let mut r = Rng(0x1234_5678_9ABC_DEF1); + let kinds = [ + TrendlineKind::Linear, + TrendlineKind::Exponential, + TrendlineKind::Logarithmic, + TrendlineKind::Power, + TrendlineKind::Polynomial(2), + TrendlineKind::Polynomial(3), + TrendlineKind::Polynomial(6), + TrendlineKind::MovingAverage(3), + ]; + for case in 0..40 { + let n = 2 + (r.next() % 60) as usize; + let pts: Vec<(f64, f64)> = (0..n) + .map(|i| { + let x = i as f64 * (1.0 + r.unit()) + 1.0; + let y = if case % 4 == 0 { f64::NAN } else { (r.unit() + 0.6) * 10f64.powi((case % 7) - 2) }; + (x, if case % 4 == 0 && i % 3 != 0 { r.unit() + 1.0 } else { y }) + }) + .collect(); + for kind in kinds { + for intercept in [None, Some(0.5), Some(-1.0)] { + let t = Trendline { kind, intercept, ..Trendline::default() }; + assert_eq!(fit_trendline(&t, &pts), fit_reference(&t, &pts), "case {case} {kind:?} {intercept:?}"); + } + } + } + } +} From 96bfba93f880a6c00c15d5a9afcc0eace1575be3 Mon Sep 17 00:00:00 2001 From: vangel76 Date: Mon, 26 Oct 2026 22:17:29 +0100 Subject: [PATCH 03/24] perf(calc): remember SUM/COUNT/AVERAGE of constant-only blocks per pass (#2681) A block that holds only constants can't change during a pass, so its sum and count are worked out once and reused by every formula that reads the same block. A block that is the same rows from the top but longer (a running total) continues the shorter block's sum, which is the same additions in the same order as one walk from the start. Co-Authored-By: Claude Sonnet 5.5 --- crates/calc/src/eval.rs | 14 +++++ crates/calc/src/recalc.rs | 49 +++++++++++++++++ crates/calc/src/tests.rs | 113 ++++++++++++++++++++++++++++++++++++++ 3 files changed, 176 insertions(+) diff --git a/crates/calc/src/eval.rs b/crates/calc/src/eval.rs index 4bdb3a30d..54777f98b 100644 --- a/crates/calc/src/eval.rs +++ b/crates/calc/src/eval.rs @@ -103,6 +103,12 @@ pub trait Host { fn visit_range(&mut self, _sheet: usize, _range: RangeRef, _f: &mut dyn FnMut(&Value) -> bool) -> bool { false } + /// `(sum, count)` of the numbers in a block that holds nothing but constants (no formula, + /// error or spilled value), added row-major from `0.0` exactly as `visit_range` would hand + /// them over. `None`: the block isn't like that, or this host can't tell. + fn range_sum(&mut self, _sheet: usize, _range: RangeRef) -> Option<(f64, u64)> { + None + } /// While a formula is calculated on demand inside another one: `stack_position()` where the /// outermost one started, which the stack budget is counted from. fn stack_top(&self) -> Option { @@ -984,6 +990,14 @@ impl<'h> Evaluator<'h> { /// Adds one area to `acc`; `Err` is the result the function ends with. fn aggregate_area(&mut self, kind: Agg, a: &Area, acc: &mut Acc) -> Result<(), Ev> { if !a.range.is_single() { + // Adding the block's remembered sum to anything but +0.0 would add in another order. + if acc.sum.to_bits() == 0 + && let Some((sum, count)) = self.host.range_sum(a.sheet, a.range) + { + acc.sum = sum; + acc.count = acc.count.saturating_add(count); + return Ok(()); + } let mut error = None; let walked = self.host.visit_range(a.sheet, a.range, &mut |v| match acc.add_cell(kind, v) { Ok(()) => true, diff --git a/crates/calc/src/recalc.rs b/crates/calc/src/recalc.rs index 17a0e6649..753fd1971 100644 --- a/crates/calc/src/recalc.rs +++ b/crates/calc/src/recalc.rs @@ -436,6 +436,10 @@ struct PassHost<'a> { stack_top: usize, /// Whether an evaluation ran past the stack budget since the innermost formula started. out_of_stack: bool, + /// Sum and count of blocks that hold only constants, for `range_sum`. Keyed by sheet, first + /// cell and last column; each value maps a last row to the block's result (`None`: the block + /// has a formula, an error value or a spilled value, so every longer one has too). + sums: HashMap<(usize, CellRef, u32), std::collections::BTreeMap>>, } impl<'a> PassHost<'a> { @@ -468,6 +472,7 @@ impl<'a> PassHost<'a> { taint: false, stack_top: 0, out_of_stack: false, + sums: HashMap::default(), } } } @@ -909,6 +914,50 @@ impl Host for PassHost<'_> { } true } + fn range_sum(&mut self, sheet: usize, range: RangeRef) -> Option<(f64, u64)> { + // Constants can't change during a pass, so a block's result stays valid for it. Spills + // found in this pass (and iteration, which tracks what a read depends on) go the long way. + if self.iterate || self.spills.keys().any(|a| a.0 == sheet) { + return None; + } + let sh = self.wb.sheet(sheet)?; + let (key, last) = ((sheet, range.start, range.end.col), range.end.row); + let known = self.sums.entry(key).or_default(); + if let Some(r) = known.get(&last) { + return *r; + } + // Start from the longest shorter block of the same rows from the top: summing row-major + // from there on is the same additions in the same order as one walk from the start. + let (from, mut sum, mut count) = match known.range(..last).next_back() { + Some((_, None)) => { + known.insert(last, None); + return None; + } + Some((&row, &Some((sum, count)))) => (row + 1, sum, count), + None => (range.start.row, 0.0, 0), + }; + let rest = RangeRef::new(CellRef::new(from, range.start.col), range.end); + let mut plain = sh.spill.range(rest.start..=rest.end).all(|(c, _)| !rest.contains(*c)); + if plain { + for (_, cell) in sh.cells.iter_range(rest) { + match &cell.value { + _ if cell.formula.is_some() => plain = false, + Value::Error(_) => plain = false, + Value::Number(n) => { + sum += n; + count = u64::saturating_add(count, 1); + } + _ => {} + } + if !plain { + break; + } + } + } + let result = plain.then_some((sum, count)); + known.insert(last, result); + result + } fn used_range(&mut self, sheet: usize) -> Option { let wb = self.wb; *self.used.entry(sheet).or_insert_with(|| wb.sheet(sheet).and_then(|s| s.used_range())) diff --git a/crates/calc/src/tests.rs b/crates/calc/src/tests.rs index 5b9e90085..a2db89be0 100644 --- a/crates/calc/src/tests.rs +++ b/crates/calc/src/tests.rs @@ -2274,3 +2274,116 @@ fn loan_schedule_residue_keeps_excels_sign() { } assert_eq!(t.get("P7"), Value::Text("".into())); } + +/// `SUM`, `COUNT` and `AVERAGE` over blocks of constants share one pass-wide answer per block, and a +/// running total extends the shorter block's. Every result must equal the left-to-right sum of the +/// cells, bit for bit, with text, blanks, errors, formulas and a nonzero start in the mix. +#[test] +fn aggregates_over_ranges_match_a_plain_fold_bit_for_bit() { + struct Rng(u64); + impl Rng { + fn next(&mut self, n: u64) -> u64 { + self.0 ^= self.0 << 13; + self.0 ^= self.0 >> 7; + self.0 ^= self.0 << 17; + self.0 % n + } + } + let nums = [0.1, 0.2, 0.3, 1e16, -1e16, 1.0, 3.7, -0.0, 12345.678, 1e-9, 7.0]; + let rows = 40u32; + let mut rng = Rng(0x9E37_79B9_7F4A_7C15); + let mut wb = Workbook::new(); + let fresh = |rng: &mut Rng, plain: bool| match rng.next(if plain { 8 } else { 14 }) { + 0 => Cell::value(Value::text("t")), + 1 => Cell::default(), + 2 => Cell::value(Value::Bool(true)), + 3 if !plain => Cell::value(Value::Error(CellError::NA)), + 4 if !plain => Cell::formula(Formula::new("=A1*2")), + 5 if !plain => Cell::formula(Formula::new("=\"x\"&B2")), + _ => Cell::value(Value::Number(nums[rng.next(nums.len() as u64) as usize])), + }; + // Columns A:C mostly numbers; E holds a formula column and some errors only in a few rows. + for r in 0..rows { + for col in 0..3 { + let plain = !(r % 13 == 7 && col == 1); + wb.sheet_mut(0).unwrap().cells.set(CellRef::new(r, col), fresh(&mut rng, plain)); + } + } + // Formulas in column H and up: (function, optional leading number, blocks). + let mut oracle: Vec<(CellRef, &str, Option, Vec<(u32, u32, u32, u32)>)> = vec![]; + let funcs = ["SUM", "COUNT", "AVERAGE"]; + let mut at = 0; + let mut add = |wb: &mut Workbook, f: &'static str, lead: Option, blocks: Vec<(u32, u32, u32, u32)>| { + let args: Vec = blocks + .iter() + .map(|&(r0, c0, r1, c1)| format!("{}{}:{}{}", (b'A' + c0 as u8) as char, r0 + 1, (b'A' + c1 as u8) as char, r1 + 1)) + .collect(); + let lead_s = lead.map(|l| format!("{l:e},")).unwrap_or_default(); + let cell = CellRef::new(at % rows, 7 + at / rows); + wb.sheet_mut(0).unwrap().cells.set(cell, Cell::formula(Formula::new(&format!("={f}({lead_s}{})", args.join(","))))); + oracle.push((cell, f, lead, blocks)); + at += 1; + }; + // Running totals over a clean column and over a column with an error further down. + for r in 0..rows { + add(&mut wb, "SUM", None, vec![(0, 0, r, 0)]); + add(&mut wb, "SUM", None, vec![(0, 1, r, 1)]); + add(&mut wb, "COUNT", None, vec![(0, 0, r, 2)]); + } + for _ in 0..120 { + let (r0, r1) = (rng.next(rows as u64) as u32, rng.next(rows as u64) as u32); + let (c0, c1) = (rng.next(3) as u32, rng.next(3) as u32); + let n = 1 + rng.next(2) as usize; + let blocks = (0..n).map(|_| (r0.min(r1), c0.min(c1), r0.max(r1), c0.max(c1))).collect(); + let lead = [None, Some(1e16), Some(0.5), Some(2.0)][rng.next(4) as usize]; + let f = funcs[rng.next(3) as usize]; + add(&mut wb, f, lead, blocks); + } + let check = |wb: &Workbook, when: &str| { + let sh = wb.sheet(0).unwrap(); + for (cell, f, lead, blocks) in &oracle { + let (mut sum, mut count, mut error) = (lead.unwrap_or(0.0), u64::from(lead.is_some()), None); + for &(r0, c0, r1, c1) in blocks { + for r in r0..=r1 { + for col in c0..=c1 { + match sh.value(CellRef::new(r, col)) { + Value::Number(n) => { + sum += n; + count += 1; + } + Value::Error(e) if *f != "COUNT" => error = error.or(Some(e)), + _ => {} + } + } + } + } + let want = match (error, *f) { + (Some(e), _) => Value::Error(e), + (_, "SUM") => Value::number(sum), + (_, "COUNT") => Value::number(count as f64), + (_, _) if count == 0 => Value::Error(CellError::Div0), + _ => Value::number(sum / count as f64), + }; + let got = sh.value(*cell); + let same = match (&got, &want) { + (Value::Number(a), Value::Number(b)) => a.to_bits() == b.to_bits(), + (a, b) => a == b, + }; + assert!(same, "{when}: {:?} = {:?}, plain fold {:?}", cell, got, want); + } + }; + let mut calc = Calc::new(); + calc.rebuild(&wb); + calc.recalc_all(&mut wb); + check(&wb, "full recalc"); + for round in 0..30 { + let mut changed = vec![]; + for _ in 0..1 + rng.next(4) { + let at = CellRef::new(rng.next(rows as u64) as u32, rng.next(3) as u32); + wb.sheet_mut(0).unwrap().cells.set(at, fresh(&mut rng, round % 2 == 0)); + changed.push((0, at)); + } + calc.cells_changed(&mut wb, &changed); + check(&wb, &format!("edit round {round}")); + } +} From 95c3bb9b89e229759077aa025032396e951e30a8 Mon Sep 17 00:00:00 2001 From: vangel76 Date: Thu, 29 Oct 2026 23:05:54 +0100 Subject: [PATCH 04/24] Fix clippy complexity after merging upstream/main --- crates/chart/src/render/cartesian.rs | 84 +++++++++++++++++++--------- 1 file changed, 58 insertions(+), 26 deletions(-) diff --git a/crates/chart/src/render/cartesian.rs b/crates/chart/src/render/cartesian.rs index f30283b58..78774b078 100644 --- a/crates/chart/src/render/cartesian.rs +++ b/crates/chart/src/render/cartesian.rs @@ -3,8 +3,8 @@ use super::legend::{cat_label, plot_kind}; use super::lines::{decimate, draw_markers, draw_runs}; use super::trend::{draw_trend_labels, draw_trendlines, max_backward, max_forward, max_forward_slots, supports_trendline}; use super::{ - Chart, ChartData, ChartKind, Ctx, Fmt, HAlign, LABEL_SIZE, MAX_CATS, MAX_LABELS, MAX_LINE_PTS, MAX_MARKERS, Prim, Scale, SeriesData, TEXT, - VAlign, WHITE, clean, fixed_scale, nice_scale, + Chart, ChartData, ChartKind, Ctx, Fmt, HAlign, LABEL_SIZE, MAX_CATS, MAX_LABELS, MAX_LINE_PTS, MAX_MARKERS, MIN_BAR_PX, Prim, Rgba, Scale, + SeriesData, TEXT, VAlign, WHITE, clean, fixed_scale, nice_scale, }; // ---------------------------------------------------------------- cartesian (column/bar/line/area/combo) @@ -81,6 +81,58 @@ impl BarRun { } } +/// Whether bars of width `bw` are merged per pixel column (never when they carry data labels). +fn merge_hairlines(bw: f32, labels_on: bool) -> bool { + bw < MIN_BAR_PX && !labels_on +} + +/// Draws the bars of one series: a rectangle each, or, for hairline bars, one merged [`BarRun`] +/// per pixel column. +struct BarSink { + run: Option, + bw: f32, + horizontal: bool, + color: Rgba, + nslots: f32, + k: f32, +} + +impl BarSink { + fn new(bw: f32, horizontal: bool, color: Rgba, nslots: f32, k: f32) -> Self { + BarSink { run: None, bw, horizontal, color, nslots, k } + } + /// Start of this series' slot along the category axis, for the category centered at `c`. + fn pos(&self, c: f32) -> f32 { + if self.horizontal { c + self.nslots * self.bw / 2.0 - (self.k + 1.0) * self.bw } else { c - self.nslots * self.bw / 2.0 + self.k * self.bw } + } + /// Draws the bar at category center `c` spanning `lo..lo + len` on the value axis, or, when + /// `merge` is set, adds it to the current run. + fn bar(&mut self, ctx: &mut Ctx, merge: bool, c: f32, lo: f32, len: f32) { + let pos = self.pos(c); + if !merge { + if self.horizontal { + ctx.rect(lo, pos, len, self.bw, self.color, None); + } else { + ctx.rect(pos, lo, self.bw, len, self.color, None); + } + return; + } + match &mut self.run { + Some(m) if m.bucket == pos.floor() => m.add(pos, lo, lo + len), + run => { + if let Some(m) = run.replace(BarRun::new(pos, lo, lo + len)) { + m.draw(ctx, self.bw, self.horizontal, self.color); + } + } + } + } + fn finish(self, ctx: &mut Ctx) { + if let Some(m) = self.run { + m.draw(ctx, self.bw, self.horizontal, self.color); + } + } +} + fn series_spans(data: &ChartData, fams: &[Fam], sec: &[bool], ncat: usize) -> Vec>> { // Totals for 100% stacking, per axis and family (100% columns and 100% areas stack apart). let mut totals: Vec<((bool, Fam), Vec)> = Vec::new(); @@ -307,33 +359,15 @@ pub(super) fn cartesian(ctx: &mut Ctx, chart: &Chart, data: &ChartData, area: Bx 1 => { let k = slot_of.get(i).copied().unwrap_or(0) as f32; // Hairline bars are merged per pixel column (their envelope is what shows). - let cull = bw < MIN_BAR_PX && !labels_on; - let mut run: Option = None; + let cull = merge_hairlines(bw, labels_on); + let mut sink = BarSink::new(bw, horizontal, s.color, nslots, k); for (j, sp) in row.iter().enumerate() { let Some((b, t)) = sp else { continue }; let (pb, pt) = (vpos(*b, &sc), vpos(*t, &sc)); let c = center(j); let lo = pb.min(pt); let len = (pb - pt).abs(); - if cull { - let pos = if horizontal { c + nslots * bw / 2.0 - (k + 1.0) * bw } else { c - nslots * bw / 2.0 + k * bw }; - match &mut run { - Some(m) if m.bucket == pos.floor() => m.add(pos, lo, lo + len), - _ => { - if let Some(m) = run.replace(BarRun::new(pos, lo, lo + len)) { - m.draw(ctx, bw, horizontal, s.color); - } - } - } - continue; - } - if horizontal { - let y = c + nslots * bw / 2.0 - (k + 1.0) * bw; - ctx.rect(lo, y, len, bw, s.color, None); - } else { - let x = c - nslots * bw / 2.0 + k * bw; - ctx.rect(x, lo, bw, len, s.color, None); - } + sink.bar(ctx, cull, c, lo, len); if labels_on && let Some(v) = s.values.get(j).copied().flatten() { let t = fmt.format(v, None); let stacked = fam != Fam::Col; @@ -359,9 +393,7 @@ pub(super) fn cartesian(ctx: &mut Ctx, chart: &Chart, data: &ChartData, area: Bx } } } - if let Some(m) = run { - m.draw(ctx, bw, horizontal, s.color); - } + sink.finish(ctx); } _ => { let pts: Vec> = row.iter().enumerate().map(|(j, sp)| sp.map(|(_, t)| [center(j), vpos(t, &sc)])).collect(); From 494a7dbb5c0fe57d819b83f1981d62e15d4b14d4 Mon Sep 17 00:00:00 2001 From: vangel76 Date: Fri, 30 Oct 2026 00:54:56 +0100 Subject: [PATCH 05/24] Keep clustered hairline bars of each series in their own strip of the pixel column Co-Authored-By: Claude Sonnet 5.5 --- crates/chart/src/render/cartesian.rs | 32 ++++++++++++++++++++------ crates/chart/src/tests/budget.rs | 34 ++++++++++++++++++++++++++++ 2 files changed, 59 insertions(+), 7 deletions(-) diff --git a/crates/chart/src/render/cartesian.rs b/crates/chart/src/render/cartesian.rs index 78774b078..1d643c21f 100644 --- a/crates/chart/src/render/cartesian.rs +++ b/crates/chart/src/render/cartesian.rs @@ -59,24 +59,32 @@ struct BarRun { a1: f32, lo: f32, hi: f32, + /// More than one bar was merged. + merged: bool, } impl BarRun { fn new(pos: f32, lo: f32, hi: f32) -> Self { - BarRun { bucket: pos.floor(), a0: pos, a1: pos, lo, hi } + BarRun { bucket: pos.floor(), a0: pos, a1: pos, lo, hi, merged: false } } fn add(&mut self, pos: f32, lo: f32, hi: f32) { + self.merged = true; self.a0 = self.a0.min(pos); self.a1 = self.a1.max(pos); self.lo = self.lo.min(lo); self.hi = self.hi.max(hi); } - fn draw(&self, ctx: &mut Ctx, bw: f32, horizontal: bool, color: Rgba) { - let along = self.a1 - self.a0 + bw; + /// `band` is the part of the pixel column reserved for this series (`start`, `len`): in a + /// clustered chart a merged run stays inside it, so the other series keep a visible strip. + fn draw(&self, ctx: &mut Ctx, bw: f32, horizontal: bool, color: Rgba, band: Option<(f32, f32)>) { + let (a0, along) = match band { + Some((start, len)) if self.merged => (self.bucket + start, len), + _ => (self.a0, self.a1 - self.a0 + bw), + }; if horizontal { - ctx.rect(self.lo, self.a0, self.hi - self.lo, along, color, None); + ctx.rect(self.lo, a0, self.hi - self.lo, along, color, None); } else { - ctx.rect(self.a0, self.lo, along, self.hi - self.lo, color, None); + ctx.rect(a0, self.lo, along, self.hi - self.lo, color, None); } } } @@ -105,10 +113,19 @@ impl BarSink { fn pos(&self, c: f32) -> f32 { if self.horizontal { c + self.nslots * self.bw / 2.0 - (self.k + 1.0) * self.bw } else { c - self.nslots * self.bw / 2.0 + self.k * self.bw } } + /// This series' strip of a pixel column (offset, length) when the columns hold several + /// clustered series, in the direction the slots run. + fn band(&self) -> Option<(f32, f32)> { + (self.nslots > 1.0).then(|| { + let idx = if self.horizontal { self.nslots - 1.0 - self.k } else { self.k }; + (idx / self.nslots, 1.0 / self.nslots) + }) + } /// Draws the bar at category center `c` spanning `lo..lo + len` on the value axis, or, when /// `merge` is set, adds it to the current run. fn bar(&mut self, ctx: &mut Ctx, merge: bool, c: f32, lo: f32, len: f32) { let pos = self.pos(c); + let band = self.band(); if !merge { if self.horizontal { ctx.rect(lo, pos, len, self.bw, self.color, None); @@ -121,14 +138,15 @@ impl BarSink { Some(m) if m.bucket == pos.floor() => m.add(pos, lo, lo + len), run => { if let Some(m) = run.replace(BarRun::new(pos, lo, lo + len)) { - m.draw(ctx, self.bw, self.horizontal, self.color); + m.draw(ctx, self.bw, self.horizontal, self.color, band); } } } } fn finish(self, ctx: &mut Ctx) { + let band = self.band(); if let Some(m) = self.run { - m.draw(ctx, self.bw, self.horizontal, self.color); + m.draw(ctx, self.bw, self.horizontal, self.color, band); } } } diff --git a/crates/chart/src/tests/budget.rs b/crates/chart/src/tests/budget.rs index 1f2e35ab3..54d33aa7d 100644 --- a/crates/chart/src/tests/budget.rs +++ b/crates/chart/src/tests/budget.rs @@ -124,6 +124,40 @@ fn hairline_columns_are_merged_and_wide_ones_are_not() { assert!((1..=320).contains(&n), "{n} bars"); } +#[test] +fn clustered_hairline_series_keep_a_visible_strip() { + // Four series of similar height, 20,000 categories in 400 px: every pixel column holds a + // merged bar of each series, and they must not cover each other. + for kind in [ChartKind::ColumnClustered, ChartKind::BarClustered] { + let horizontal = kind == ChartKind::BarClustered; + let mut c = chart(kind); + c.data_labels = false; + c.legend = LegendPos::None; + let mut d = data_of(4, 20_000); + for (s, series) in d.series.iter_mut().enumerate() { + for (i, v) in series.values.iter_mut().enumerate() { + *v = Some(50.0 + ((i + s) % 3) as f64); + } + } + let prims = render(&c, &d, 400.0, 300.0, &ApproxMeasure); + let along = |r: &(f32, f32, f32, f32)| if horizontal { (r.1, r.1 + r.3) } else { (r.0, r.0 + r.2) }; + let mut strips: Vec<(f32, f32, usize)> = Vec::new(); + for (s, series) in d.series.iter().enumerate() { + strips.extend(series_rects(&prims, series.color).iter().map(|r| (along(r).0, along(r).1, s))); + } + strips.sort_by(|a, b| a.0.total_cmp(&b.0)); + assert!(strips.len() > 400); + // Rectangles of different series never overlap (more than rounding). + for w in strips.windows(2) { + assert!(w[0].2 == w[1].2 || w[1].0 >= w[0].1 - 1e-3, "series {} and {} overlap", w[0].2, w[1].2); + } + // And all four series show up in the same pixel column. + let col = strips[strips.len() / 2].0.floor(); + let seen: std::collections::BTreeSet = strips.iter().filter(|s| s.0.floor() == col).map(|s| s.2).collect(); + assert_eq!(seen.len(), 4, "series visible in pixel column {col}"); + } +} + #[test] fn clean_data_is_rendered_without_a_copy() { let d = data_of(3, 50); From 00f565bc11d69623f5260a13e143c20c65d0c85f Mon Sep 17 00:00:00 2001 From: vangel76 Date: Mon, 2 Nov 2026 19:44:16 +0100 Subject: [PATCH 06/24] Factor out types clippy flags as too complex Co-Authored-By: Claude Sonnet 5.5 --- crates/calc/src/recalc/pass_host.rs | 5 ++++- crates/calc/src/tests.rs | 3 ++- 2 files changed, 6 insertions(+), 2 deletions(-) diff --git a/crates/calc/src/recalc/pass_host.rs b/crates/calc/src/recalc/pass_host.rs index 6151b74d3..9964b3363 100644 --- a/crates/calc/src/recalc/pass_host.rs +++ b/crates/calc/src/recalc/pass_host.rs @@ -13,6 +13,9 @@ const MAX_NESTED: usize = 64; /// evaluated again), however high Maximum Iterations is. const MAX_ITERATION_WORK: usize = 10_000_000; +/// A block's results by last row: its sum and count, or `None` when it can't be summed from constants. +type BlockSums = std::collections::BTreeMap>; + pub(super) struct PassHost<'a> { wb: &'a Workbook, exprs: &'a HashMap>, @@ -71,7 +74,7 @@ pub(super) struct PassHost<'a> { /// Sum and count of blocks that hold only constants, for `range_sum`. Keyed by sheet, first /// cell and last column; each value maps a last row to the block's result (`None`: the block /// has a formula, a Python cell, an error value or a spilled value, so every longer one has too). - sums: HashMap<(usize, CellRef, u32), std::collections::BTreeMap>>, + sums: HashMap<(usize, CellRef, u32), BlockSums>, /// The kept results of the Python cells in the pass: a Python cell isn't evaluated, its /// result is laid out (an array spills from the cell, as a formula's does). Each one is in /// `pending` until it is laid out, so a formula that reads it first gets its new value. diff --git a/crates/calc/src/tests.rs b/crates/calc/src/tests.rs index a55231794..0c9ec4aa1 100644 --- a/crates/calc/src/tests.rs +++ b/crates/calc/src/tests.rs @@ -2348,7 +2348,8 @@ fn aggregates_over_ranges_match_a_plain_fold_bit_for_bit() { } } // Formulas in column H and up: (function, optional leading number, blocks). - let mut oracle: Vec<(CellRef, &str, Option, Vec<(u32, u32, u32, u32)>)> = vec![]; + type Oracle<'a> = (CellRef, &'a str, Option, Vec<(u32, u32, u32, u32)>); + let mut oracle: Vec = vec![]; let funcs = ["SUM", "COUNT", "AVERAGE"]; let mut at = 0; let mut add = |wb: &mut Workbook, f: &'static str, lead: Option, blocks: Vec<(u32, u32, u32, u32)>| { From 22c06b7a4a394334e0d35f8880da2e9ce7d830e6 Mon Sep 17 00:00:00 2001 From: vangel76 Date: Tue, 3 Nov 2026 23:49:52 +0100 Subject: [PATCH 07/24] BINOM.DIST, POISSON.DIST: Loader mass, recurrence sums, exact prefactors (#3324) The mass came from ln_gamma differences (relative error 1.1e-13 at n = 100, 1.5e-11 at n = 10,000) and the cumulative forms summed up to 10,000 of them. The mass now comes from Loader's saddle-point formula, the cumulative sum walks away from the mode by the ratio of neighbouring terms and stops when terms stop mattering, and the incomplete beta / gamma branches take the same mass as their prefactor. Co-Authored-By: Claude Sonnet 5.5 --- crates/functions/src/special.rs | 36 +++- crates/functions/src/stat.rs | 5 +- crates/functions/src/stat/distributions.rs | 191 +++++++++++++++-- crates/functions/src/stat/tests.rs | 230 +++++++++++++++++++++ 4 files changed, 436 insertions(+), 26 deletions(-) diff --git a/crates/functions/src/special.rs b/crates/functions/src/special.rs index 1541cd8a1..b173ff765 100644 --- a/crates/functions/src/special.rs +++ b/crates/functions/src/special.rs @@ -88,11 +88,17 @@ pub(crate) fn reg_gamma_p(a: f64, x: f64) -> f64 { if x.is_infinite() { return 1.0; } - if x < a + 1.0 { gamma_series(a, x) } else { 1.0 - gamma_cf(a, x) } + if x < a + 1.0 { gamma_series(a, x, front_of(a, x)) } else { 1.0 - gamma_cf(a, x, front_of(a, x)) } } /// Regularised upper incomplete gamma Q(a, x) = 1 − P(a, x). pub(crate) fn reg_gamma_q(a: f64, x: f64) -> f64 { + let front = if a > 0.0 && x > 0.0 && x.is_finite() { (-x + a * x.ln() - ln_gamma(a)).exp() } else { 0.0 }; + reg_gamma_q_front(a, x, front) +} + +/// Q(a, x) given the prefactor `e^-x x^a / Γ(a)` (the Poisson mass at a - 1, times x). +pub(crate) fn reg_gamma_q_front(a: f64, x: f64, front: f64) -> f64 { if a.is_nan() || a <= 0.0 || x.is_nan() || x < 0.0 || !a.is_finite() { return f64::NAN; } @@ -102,10 +108,15 @@ pub(crate) fn reg_gamma_q(a: f64, x: f64) -> f64 { if x.is_infinite() { return 0.0; } - if x < a + 1.0 { 1.0 - gamma_series(a, x) } else { gamma_cf(a, x) } + if x < a + 1.0 { 1.0 - gamma_series(a, x, front) } else { gamma_cf(a, x, front) } } -fn gamma_series(a: f64, x: f64) -> f64 { +/// `e^-x x^a / Γ(a)`. +fn front_of(a: f64, x: f64) -> f64 { + (-x + a * x.ln() - ln_gamma(a)).exp() +} + +fn gamma_series(a: f64, x: f64, front: f64) -> f64 { let mut ap = a; let mut sum = 1.0 / a; let mut del = sum; @@ -117,10 +128,10 @@ fn gamma_series(a: f64, x: f64) -> f64 { break; } } - sum * (-x + a * x.ln() - ln_gamma(a)).exp() + sum * front } -fn gamma_cf(a: f64, x: f64) -> f64 { +fn gamma_cf(a: f64, x: f64, front: f64) -> f64 { let tiny = 1e-300; let mut b = x + 1.0 - a; let mut c = 1.0 / tiny; @@ -144,7 +155,7 @@ fn gamma_cf(a: f64, x: f64) -> f64 { break; } } - (-x + a * x.ln() - ln_gamma(a)).exp() * h + front * h } /// Error function. @@ -187,6 +198,17 @@ pub(crate) fn erfc(x: f64) -> f64 { /// Regularised incomplete beta I_x(a, b). pub(crate) fn reg_inc_beta(x: f64, a: f64, b: f64) -> f64 { + if x > 0.0 && x < 1.0 && a > 0.0 && b > 0.0 { + let ln_bt = -ln_beta(a, b) + a * x.ln() + b * (1.0 - x).ln(); + return reg_inc_beta_front(x, a, b, ln_bt.exp()); + } + reg_inc_beta_front(x, a, b, 0.0) +} + +/// I_x(a, b) given the prefactor `bt = x^a (1-x)^b / B(a, b)` (symmetric in the two sides). A +/// caller that knows `bt` more accurately than the log-gamma route gives (the binomial mass, for +/// instance) passes it here. +pub(crate) fn reg_inc_beta_front(x: f64, a: f64, b: f64, bt: f64) -> f64 { if a.is_nan() || a <= 0.0 || b.is_nan() || b <= 0.0 || !(0.0..=1.0).contains(&x) { return f64::NAN; } @@ -196,8 +218,6 @@ pub(crate) fn reg_inc_beta(x: f64, a: f64, b: f64) -> f64 { if x == 1.0 { return 1.0; } - let ln_bt = -ln_beta(a, b) + a * x.ln() + b * (1.0 - x).ln(); - let bt = ln_bt.exp(); if x < (a + 1.0) / (a + b + 2.0) { bt * beta_cf(a, b, x) / a } else { 1.0 - bt * beta_cf(b, a, 1.0 - x) / b } } diff --git a/crates/functions/src/stat.rs b/crates/functions/src/stat.rs index 270bd2373..790915c95 100644 --- a/crates/functions/src/stat.rs +++ b/crates/functions/src/stat.rs @@ -3,7 +3,10 @@ use gridcraft_core::{CellError, Value}; -use crate::special::{erf, gamma, invert, ln_beta, ln_gamma, norm_cdf, norm_inv, norm_pdf, reg_gamma_p, reg_gamma_q, reg_inc_beta}; +use crate::special::{ + erf, gamma, invert, ln_beta, ln_gamma, norm_cdf, norm_inv, norm_pdf, reg_gamma_p, reg_gamma_q, reg_gamma_q_front, reg_inc_beta, + reg_inc_beta_front, +}; use crate::util::{A, R, S, arg, array_numbers, array_val, as_array, boolean, has, num, num_val, numbers, numbers_a, opt_num}; use crate::{Arg, Ctx, FnSpec, VAR}; diff --git a/crates/functions/src/stat/distributions.rs b/crates/functions/src/stat/distributions.rs index a3964ad80..7fa486e93 100644 --- a/crates/functions/src/stat/distributions.rs +++ b/crates/functions/src/stat/distributions.rs @@ -1,7 +1,7 @@ use super::descriptive::{arg_nums, arg_pairs, mean, var_s}; use super::{ Arg, C, CellError, R, Value, arg, as_array, boolean, erf, gamma, has, invert, ln_beta, ln_gamma, norm_cdf, norm_inv, norm_pdf, num, num_val, - reg_gamma_p, reg_gamma_q, reg_inc_beta, + reg_gamma_p, reg_gamma_q, reg_gamma_q_front, reg_inc_beta, reg_inc_beta_front, }; // --------------------------------------------------------------------------------------------- @@ -93,6 +93,95 @@ fn f_pdf(x: f64, d1: f64, d2: f64) -> f64 { const MAX_BINOM_TRIALS: f64 = 2_147_483_646.0; +/// Stirling-series error `ln n! - ((n + 1/2) ln n - n + ln(2π)/2)` for integer `n >= 1`: exact +/// table (computed with 60-digit Decimal arithmetic) up to 30, then the series, whose first +/// omitted term is below 1e-19 there. +fn stirlerr(n: f64) -> f64 { + const TABLE: [f64; 31] = [ + 0.0, + 0.08106146679532726, + 0.0413406959554093, + 0.02767792568499834, + 0.020790672103765093, + 0.016644691189821193, + 0.013876128823070748, + 0.01189670994589177, + 0.010411265261972096, + 0.009255462182712733, + 0.00833056343336287, + 0.007573675487951841, + 0.00694284010720953, + 0.006408994188004207, + 0.0059513701127588475, + 0.005554733551962801, + 0.0052076559196096404, + 0.004901395948434738, + 0.004629153749334028, + 0.004385560249232324, + 0.004166319691996922, + 0.00396795421864086, + 0.0037876180684444346, + 0.0036229602246830948, + 0.003472021382978767, + 0.003333155636728093, + 0.003204970228055038, + 0.0030862786826087773, + 0.002976063983550409, + 0.0028734493623524663, + 0.0027776749297526936, + ]; + if n <= 30.0 { + return TABLE.get(n as usize).copied().unwrap_or(0.0); + } + let r = 1.0 / n; + let r2 = r * r; + r * (1.0 / 12.0 - r2 * (1.0 / 360.0 - r2 * (1.0 / 1260.0 - r2 * (1.0 / 1680.0 - r2 * (1.0 / 1188.0))))) +} + +/// `x ln(x/np) + np - x` without the cancellation of the direct formula when `x` is near `np` +/// (Loader, 2000). +fn bd0(x: f64, np: f64) -> f64 { + if (x - np).abs() < 0.1 * (x + np) { + let v = (x - np) / (x + np); + let mut s = (x - np) * v; + if s.abs() < f64::MIN_POSITIVE { + return s; + } + let mut ej = 2.0 * x * v; + let v2 = v * v; + for j in 1..1000 { + ej *= v2; + let s1 = s + ej / f64::from(2 * j + 1); + if s1 == s { + return s1; + } + s = s1; + } + } + x * (x / np).ln() + np - x +} + +/// `(1 - p)^e` with the rounding error of `1 - p` (it is not exact for `p < 1/2`) folded back in, +/// which `e` would otherwise multiply. +fn pow_one_minus(p: f64, e: f64) -> f64 { + let hi = 1.0 - p; + let lo = (1.0 - hi) - p; + hi.powf(e) * (e * (lo / hi).ln_1p()).exp() +} + +/// `C(n, k)` for a small `k` as a product of `k` ratios. +fn choose_small(n: f64, k: f64) -> f64 { + let mut c = 1.0; + let mut j = 0.0; + while j < k { + c *= (n - j) / (j + 1.0); + j += 1.0; + } + c +} + +/// Binomial mass at integer `k`, `0 <= k <= n`, by Loader's saddle-point formula: no log-gamma +/// of large numbers, so the relative error stays near 1e-15 for any `n`. fn binom_pmf(k: f64, n: f64, p: f64) -> f64 { if k < 0.0 || k > n { return 0.0; @@ -103,7 +192,60 @@ fn binom_pmf(k: f64, n: f64, p: f64) -> f64 { if p == 1.0 { return if k == n { 1.0 } else { 0.0 }; } - (ln_comb(n, k) + k * p.ln() + (n - k) * (1.0 - p).ln()).exp() + if k == 0.0 { + return if n == 0.0 { 1.0 } else { pow_one_minus(p, n) }; + } + if k == n { + return p.powf(n); + } + // A few successes or a few failures: the product form has no large cancelling terms. + if k < 16.0 { + return choose_small(n, k) * p.powi(k as i32) * pow_one_minus(p, n - k); + } + if n - k < 16.0 { + return choose_small(n, n - k) * p.powf(k) * pow_one_minus(p, n - k); + } + let q = 1.0 - p; + let lc = stirlerr(n) - stirlerr(k) - stirlerr(n - k) - bd0(k, n * p) - bd0(n - k, n * q); + let lf = std::f64::consts::TAU.ln() + k.ln() + (-k / n).ln_1p(); + (lc - 0.5 * lf).exp() +} + +/// Sum of the binomial masses 0..=k for `n <= 10_000`: starts at the term nearest the mode +/// (accurate by the formula above) and walks away from it by the ratio of neighbouring masses, +/// one multiply per term, stopping once the terms no longer matter. Above the mode it sums the +/// upper tail instead and subtracts it from 1. +fn binom_cdf_sum(k: f64, n: f64, p: f64) -> f64 { + let q = 1.0 - p; + if p == 0.0 { + return 1.0; + } + if p == 1.0 { + return 0.0; + } + if k <= n * p { + // mass(i - 1) = mass(i) * i / (n - i + 1) * q / p + let mut t = binom_pmf(k, n, p); + let mut s = t; + let mut i = k; + while i >= 1.0 && t > s * 1e-18 { + t *= i / (n - i + 1.0) * (q / p); + s += t; + i -= 1.0; + } + s.min(1.0) + } else { + // mass(i + 1) = mass(i) * (n - i) / (i + 1) * p / q + let mut t = binom_pmf(k + 1.0, n, p); + let mut s = t; + let mut i = k + 1.0; + while i < n && t > s * 1e-18 { + t *= (n - i) / (i + 1.0) * (p / q); + s += t; + i += 1.0; + } + (1.0 - s).max(0.0) + } } fn binom_cdf(k: f64, n: f64, p: f64) -> f64 { @@ -114,13 +256,7 @@ fn binom_cdf(k: f64, n: f64, p: f64) -> f64 { return 1.0; } if n <= 10_000.0 { - let mut s = 0.0; - let mut i = 0.0; - while i <= k { - s += binom_pmf(i, n, p); - i += 1.0; - } - return s.min(1.0); + return binom_cdf_sum(k, n, p); } if p == 0.0 { return 1.0; @@ -128,14 +264,19 @@ fn binom_cdf(k: f64, n: f64, p: f64) -> f64 { if p == 1.0 { return 0.0; } - reg_inc_beta(1.0 - p, n - k, k + 1.0) + // I_(1-p)(n-k, k+1) has prefactor (n-k) p mass(k). + reg_inc_beta_front(1.0 - p, n - k, k + 1.0, (n - k) * p * binom_pmf(k, n, p)) } +/// Poisson mass at integer `x >= 0` by Loader's formula. fn poisson_pmf(x: f64, m: f64) -> f64 { if m == 0.0 { return if x == 0.0 { 1.0 } else { 0.0 }; } - (-m + x * m.ln() - ln_gamma(x + 1.0)).exp() + if x == 0.0 { + return (-m).exp(); + } + (-stirlerr(x) - bd0(x, m)).exp() / (std::f64::consts::TAU * x).sqrt() } fn poisson_cdf(x: f64, m: f64) -> f64 { @@ -143,15 +284,31 @@ fn poisson_cdf(x: f64, m: f64) -> f64 { return 1.0; } if x <= 1000.0 { - let mut s = 0.0; - let mut i = 0.0; - while i <= x { - s += poisson_pmf(i, m); + if x <= m { + // mass(i - 1) = mass(i) * i / m, downwards from x while the terms matter + let mut t = poisson_pmf(x, m); + let mut s = t; + let mut i = x; + while i >= 1.0 && t > s * 1e-18 { + t *= i / m; + s += t; + i -= 1.0; + } + return s.min(1.0); + } + // Above the mean: 1 minus the upper tail, mass(i + 1) = mass(i) * m / (i + 1). + let mut t = poisson_pmf(x + 1.0, m); + let mut s = t; + let mut i = x + 1.0; + while t > s * 1e-18 { + t *= m / (i + 1.0); + s += t; i += 1.0; } - return s.min(1.0); + return (1.0 - s).max(0.0); } - reg_gamma_q(x + 1.0, m) + // Q(x+1, m) has prefactor m mass(x). + reg_gamma_q_front(x + 1.0, m, m * poisson_pmf(x, m)) } pub(super) fn norm_dist(a: &[Arg], _: C) -> R { diff --git a/crates/functions/src/stat/tests.rs b/crates/functions/src/stat/tests.rs index ba59116cd..f8a8c1890 100644 --- a/crates/functions/src/stat/tests.rs +++ b/crates/functions/src/stat/tests.rs @@ -617,3 +617,233 @@ fn aggregate_helper() { close(aggregate_values_k(9, &vals, &nv(2.0)), 9.0); is_err(aggregate_values_k(14, &vals, &nv(10.0)), CellError::Num); } + +// B1: BINOM.DIST and POISSON.DIST went through ln_gamma differences (relative error 1.1e-13 at +// n = 100, 1.5e-11 at n = 10,000). Reference values: 70-digit Decimal arithmetic on the exact +// double `p` (mass by the neighbour-ratio recurrence from q^n, cdf as the running sum), rounded +// once to f64. Columns: k, n, p, mass, cdf. +#[rustfmt::skip] +const BINOM_REF: &[(f64, f64, f64, f64, f64)] = &[ + (0.0, 10.0, 0.5, 0.0009765625, 0.0009765625), + (1.0, 10.0, 0.5, 0.009765625, 0.0107421875), + (3.0, 10.0, 0.5, 0.1171875, 0.171875), + (5.0, 10.0, 0.5, 0.24609375, 0.623046875), + (7.0, 10.0, 0.5, 0.1171875, 0.9453125), + (9.0, 10.0, 0.5, 0.009765625, 0.9990234375), + (0.0, 10.0, 0.3, 0.028247524900000005, 0.028247524900000005), + (1.0, 10.0, 0.3, 0.12106082100000001, 0.14930834590000003), + (2.0, 10.0, 0.3, 0.23347444050000002, 0.3827827864), + (3.0, 10.0, 0.3, 0.266827932, 0.6496107184000001), + (4.0, 10.0, 0.3, 0.200120949, 0.8497316674000001), + (5.0, 10.0, 0.3, 0.10291934519999998, 0.9526510126), + (7.0, 10.0, 0.3, 0.009001691999999999, 0.9984096136), + (9.0, 10.0, 0.3, 0.00013778099999999996, 0.9999940951), + (0.0, 10.0, 0.999, 1.0000000000000088e-30, 1.0000000000000088e-30), + (1.0, 10.0, 0.999, 9.99000000000008e-27, 9.99100000000008e-27), + (5.0, 10.0, 0.999, 2.507425174812609e-13, 2.50951798425701e-13), + (9.0, 10.0, 0.999, 0.00991035916125875, 0.009955119790251798), + (0.0, 100.0, 0.5, 7.888609052210118e-31, 7.888609052210118e-31), + (1.0, 100.0, 0.5, 7.888609052210118e-29, 7.967495142732219e-29), + (35.0, 100.0, 0.5, 0.0008638556657416528, 0.0017588208614850792), + (45.0, 100.0, 0.5, 0.048474296626430755, 0.18410080866334813), + (50.0, 100.0, 0.5, 0.07958923738717877, 0.5397946186935894), + (55.0, 100.0, 0.5, 0.048474296626430755, 0.8643734879630827), + (65.0, 100.0, 0.5, 0.0008638556657416528, 0.9991050348042566), + (99.0, 100.0, 0.5, 7.888609052210118e-29, 1.0), + (0.0, 100.0, 0.01, 0.3660323412732295, 0.3660323412732295), + (1.0, 100.0, 0.01, 0.3697296376497268, 0.7357619789229563), + (2.0, 100.0, 0.01, 0.1848648188248634, 0.9206267977478196), + (4.0, 100.0, 0.01, 0.014941714856895142, 0.9965676784122455), + (50.0, 100.0, 0.01, 6.103987557173001e-72, 1.0), + (99.0, 100.0, 0.01, 9.900000000000021e-197, 1.0), + (0.0, 100.0, 0.7, 5.15377520732019e-53, 5.15377520732019e-53), + (1.0, 100.0, 0.7, 1.2025475483747105e-50, 1.207701323582031e-50), + (50.0, 100.0, 0.7, 1.302622713144541e-05, 2.2060913327166086e-05), + (56.0, 100.0, 0.7, 0.001028871216633018, 0.0021146172813184595), + (65.0, 100.0, 0.7, 0.04677968235272982, 0.16285828837178737), + (70.0, 100.0, 0.7, 0.08678386475342809, 0.5376602639846381), + (75.0, 100.0, 0.7, 0.04955992276217012, 0.8864298182958186), + (84.0, 100.0, 0.7, 0.0005638655029474704, 0.9995950004580563), + (99.0, 100.0, 0.7, 1.3862042184106021e-14, 0.9999999999999997), + (0.0, 1000.0, 0.5, 9.332636185032189e-302, 9.332636185032189e-302), + (1.0, 1000.0, 0.5, 9.332636185032189e-299, 9.341968821217221e-299), + (453.0, 1000.0, 0.5, 0.00030353498700958767, 0.00162629990658169), + (484.0, 1000.0, 0.5, 0.015123665829158384, 0.1634683412157977), + (500.0, 1000.0, 0.5, 0.0252250181783608, 0.5126125090891804), + (516.0, 1000.0, 0.5, 0.015123665829158384, 0.8516553246133607), + (547.0, 1000.0, 0.5, 0.00030353498700958767, 0.9986772350804279), + (999.0, 1000.0, 0.5, 9.332636185032189e-299, 1.0), + (0.0, 1000.0, 0.3, 1.2532566399657381e-155, 1.2532566399657381e-155), + (1.0, 1000.0, 0.3, 5.371099885567449e-153, 5.383632451967106e-153), + (257.0, 1000.0, 0.3, 0.00030959977077928017, 0.0014835731103312242), + (286.0, 1000.0, 0.3, 0.0174263984921581, 0.1759288949498673), + (300.0, 1000.0, 0.3, 0.027521003821268385, 0.5155935198141203), + (314.0, 1000.0, 0.3, 0.017109129819617666, 0.8414952343056957), + (343.0, 1000.0, 0.3, 0.00036310112605077133, 0.9985016810663147), + (500.0, 1000.0, 0.3, 3.4791609391626546e-40, 1.0), + (999.0, 1000.0, 0.3, 0.0, 1.0), + (0.0, 1000.0, 0.001, 0.36769542477096406, 0.36769542477096406), + (1.0, 1000.0, 0.001, 0.36806348825922325, 0.7357589130301873), + (2.0, 1000.0, 0.001, 0.18403174412961162, 0.9197906571597989), + (4.0, 1000.0, 0.001, 0.01528995542083862, 0.9963631219704782), + (500.0, 1000.0, 0.001, 0.0, 1.0), + (999.0, 1000.0, 0.001, 0.0, 1.0), + (0.0, 10000.0, 0.5, 0.0, 0.0), + (1.0, 10000.0, 0.5, 0.0, 0.0), + (4850.0, 10000.0, 0.5, 8.861480551702583e-05, 0.0013939837588563594), + (4950.0, 10000.0, 0.5, 0.004839495142316416, 0.161087099897656), + (5000.0, 10000.0, 0.5, 0.007978646139382154, 0.5039893230696911), + (5050.0, 10000.0, 0.5, 0.004839495142316416, 0.8437523952446604), + (5150.0, 10000.0, 0.5, 8.861480551702583e-05, 0.9986946310466607), + (9999.0, 10000.0, 0.5, 0.0, 1.0), + (0.0, 10000.0, 0.3, 0.0, 0.0), + (1.0, 10000.0, 0.3, 0.0, 0.0), + (2863.0, 10000.0, 0.3, 9.71726742023389e-05, 0.0013927715709303387), + (2954.0, 10000.0, 0.3, 0.0052756073908582085, 0.16038574844431697), + (3000.0, 10000.0, 0.3, 0.008705361365067056, 0.504932983819298), + (3046.0, 10000.0, 0.3, 0.005244996814931252, 0.8448693061622872), + (3137.0, 10000.0, 0.3, 0.00010232641238302388, 0.9986013755801396), + (5000.0, 10000.0, 0.3, 0.0, 1.0), + (9999.0, 10000.0, 0.3, 0.0, 1.0), + (0.0, 10000.0, 0.0001, 0.3678610464329299, 0.3678610464329299), + (1.0, 10000.0, 0.0001, 0.3678978362165516, 0.7357588826494815), + (2.0, 10000.0, 0.0001, 0.1839489181082758, 0.9197078007577573), + (4.0, 10000.0, 0.0001, 0.015324477632794846, 0.996342452182816), + (5000.0, 10000.0, 0.0001, 0.0, 1.0), + (9999.0, 10000.0, 0.0001, 0.0, 1.0), + (0.0, 10000.0, 0.999, 0.0, 0.0), + (1.0, 10000.0, 0.999, 0.0, 0.0), + (5000.0, 10000.0, 0.999, 0.0, 0.0), + (9981.0, 10000.0, 0.999, 0.0037205945263045544, 0.007158161342006954), + (9987.0, 10000.0, 0.999, 0.07292252318084555, 0.20834867055478343), + (9990.0, 10000.0, 0.999, 0.12517263665023892, 0.5421328760209826), + (9993.0, 10000.0, 0.999, 0.09007019224850273, 0.8699847268232276), + (9999.0, 10000.0, 0.999, 0.00045218564541589875, 0.9999548266540229), + (0.0, 100000.0, 0.5, 0.0, 0.0), + (1.0, 100000.0, 0.5, 0.0, 0.0), + (49526.0, 100000.0, 0.5, 2.821098952769392e-05, 0.00137358872189177), + (49842.0, 100000.0, 0.5, 0.0015314620321580088, 0.15959653953914368), + (50000.0, 100000.0, 0.5, 0.00252312621419674, 0.5012615631070984), + (50158.0, 100000.0, 0.5, 0.0015314620321580088, 0.8419349224930143), + (50474.0, 100000.0, 0.5, 2.821098952769392e-05, 0.9986546222676359), + (99999.0, 100000.0, 0.5, 0.0, 1.0), + (0.0, 100000.0, 0.3, 0.0, 0.0), + (1.0, 100000.0, 0.3, 0.0, 0.0), + (29565.0, 100000.0, 0.3, 3.016598258904529e-05, 0.0013408650955754554), + (29855.0, 100000.0, 0.3, 0.0016703029966719045, 0.15934769308464344), + (30000.0, 100000.0, 0.3, 0.002752954648397428, 0.5015600059172672), + (30145.0, 100000.0, 0.3, 0.0016672321665529399, 0.8423208126454166), + (30435.0, 100000.0, 0.3, 3.067097885268188e-05, 0.9986570625478413), + (50000.0, 100000.0, 0.3, 0.0, 1.0), + (99999.0, 100000.0, 0.3, 0.0, 1.0), + (0.0, 100000.0, 0.99, 0.0, 0.0), + (1.0, 100000.0, 0.99, 0.0, 0.0), + (50000.0, 100000.0, 0.99, 0.0, 0.0), + (98906.0, 100000.0, 0.99, 0.00015962501729101418, 0.001680685850467731), + (98969.0, 100000.0, 0.99, 0.007723834924565287, 0.1660975153530963), + (99000.0, 100000.0, 0.99, 0.012678161323544589, 0.5042686879721229), + (99031.0, 100000.0, 0.99, 0.007885853372194001, 0.8416321140066325), + (99094.0, 100000.0, 0.99, 0.00013276363518518057, 0.9988439452168577), + (99999.0, 100000.0, 0.99, 0.0, 1.0), +]; + +// Columns: x, mean, mass, cdf. The last block has x > 1000, which takes the incomplete-gamma branch. +#[rustfmt::skip] +const POISSON_REF: &[(f64, f64, f64, f64)] = &[ + (0.0, 0.5, 0.6065306597126334, 0.6065306597126334), + (1.0, 0.5, 0.3032653298563167, 0.9097959895689501), + (3.0, 0.5, 0.012636055410679864, 0.9982483774437092), + (0.0, 5.0, 0.006737946999085467, 0.006737946999085467), + (2.0, 5.0, 0.08422433748856833, 0.12465201948308115), + (5.0, 5.0, 0.1754673697678507, 0.6159606548330632), + (10.0, 5.0, 0.018132788707821874, 0.986304731401617), + (25.0, 5.0, 1.294589320478486e-10, 0.9999999999695003), + (70.0, 100.0, 0.00031056089812001654, 0.0009714440282773882), + (100.0, 100.0, 0.039860996809147134, 0.5265621985299984), + (130.0, 100.0, 0.000575252683815463, 0.9982931596294985), + (900.0, 1000.0, 7.516954352125952e-05, 0.0006977673277963068), + (1000.0, 1000.0, 0.012614611348721499, 0.508409367168506), + (1100.0, 1000.0, 9.498944242299507e-05, 0.9991323590365564), + (1050.0, 1000.0, 0.003599490404803648, 0.9439711616363267), + (1200.0, 1000.0, 7.992642848843571e-11, 0.9999999996115061), + (1500.0, 2000.0, 1.878037542328086e-32, 7.468034373687017e-32), + (2000.0, 2000.0, 0.00892024889598624, 0.5059467004517837), + (2100.0, 2000.0, 0.0007442234118916615, 0.9872092799080792), + (2300.0, 2000.0, 4.011826753210622e-12, 0.999999999973895), + (4800.0, 5000.0, 9.987733294132511e-05, 0.002269403878343364), + (5000.0, 5000.0, 0.005641801804664022, 0.5037611677708467), +]; + +fn rel_err(got: Value, want: f64) -> f64 { + let g = got.as_f64().unwrap_or(f64::NAN); + if want == 0.0 { g.abs() } else { ((g - want) / want).abs() } +} + +/// Error bound the grid below must meet: 1e-13, plus 2e-16 per unit of |ln value| for far-tail +/// results (an exp of a sum of large logs cannot do better in double precision). Measured: at +/// most 9e-14 on the grid (the n = 100,000 cumulative cases), 2e-14 for n <= 10,000 away from the +/// far tails, 2e-15 on the Excel-sized central cases. The log-gamma code was off by 1.5e-11. +fn allowed(want: f64) -> f64 { + 1e-13 + 2e-16 * want.abs().ln().abs() +} + +#[test] +fn binom_and_poisson_match_exact_decimal_references() { + // Worst error relative to the allowance (above 1 is a failure) and plain worst error. + let (mut worst, mut worst_plain) = (0.0f64, 0.0f64); + let mut check = |what: &str, got: Value, want: f64| { + let e = rel_err(got, want); + let ratio = e / allowed(want); + if ratio > 1.0 { + println!("{what}: error {e:e}, allowed {:e}", allowed(want)); + } + worst = worst.max(ratio); + if want > 1e-3 { + worst_plain = worst_plain.max(e); + } + }; + for &(k, nn, p, pmf, cdf) in BINOM_REF { + check(&format!("BINOM.DIST({k},{nn},{p},FALSE)"), ev("BINOM.DIST", vec![n(k), n(nn), n(p), b(false)]), pmf); + check(&format!("BINOM.DIST({k},{nn},{p},TRUE)"), ev("BINOM.DIST", vec![n(k), n(nn), n(p), b(true)]), cdf); + } + for &(x, m, pmf, cdf) in POISSON_REF { + check(&format!("POISSON.DIST({x},{m},FALSE)"), ev("POISSON.DIST", vec![n(x), n(m), b(false)]), pmf); + check(&format!("POISSON.DIST({x},{m},TRUE)"), ev("POISSON.DIST", vec![n(x), n(m), b(true)]), cdf); + } + println!("worst error / allowance {worst:.3}; worst relative error where the result is above 1e-3: {worst_plain:e}"); + assert!(worst <= 1.0); + assert!(worst_plain < 1e-13); +} + +#[test] +fn binom_and_poisson_edge_cases() { + for cum in [false, true] { + let pm = |k: f64, nn: f64, p: f64| ev("BINOM.DIST", vec![n(k), n(nn), n(p), b(cum)]); + // p = 0 and p = 1 + close(pm(0.0, 50.0, 0.0), 1.0); + close(pm(3.0, 50.0, 1.0), 0.0); + close(pm(50.0, 50.0, 1.0), 1.0); + // n = 0 + close(pm(0.0, 0.0, 0.3), 1.0); + // k = n is certain when cumulative and p^n otherwise + close(pm(20.0, 20.0, 0.5), if cum { 1.0 } else { 0.5f64.powi(20) }); + close_tol(pm(10_000.0, 10_000.0, 0.999), if cum { 1.0 } else { 0.999f64.powi(10_000) }, 1e-12); + // k = 0 is q^n + close_tol(pm(0.0, 1000.0, 0.001), 0.999f64.powi(1000), 1e-12); + close_tol(pm(0.0, 100_000.0, 0.00001), (1.0f64 - 1e-5).powi(100_000), 1e-9); + } + close(ev("POISSON.DIST", vec![n(0.0), n(0.0), b(false)]), 1.0); + close(ev("POISSON.DIST", vec![n(4.0), n(0.0), b(false)]), 0.0); + close(ev("POISSON.DIST", vec![n(4.0), n(0.0), b(true)]), 1.0); + close_tol(ev("POISSON.DIST", vec![n(0.0), n(3.0), b(true)]), (-3.0f64).exp(), 1e-15); + // Fractions truncate as before. + close(ev("BINOM.DIST", vec![n(6.9), n(10.9), n(0.5), b(false)]), 0.205078125); + // The 10,000 trial switch of the method no longer shows: neighbours agree to 1e-14. + let lo = ev("BINOM.DIST", vec![n(5000.0), n(10_000.0), n(0.5), b(true)]).as_f64().unwrap(); + let hi = ev("BINOM.DIST", vec![n(5000.0), n(10_001.0), n(0.5), b(true)]).as_f64().unwrap(); + let pmf = ev("BINOM.DIST", vec![n(5000.0), n(10_000.0), n(0.5), b(false)]).as_f64().unwrap(); + // cdf(5000; 10,001) = cdf(5000; 10,000) - p * mass(5000; 10,000) by the recurrence on n + assert!(((hi - (lo - 0.5 * pmf)) / hi).abs() < 1e-13, "{lo} {hi}"); + // Inverse stays consistent: the smallest k with cdf >= alpha. + close(ev("BINOM.INV", vec![n(10_000.0), n(0.5), n(0.5)]), 5000.0); +} From 7e8b94fc39f345d8061402c2bdbda5b998018407 Mon Sep 17 00:00:00 2001 From: vangel76 Date: Tue, 3 Nov 2026 23:54:53 +0100 Subject: [PATCH 08/24] BINOM.DIST: keep tiny masses and huge n out of the product form Co-Authored-By: Claude Sonnet 5.5 --- crates/functions/src/stat/distributions.rs | 18 ++++++++++++------ 1 file changed, 12 insertions(+), 6 deletions(-) diff --git a/crates/functions/src/stat/distributions.rs b/crates/functions/src/stat/distributions.rs index 7fa486e93..d2e4e3421 100644 --- a/crates/functions/src/stat/distributions.rs +++ b/crates/functions/src/stat/distributions.rs @@ -166,7 +166,9 @@ fn bd0(x: f64, np: f64) -> f64 { fn pow_one_minus(p: f64, e: f64) -> f64 { let hi = 1.0 - p; let lo = (1.0 - hi) - p; - hi.powf(e) * (e * (lo / hi).ln_1p()).exp() + let r = hi.powf(e) * (e * (lo / hi).ln_1p()).exp(); + // An exponent so large that the two factors under- and overflow: the plain form is enough. + if r.is_nan() { (e * (-p).ln_1p()).exp() } else { r } } /// `C(n, k)` for a small `k` as a product of `k` ratios. @@ -199,11 +201,15 @@ fn binom_pmf(k: f64, n: f64, p: f64) -> f64 { return p.powf(n); } // A few successes or a few failures: the product form has no large cancelling terms. - if k < 16.0 { - return choose_small(n, k) * p.powi(k as i32) * pow_one_minus(p, n - k); - } - if n - k < 16.0 { - return choose_small(n, n - k) * p.powf(k) * pow_one_minus(p, n - k); + // (Not for astronomically large n or a power of p near underflow: the saddle-point form + // below handles those through logs.) + if n < 1e30 && p.powf(k) > 1e-280 { + if k < 16.0 { + return choose_small(n, k) * p.powi(k as i32) * pow_one_minus(p, n - k); + } + if n - k < 16.0 { + return choose_small(n, n - k) * p.powf(k) * pow_one_minus(p, n - k); + } } let q = 1.0 - p; let lc = stirlerr(n) - stirlerr(k) - stirlerr(n - k) - bd0(k, n * p) - bd0(n - k, n * q); From 40e75c79127ffa545040d2d97e6cc7d596bafe9b Mon Sep 17 00:00:00 2001 From: vangel76 Date: Wed, 4 Nov 2026 00:05:58 +0100 Subject: [PATCH 09/24] binom/poisson reference test: fail on errors, NaN and non-zero results for zero references (review point) CodeRabbit pointed out that an error or NaN result passed the accuracy test and that a zero reference allowed any finite value (the bound was infinite). Errors now count as infinite error and a zero reference allows 1e-300. Co-Authored-By: Claude Sonnet 5.5 --- crates/functions/src/stat/tests.rs | 7 ++++++- 1 file changed, 6 insertions(+), 1 deletion(-) diff --git a/crates/functions/src/stat/tests.rs b/crates/functions/src/stat/tests.rs index f8a8c1890..67e626251 100644 --- a/crates/functions/src/stat/tests.rs +++ b/crates/functions/src/stat/tests.rs @@ -775,7 +775,8 @@ const POISSON_REF: &[(f64, f64, f64, f64)] = &[ ]; fn rel_err(got: Value, want: f64) -> f64 { - let g = got.as_f64().unwrap_or(f64::NAN); + // An error or a non-finite result is the worst possible miss, not a skipped one. + let Some(g) = got.as_f64().filter(|g| g.is_finite()) else { return f64::INFINITY }; if want == 0.0 { g.abs() } else { ((g - want) / want).abs() } } @@ -784,6 +785,10 @@ fn rel_err(got: Value, want: f64) -> f64 { /// most 9e-14 on the grid (the n = 100,000 cumulative cases), 2e-14 for n <= 10,000 away from the /// far tails, 2e-15 on the Excel-sized central cases. The log-gamma code was off by 1.5e-11. fn allowed(want: f64) -> f64 { + // A reference that underflows to 0 must come back as (almost) 0, not as any finite value. + if want == 0.0 { + return 1e-300; + } 1e-13 + 2e-16 * want.abs().ln().abs() } From 7e828a982c608c5c1e418065af37bc09729ea53c Mon Sep 17 00:00:00 2001 From: vangel76 Date: Thu, 5 Nov 2026 01:02:21 +0100 Subject: [PATCH 10/24] Tabs: Ctrl-click on the sheet arrows jumps to the first or last visible sheet (#3350) Co-Authored-By: Claude Sonnet 5.5 --- crates/ui-egui/src/tabs.rs | 35 ++++++++++++++++++++++++++++++++--- 1 file changed, 32 insertions(+), 3 deletions(-) diff --git a/crates/ui-egui/src/tabs.rs b/crates/ui-egui/src/tabs.rs index ea11615d0..917dc3b7f 100644 --- a/crates/ui-egui/src/tabs.rs +++ b/crates/ui-egui/src/tabs.rs @@ -22,6 +22,12 @@ fn navigator_items(sheets: &[Arc]) -> Vec<(String, String)> { sheets.iter().enumerate().filter(|(_, s)| s.visibility == Visibility::Visible).map(|(i, s)| (i.to_string(), s.name.clone())).collect() } +/// Workbook index of the first (or, with `last`, the last) visible sheet: where Ctrl-clicking a tab arrow lands. +fn edge_sheet(sheets: &[Arc], last: bool) -> Option { + let mut visible = sheets.iter().enumerate().filter(|(_, s)| s.visibility == Visibility::Visible).map(|(i, _)| i); + if last { visible.next_back() } else { visible.next() } +} + /// Switch to sheet `i` from the tab strip or the navigator. In point mode the formula being typed keeps going on the /// other sheet; otherwise the edit is committed first. fn activate_sheet(app: &mut SheetApp, i: usize) { @@ -117,10 +123,22 @@ pub fn sheet_tabs(app: &mut SheetApp, ui: &mut Ui) { ui.add_space(8.0); if view.show_sheet_tabs { if crate::widgets::icon_button(ui, Icon::Left, t.text_dim, app.ui.language.tr("Previous sheet"), vec2(24.0, 24.0)).clicked() { - app.run_or_alert("sheet.previous", json!({})); + // Ctrl-click jumps to the first visible sheet, like Excel's tab scroll buttons. + let edge = + if ui.input(|i| i.modifiers.command) { app.session.active().and_then(|d| edge_sheet(&d.wb.sheets, false)) } else { None }; + match edge { + Some(i) => activate_sheet(app, i), + None => app.run_or_alert("sheet.previous", json!({})), + } } if crate::widgets::icon_button(ui, Icon::Right, t.text_dim, app.ui.language.tr("Next sheet"), vec2(24.0, 24.0)).clicked() { - app.run_or_alert("sheet.next", json!({})); + // Ctrl-click jumps to the last visible sheet. + let edge = + if ui.input(|i| i.modifiers.command) { app.session.active().and_then(|d| edge_sheet(&d.wb.sheets, true)) } else { None }; + match edge { + Some(i) => activate_sheet(app, i), + None => app.run_or_alert("sheet.next", json!({})), + } } let all = crate::widgets::icon_button(ui, Icon::SheetList, t.text_dim, app.ui.language.tr("All sheets"), vec2(24.0, 24.0)); all.widget_info(|| egui::WidgetInfo::labeled(egui::Role::Button, true, app.ui.language.tr("All sheets"))); @@ -522,7 +540,8 @@ fn status_bar_menu(app: &mut SheetApp, ui: &mut Ui) { #[cfg(test)] mod tests { use super::{ - MIN_TRACK, NEW_SHEET_WIDTH, SPLIT_WIDTH, activate_sheet, dragged_share, drop_copies_sheet, navigator_items, rename_needs_focus, tab_drop, + MIN_TRACK, NEW_SHEET_WIDTH, SPLIT_WIDTH, activate_sheet, dragged_share, drop_copies_sheet, edge_sheet, navigator_items, rename_needs_focus, + tab_drop, }; use crate::SheetApp; use egui::Modifiers; @@ -614,6 +633,16 @@ mod tests { assert_eq!(navigator_items(&sheets), vec![("0".to_string(), "Data".to_string()), ("3".to_string(), "Summary".to_string())]); } + #[test] + fn ctrl_click_on_the_tab_arrows_lands_on_the_first_and_last_visible_sheet() { + let mut hidden = Sheet::new("Hidden"); + hidden.visibility = Visibility::Hidden; + let sheets = [hidden.clone(), Sheet::new("A"), Sheet::new("B"), Sheet::new("C"), hidden].map(Arc::new); + assert_eq!(edge_sheet(&sheets, false), Some(1)); + assert_eq!(edge_sheet(&sheets, true), Some(3)); + assert_eq!(edge_sheet(&[], true), None); + } + #[test] fn a_rejected_commit_keeps_the_sheet_and_the_edit() { let mut session = gridcraft_engine::Session::new(); From 23c909e4775ac8d28e722ada888980f4faefe42f Mon Sep 17 00:00:00 2001 From: vangel76 Date: Thu, 5 Nov 2026 01:53:24 +0100 Subject: [PATCH 11/24] chart perf tests: import LegendPos (legend module now imports it itself) Co-Authored-By: Claude Sonnet 5.5 --- crates/chart/src/render_perf_tests.rs | 1 + 1 file changed, 1 insertion(+) diff --git a/crates/chart/src/render_perf_tests.rs b/crates/chart/src/render_perf_tests.rs index 461c2f159..ce15e7861 100644 --- a/crates/chart/src/render_perf_tests.rs +++ b/crates/chart/src/render_perf_tests.rs @@ -7,6 +7,7 @@ use super::kinds::squarify; use super::legend::swatch_w; use super::*; use crate::{ApproxMeasure, Measure}; +use gridcraft_model::LegendPos; /// The pre-rewrite `Ctx::fit`: one `Measure` call per candidate length, longest first. fn fit_reference(ctx: &Ctx, s: &str, size: f32, maxw: f32) -> String { From bc1625248e325179f8da7d14f06b99b04c99a8d7 Mon Sep 17 00:00:00 2001 From: vangel76 Date: Sat, 7 Nov 2026 16:25:23 +0100 Subject: [PATCH 12/24] Chart: cap category levels in the clean-data check, recompute overflowing moving-average windows, keep disjoint stacked hairline spans apart Co-Authored-By: Claude Sonnet 5.5 --- crates/chart/src/render.rs | 2 ++ crates/chart/src/render/cartesian.rs | 6 +++++- crates/chart/src/tests/budget.rs | 27 +++++++++++++++++++++++++++ crates/chart/src/trend.rs | 15 ++++++++++++++- 4 files changed, 48 insertions(+), 2 deletions(-) diff --git a/crates/chart/src/render.rs b/crates/chart/src/render.rs index 3fa9f8431..3abf8bf14 100644 --- a/crates/chart/src/render.rs +++ b/crates/chart/src/render.rs @@ -141,6 +141,8 @@ fn needs_sanitizing(d: &ChartData) -> bool { let mut trendlines = 0usize; d.palette.len() > MAX_POINTS || long(d.categories.len()) + || d.category_levels.len() > 3 + || d.category_levels.iter().any(|l| l.len() > MAX_POINTS) || d.series.len() > MAX_SERIES || d.series.iter().any(|s| { trendlines += s.trendlines.len(); diff --git a/crates/chart/src/render/cartesian.rs b/crates/chart/src/render/cartesian.rs index 6689bf202..44c7a3dc9 100644 --- a/crates/chart/src/render/cartesian.rs +++ b/crates/chart/src/render/cartesian.rs @@ -67,6 +67,10 @@ impl BarRun { fn new(pos: f32, lo: f32, hi: f32) -> Self { BarRun { bucket: pos.floor(), a0: pos, a1: pos, lo, hi, merged: false } } + /// Whether `lo..hi` touches the run's value span; a disjoint bar would be joined by a filled gap. + fn touches(&self, lo: f32, hi: f32) -> bool { + lo <= self.hi && hi >= self.lo + } fn add(&mut self, pos: f32, lo: f32, hi: f32) { self.merged = true; self.a0 = self.a0.min(pos); @@ -135,7 +139,7 @@ impl BarSink { return; } match &mut self.run { - Some(m) if m.bucket == pos.floor() => m.add(pos, lo, lo + len), + Some(m) if m.bucket == pos.floor() && m.touches(lo, lo + len) => m.add(pos, lo, lo + len), run => { if let Some(m) = run.replace(BarRun::new(pos, lo, lo + len)) { m.draw(ctx, self.bw, self.horizontal, self.color, band); diff --git a/crates/chart/src/tests/budget.rs b/crates/chart/src/tests/budget.rs index 54d33aa7d..6c5a10879 100644 --- a/crates/chart/src/tests/budget.rs +++ b/crates/chart/src/tests/budget.rs @@ -158,6 +158,26 @@ fn clustered_hairline_series_keep_a_visible_strip() { } } +#[test] +fn stacked_hairline_bars_do_not_fill_the_gap_between_disjoint_spans() { + // Series 1 alternates +10 (stacked on the 100) and -10 (below zero): its two spans in one + // pixel column are 100..110 and -10..0, and the 100 units between them must stay empty. + let mut c = chart(ChartKind::ColumnStacked); + c.data_labels = false; + c.legend = LegendPos::None; + let mut d = data_of(2, 20_000); + for v in d.series[0].values.iter_mut() { + *v = Some(100.0); + } + for (i, v) in d.series[1].values.iter_mut().enumerate() { + *v = Some(if i % 2 == 0 { 10.0 } else { -10.0 }); + } + let prims = render(&c, &d, 400.0, 300.0, &ApproxMeasure); + let rects = series_rects(&prims, d.series[1].color); + let tallest = rects.iter().map(|r| r.3).fold(0.0f32, f32::max); + assert!(!rects.is_empty() && tallest < 60.0, "tallest merged bar is {tallest} px"); +} + #[test] fn clean_data_is_rendered_without_a_copy() { let d = data_of(3, 50); @@ -181,6 +201,13 @@ fn clean_data_is_rendered_without_a_copy() { let got = sanitized(&many); assert_eq!(got.series.len(), 255); assert_eq!(*got, sanitize(&many)); + + // Too many category levels take the copying path too. + let mut levels = data_of(3, 50); + levels.category_levels = vec![vec!["a".into()]; 5]; + let got = sanitized(&levels); + assert!(matches!(got, Cow::Owned(_))); + assert_eq!(got.category_levels.len(), 3); } #[test] diff --git a/crates/chart/src/trend.rs b/crates/chart/src/trend.rs index 6b22ed7ad..818c75796 100644 --- a/crates/chart/src/trend.rs +++ b/crates/chart/src/trend.rs @@ -294,7 +294,11 @@ pub(crate) fn moving_average(pts: &[(f64, f64)], period: u32) -> Vec<(f64, f64)> (p - 1..n) .filter_map(|i| { let s = i + 1 - p; - let sum = tail.get(s)? + if s.is_multiple_of(p) { 0.0 } else { *head.get(i)? }; + let mut sum = tail.get(s)? + if s.is_multiple_of(p) { 0.0 } else { *head.get(i)? }; + if !sum.is_finite() { + // A block sum overflowed although the window itself may not: sum it directly. + sum = pts.get(s..=i)?.iter().map(|q| q.1).sum(); + } Some((pts.get(i)?.0, sum / p as f64)) }) .collect() @@ -390,6 +394,15 @@ mod tests { } } + #[test] + fn moving_average_recomputes_a_window_whose_block_sums_overflow() { + let pts: Vec<(f64, f64)> = [0.0, -1e308, 1e308, 1e308, -1e308].iter().enumerate().map(|(i, y)| (i as f64, *y)).collect(); + let fast = moving_average(&pts, 4); + let slow = moving_average_naive(&pts, 4); + assert_eq!(fast, slow); + assert!(fast.iter().all(|q| q.1.is_finite())); + } + /// The trendline fit as it was written with a `Vec` of powers per data point. fn fit_reference(t: &Trendline, pts: &[(f64, f64)]) -> Option { let kind = t.kind; From 76002fe0bd1eadad4c59fe40778fbde126297a67 Mon Sep 17 00:00:00 2001 From: vangel76 Date: Sat, 7 Nov 2026 16:58:19 +0100 Subject: [PATCH 13/24] Reuse remembered range sums for every COUNT block COUNT never reads the running sum, so the zero-sum gate that keeps SUM and AVERAGE additions in order does not apply to it. Co-Authored-By: Claude Sonnet 5.5 --- crates/calc/src/eval/call.rs | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/crates/calc/src/eval/call.rs b/crates/calc/src/eval/call.rs index 46f0d02c9..8ac58c8d1 100644 --- a/crates/calc/src/eval/call.rs +++ b/crates/calc/src/eval/call.rs @@ -384,7 +384,8 @@ impl<'h> Evaluator<'h> { fn aggregate_area(&mut self, kind: Agg, a: &Area, acc: &mut Acc) -> Result<(), Ev> { if !a.range.is_single() { // Adding the block's remembered sum to anything but +0.0 would add in another order. - if acc.sum.to_bits() == 0 + // COUNT never reads the sum, so it may reuse every block. + if (kind == Agg::Count || acc.sum.to_bits() == 0) && let Some((sum, count)) = self.host.range_sum(a.sheet, a.range) { acc.sum = sum; From 12f69f99bf40306008c243a89b44c2f122d30f06 Mon Sep 17 00:00:00 2001 From: vangel76 Date: Sat, 7 Nov 2026 17:20:53 +0100 Subject: [PATCH 14/24] BINOM.DIST: keep small-k masses whose complement power underflows The small-k product also needs (1 - p)^(n - k) to stay a normal number; otherwise it returned 0 for BINOM.DIST(15,1100,0.5,FALSE) (exact 2.1367646606291324e-298) and the cumulative form built on it. Those cases now use the logarithmic formula. Reference grid gets k = 10..25 at n = 1100. Co-Authored-By: Claude Sonnet 5.5 --- crates/functions/src/stat/distributions.rs | 9 ++++++--- crates/functions/src/stat/tests.rs | 7 +++++++ 2 files changed, 13 insertions(+), 3 deletions(-) diff --git a/crates/functions/src/stat/distributions.rs b/crates/functions/src/stat/distributions.rs index d2e4e3421..0853675a6 100644 --- a/crates/functions/src/stat/distributions.rs +++ b/crates/functions/src/stat/distributions.rs @@ -201,11 +201,14 @@ fn binom_pmf(k: f64, n: f64, p: f64) -> f64 { return p.powf(n); } // A few successes or a few failures: the product form has no large cancelling terms. - // (Not for astronomically large n or a power of p near underflow: the saddle-point form - // below handles those through logs.) + // (Not for astronomically large n, or when either power underflows: the + // saddle-point form below handles those through logs.) if n < 1e30 && p.powf(k) > 1e-280 { if k < 16.0 { - return choose_small(n, k) * p.powi(k as i32) * pow_one_minus(p, n - k); + let rest = pow_one_minus(p, n - k); + if rest >= f64::MIN_POSITIVE { + return choose_small(n, k) * p.powi(k as i32) * rest; + } } if n - k < 16.0 { return choose_small(n, n - k) * p.powf(k) * pow_one_minus(p, n - k); diff --git a/crates/functions/src/stat/tests.rs b/crates/functions/src/stat/tests.rs index 0eb0de0d1..94b35e1f2 100644 --- a/crates/functions/src/stat/tests.rs +++ b/crates/functions/src/stat/tests.rs @@ -745,6 +745,13 @@ const BINOM_REF: &[(f64, f64, f64, f64, f64)] = &[ (99031.0, 100000.0, 0.99, 0.007885853372194001, 0.8416321140066325), (99094.0, 100000.0, 0.99, 0.00013276363518518057, 0.9988439452168577), (99999.0, 100000.0, 0.99, 0.0, 1.0), + // Small k where (1 - p)^(n - k) underflows although the mass is a normal number. + (10.0, 1100.0, 0.5, 5.050689010687844e-308, 5.097367491037505e-308), + (14.0, 1100.0, 0.5, 2.9513324041838846e-300, 2.989803284040754e-300), + (15.0, 1100.0, 0.5, 2.1367646606291324e-298, 2.16666269346954e-298), + (16.0, 1100.0, 0.5, 1.4489935354891303e-296, 1.4706601624238258e-296), + (20.0, 1100.0, 0.5, 1.711087104904223e-289, 1.7433098849775326e-289), + (25.0, 1100.0, 0.5, 3.906996229178926e-281, 3.999839082741416e-281), ]; // Columns: x, mean, mass, cdf. The last block has x > 1000, which takes the incomplete-gamma branch. From 87cb0f11d2a65e929c0d1a2af2e81f58720365ae Mon Sep 17 00:00:00 2001 From: vangel76 Date: Sat, 7 Nov 2026 17:22:54 +0100 Subject: [PATCH 15/24] Tab strip: reveal the edge tab after a repeated Ctrl-click on an arrow When the edge sheet was already active nothing changed, so the strip was not scrolled back to it. A Ctrl-click on the arrow now asks for the reveal itself. Co-Authored-By: Claude Sonnet 5.5 --- crates/ui-egui/src/tabs.rs | 27 ++++++++++++++-- crates/ui-egui/tests/it/sheet_tabs.rs | 44 +++++++++++++++++++++++++++ 2 files changed, 68 insertions(+), 3 deletions(-) diff --git a/crates/ui-egui/src/tabs.rs b/crates/ui-egui/src/tabs.rs index 917dc3b7f..462dc2e4c 100644 --- a/crates/ui-egui/src/tabs.rs +++ b/crates/ui-egui/src/tabs.rs @@ -61,6 +61,18 @@ fn tab_drop(i: usize, steps: i64, copy: bool) -> Option { Some(json!({"sheet": i, "to": to})) } +/// Where a request to scroll the active tab into view is kept until the tabs are drawn. +fn reveal_request_id(scope: egui::Id) -> egui::Id { + scope.with("worksheet_tabs_reveal_request") +} + +/// Asks the tab strip to bring the active tab into view in this frame, even when the active sheet +/// did not change (a Ctrl-click on an arrow whose edge sheet is already active). +fn request_reveal(ui: &Ui) { + let id = reveal_request_id(ui.scope_id()); + ui.ctx().data_mut(|data| data.insert_temp(id, true)); +} + /// The New sheet (+) button and the gap before it. const NEW_SHEET_WIDTH: f32 = 32.0; /// The handle between the tabs and the horizontal scrollbar. @@ -127,7 +139,10 @@ pub fn sheet_tabs(app: &mut SheetApp, ui: &mut Ui) { let edge = if ui.input(|i| i.modifiers.command) { app.session.active().and_then(|d| edge_sheet(&d.wb.sheets, false)) } else { None }; match edge { - Some(i) => activate_sheet(app, i), + Some(i) => { + activate_sheet(app, i); + request_reveal(ui); + } None => app.run_or_alert("sheet.previous", json!({})), } } @@ -136,7 +151,10 @@ pub fn sheet_tabs(app: &mut SheetApp, ui: &mut Ui) { let edge = if ui.input(|i| i.modifiers.command) { app.session.active().and_then(|d| edge_sheet(&d.wb.sheets, true)) } else { None }; match edge { - Some(i) => activate_sheet(app, i), + Some(i) => { + activate_sheet(app, i); + request_reveal(ui); + } None => app.run_or_alert("sheet.next", json!({})), } } @@ -181,10 +199,13 @@ pub fn sheet_tabs(app: &mut SheetApp, ui: &mut Ui) { // allow browsing other tabs without snapping back on every frame. let reveal_key = egui::Id::unique((d.uid, d.wb.active_sheet(), &sheets, tabs_width.to_bits())); let reveal_id = ui.scope_id().with("worksheet_tabs_reveal"); + let request_id = reveal_request_id(ui.scope_id()); let reveal_active = ui.ctx().data_mut(|data| { let changed = data.get_temp::(reveal_id) != Some(reveal_key); data.insert_temp(reveal_id, reveal_key); - changed + // A Ctrl-click on an arrow whose edge sheet is already active changes nothing above, + // yet asks to bring that tab into view. + changed | data.remove_temp::(request_id).unwrap_or(false) }); if view.show_sheet_tabs { ui.allocate_ui_with_layout(vec2(tabs_width, 30.0), egui::Layout::left_to_right(egui::Align::Center), |ui| { diff --git a/crates/ui-egui/tests/it/sheet_tabs.rs b/crates/ui-egui/tests/it/sheet_tabs.rs index 460162e67..725e761c6 100644 --- a/crates/ui-egui/tests/it/sheet_tabs.rs +++ b/crates/ui-egui/tests/it/sheet_tabs.rs @@ -58,6 +58,20 @@ fn click(h: &mut Harness, at: egui::Pos2) { h.run_steps(20); } +fn ctrl_click(h: &mut Harness, at: egui::Pos2) { + let m = Modifiers::COMMAND; + h.input_mut().events.extend([ + Event::ModifiersChanged(m), + Event::PointerMoved(at), + Event::PointerButton { pos: at, button: PointerButton::Primary, pressed: true, modifiers: m }, + ]); + h.step(); + h.input_mut().events.push(Event::PointerButton { pos: at, button: PointerButton::Primary, pressed: false, modifiers: m }); + h.run_steps(20); + h.input_mut().events.push(Event::ModifiersChanged(Modifiers::NONE)); + h.run_steps(2); +} + #[test] fn opening_and_activating_late_sheets_reveals_their_tabs() { for width in [500.0, 1000.0] { @@ -85,6 +99,36 @@ fn navigation_arrows_reveal_the_selected_tab() { assert_visible(&h, "Tab29"); } +/// A Ctrl/Cmd-click on the Next arrow jumps to the last sheet; repeated after the strip was scrolled +/// away from it, the already active edge tab is brought back into view. +#[test] +fn repeated_ctrl_click_on_the_arrow_reveals_the_active_edge_tab() { + let mut h = harness(700.0, 0); + let (_, clip) = tab(&h, "Tab00").unwrap(); + let next = pos2(clip.left() - 44.0, clip.center().y); + ctrl_click(&mut h, next); + assert_eq!(h.state().session.doc().unwrap().wb.active_sheet(), 29); + assert_visible(&h, "Tab29"); + // Browse the strip back to the start without changing the selection. + h.input_mut().events.extend([Event::PointerMoved(clip.center())]); + for _ in 0..3 { + h.input_mut().events.push(Event::MouseWheel { + unit: egui::MouseWheelUnit::Point, + source: egui::MouseWheelSource::Unknown, + phase: egui::TouchPhase::Move, + delta: vec2(0.0, 600.0), + modifiers: Modifiers::NONE, + }); + h.run_steps(10); + } + h.run_steps(40); + assert!(!tab(&h, "Tab29").is_some_and(|(r, c)| c.contains_rect(r)), "the strip was scrolled away from the active tab"); + assert_eq!(h.state().session.doc().unwrap().wb.active_sheet(), 29); + ctrl_click(&mut h, next); + assert_eq!(h.state().session.doc().unwrap().wb.active_sheet(), 29); + assert_visible(&h, "Tab29"); +} + #[test] /// Horizontal wheel/trackpad browsing leaves the workbook unchanged, does not snap back on idle frames, and activating a sheet still reveals its tab. fn manual_wheel_and_trackpad_browsing_does_not_change_selection_or_snap_back() { From 05cffdceb1f8356c64ade34858edbcfde32ea608 Mon Sep 17 00:00:00 2001 From: vangel76 Date: Sat, 7 Nov 2026 17:27:07 +0100 Subject: [PATCH 16/24] Tab strip: test the Previous arrow reveal for the already active first tab Co-Authored-By: Claude Sonnet 5.5 --- crates/ui-egui/tests/it/sheet_tabs.rs | 29 +++++++++++++++++++++++++++ 1 file changed, 29 insertions(+) diff --git a/crates/ui-egui/tests/it/sheet_tabs.rs b/crates/ui-egui/tests/it/sheet_tabs.rs index 725e761c6..6703531a0 100644 --- a/crates/ui-egui/tests/it/sheet_tabs.rs +++ b/crates/ui-egui/tests/it/sheet_tabs.rs @@ -129,6 +129,35 @@ fn repeated_ctrl_click_on_the_arrow_reveals_the_active_edge_tab() { assert_visible(&h, "Tab29"); } +/// The same for the Previous arrow and the first sheet. +#[test] +fn repeated_ctrl_click_on_the_previous_arrow_reveals_the_first_tab() { + let mut h = harness(700.0, 29); + let (_, clip) = tab(&h, "Tab29").unwrap(); + let prev = pos2(clip.left() - 68.0, clip.center().y); + ctrl_click(&mut h, prev); + assert_eq!(h.state().session.doc().unwrap().wb.active_sheet(), 0); + assert_visible(&h, "Tab00"); + // Browse the strip to its end without changing the selection. + h.input_mut().events.extend([Event::PointerMoved(clip.center())]); + for _ in 0..3 { + h.input_mut().events.push(Event::MouseWheel { + unit: egui::MouseWheelUnit::Point, + source: egui::MouseWheelSource::Unknown, + phase: egui::TouchPhase::Move, + delta: vec2(0.0, -600.0), + modifiers: Modifiers::NONE, + }); + h.run_steps(10); + } + h.run_steps(40); + assert!(!tab(&h, "Tab00").is_some_and(|(r, c)| c.contains_rect(r)), "the strip was scrolled away from the active tab"); + assert_eq!(h.state().session.doc().unwrap().wb.active_sheet(), 0); + ctrl_click(&mut h, prev); + assert_eq!(h.state().session.doc().unwrap().wb.active_sheet(), 0); + assert_visible(&h, "Tab00"); +} + #[test] /// Horizontal wheel/trackpad browsing leaves the workbook unchanged, does not snap back on idle frames, and activating a sheet still reveals its tab. fn manual_wheel_and_trackpad_browsing_does_not_change_selection_or_snap_back() { From 6a7cf9cf667956b486a719977106ff1f9152a7c4 Mon Sep 17 00:00:00 2001 From: vangel76 Date: Sat, 7 Nov 2026 17:31:27 +0100 Subject: [PATCH 17/24] Chart: bound the category axis by the point cap, not the raw cell count Read only the cells the category levels use and cut the category dimension at the per-series cap, so a multi-level axis keeps all its levels. Co-Authored-By: Claude Sonnet 5.5 --- crates/chart/src/resolve.rs | 36 +++++++++++++++++++++---- crates/chart/src/tests/budget.rs | 45 ++++++++++++++++++++++++++++++++ 2 files changed, 76 insertions(+), 5 deletions(-) diff --git a/crates/chart/src/resolve.rs b/crates/chart/src/resolve.rs index 1ee635a14..bd6849435 100644 --- a/crates/chart/src/resolve.rs +++ b/crates/chart/src/resolve.rs @@ -189,8 +189,29 @@ fn value_text(v: &Value) -> String { /// taller than a row of values) groups its innermost labels, year beside month. Each outer level /// has one entry per category, non-empty where a group starts, as the cells are. fn category_table(wb: &Workbook, sheet: usize, formula: &str, values: &str, limit: usize) -> (Vec, Vec>) { - let vals = eval_list_in(wb, sheet, formula, limit, 0); let cells: Option<(usize, RangeRef)> = parse_ref(wb, sheet, formula); + let value_range = parse_ref(wb, sheet, values).map(|(_, r)| r); + // Bound the category dimension by `limit` and read only the cells the levels use (the + // innermost label line and up to three outer ones), so that every kept category keeps its + // cells in every level. Counting raw cells would cut a two-level axis in the middle. + let levels = cells.and_then(|(_, r)| category_levels(&r, value_range.as_ref())); + let (cells, read) = match (cells, levels) { + (Some((sh, r)), Some(lv)) => { + let cut = u32::try_from(limit.saturating_sub(1)).unwrap_or(u32::MAX); + let outer = lv.outer; + let end = if lv.by_columns { CellRef::new(r.end.row.min(r.start.row.saturating_add(cut)), r.end.col) } else { CellRef::new(r.end.row, r.end.col.min(r.start.col.saturating_add(cut))) }; + let start = if lv.by_columns { CellRef::new(r.start.row, r.start.col.max(end.col.saturating_sub(outer))) } else { CellRef::new(r.start.row.max(end.row.saturating_sub(outer)), r.start.col) }; + let r = RangeRef::new(start, end); + let prefix = formula.rfind('!').and_then(|i| formula.get(..=i)).unwrap_or(""); + let cap = (r.width() as usize).saturating_mul(r.height() as usize); + (Some((sh, r)), Some((format!("{prefix}{r}"), cap))) + } + (c, _) => (c, None), + }; + let vals = match &read { + Some((f, cap)) => eval_list_in(wb, sheet, f, *cap, 0), + None => eval_list_in(wb, sheet, formula, limit, 0), + }; let src = cells.and_then(|(sh, r)| wb.sheet(sh).map(|s| (s, r))); let texts: Vec = vals .iter() @@ -210,8 +231,7 @@ fn category_table(wb: &Workbook, sheet: usize, formula: &str, values: &str, limi }) .collect(); let Some((_, r)) = src else { return (texts, vec![]) }; - let value_range = parse_ref(wb, sheet, values).map(|(_, r)| r); - let Some(levels) = category_levels(&r, value_range.as_ref()) else { return (texts, vec![]) }; + let Some(levels) = levels else { return (texts, vec![]) }; let (w, h) = (r.width().max(1) as usize, r.height().max(1) as usize); if texts.len() != w.saturating_mul(h) { return (texts, vec![]); // truncated or not a plain block: a flat list @@ -252,7 +272,7 @@ fn series_name(wb: &Workbook, sheet: usize, name: &str) -> String { t.to_string() } -fn effective_kind(chart_kind: ChartKind, series_kind: Option, i: usize) -> ChartKind { +pub(crate) fn effective_kind(chart_kind: ChartKind, series_kind: Option, i: usize) -> ChartKind { match (chart_kind, series_kind) { (ChartKind::Combo, Some(k)) if k != ChartKind::Combo => k, (ChartKind::Combo, _) => { @@ -267,9 +287,15 @@ fn effective_kind(chart_kind: ChartKind, series_kind: Option, i: usiz } } +/// The kind series `i` is drawn as: its own kind in a combo chart (first series a column, the +/// rest lines when it has none), else the chart's. The one rule shared by resolving and drawing. +pub(crate) fn plot_kind(chart_kind: ChartKind, series_kind: Option, i: usize) -> ChartKind { + if chart_kind == ChartKind::Combo { effective_kind(chart_kind, series_kind, i) } else { chart_kind } +} + /// The kind series `i` of `chart` is drawn as: its own kind in a combo chart, else the chart's. pub fn series_plot_kind(chart: &Chart, i: usize) -> ChartKind { - if chart.kind == ChartKind::Combo { effective_kind(chart.kind, chart.series.get(i).and_then(|s| s.kind), i) } else { chart.kind } + plot_kind(chart.kind, chart.series.get(i).and_then(|s| s.kind), i) } /// Evaluates a chart's series formulas: names, values, categories (default "1", "2", …), X diff --git a/crates/chart/src/tests/budget.rs b/crates/chart/src/tests/budget.rs index 6c5a10879..fad349ba6 100644 --- a/crates/chart/src/tests/budget.rs +++ b/crates/chart/src/tests/budget.rs @@ -232,3 +232,48 @@ fn resolve_keeps_every_series_within_the_point_budget() { let counts: Vec = resolve(&wb, 0, &c).series.iter().map(|s| s.trendlines.len()).collect(); assert_eq!(counts, vec![9, 7, 0, 0]); } + +/// 100 series lower the per-series cap to 20,000, below the number of category cells of a +/// two-level axis (#1155). Each kept category keeps its inner label and its outer level. +fn two_level_chart(n: u32, vertical: bool) -> ChartData { + let at = |i: u32, level: u32| if vertical { CellRef::new(i, level) } else { CellRef::new(level, i) }; + let mut wb = Workbook::new(); + if let Some(s) = wb.sheet_mut(0) { + for i in 0..n { + s.set_value(at(i, 0), Value::Text(format!("Y{}", i / 3).into())); + s.set_value(at(i, 1), Value::Text(format!("C{i}").into())); + s.set_value(at(i, 2), Value::Number(f64::from(i))); + } + } + let cats = format!("{}:{}", at(0, 0), at(n - 1, 1)); + let values = format!("{}:{}", at(0, 2), at(n - 1, 2)); + let mut c = chart(ChartKind::ColumnClustered); + c.series = (0..100).map(|_| Series { values: values.clone(), categories: Some(cats.clone()), ..Series::default() }).collect(); + resolve(&wb, 0, &c) +} + +#[test] +fn multi_level_categories_keep_their_levels_under_the_point_cap() { + for vertical in [true, false] { + let d = two_level_chart(15_000, vertical); + assert_eq!(d.series[0].values.len(), 15_000, "vertical {vertical}"); + assert_eq!(d.categories.len(), 15_000, "vertical {vertical}"); + assert_eq!(d.categories[0], "C0"); + assert_eq!(d.categories[14_999], "C14999"); + assert_eq!(d.category_levels.len(), 1, "one outer level, vertical {vertical}"); + assert_eq!(d.category_levels[0].len(), 15_000); + assert_eq!(d.category_levels[0][3], "Y1"); + } +} + +#[test] +fn multi_level_categories_longer_than_the_cap_are_cut_together_with_their_levels() { + // 25,000 categories against a cap of 20,000. + let d = two_level_chart(25_000, true); + let cap = points_per_series(100); + assert_eq!(d.categories.len(), cap); + assert_eq!(d.category_levels.len(), 1); + assert_eq!(d.category_levels[0].len(), cap); + assert_eq!(d.categories[cap - 1], format!("C{}", cap - 1)); + assert_eq!(d.category_levels[0][cap - 3], format!("Y{}", (cap - 3) / 3)); +} From 9ba79c3e3850fc14e4ea12e1c2c76d3ef2952206 Mon Sep 17 00:00:00 2001 From: vangel76 Date: Sat, 7 Nov 2026 17:31:41 +0100 Subject: [PATCH 18/24] fmt --- crates/chart/src/resolve.rs | 12 ++++++++++-- 1 file changed, 10 insertions(+), 2 deletions(-) diff --git a/crates/chart/src/resolve.rs b/crates/chart/src/resolve.rs index bd6849435..11213a5e7 100644 --- a/crates/chart/src/resolve.rs +++ b/crates/chart/src/resolve.rs @@ -199,8 +199,16 @@ fn category_table(wb: &Workbook, sheet: usize, formula: &str, values: &str, limi (Some((sh, r)), Some(lv)) => { let cut = u32::try_from(limit.saturating_sub(1)).unwrap_or(u32::MAX); let outer = lv.outer; - let end = if lv.by_columns { CellRef::new(r.end.row.min(r.start.row.saturating_add(cut)), r.end.col) } else { CellRef::new(r.end.row, r.end.col.min(r.start.col.saturating_add(cut))) }; - let start = if lv.by_columns { CellRef::new(r.start.row, r.start.col.max(end.col.saturating_sub(outer))) } else { CellRef::new(r.start.row.max(end.row.saturating_sub(outer)), r.start.col) }; + let end = if lv.by_columns { + CellRef::new(r.end.row.min(r.start.row.saturating_add(cut)), r.end.col) + } else { + CellRef::new(r.end.row, r.end.col.min(r.start.col.saturating_add(cut))) + }; + let start = if lv.by_columns { + CellRef::new(r.start.row, r.start.col.max(end.col.saturating_sub(outer))) + } else { + CellRef::new(r.start.row.max(end.row.saturating_sub(outer)), r.start.col) + }; let r = RangeRef::new(start, end); let prefix = formula.rfind('!').and_then(|i| formula.get(..=i)).unwrap_or(""); let cap = (r.width() as usize).saturating_mul(r.height() as usize); From 9354fdb0c4d155b1124facda01142564b24d7887 Mon Sep 17 00:00:00 2001 From: vangel76 Date: Sat, 7 Nov 2026 17:37:20 +0100 Subject: [PATCH 19/24] Chart: fit labels with a line break by a linear scan The width of a prefix plus an ellipsis is not monotone when the text has a line break, so the binary search could drop a label that fits. Also pass the series index to plot_kind in the data table (upstream/main does not compile without it). Co-Authored-By: Claude Sonnet 5.5 --- crates/chart/src/render.rs | 6 ++++ crates/chart/src/render/table.rs | 2 +- crates/chart/src/render_perf_tests.rs | 41 +++++++++++++++++++++++++++ crates/ui-egui/src/chartview.rs | 27 ++++++++++++++++++ 4 files changed, 75 insertions(+), 1 deletion(-) diff --git a/crates/chart/src/render.rs b/crates/chart/src/render.rs index 1175ffc19..65b6d4f1a 100644 --- a/crates/chart/src/render.rs +++ b/crates/chart/src/render.rs @@ -71,6 +71,8 @@ impl Ctx<'_> { /// shrink as the prefix grows, which holds for per-character advance sums (`ApproxMeasure`, the /// PDF metrics) and for the UI's text layout up to kerning of a few hundredths of a point; with /// such a non-monotone measure the result may differ from the old linear scan by a character. + /// A label with a line break is not monotone (the ellipsis moves to a new line and the widest + /// line can get shorter), so it takes the linear scan, longest prefix first. fn fit(&self, s: &str, size: f32, maxw: f32) -> String { if self.tw(s, size) <= maxw { return s.to_string(); @@ -86,6 +88,10 @@ impl Ctx<'_> { t.push('…'); self.tw(&t, size) <= maxw }; + if s.contains('\n') { + let k = (1..=top).rev().find(|&k| fits(k)); + return k.map_or_else(String::new, |k| format!("{}…", s.get(..starts.get(k).copied().unwrap_or(s.len())).unwrap_or(""))); + } let (mut lo, mut hi) = (0usize, top); while lo < hi { let mid = lo + (hi - lo).div_ceil(2); diff --git a/crates/chart/src/render/table.rs b/crates/chart/src/render/table.rs index f0839dcfe..4f641afd2 100644 --- a/crates/chart/src/render/table.rs +++ b/crates/chart/src/render/table.rs @@ -70,7 +70,7 @@ pub(super) fn draw(ctx: &mut Ctx, chart: &Chart, data: &ChartData, plot: Bx, lay let y = top + (layout.headers as f32 + i as f32 + 0.5) * layout.row_height; let name_x = if settings.show_keys { x0 + 25.0 } else { x0 + 4.0 }; if settings.show_keys { - let swatch = match plot_kind(chart, series) { + let swatch = match plot_kind(chart, series, i) { ChartKind::Line | ChartKind::LineStacked => Swatch::Line, ChartKind::LineMarkers => Swatch::LineMarker, _ => Swatch::Box, diff --git a/crates/chart/src/render_perf_tests.rs b/crates/chart/src/render_perf_tests.rs index ce15e7861..9d2d7177e 100644 --- a/crates/chart/src/render_perf_tests.rs +++ b/crates/chart/src/render_perf_tests.rs @@ -60,6 +60,30 @@ impl Measure for Counting<'_> { } } +/// Like a text layout that breaks at `\n`: the width is that of the widest line. Adding a +/// newline can make the text narrower, so the width of `prefix + "…"` is not monotone. +struct WidestLine; + +impl Measure for WidestLine { + fn text_width(&self, text: &str, size: f32, _bold: bool) -> f32 { + text.split('\n') + .map(|line| { + line.chars() + .map(|c| { + size * if c == 'i' { + 0.25 + } else if c == 'M' { + 0.9 + } else { + 0.5 + } + }) + .sum::() + }) + .fold(0.0, f32::max) + } +} + fn corpus() -> Vec { let mut v: Vec = ["", "a", "ab", "Hello, world", "AVAVAV tight", "日本語のラベルです", "👨‍👩‍👧 family 🎉 party", "e\u{301}e\u{301}e\u{301} combining", "…", "……"] @@ -71,6 +95,9 @@ fn corpus() -> Vec { v.push("x".repeat(401)); v.push("é🎉".repeat(500)); v.push("long label ".repeat(100)); + for s in ["iiii\nMM", "iiii\n", "\niiii", "ab\ncd\nef", "iiiiiiii\niiiiMMMM\nMM", "line one\nline two is longer\nx"] { + v.push(s.to_string()); + } v } @@ -89,6 +116,20 @@ fn run_fit(m: &dyn Measure) { fn fit_matches_linear_reference() { run_fit(&ApproxMeasure); run_fit(&Varied { calls: Cell::new(0) }); + run_fit(&WidestLine); +} + +#[test] +fn a_multiline_label_keeps_its_fitting_prefix() { + // The whole label is 18 wide and "iii…" is 12.5, both too wide for 10.5, but "iiii\n…" is + // 10 wide and fits. A binary search that stops at the failing midpoint would return nothing. + let ctx = Ctx { out: vec![], m: &WidestLine }; + let s = "iiii\nMM"; + let size = 10.0; + let maxw = 10.5; + assert!(ctx.tw(s, size) > maxw); + assert!(ctx.tw("iii…", size) > maxw && ctx.tw("iiii\n…", size) <= maxw); + assert_eq!(ctx.fit(s, size, maxw), "iiii\n…"); } #[test] diff --git a/crates/ui-egui/src/chartview.rs b/crates/ui-egui/src/chartview.rs index f1a366d9e..dbfba039c 100644 --- a/crates/ui-egui/src/chartview.rs +++ b/crates/ui-egui/src/chartview.rs @@ -581,3 +581,30 @@ pub fn hit(app: &SheetApp, geo: &Geo, sh: &Sheet, p: Pos2) -> Option<(ObjRef, Re pub(crate) fn ink_color(line: Color32, width_pt: f32) -> Color32 { if width_pt > 8.0 { line.gamma_multiply(0.45) } else { line } } + +#[cfg(test)] +mod tests { + use gridcraft_chart::Measure; + + use super::*; + + /// The production measure is not monotone for labels with a line break: egui lays the text out + /// line by line, so ending a prefix with `\n…` can be narrower than the shorter `iii…`. The + /// chart's label fitting must not assume otherwise (#1152). + #[test] + fn egui_measure_shrinks_when_a_prefix_ends_in_a_line_break() { + let ctx = egui::Context::default(); + ctx.set_fonts(theme::pinned_font_definitions()); + let mut checked = false; + // Fonts are ready from the second pass on. + for _ in 0..2 { + let mut out = ctx.run_ui(Default::default(), |ui| { + let m = EguiMeasure { painter: ui.painter(), z: 1.0 }; + let (short, long) = (m.text_width("iii…", 11.0, false), m.text_width("iiii\n…", 11.0, false)); + checked = long < short; + }); + out.textures_delta.clear(); + } + assert!(checked); + } +} From 640fbf22f64e8726188ff604480a2c6f3e6aad42 Mon Sep 17 00:00:00 2001 From: richard1912 <59070720+richard1912@users.noreply.github.com> Date: Sun, 8 Nov 2026 12:30:06 +1100 Subject: [PATCH 20/24] fix(edit): paste, Ctrl+Enter, Fill and Copy skip rows a filter hides (#1101) (#1175) * Paste, Ctrl+Enter and Fill skip rows hidden by a filter (#1101) Repeating one cell or row down a filtered selection (paste of a single cell or row, Ctrl+Enter, Fill Down/Up/Right/Left) now leaves the rows an applied filter hides alone, as Excel does. A taller pasted block still goes straight down, and rows hidden by hand are still written, also as in Excel. Each case was measured in Excel for Microsoft 365. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01A2bJ3LxrZ6YnmppgrfwHwr * Copying a filtered range takes only the rows showing (#1101) A copied range with rows hidden by an applied filter is split into its runs of showing rows and packed like a multi-area copy (formulas as values), so pasting it inside GridCraft no longer brings the hidden rows back. One run is copied as a plain range, keeping formulas. Cut still moves every row. All as measured in Excel for Microsoft 365. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01A2bJ3LxrZ6YnmppgrfwHwr * Find the rows a filter hides without rescanning every table per row filtered_rows asks the sheet for its filter row spans once instead of checking every table for each hidden row. * Paste over filtered rows leaves their validation rules alone When one row is tiled over filtered rows, paste_validations no longer clears or copies rules on the rows the filter hides. * Filtered copy packs links and validation with the rows showing slide_together now moves hyperlinks and validation rules with their rows, and a filtered whole-row copy records the packed rows as its selection so Insert Copied Cells inserts only the rows copied. * Group validation paste options and format the new tests * Pack a filtered copy in one pass over notes, merges, links and spills slide_together finds each entry's area by search instead of scanning every map once per area, so a filter leaving thousands of separate runs no longer costs runs times map size on every copy. --------- Co-authored-by: Richard Co-authored-by: Claude Opus 5.5 Co-authored-by: skylar --- crates/engine/src/cmd/edit.rs | 7 +- crates/engine/src/cmd/edit/cell.rs | 4 + crates/engine/src/cmd/edit/clipboard.rs | 142 +++++++++-- crates/engine/src/cmd/edit/fill.rs | 3 +- crates/engine/src/cmd/edit/paste_special.rs | 48 +++- crates/engine/src/fill.rs | 15 +- crates/engine/src/tests.rs | 266 ++++++++++++++++++++ crates/model/src/sheet.rs | 14 +- 8 files changed, 469 insertions(+), 30 deletions(-) diff --git a/crates/engine/src/cmd/edit.rs b/crates/engine/src/cmd/edit.rs index 83133c01c..8c7072f52 100644 --- a/crates/engine/src/cmd/edit.rs +++ b/crates/engine/src/cmd/edit.rs @@ -239,7 +239,12 @@ fn redo(s: &mut Session, p: &Json) -> Result { /// The rows of `range` an applied filter hides (rows hidden by hand are not among them). fn filtered_rows(sh: &gridcraft_model::Sheet, range: RangeRef) -> std::collections::BTreeSet { - sh.rows.range(range.start.row..=range.end.row).filter(|(r, info)| info.hidden && sh.is_row_filtered(**r)).map(|(r, _)| *r).collect() + let spans = sh.filter_row_spans(); + if spans.is_empty() { + return Default::default(); + } + let filtered = |row: u32| spans.iter().any(|&(first, last)| row >= first && row <= last); + sh.rows.range(range.start.row..=range.end.row).filter(|(r, info)| info.hidden && filtered(**r)).map(|(r, _)| *r).collect() } /// The Clear commands (Delete is Clear Contents). As in Excel, the rows an applied filter hides diff --git a/crates/engine/src/cmd/edit/cell.rs b/crates/engine/src/cmd/edit/cell.rs index 610135682..c3fe44385 100644 --- a/crates/engine/src/cmd/edit/cell.rs +++ b/crates/engine/src/cmd/edit/cell.rs @@ -328,7 +328,11 @@ pub(super) fn range_fill(s: &mut Session, p: &Json) -> Result { if r.count() > 2_000_000 { return Err(bad("range.fill", "range too large")); } + let skip = cx.wb.sheet(sheet).map(|sh| super::filtered_rows(sh, *r)).unwrap_or_default(); for at in r.iter() { + if skip.contains(&at.row) { + continue; + } let style = cx.wb.sheet(sheet).map(|sh| sh.style_id(at)).unwrap_or_default(); let base = match by_style.get(&style) { Some(c) => c.clone(), diff --git a/crates/engine/src/cmd/edit/clipboard.rs b/crates/engine/src/cmd/edit/clipboard.rs index da34c017a..9feb380a0 100644 --- a/crates/engine/src/cmd/edit/clipboard.rs +++ b/crates/engine/src/cmd/edit/clipboard.rs @@ -1,8 +1,8 @@ use super::cell::input_to_cell; use super::paste_special::paste_special; use super::{ - Cell, Clipboard, EngineError, Json, RangeRef, Result, Session, Value, bad, bool_param, cell_param_on, edit, json, ok, str_param, target_ranges, - target_sheet, + Cell, CellRef, Clipboard, EngineError, Json, RangeRef, Result, Session, Value, bad, bool_param, cell_param_on, edit, json, ok, str_param, + target_ranges, target_sheet, }; // ---------------------------------------------------------------- clipboard @@ -15,7 +15,7 @@ fn copy_impl(s: &mut Session, p: &Json, cut: bool) -> Result { return Err(multiple()); } // The selection as made (untrimmed), for Insert Copied Cells. - let selected = ranges.first().copied().unwrap_or_default(); + let mut selected = ranges.first().copied().unwrap_or_default(); let d = s.doc()?; let Some(sh) = d.wb.sheet(sheet) else { return Err(EngineError::NoDocument) }; let trim = |range: RangeRef| { @@ -31,7 +31,27 @@ fn copy_impl(s: &mut Session, p: &Json, cut: bool) -> Result { let (wb, block) = slide_together(&d.wb, sheet, &areas).ok_or_else(multiple)?; (std::sync::Arc::new(wb), block, areas) } else { - (d.wb.clone(), trim(ranges.first().copied().unwrap_or_default()), Vec::new()) + let range = trim(ranges.first().copied().unwrap_or_default()); + // A copied range with rows a filter hides takes only the rows showing: they are copied + // like a multi-area selection stacked top to bottom (formulas as values), as in Excel. + // A cut still takes every row. + let hidden = if cut { Default::default() } else { super::filtered_rows(sh, range) }; + let visible = if hidden.is_empty() { Vec::new() } else { visible_row_runs(range, &hidden) }; + let any_visible = !visible.is_empty(); + let packed = match visible.as_slice() { + [] => (d.wb.clone(), range, Vec::new()), + [only] => (d.wb.clone(), *only, Vec::new()), + _ => { + let (wb, block) = slide_together(&d.wb, sheet, &visible).ok_or_else(multiple)?; + (std::sync::Arc::new(wb), block, visible) + } + }; + // Whole rows copied with some hidden by a filter come as the block of rows showing, so + // Insert Copied Cells makes room for those rows, not for the hidden ones too. + if !hidden.is_empty() && any_visible && selected.is_full_rows() { + selected = RangeRef::rows(packed.1.start.row, packed.1.end.row); + } + packed }; let text = crate::display::range_text(&wb, sheet, range); // The HTML rendering is for the host clipboard (UI); it can reach 4 MB, so agents and @@ -46,6 +66,56 @@ fn copy_impl(s: &mut Session, p: &Json, cut: bool) -> Result { Ok(result) } +/// The runs of consecutive rows of `range` not in `hidden`, as ranges of `range`'s columns. +fn visible_row_runs(range: RangeRef, hidden: &std::collections::BTreeSet) -> Vec { + let mut runs = Vec::new(); + let mut start = None; + for row in range.start.row..=range.end.row { + match (hidden.contains(&row), start) { + (false, None) => start = Some(row), + (true, Some(first)) => { + runs.push(RangeRef::new(CellRef::new(first, range.start.col), CellRef::new(row - 1, range.end.col))); + start = None; + } + _ => {} + } + } + if let Some(first) = start { + runs.push(RangeRef::new(CellRef::new(first, range.start.col), CellRef::new(range.end.row, range.end.col))); + } + runs +} + +/// The validation rules of `src` cut into the areas of a packed block: one rule per area a rule +/// touches, over that area's part moved by the area's shift. `sorted` areas run in order. +fn pack_validations(src: &gridcraft_model::Sheet, sorted: &[RangeRef], shifts: &[(i64, i64)], across: bool) -> Vec { + let key = |c: CellRef| if across { c.col } else { c.row }; + let mut packed_rules = Vec::new(); + for dv in &src.validations { + let Some(from) = dv.ranges.first().map(|r| r.start) else { continue }; + // The areas each of the rule's ranges touches lie side by side in the sorted list. + let end_key = |a: &RangeRef| if across { a.end.col } else { a.end.row }; + let mut by_area: std::collections::BTreeMap> = Default::default(); + for r in &dv.ranges { + let first = sorted.partition_point(|a| end_key(a) < key(r.start)); + for (i, a) in sorted.iter().enumerate().skip(first).take_while(|(_, a)| key(a.start) <= end_key(r)) { + let (dr, dc) = shifts[i]; + let moved = r.intersection(a).and_then(|x| Some(RangeRef::new(x.start.offset(dr, dc)?, x.end.offset(dr, dc)?))); + by_area.entry(i).or_default().extend(moved); + } + } + for ranges in by_area.into_values() { + let Some(to) = ranges.first().map(|r| r.start) else { continue }; + let mut rule = dv.clone(); + rule.f1 = super::super::data::move_rule_formula(&dv.f1, from, to); + rule.f2 = dv.f2.as_ref().map(|f| super::super::data::move_rule_formula(f, from, to)); + rule.ranges = ranges; + packed_rules.push(rule); + } + } + packed_rules +} + /// Excel copies a multi-area selection only when the areas line up: all cover the same rows (they /// slide together side by side, left to right) or all cover the same columns (stacked, top to /// bottom), without overlapping. Returns a copy of the workbook whose sheet holds the areas packed @@ -80,17 +150,35 @@ fn slide_together(wb: &gridcraft_model::Workbook, sheet: usize, areas: &[RangeRe dst.spill.retain(|c, _| !block.contains(*c)); dst.spill_ranges.retain(|c, _| !block.contains(*c)); dst.comments.retain(|c, _| !block.contains(*c)); + dst.hyperlinks.retain(|c, _| !block.contains(*c)); + super::super::data::remove_validation_cells(&mut dst.validations, &[block]); dst.merges.retain(|m| !m.intersects(&block)); if across { dst.cols.retain(|c, _| !(block.start.col..=block.end.col).contains(c)); } else { dst.rows.retain(|r, _| !(block.start.row..=block.end.row).contains(r)); } + // Where each area lands: the shift of its cells, and its first row or column in the block. + let key = |c: CellRef| if across { c.col } else { c.row }; let mut next = if across { anchor.col } else { anchor.row }; + let mut shifts: Vec<(i64, i64)> = Vec::with_capacity(sorted.len()); for a in &sorted { - let (dr, dc) = if across { (0, i64::from(next) - i64::from(a.start.col)) } else { (i64::from(next) - i64::from(a.start.row), 0) }; - let cells = src.cells.iter_range(*a).map(|(c, _)| c).chain(src.spill.keys().copied().filter(|c| a.contains(*c))); - for c in cells { + let to = i64::from(next); + shifts.push(if across { (0, to - i64::from(a.start.col)) } else { (to - i64::from(a.start.row), 0) }); + next = next.saturating_add(if across { a.width() } else { a.height() }); + } + // The areas never overlap and run in order, so the area holding a cell is found by search: + // each map below is read once, whatever the number of areas. + let area_of = |c: CellRef| -> Option { + let i = sorted.partition_point(|a| key(a.start) <= key(c)).checked_sub(1)?; + sorted.get(i)?.contains(c).then_some(i) + }; + let moved = |c: CellRef| -> Option { + let (dr, dc) = shifts[area_of(c)?]; + c.offset(dr, dc) + }; + for (a, &(dr, dc)) in sorted.iter().zip(&shifts) { + for (c, _) in src.cells.iter_range(*a) { let Some(to) = c.offset(dr, dc) else { continue }; let style = src.cell(c).map(|x| x.style).unwrap_or_default(); dst.set_cell(to, Cell { value: src.value(c), formula: None, style, py: None }); @@ -98,17 +186,38 @@ fn slide_together(wb: &gridcraft_model::Workbook, sheet: usize, areas: &[RangeRe dst.cell_pictures.insert(to, picture.clone()); } } - for (c, comment) in src.comments.iter().filter(|(c, _)| a.contains(**c)) { - if let Some(to) = c.offset(dr, dc) { - dst.comments.insert(to, comment.clone()); - } + } + for c in src.spill.keys().copied() { + let Some(to) = moved(c) else { continue }; + let style = src.cell(c).map(|x| x.style).unwrap_or_default(); + dst.set_cell(to, Cell { value: src.value(c), formula: None, style, py: None }); + if let Some(picture) = src.cell_pictures.get(&c) { + dst.cell_pictures.insert(to, picture.clone()); } - for m in src.merges.iter().filter(|m| a.contains_range(m)) { - if let (Some(s), Some(e)) = (m.start.offset(dr, dc), m.end.offset(dr, dc)) { - dst.merges.push(RangeRef::new(s, e)); - } + } + // Links and validation rules follow their cells too, so a row the filter hides leaves its own + // behind. + for (c, link) in &src.hyperlinks { + if let Some(to) = moved(*c) { + dst.hyperlinks.insert(to, link.clone()); } - // Column widths (side by side) or row heights and hidden rows (stacked) follow their cells. + } + let packed_rules = pack_validations(src, &sorted, &shifts, across); + for (c, comment) in &src.comments { + if let Some(to) = moved(*c) { + dst.comments.insert(to, comment.clone()); + } + } + for m in &src.merges { + let Some(i) = area_of(m.start).filter(|&i| sorted[i].contains_range(m)) else { continue }; + let (dr, dc) = shifts[i]; + if let (Some(s), Some(e)) = (m.start.offset(dr, dc), m.end.offset(dr, dc)) { + dst.merges.push(RangeRef::new(s, e)); + } + } + // Column widths (side by side) or row heights and hidden rows (stacked) follow their cells. + let mut next = if across { anchor.col } else { anchor.row }; + for a in &sorted { if across { for (c, info) in src.cols.range(a.start.col..=a.end.col) { dst.cols.insert(next.saturating_add(c.saturating_sub(a.start.col)), *info); @@ -120,6 +229,7 @@ fn slide_together(wb: &gridcraft_model::Workbook, sheet: usize, areas: &[RangeRe } next = next.saturating_add(if across { a.width() } else { a.height() }); } + dst.validations.extend(packed_rules); Some((out, block)) } diff --git a/crates/engine/src/cmd/edit/fill.rs b/crates/engine/src/cmd/edit/fill.rs index 8cd1e3d88..489eca2a4 100644 --- a/crates/engine/src/cmd/edit/fill.rs +++ b/crates/engine/src/cmd/edit/fill.rs @@ -31,8 +31,9 @@ pub(super) fn fill_dir(s: &mut Session, p: &Json, dr: i64, dc: i64) -> Result Result { && !sel.is_full_rows() && sel.count() <= 4_000_000; let (tiles_r, tiles_c) = if tile { (sel.height() / h, sel.width() / w) } else { (1, 1) }; + // A single cell or row repeated down a filtered selection skips the rows the filter hides; a + // taller block is pasted straight down, hidden rows included, as in Excel. + let skip_rows = match s.doc()?.wb.sheet(dest_sheet) { + Some(sh) if tile && h == 1 => super::filtered_rows(sh, sel), + _ => Default::default(), + }; // Excel refuses a paste that would run off the sheet (whole columns pasted below row 1, whole // rows right of column A, a whole column transposed) instead of pasting part of it. if at.offset(i64::from(tiles_r * h) - 1, i64::from(tiles_c * w) - 1).is_none() { @@ -134,6 +140,9 @@ pub(super) fn paste_special(s: &mut Session, p: &Json) -> Result { let sc = CellRef::new(src.start.row + r, src.start.col + c); let (dr, dc) = if transpose { (c, r) } else { (r, c) }; let Some(dest) = at.offset((tr * h + dr) as i64, (tc * w + dc) as i64) else { continue }; + if skip_rows.contains(&dest.row) { + continue; + } let src_cell = if cut_move { moved_formulas.get(&sc).cloned() } else { src_sheet.cell(sc).cloned() }; let src_val = src_sheet.value(sc); let src_picture = if cut_move { moved_pictures.get(&sc) } else { src_sheet.cell_pictures.get(&sc) }.cloned(); @@ -326,7 +335,7 @@ pub(super) fn paste_special(s: &mut Session, p: &Json) -> Result { } let pasted = RangeRef::new(at, at.offset_clamped(i64::from(tiles_r * h) - 1, i64::from(tiles_c * w) - 1)); if !link && matches!(what, What::All | What::AllExceptBorders | What::Validation) { - paste_validations(cx, src_sheet, src, dest_sheet, pasted, transpose, skip_blanks)?; + paste_validations(cx, src_sheet, src, dest_sheet, pasted, PasteMode { transpose, skip_blanks, skip_rows: &skip_rows })?; } if what == What::ColumnWidths { // Each pasted column takes the width of the column copied to it (transposed, the @@ -394,6 +403,31 @@ const MAX_PASTE_LINK_CELLS: u64 = 1_048_576; /// selection makes one range per tile). const MAX_PASTE_VALIDATION_RANGES: usize = 100_000; +/// How a validation paste lays the rules out: transposed or not, keeping the destination's rule under +/// blank source cells, and leaving the rows a filter hides as they are. +#[derive(Clone, Copy)] +struct PasteMode<'a> { + transpose: bool, + skip_blanks: bool, + skip_rows: &'a std::collections::BTreeSet, +} + +/// `range` less the rows in `skip`: the runs of rows between them, as ranges of its columns. +fn without_rows(range: RangeRef, skip: &std::collections::BTreeSet) -> Vec { + let mut out = Vec::new(); + let mut next = range.start.row; + for &row in skip.range(range.start.row..=range.end.row) { + if row > next { + out.push(RangeRef::new(CellRef::new(next, range.start.col), CellRef::new(row - 1, range.end.col))); + } + next = row + 1; + } + if next <= range.end.row { + out.push(RangeRef::new(CellRef::new(next, range.start.col), CellRef::new(range.end.row, range.end.col))); + } + out +} + /// Pasting validation works on ranges, as Excel stores it: the pasted rectangle is cut out of the /// destination's rules (cells whose source had no rule end up with none), then each source rule /// that covers part of the copied block `src` comes along as one rule over the translated parts. @@ -404,9 +438,9 @@ fn paste_validations( src: RangeRef, dest_sheet: usize, pasted: RangeRef, - transpose: bool, - skip_blanks: bool, + mode: PasteMode<'_>, ) -> Result<()> { + let PasteMode { transpose, skip_blanks, skip_rows } = mode; let (h, w) = if transpose { (src.width(), src.height()) } else { (src.height(), src.width()) }; let (tiles_r, tiles_c) = (pasted.height() / h.max(1), pasted.width() / w.max(1)); let mut rules = Vec::new(); @@ -426,7 +460,9 @@ fn paste_validations( let (r, c) = if transpose { (c, r) } else { (r, c) }; for tr in 0..tiles_r { for tc in 0..tiles_c { - if let Some(d) = pasted.start.offset(i64::from(tr * h + r), i64::from(tc * w + c)) { + if let Some(d) = pasted.start.offset(i64::from(tr * h + r), i64::from(tc * w + c)) + && !skip_rows.contains(&d.row) + { cells.push((d.row, d.col)); } } @@ -490,6 +526,8 @@ fn paste_validations( } } } + // The rows a filter hides keep what they have. + let ranges: Vec = ranges.into_iter().flat_map(|r| without_rows(r, skip_rows)).collect(); let Some(anchor) = ranges.first().map(|r| r.start) else { continue }; let mut rule = dv.clone(); // Relative references move with the rule, as pasted formulas do: they are written for the @@ -507,7 +545,7 @@ fn paste_validations( cleared.extend(rules.iter().flat_map(|r| r.ranges.iter().copied())); super::super::data::remove_validation_cells(&mut sh.validations, &cleared); } else { - super::super::data::remove_validation_cells(&mut sh.validations, &[pasted]); + super::super::data::remove_validation_cells(&mut sh.validations, &without_rows(pasted, skip_rows)); } sh.validations.extend(rules); Ok(()) diff --git a/crates/engine/src/fill.rs b/crates/engine/src/fill.rs index efa09310a..25e137878 100644 --- a/crates/engine/src/fill.rs +++ b/crates/engine/src/fill.rs @@ -87,6 +87,18 @@ pub fn holds_source_in_corner(src: RangeRef, target: RangeRef) -> bool { /// Fills `target` (which contains `src` at one edge) from the pattern in `src`. pub fn fill(cx: &mut Ctx, sheet: usize, src: RangeRef, target: RangeRef, mode: FillMode) -> Result<()> { + fill_except(cx, sheet, src, target, mode, &std::collections::BTreeSet::new()) +} + +/// [`fill`], leaving the rows in `skip_rows` untouched (Fill Down over a filtered range). +pub fn fill_except( + cx: &mut Ctx, + sheet: usize, + src: RangeRef, + target: RangeRef, + mode: FillMode, + skip_rows: &std::collections::BTreeSet, +) -> Result<()> { let custom = cx.wb.custom_lists.clone(); let Some(sh) = cx.wb.sheet(sheet) else { return Ok(()) }; // Direction: down/up if target extends rows, else right/left. @@ -119,7 +131,7 @@ pub fn fill(cx: &mut Ctx, sheet: usize, src: RangeRef, target: RangeRef, mode: F (c, sh.cell(c).cloned()) }) .collect(); - let targets: Vec = if vertical { + let mut targets: Vec = if vertical { let rows: Vec = if forward { (src.end.row + 1..=target.end.row).collect() } else { (target.start.row..src.start.row).rev().collect() }; rows.into_iter().map(|r| CellRef::new(r, lane)).collect() @@ -128,6 +140,7 @@ pub fn fill(cx: &mut Ctx, sheet: usize, src: RangeRef, target: RangeRef, mode: F if forward { (src.end.col + 1..=target.end.col).collect() } else { (target.start.col..src.start.col).rev().collect() }; cols.into_iter().map(|c| CellRef::new(lane, c)).collect() }; + targets.retain(|t| !skip_rows.contains(&t.row)); let fmt = |c: &(CellRef, Option)| cx.wb.styles.get(c.1.as_ref().map(|c| c.style).unwrap_or_default()).num_fmt.as_str(); let pattern = if mode.extends_series() || mode == FillMode::Alternate { let first_fmt = src_cells.first().map(fmt).unwrap_or_default(); diff --git a/crates/engine/src/tests.rs b/crates/engine/src/tests.rs index 445ccf545..f97e1bc10 100644 --- a/crates/engine/src/tests.rs +++ b/crates/engine/src/tests.rs @@ -4997,6 +4997,246 @@ fn precision_as_displayed_rounds_numbers_written_to_a_new_sheet() { assert!(rounded > 0, "the histogram's formatted cells should be rounded"); } +/// #1101: repeating one cell or row down a filtered selection (paste, Ctrl+Enter, Fill Down) skips +/// the rows the filter hides, as in Excel. A taller pasted block goes straight down, and rows +/// hidden by hand are written, also as in Excel (each case measured in Excel for Microsoft 365). +#[test] +fn fills_and_single_row_pastes_skip_filtered_rows() { + let fresh = s; + let filtered = || { + let mut s = fresh(); + let mut rows = vec![json!(["Val", "Tag"])]; + for i in 2..=11 { + rows.push(json!([format!("v{i}"), if [3, 5, 7, 9].contains(&i) { "hide" } else { "keep" }])); + } + s.execute("range.setValues", json!({"range": "A1", "values": rows})).unwrap(); + s.execute("range.setValues", json!({"range": "E1", "values": [["X", "W"], ["Y", ""]]})).unwrap(); + s.execute("selection.set", json!({"range": "A1:B1"})).unwrap(); + s.execute("data.filter", json!({})).unwrap(); + assert_eq!(s.execute("data.filterBy", json!({"column": "B", "values": ["keep"]})).unwrap()["hiddenRows"], 4); + s + }; + let col = |s: &Session, c: &str| -> Vec { (2..=11).map(|r| v(s, &format!("{c}{r}"))).collect() }; + let skipped = + |x: &str| -> Vec { (2..=11).map(|r| Value::from(if [3, 5, 7, 9].contains(&r) { format!("v{r}") } else { x.to_string() })).collect() }; + + // One cell pasted over the filtered rows. + let mut s = filtered(); + s.execute("edit.copy", json!({"range": "E1"})).unwrap(); + s.execute("selection.set", json!({"range": "A2:A11"})).unwrap(); + s.execute("edit.paste", json!({})).unwrap(); + assert_eq!(col(&s, "A"), skipped("X")); + + // One row (two cells) pasted over the filtered rows. + let mut s = filtered(); + s.execute("edit.copy", json!({"range": "E1:F1"})).unwrap(); + s.execute("selection.set", json!({"range": "A2:B11"})).unwrap(); + s.execute("edit.paste", json!({})).unwrap(); + assert_eq!(col(&s, "A"), skipped("X")); + assert_eq!(v(&s, "B3"), Value::from("hide")); + assert_eq!(v(&s, "B4"), Value::from("W")); + + // A two-row block is pasted straight down, hidden rows included. + let mut s = filtered(); + s.execute("edit.copy", json!({"range": "E1:E2"})).unwrap(); + s.execute("selection.set", json!({"range": "A2:A11"})).unwrap(); + s.execute("edit.paste", json!({})).unwrap(); + assert_eq!(col(&s, "A"), ["X", "Y", "X", "Y", "X", "Y", "X", "Y", "X", "Y"].map(Value::from)); + + // Ctrl+Enter. + let mut s = filtered(); + s.execute("selection.set", json!({"range": "A2:A11"})).unwrap(); + s.execute("range.fill", json!({"input": "Z"})).unwrap(); + assert_eq!(col(&s, "A"), skipped("Z")); + + // Fill Down, and a formula filled down still adjusts to the row it lands in. + let mut s = filtered(); + s.execute("edit.fillDown", json!({"range": "A2:A11"})).unwrap(); + assert_eq!(col(&s, "A"), skipped("v2")); + s.execute("cell.set", json!({"cell": "C2", "input": "=ROW()"})).unwrap(); + s.execute("edit.fillDown", json!({"range": "C2:C11"})).unwrap(); + assert_eq!(v(&s, "C4"), Value::Number(4.0)); + assert!(v(&s, "C3").is_empty()); + + // Fill Up and Fill Right skip them too. + let mut s = filtered(); + s.execute("edit.fillUp", json!({"range": "A2:A11"})).unwrap(); + assert_eq!(col(&s, "A"), skipped("v11")); + let mut s = filtered(); + s.execute("range.setValues", json!({"range": "C2", "values": [["c2"], ["c3"], ["c4"]]})).unwrap(); + s.execute("edit.fillRight", json!({"range": "A2:C4"})).unwrap(); + assert_eq!([v(&s, "C2"), v(&s, "C3"), v(&s, "C4")], [Value::from("v2"), Value::from("c3"), Value::from("v4")]); + s.execute("edit.undo", json!({})).unwrap(); + assert_eq!(v(&s, "C4"), Value::from("c4")); + + // Copying the filtered range takes only the rows showing, formulas as their values; a cut + // still moves every row. + let mut s = filtered(); + s.execute("cell.set", json!({"cell": "C2", "input": "=ROW()"})).unwrap(); + s.execute("edit.fillDown", json!({"range": "C2:C11"})).unwrap(); + let r = s.execute("edit.copy", json!({"range": "A1:C11"})).unwrap(); + assert_eq!( + r["text"], + "Val Tag +v2 keep 2 +v4 keep 4 +v6 keep 6 +v8 keep 8 +v10 keep 10 +v11 keep 11 +" + ); + s.execute("edit.paste", json!({"at": "H20"})).unwrap(); + let pasted: Vec = (20..=27).map(|r| v(&s, &format!("H{r}"))).collect(); + assert_eq!(pasted, ["Val", "v2", "v4", "v6", "v8", "v10", "v11"].map(Value::from).into_iter().chain([Value::Empty]).collect::>()); + assert_eq!(v(&s, "J22"), Value::Number(4.0)); + assert!(s.doc().unwrap().wb.active().unwrap().cell(CellRef::parse("J22").unwrap()).unwrap().formula.is_none()); + let mut s = filtered(); + s.execute("edit.cut", json!({"range": "A1:B11"})).unwrap(); + s.execute("edit.paste", json!({"at": "H20"})).unwrap(); + assert_eq!(v(&s, "H22"), Value::from("v3")); + + // Rows hidden by hand (no filter) are written, as in Excel. + let mut s = fresh(); + s.execute("range.setValues", json!({"range": "A1", "values": [["a"], ["b"], ["c"]]})).unwrap(); + s.execute("home.hideRows", json!({"rows": "2:2"})).unwrap(); + s.execute("selection.set", json!({"range": "A1:A3"})).unwrap(); + s.execute("range.fill", json!({"input": "Z"})).unwrap(); + assert_eq!(v(&s, "A2"), Value::from("Z")); +} + +/// A table of ten rows (A1:B11) with a link on every value cell, four of its rows hidden by a +/// filter (3, 5, 7, 9), and a spare corner at E1 to copy from. +fn filtered_rows_with_links() -> Session { + let mut s = s(); + let mut rows = vec![json!(["Val", "Tag"])]; + for i in 2..=11 { + rows.push(json!([format!("v{i}"), if [3, 5, 7, 9].contains(&i) { "hide" } else { "keep" }])); + } + s.execute("range.setValues", json!({"range": "A1", "values": rows})).unwrap(); + s.execute("range.setValues", json!({"range": "E1", "values": [["X"]]})).unwrap(); + let sh = std::sync::Arc::make_mut(&mut s.doc_mut().unwrap().wb).sheet_mut(0).unwrap(); + for r in 2..=11u32 { + sh.hyperlinks + .insert(CellRef::new(r - 1, 0), gridcraft_model::Hyperlink { target: format!("https://example.com/{r}"), tooltip: None, external: true }); + } + s.execute("selection.set", json!({"range": "A1:B1"})).unwrap(); + s.execute("data.filter", json!({})).unwrap(); + s.execute("data.filterBy", json!({"column": "B", "values": ["keep"]})).unwrap(); + s +} + +fn cell_link_target(s: &Session, cell: &str) -> Option { + s.doc().unwrap().wb.active().unwrap().hyperlinks.get(&CellRef::parse(cell).unwrap()).map(|h| h.target.clone()) +} + +fn cell_validation(s: &Session, cell: &str) -> Option { + let c = CellRef::parse(cell).unwrap(); + s.doc().unwrap().wb.active().unwrap().validations.iter().find(|d| d.ranges.iter().any(|r| r.contains(c))).map(|d| d.f1.clone()) +} + +/// Copying a filtered range packs the rows showing into one block; their links and validation +/// rules travel with them, and those of the hidden rows are left behind. +#[test] +fn filtered_copy_pastes_links_and_validation_of_the_rows_showing_only() { + let mut s = filtered_rows_with_links(); + for (cell, source) in [("A3", "\"hidden\""), ("A4", "\"showing\""), ("A9", "\"hidden too\"")] { + s.execute("data.validation", json!({"range": cell, "type": "list", "formula1": source})).unwrap(); + } + s.execute("edit.copy", json!({"range": "A1:B11"})).unwrap(); + s.execute("edit.paste", json!({"at": "H20"})).unwrap(); + let links: Vec> = (20..=26).map(|r| cell_link_target(&s, &format!("H{r}"))).collect(); + let expected: Vec> = + [None, Some(2u32), Some(4), Some(6), Some(8), Some(10), Some(11)].iter().map(|n| n.map(|n| format!("https://example.com/{n}"))).collect(); + assert_eq!(links, expected); + let rules: Vec> = (20..=26).map(|r| cell_validation(&s, &format!("H{r}"))).collect(); + assert_eq!(rules, [None, None, Some("\"showing\"".to_string()), None, None, None, None]); + // The source keeps its own links and rules where they were. + assert_eq!(cell_link_target(&s, "A3").as_deref(), Some("https://example.com/3")); + assert_eq!(cell_validation(&s, "A3").as_deref(), Some("\"hidden\"")); +} + +/// One cell pasted over a filtered column leaves the rules of the rows the filter hides alone, and +/// does not give them the copied rule, with or without Skip Blanks. +#[test] +fn single_row_paste_over_filtered_rows_leaves_their_validation_alone() { + for skip_blanks in [false, true] { + let mut s = filtered_rows_with_links(); + s.execute("data.validation", json!({"range": "E1", "type": "list", "formula1": "\"copied\""})).unwrap(); + s.execute("data.validation", json!({"range": "A2:A5", "type": "list", "formula1": "\"old\""})).unwrap(); + s.execute("edit.copy", json!({"range": "E1"})).unwrap(); + s.execute("selection.set", json!({"range": "A2:A11"})).unwrap(); + s.execute("edit.pasteSpecial", json!({"what": "all", "skipBlanks": skip_blanks})).unwrap(); + let rules: Vec> = (2..=11).map(|r| cell_validation(&s, &format!("A{r}"))).collect(); + let showing = Some("\"copied\"".to_string()); + let old = Some("\"old\"".to_string()); + assert_eq!( + rules, + [showing.clone(), old.clone(), showing.clone(), old, showing.clone(), None, showing.clone(), None, showing.clone(), showing], + "skip blanks: {skip_blanks}" + ); + } +} + +/// Insert Copied Cells after a filtered copy of whole rows makes room for the rows copied, not for +/// the rows that were selected. +#[test] +fn insert_copied_cells_after_a_filtered_row_copy_inserts_only_the_rows_showing() { + let mut s = filtered_rows_with_links(); + s.execute("cell.set", json!({"cell": "A30", "input": "below"})).unwrap(); + s.execute("edit.copy", json!({"range": "1:11"})).unwrap(); + s.execute("selection.set", json!({"range": "A20"})).unwrap(); + s.execute("home.insertCopiedCells", json!({})).unwrap(); + // Seven rows were copied (the header and six rows showing). + assert_eq!(v(&s, "A37"), Value::from("below")); + assert_eq!(v(&s, "A20"), Value::from("Val")); + assert_eq!(v(&s, "A26"), Value::from("v11")); + assert_eq!(v(&s, "A27"), Value::Empty); +} + +/// Copying a filtered range with many separate runs of rows showing moves each row's note, link and +/// merge with it, and leaves the hidden rows' behind. +#[test] +fn filtered_copy_with_many_runs_moves_notes_and_merges_with_their_rows() { + let mut s = s(); + let mut rows = vec![json!(["Val", "Tag"])]; + for i in 2..=2001 { + rows.push(json!([format!("v{i}"), if i % 2 == 1 { "hide" } else { "keep" }])); + } + s.execute("range.setValues", json!({"range": "A1", "values": rows})).unwrap(); + let sh = std::sync::Arc::make_mut(&mut s.doc_mut().unwrap().wb).sheet_mut(0).unwrap(); + for r in 2..=2001u32 { + let note = gridcraft_model::Comment { + author: "a".into(), + text: format!("note {r}"), + replies: vec![], + threaded: false, + resolved: false, + visible: false, + dates: vec![], + note_box: None, + }; + sh.comments.insert(CellRef::new(r - 1, 0), note); + } + sh.merges.push(gridcraft_core::RangeRef::new(CellRef::new(3, 1), CellRef::new(3, 2))); + sh.merges.push(gridcraft_core::RangeRef::new(CellRef::new(2, 1), CellRef::new(2, 2))); + s.execute("selection.set", json!({"range": "A1:B1"})).unwrap(); + s.execute("data.filter", json!({})).unwrap(); + s.execute("data.filterBy", json!({"column": "B", "values": ["keep"]})).unwrap(); + s.execute("edit.copy", json!({"range": "A1:C2001"})).unwrap(); + s.execute("edit.paste", json!({"at": "H1"})).unwrap(); + let sh = s.doc().unwrap().wb.active().unwrap(); + let note = |row: u32| sh.comments.get(&CellRef::new(row - 1, 7)).map(|c| c.text.clone()); + assert_eq!(note(1), None); + assert_eq!(note(2).as_deref(), Some("note 2")); + assert_eq!(note(3).as_deref(), Some("note 4")); + assert_eq!(note(1001).as_deref(), Some("note 2000")); + assert_eq!(sh.comments.keys().filter(|c| c.col == 7).count(), 1000); + // Only the merge on a row showing (row 4) came along, packed to the row it landed in (row 3). + assert!(sh.merges.contains(&gridcraft_core::RangeRef::new(CellRef::new(2, 8), CellRef::new(2, 9)))); + assert_eq!(sh.merges.iter().filter(|m| m.start.col >= 7).count(), 1); +} + #[test] fn editing_quote_prefixed_text_keeps_it_text() { let mut s = s(); @@ -5257,6 +5497,32 @@ fn paste_validation_skip_blanks_large_populated_paste() { assert!(validation_areas(&s).is_empty()); } +/// A whole-column copy of a filtered range takes only the rows showing, also when they form one +/// block (the hidden rows are all at the end) and the selection covers the whole column. +#[test] +fn whole_column_copy_of_a_filtered_range_leaves_the_hidden_rows_out() { + for hidden_tail in [false, true] { + let mut s = s(); + let mut rows = vec![json!(["Val", "Tag"])]; + for i in 2..=11 { + let hide = if hidden_tail { i >= 8 } else { [3, 5, 7, 9].contains(&i) }; + rows.push(json!([format!("v{i}"), if hide { "hide" } else { "keep" }])); + } + s.execute("range.setValues", json!({"range": "A1", "values": rows})).unwrap(); + s.execute("selection.set", json!({"range": "A1:B1"})).unwrap(); + s.execute("data.filter", json!({})).unwrap(); + s.execute("data.filterBy", json!({"column": "B", "values": ["keep"]})).unwrap(); + s.execute("cell.set", json!({"cell": "H30", "input": "old"})).unwrap(); + s.execute("edit.copy", json!({"range": "A:B"})).unwrap(); + s.execute("edit.paste", json!({"at": "H1"})).unwrap(); + let col: Vec = (1..=11).map(|r| v(&s, &format!("H{r}"))).collect(); + let want: Vec<&str> = + if hidden_tail { vec!["Val", "v2", "v3", "v4", "v5", "v6", "v7"] } else { vec!["Val", "v2", "v4", "v6", "v8", "v10", "v11"] }; + let want: Vec = want.into_iter().map(Value::from).chain(std::iter::repeat(Value::Empty)).take(11).collect(); + assert_eq!(col, want, "hidden rows stay out of a whole-column copy (hidden tail: {hidden_tail})"); + } +} + /// Open a workbook by path, and `CELL("filename")` names the file it came from, as Excel does; /// Save As makes the new file the one it names. The sheet-name formula of #1489 depends on it. #[test] diff --git a/crates/model/src/sheet.rs b/crates/model/src/sheet.rs index 67ca1f13e..440647d42 100644 --- a/crates/model/src/sheet.rs +++ b/crates/model/src/sheet.rs @@ -491,12 +491,14 @@ impl Sheet { /// sheet AutoFilter while it has criteria, or a hidden data row of a filtered table. /// Other hidden rows were hidden by hand. pub fn is_row_filtered(&self, row: u32) -> bool { - if !self.is_row_hidden(row) { - return false; - } - let in_sheet_filter = - self.autofilter.as_ref().is_some_and(|af| !af.criteria.is_empty() && row > af.range.start.row && row <= af.range.end.row); - in_sheet_filter || self.tables.iter().any(|t| !t.filter.is_empty() && t.data_range().is_some_and(|r| row >= r.start.row && row <= r.end.row)) + self.is_row_hidden(row) && self.filter_row_spans().iter().any(|&(first, last)| row >= first && row <= last) + } + /// The row spans (first, last) an applied filter can hide: the data rows of the sheet + /// AutoFilter while it has criteria, and of each filtered table. + pub fn filter_row_spans(&self) -> Vec<(u32, u32)> { + let sheet = self.autofilter.as_ref().filter(|af| !af.criteria.is_empty()).map(|af| (af.range.start.row.saturating_add(1), af.range.end.row)); + let tables = self.tables.iter().filter(|t| !t.filter.is_empty()).filter_map(|t| t.data_range()).map(|r| (r.start.row, r.end.row)); + sheet.into_iter().chain(tables).collect() } pub fn is_col_hidden(&self, col: u32) -> bool { self.cols.get(&col).is_some_and(|r| r.hidden) From 4bba98fd500665fd3c24511302c255d3eb2b424e Mon Sep 17 00:00:00 2001 From: vangel76 Date: Sun, 8 Nov 2026 02:57:51 +0100 Subject: [PATCH 21/24] perf(model): cache formula trees when reading a workbook from JSON A Formula read from JSON had no cached tree (`expr` is not saved), so every graph rebuild and every recalculation parsed each formula again. Reading one now parses its text once and keeps the tree, as Formula::new does. The saved text is kept as written. Fixes #3577 Co-Authored-By: Claude Opus 5.5 --- crates/model/src/cell.rs | 82 ++++++++++++++++++++++++++++++++++++++-- 1 file changed, 79 insertions(+), 3 deletions(-) diff --git a/crates/model/src/cell.rs b/crates/model/src/cell.rs index 0ca1dbe74..33186bd26 100644 --- a/crates/model/src/cell.rs +++ b/crates/model/src/cell.rs @@ -11,6 +11,7 @@ use crate::style::StyleId; /// A parsed formula. `text` is the canonical formula text without the leading `=`. #[derive(Clone, Debug, PartialEq, Serialize, Deserialize)] +#[serde(from = "SavedFormula")] pub struct Formula { pub text: String, #[serde(skip)] @@ -39,15 +40,15 @@ impl Formula { pub fn from_expr(e: Expr) -> Formula { Formula { text: gridcraft_formula::print(&e), expr: Some(Arc::new(e)), array: None, dynamic: false } } - /// The parsed expression, re-parsing after deserialization. + /// The parsed expression, re-parsing when no tree is cached. pub fn expr(&self) -> Option { self.expr.as_deref().cloned().or_else(|| gridcraft_formula::parse(&self.text).ok()) } - /// The parsed expression without copying it (re-parses after deserialization). + /// The parsed expression without copying it (re-parses when no tree is cached). pub fn expr_arc(&self) -> Option> { self.expr.clone().or_else(|| gridcraft_formula::parse(&self.text).ok().map(Arc::new)) } - /// Runs `f` on the parsed expression, borrowing the cached tree or, after deserialization, + /// Runs `f` on the parsed expression, borrowing the cached tree or, when none is cached, /// a freshly parsed one that is dropped afterwards. Unlike [`expr`](Self::expr) it never copies /// the tree, and unlike [`expr_arc`](Self::expr_arc) it needs no reference count or `Arc` /// allocation. `None` when the text does not parse. @@ -59,6 +60,25 @@ impl Formula { } } +/// A formula as a saved workbook holds it: the tree is not saved. Reading one parses its text +/// once and caches the tree, so a workbook opened from JSON does not parse every formula again on +/// each recalculation. The text is kept as saved, and text that doesn't parse caches nothing, as +/// [`Formula::new`] does. +#[derive(Deserialize)] +struct SavedFormula { + text: String, + array: Option, + #[serde(default)] + dynamic: bool, +} + +impl From for Formula { + fn from(f: SavedFormula) -> Formula { + let expr = gridcraft_formula::parse(&f.text).ok().map(Arc::new); + Formula { text: f.text, expr, array: f.array, dynamic: f.dynamic } + } +} + /// A static picture stored as cell content, independent of floating drawing anchors. #[derive(Clone, Debug, PartialEq, Serialize, Deserialize)] pub struct CellPicture { @@ -131,3 +151,59 @@ impl Cell { } } } + +#[cfg(test)] +mod tests { + use super::*; + use gridcraft_core::CellRef; + + #[test] + fn formulas_read_from_json_keep_their_text_and_cache_the_tree() -> Result<(), serde_json::Error> { + let mut array = Formula::new("=A1:B2*2"); + array.array = Some(RangeRef::new(CellRef::new(0, 2), CellRef::new(1, 3))); + array.dynamic = true; + for f in [Formula::new("=SUM(A1:A3)+'My Sheet'!B2"), Formula::new("=LET(x,A1*2,x+1)"), array] { + let back: Formula = serde_json::from_str(&serde_json::to_string(&f)?)?; + assert!(back.expr.is_some(), "{}: the tree is cached after reading", f.text); + assert_eq!(back, f); + } + // Saved text is kept as written, even when the canonical print differs, and its tree is + // the one the text parses to. + let back: Formula = serde_json::from_str(r#"{"text": "sum( a1 , 2 )", "array": null}"#)?; + assert_eq!(back.text, "sum( a1 , 2 )"); + assert!(!back.dynamic); + assert_eq!(back.expr.as_deref(), gridcraft_formula::parse("sum( a1 , 2 )").ok().as_ref()); + assert!(back.expr.is_some()); + // Text that doesn't parse caches nothing, as `Formula::new` does. + let bad = Formula::new("=SUM("); + let back: Formula = serde_json::from_str(&serde_json::to_string(&bad)?)?; + assert_eq!(back, bad); + assert!(back.expr.is_none()); + Ok(()) + } + + #[test] + fn a_workbook_read_from_json_has_every_formula_tree() -> Result<(), Box> { + let mut wb = crate::Workbook::new(); + let mut array = Formula::new("=A1:A3*2"); + array.array = Some(RangeRef::new(CellRef::new(0, 3), CellRef::new(2, 3))); + let sheet = wb.sheet_mut(0).ok_or("no sheet")?; + sheet.cells.set(CellRef::new(0, 0), Cell::value(Value::Number(1.0))); + sheet.cells.set(CellRef::new(0, 1), Cell::formula(Formula::new("=SUM(A1:A3)"))); + sheet.cells.set(CellRef::new(1, 1), Cell::formula(Formula::new("=IF(A1>0,\"yes\",\"no\")"))); + sheet.cells.set(CellRef::new(2, 1), Cell::formula(Formula::new("=SUM("))); + sheet.cells.set(CellRef::new(0, 3), Cell::formula(array)); + let json = serde_json::to_string(&wb)?; + let back: crate::Workbook = serde_json::from_str(&json)?; + let (before, after) = (wb.sheet(0).ok_or("no sheet")?, back.sheet(0).ok_or("no sheet")?); + let mut formulas = 0; + for (c, cell) in before.cells.iter() { + let (Some(f), Some(read)) = (&cell.formula, after.cell(c).and_then(|cell| cell.formula.as_ref())) else { continue }; + assert_eq!(read, f, "{c:?}: text, tree and flags as saved"); + assert_eq!(read.expr.is_some(), gridcraft_formula::parse(&f.text).is_ok(), "{c:?}: every formula that parses has its tree"); + formulas += 1; + } + assert_eq!(formulas, 4); + Ok(()) + } +} From 7cb0088361e1118b1df96ea008a7065bcaf4b9f8 Mon Sep 17 00:00:00 2001 From: vangel76 Date: Sun, 8 Nov 2026 03:04:24 +0100 Subject: [PATCH 22/24] perf(formula): adjust_if_changed skips formulas a structural edit leaves alone adjust had no way to report that an edit does not affect a formula, so callers copied, rebuilt and printed every formula in the workbook. adjust_if_changed borrows the formula and returns None when nothing moves. Its check reuses adjust_ref on each reference into the edited sheet, on a copy without the sheet name, so it cannot disagree with adjust. adjust_ref and shift_ref take the reference by value and no longer clone sheet names. An integration test keeps the previous implementation as an oracle over every edit kind and a randomized property test. Fixes #3576 Co-Authored-By: Claude Opus 5.5 --- crates/formula/src/adjust.rs | 73 ++-- crates/formula/tests/it/adjust_oracle.rs | 494 +++++++++++++++++++++++ crates/formula/tests/it/main.rs | 1 + 3 files changed, 545 insertions(+), 23 deletions(-) create mode 100644 crates/formula/tests/it/adjust_oracle.rs diff --git a/crates/formula/src/adjust.rs b/crates/formula/src/adjust.rs index 319474b4e..ba6616d64 100644 --- a/crates/formula/src/adjust.rs +++ b/crates/formula/src/adjust.rs @@ -8,7 +8,7 @@ use crate::ast::{Anchor, Expr, RefKind, Reference, SheetSel, clamp_anchor}; /// Copy/fill: relative parts move by `(dr, dc)`. References pushed off the sheet become `#REF!`. pub fn shift_relative(e: Expr, dr: i64, dc: i64) -> Expr { e.map(&mut |x| match x { - Expr::Ref(r) => match shift_ref(&r, dr, dc) { + Expr::Ref(r) => match shift_ref(r, dr, dc) { Some(r) => Expr::Ref(r), None => Expr::Error(CellError::Ref), }, @@ -60,14 +60,14 @@ fn shift_line(v: u32, abs: bool, d: i64, max: u32) -> Option { if (0..max as i64).contains(&n) { Some(n as u32) } else { None } } -fn shift_ref(r: &Reference, dr: i64, dc: i64) -> Option { +fn shift_ref(r: Reference, dr: i64, dc: i64) -> Option { let kind = match &r.kind { RefKind::Cell(a) => RefKind::Cell(shift_anchor(a, dr, dc)?), RefKind::Range(a, b) => RefKind::Range(shift_anchor(a, dr, dc)?, shift_anchor(b, dr, dc)?), RefKind::Rows(r0, a0, r1, a1) => RefKind::Rows(shift_line(*r0, *a0, dr, MAX_ROWS)?, *a0, shift_line(*r1, *a1, dr, MAX_ROWS)?, *a1), RefKind::Cols(c0, a0, c1, a1) => RefKind::Cols(shift_line(*c0, *a0, dc, MAX_COLS)?, *a0, shift_line(*c1, *a1, dc, MAX_COLS)?, *a1), }; - Some(Reference { sheet: r.sheet.clone(), kind }) + Some(Reference { sheet: r.sheet, kind }) } #[derive(Clone, Copy, Debug, PartialEq, Eq)] @@ -95,27 +95,54 @@ pub enum Edit { MoveToSheet { from: RangeRef, to_sheet: String, to_row: u32, to_col: u32 }, } +/// Whether `r`, in a formula living on sheet `host`, points into sheet `target`. +fn on_target(r: &Reference, host: &str, target: &str) -> bool { + match &r.sheet { + SheetSel::Current => host.eq_ignore_ascii_case(target), + SheetSel::Named(n) => n.eq_ignore_ascii_case(target), + SheetSel::Span(..) => false, + } +} + +/// Whether [`adjust`] would change `r`. It runs the same [`adjust_ref`] on a copy without the +/// sheet name, so no name is copied: the adjustment reads only the position, and the sheet a cut +/// to another sheet assigns is compared afterwards. +fn ref_changes(r: &Reference, edit: &Edit) -> bool { + let probe = Reference { sheet: SheetSel::Current, kind: r.kind.clone() }; + match adjust_ref(probe, edit) { + None => true, + Some(p) => p.kind != r.kind || (p.sheet != SheetSel::Current && p.sheet != r.sheet), + } +} + +/// Whether [`adjust`] would change the formula `e` at all, without copying or rebuilding it. +fn changes(e: &Expr, host: &str, target: &str, edit: &Edit) -> bool { + let mut changed = false; + e.walk(&mut |x| { + if !changed && let Expr::Ref(r) = x { + changed = on_target(r, host, target) && ref_changes(r, edit); + } + }); + changed +} + /// Adjusts a formula living on sheet `host` after `edit` on sheet `target`. pub fn adjust(e: Expr, host: &str, target: &str, edit: &Edit) -> Expr { e.map(&mut |x| match x { - Expr::Ref(r) => { - let on_target = match &r.sheet { - SheetSel::Current => host.eq_ignore_ascii_case(target), - SheetSel::Named(n) => n.eq_ignore_ascii_case(target), - SheetSel::Span(..) => false, - }; - if !on_target { - return Expr::Ref(r); - } - match adjust_ref(&r, edit) { - Some(r) => Expr::Ref(r), - None => Expr::Error(CellError::Ref), - } - } + Expr::Ref(r) if on_target(&r, host, target) => match adjust_ref(r, edit) { + Some(r) => Expr::Ref(r), + None => Expr::Error(CellError::Ref), + }, other => other, }) } +/// [`adjust`] for a borrowed formula: `None` when the edit leaves it exactly as it is, so a +/// caller can skip unaffected formulas without copying, rebuilding or printing them. +pub fn adjust_if_changed(e: &Expr, host: &str, target: &str, edit: &Edit) -> Option { + changes(e, host, target, edit).then(|| adjust(e.clone(), host, target, edit)) +} + /// New position of an index after an insert (`None` never) or delete (`None` = deleted). fn map_index(v: u32, at: u32, count: u32, insert: bool, max: u32) -> Option { if insert { @@ -168,7 +195,7 @@ fn map_span(lo: u32, hi: u32, at: u32, count: u32, insert: bool, max: u32) -> Op } } -fn adjust_ref(r: &Reference, edit: &Edit) -> Option { +fn adjust_ref(r: Reference, edit: &Edit) -> Option { match edit { Edit::Insert { axis, at, count } | Edit::Delete { axis, at, count } => { let insert = matches!(edit, Edit::Insert { .. }); @@ -193,7 +220,7 @@ fn adjust_ref(r: &Reference, edit: &Edit) -> Option { } (k, _) => k.clone(), }; - Some(Reference { sheet: r.sheet.clone(), kind }) + Some(Reference { sheet: r.sheet, kind }) } Edit::InsertCells { axis, range } | Edit::DeleteCells { axis, range } => { // Like inserting/deleting lines, for references that lie within the shifted @@ -204,7 +231,7 @@ fn adjust_ref(r: &Reference, edit: &Edit) -> Option { Axis::Cols => (range.start.row <= span.start.row && span.end.row <= range.end.row, range.start.col, range.width()), }; if !inside { - return Some(r.clone()); + return Some(r); } let axis = *axis; let lines = if matches!(edit, Edit::InsertCells { .. }) { Edit::Insert { axis, at, count } } else { Edit::Delete { axis, at, count } }; @@ -215,7 +242,7 @@ fn adjust_ref(r: &Reference, edit: &Edit) -> Option { let dc = *to_col as i64 - from.start.col as i64; let range = r.range(); if !from.contains_range(&range) { - return Some(r.clone()); + return Some(r); } let mv = |a: &Anchor| -> Option { let (row, col) = clamp_anchor(a.row as i64 + dr, a.col as i64 + dc)?; @@ -233,11 +260,11 @@ fn adjust_ref(r: &Reference, edit: &Edit) -> Option { RefKind::Cols(shift_line(*a, false, dc, MAX_COLS)?, *aa, shift_line(*b, false, dc, MAX_COLS)?, *ba) } }; - Some(Reference { sheet: r.sheet.clone(), kind }) + Some(Reference { sheet: r.sheet, kind }) } Edit::MoveToSheet { from, to_sheet, to_row, to_col } => { if !from.contains_range(&r.range()) { - return Some(r.clone()); + return Some(r); } let mut moved = adjust_ref(r, &Edit::Move { from: *from, to_row: *to_row, to_col: *to_col })?; moved.sheet = SheetSel::Named(to_sheet.clone()); diff --git a/crates/formula/tests/it/adjust_oracle.rs b/crates/formula/tests/it/adjust_oracle.rs new file mode 100644 index 000000000..42c5314dc --- /dev/null +++ b/crates/formula/tests/it/adjust_oracle.rs @@ -0,0 +1,494 @@ +//! `adjust`, `adjust_if_changed` and `shift_relative` against a verbatim copy of the previous +//! implementation, which cloned and rebuilt every formula. The rewrite must give the same tree +//! (and so the same text) for every formula and edit, and `adjust_if_changed` must say +//! "unchanged" exactly when the old code returned the formula as it was. + +use gridcraft_core::{CellRef, MAX_COLS, MAX_ROWS, RangeRef}; +use gridcraft_formula::adjust::{Axis, Edit, adjust, adjust_if_changed, shift_relative}; +use gridcraft_formula::{Expr, parse, print}; + +/// The implementation before the borrowed check, kept as the test oracle. +mod old { + use gridcraft_core::{CellError, MAX_COLS, MAX_ROWS}; + use gridcraft_formula::adjust::{Axis, Edit}; + use gridcraft_formula::{Anchor, Expr, RefKind, Reference, SheetSel}; + + fn clamp_anchor(row: i64, col: i64) -> Option<(u32, u32)> { + if (0..MAX_ROWS as i64).contains(&row) && (0..MAX_COLS as i64).contains(&col) { Some((row as u32, col as u32)) } else { None } + } + + pub fn shift_relative(e: Expr, dr: i64, dc: i64) -> Expr { + e.map(&mut |x| match x { + Expr::Ref(r) => match shift_ref(&r, dr, dc) { + Some(r) => Expr::Ref(r), + None => Expr::Error(CellError::Ref), + }, + other => other, + }) + } + + fn shift_anchor(a: &Anchor, dr: i64, dc: i64) -> Option { + let row = if a.row_abs { a.row as i64 } else { a.row as i64 + dr }; + let col = if a.col_abs { a.col as i64 } else { a.col as i64 + dc }; + let (row, col) = clamp_anchor(row, col)?; + Some(Anchor { row, col, ..*a }) + } + + fn shift_line(v: u32, abs: bool, d: i64, max: u32) -> Option { + if abs { + return Some(v); + } + let n = v as i64 + d; + if (0..max as i64).contains(&n) { Some(n as u32) } else { None } + } + + fn shift_ref(r: &Reference, dr: i64, dc: i64) -> Option { + let kind = match &r.kind { + RefKind::Cell(a) => RefKind::Cell(shift_anchor(a, dr, dc)?), + RefKind::Range(a, b) => RefKind::Range(shift_anchor(a, dr, dc)?, shift_anchor(b, dr, dc)?), + RefKind::Rows(r0, a0, r1, a1) => RefKind::Rows(shift_line(*r0, *a0, dr, MAX_ROWS)?, *a0, shift_line(*r1, *a1, dr, MAX_ROWS)?, *a1), + RefKind::Cols(c0, a0, c1, a1) => RefKind::Cols(shift_line(*c0, *a0, dc, MAX_COLS)?, *a0, shift_line(*c1, *a1, dc, MAX_COLS)?, *a1), + }; + Some(Reference { sheet: r.sheet.clone(), kind }) + } + + pub fn adjust(e: Expr, host: &str, target: &str, edit: &Edit) -> Expr { + e.map(&mut |x| match x { + Expr::Ref(r) => { + let on_target = match &r.sheet { + SheetSel::Current => host.eq_ignore_ascii_case(target), + SheetSel::Named(n) => n.eq_ignore_ascii_case(target), + SheetSel::Span(..) => false, + }; + if !on_target { + return Expr::Ref(r); + } + match adjust_ref(&r, edit) { + Some(r) => Expr::Ref(r), + None => Expr::Error(CellError::Ref), + } + } + other => other, + }) + } + + fn map_index(v: u32, at: u32, count: u32, insert: bool, max: u32) -> Option { + if insert { + if v >= at { + let n = v as u64 + count as u64; + if n >= max as u64 { None } else { Some(n as u32) } + } else { + Some(v) + } + } else if v < at { + Some(v) + } else if v >= at.saturating_add(count) { + Some(v - count) + } else { + None + } + } + + fn map_span(lo: u32, hi: u32, at: u32, count: u32, insert: bool, max: u32) -> Option<(u32, u32)> { + if insert { + let lo2 = if lo >= at { lo as u64 + count as u64 } else { lo as u64 }; + let hi2 = if hi >= at { hi as u64 + count as u64 } else { hi as u64 }; + let hi2 = hi2.min(max as u64 - 1); + if lo2 > hi2 { + return None; + } + Some((lo2 as u32, hi2 as u32)) + } else { + let end = at.saturating_add(count); + if lo >= at && hi < end { + return None; + } + let lo2 = if lo < at { + lo + } else if lo >= end { + lo - count + } else { + at + }; + let hi2 = if hi < at { + hi + } else if hi >= end { + hi - count + } else { + at.saturating_sub(1) + }; + Some((lo2, hi2.max(lo2))) + } + } + + fn adjust_ref(r: &Reference, edit: &Edit) -> Option { + match edit { + Edit::Insert { axis, at, count } | Edit::Delete { axis, at, count } => { + let insert = matches!(edit, Edit::Insert { .. }); + let kind = match (&r.kind, axis) { + (RefKind::Cell(a), Axis::Rows) => RefKind::Cell(Anchor { row: map_index(a.row, *at, *count, insert, MAX_ROWS)?, ..*a }), + (RefKind::Cell(a), Axis::Cols) => RefKind::Cell(Anchor { col: map_index(a.col, *at, *count, insert, MAX_COLS)?, ..*a }), + (RefKind::Range(a, b), Axis::Rows) => { + let (lo, hi) = map_span(a.row.min(b.row), a.row.max(b.row), *at, *count, insert, MAX_ROWS)?; + RefKind::Range(Anchor { row: lo, ..*a }, Anchor { row: hi, ..*b }) + } + (RefKind::Range(a, b), Axis::Cols) => { + let (lo, hi) = map_span(a.col.min(b.col), a.col.max(b.col), *at, *count, insert, MAX_COLS)?; + RefKind::Range(Anchor { col: lo, ..*a }, Anchor { col: hi, ..*b }) + } + (RefKind::Rows(r0, a0, r1, a1), Axis::Rows) => { + let (lo, hi) = map_span(*r0.min(r1), *r0.max(r1), *at, *count, insert, MAX_ROWS)?; + RefKind::Rows(lo, *a0, hi, *a1) + } + (RefKind::Cols(c0, a0, c1, a1), Axis::Cols) => { + let (lo, hi) = map_span(*c0.min(c1), *c0.max(c1), *at, *count, insert, MAX_COLS)?; + RefKind::Cols(lo, *a0, hi, *a1) + } + (k, _) => k.clone(), + }; + Some(Reference { sheet: r.sheet.clone(), kind }) + } + Edit::InsertCells { axis, range } | Edit::DeleteCells { axis, range } => { + let span = r.range(); + let (inside, at, count) = match axis { + Axis::Rows => (range.start.col <= span.start.col && span.end.col <= range.end.col, range.start.row, range.height()), + Axis::Cols => (range.start.row <= span.start.row && span.end.row <= range.end.row, range.start.col, range.width()), + }; + if !inside { + return Some(r.clone()); + } + let axis = *axis; + let lines = + if matches!(edit, Edit::InsertCells { .. }) { Edit::Insert { axis, at, count } } else { Edit::Delete { axis, at, count } }; + adjust_ref(r, &lines) + } + Edit::Move { from, to_row, to_col } => { + let dr = *to_row as i64 - from.start.row as i64; + let dc = *to_col as i64 - from.start.col as i64; + let range = r.range(); + if !from.contains_range(&range) { + return Some(r.clone()); + } + let mv = |a: &Anchor| -> Option { + let (row, col) = clamp_anchor(a.row as i64 + dr, a.col as i64 + dc)?; + Some(Anchor { row, col, ..*a }) + }; + let kind = match &r.kind { + RefKind::Cell(a) => RefKind::Cell(mv(a)?), + RefKind::Range(a, b) => RefKind::Range(mv(a)?, mv(b)?), + RefKind::Rows(a, aa, b, ba) => { + let dr = *to_row as i64 - from.start.row as i64; + RefKind::Rows(shift_line(*a, false, dr, MAX_ROWS)?, *aa, shift_line(*b, false, dr, MAX_ROWS)?, *ba) + } + RefKind::Cols(a, aa, b, ba) => { + let dc = *to_col as i64 - from.start.col as i64; + RefKind::Cols(shift_line(*a, false, dc, MAX_COLS)?, *aa, shift_line(*b, false, dc, MAX_COLS)?, *ba) + } + }; + Some(Reference { sheet: r.sheet.clone(), kind }) + } + Edit::MoveToSheet { from, to_sheet, to_row, to_col } => { + if !from.contains_range(&r.range()) { + return Some(r.clone()); + } + let mut moved = adjust_ref(r, &Edit::Move { from: *from, to_row: *to_row, to_col: *to_col })?; + moved.sheet = SheetSel::Named(to_sheet.clone()); + Some(moved) + } + } + } +} + +/// Formulas with every reference shape the adjustment distinguishes. +const FORMULAS: &[&str] = &[ + "A1", + "B2+C3*D4", + "$B$2+B$2+$B2", + "SUM(A1:A10)+SUM(B2:C3)+SUM($B$2:$C$3)", + "SUM(C3:B2)", + "SUM(A:A)+SUM($B:$C)+SUM(2:2)+SUM($3:$5)", + "SUM(A1:.A7)+SUM(B:.B)+ROWS(2.:.6)+SUM(A1.:B1)", + "S!B2+s!C3+'S'!A1:B9", + "Other!B2+Other!A:A+Other!2:3", + "'My Sheet'!B2+'my sheet'!C3:D4", + "Dest!B2+S!B2", + "SUM(S:Other!B2)+SUM(Other:S!A1:C3)", + "XFD1048576+XFD1+A1048576+SUM(XFC1048575:XFD1048576)", + "SUM(1048576:1048576)+SUM(XFD:XFD)", + "VLOOKUP($B2,Lookup!$A$1:$D$500,3,FALSE)", + "IF(AND(C2>0,D2<100),C2*1.07,IF(E2=\"x\",\"yes\",\"no\"))", + "Table1[@Qty]*Table1[@Price]+SUM(Table1[[#Data],[Qty]])", + "SUMPRODUCT(A2:C2,{1,2,3})", + "LET(x,A2*2,y,B2+x,x*y)", + "LAMBDA(x,x+B2)(C3)", + "IFERROR(INDEX(Lookup!$A:$A,MATCH(F2,Lookup!$B:$B,0)),\"\")", + "-B2%+(C3)^2&\"a\"", + "A1#+S!B2#", + "S!MyName+MyName+#REF!+S!#REF!", + "COUNTIFS(Data!$A:$A,A2,Data!$C:$C,\">=\"&$G$2)", + "1+2", + "\"text\"", + "B2:C3 B3:D4", + "(B2:C3,E5)", + "INDIRECT(\"B2\")+OFFSET(B2,1,1)", + // From the parser edge cases (prefixes and LET parameter names). + "LET(_xlpm.TRUE,1,_xlpm.TRUE+1)", + "LET(_xlpm.A1,1,_xlpm.A1)", + "LAMBDA(_xlpm.x,_xlpm.x+1)(2)", + "_xlfn.LET(_xlpm.R1C1,1,_xlpm.R1C1*2)", + "LET(_xlpm.FALSE,2,_xlpm.FALSE)", + "_xlpm.SUM(1)", + "LET(_xlpm.x_1,1,_xlpm.x_1)", + "_xlfn.ANCHORARRAY(A1)+_xlfn.SINGLE(B1:B3)", + "SUM(_xlfn._TRO_ALL(A1:E10))", +]; + +fn col(c: u32) -> String { + gridcraft_core::addr::col_to_letters(c) +} + +/// The realistic mix the structural-edit benchmark uses, row-dependent. +fn mix(i: usize, r: u32) -> String { + let r1 = r + 1; + match i % 12 { + 0 => format!("SUM(A{r1}:A{})", r1 + 9), + 1 => format!("VLOOKUP($B{r1},Lookup!$A$1:$D$500,3,FALSE)"), + 2 => format!("IF(AND(C{r1}>0,D{r1}<100),C{r1}*1.07,IF(E{r1}=\"x\",\"yes\",\"no\"))"), + 3 => format!("'My Sheet'!B{r1}+Data!C{r1}*2-1"), + 4 => "Table1[@Qty]*Table1[@Price]".to_string(), + 5 => format!("SUMPRODUCT(A{r1}:C{r1},{{1,2,3}})+SUM(Table1[[#Data],[Qty]])"), + 6 => format!("\"Lorem ipsum\"&A{r1}&\" units\""), + 7 => format!("IFERROR(INDEX(Lookup!$A:$A,MATCH(F{r1},Lookup!$B:$B,0)),\"\")"), + 8 => (0..20).map(|k| format!("{}{r1}", col(k))).collect::>().join("+"), + 9 => format!("LET(x,A{r1}*2,y,B{r1}+x,x*y)"), + 10 => format!("ROUND(AVERAGE(B{r1}:F{r1})*(1+$H$1)^2,2)"), + _ => format!("COUNTIFS(Data!$A:$A,A{r1},Data!$C:$C,\">=\"&$G$2)"), + } +} + +fn range(r0: u32, c0: u32, r1: u32, c1: u32) -> RangeRef { + RangeRef::new(CellRef::new(r0, c0), CellRef::new(r1, c1)) +} + +/// Every edit kind, at the sheet's start, middle and edges, with counts that run past the end. +fn edits() -> Vec { + let mut v = Vec::new(); + for axis in [Axis::Rows, Axis::Cols] { + let max = if axis == Axis::Rows { MAX_ROWS } else { MAX_COLS }; + for at in [0, 1, 2, 3, 9, max - 2, max - 1] { + for count in [0, 1, 2, 5, max - 1, max, u32::MAX] { + v.push(Edit::Insert { axis, at, count }); + v.push(Edit::Delete { axis, at, count }); + } + } + for r in [ + range(1, 1, 2, 2), + range(0, 0, 0, 0), + range(1, 0, 1, MAX_COLS - 1), + range(0, 1, MAX_ROWS - 1, 1), + range(MAX_ROWS - 2, MAX_COLS - 2, MAX_ROWS - 1, MAX_COLS - 1), + ] { + v.push(Edit::InsertCells { axis, range: r }); + v.push(Edit::DeleteCells { axis, range: r }); + } + } + let blocks = [ + range(1, 1, 2, 2), + range(0, 0, 0, 0), + range(0, 0, MAX_ROWS - 1, MAX_COLS - 1), + range(0, 0, MAX_ROWS - 1, 2), + range(1, 0, 4, MAX_COLS - 1), + range(0, 0, 20, 10), + ]; + let targets = [(9, 3), (0, 0), (1, 1), (MAX_ROWS - 1, MAX_COLS - 1), (MAX_ROWS - 2, 0)]; + for from in blocks { + for (to_row, to_col) in targets { + v.push(Edit::Move { from, to_row, to_col }); + for to_sheet in ["Dest", "S", "s", "Other"] { + v.push(Edit::MoveToSheet { from, to_sheet: to_sheet.to_string(), to_row, to_col }); + } + } + } + v +} + +/// Compares the new and the old adjustment of `src` for every host and edit; returns the +/// number of (formula, host, edit) cases checked, or 0 when `src` does not parse. +fn check(src: &str, edits: &[Edit]) -> usize { + let Ok(e) = parse(src) else { return 0 }; + let mut n = 0; + for host in ["S", "s", "Other", "My Sheet"] { + for edit in edits { + let want = old::adjust(e.clone(), host, "S", edit); + let got = adjust(e.clone(), host, "S", edit); + assert!(got == want, "{src} on {host} after {edit:?}: {} vs {}", print(&got), print(&want)); + assert_eq!(print(&got), print(&want), "{src} on {host} after {edit:?}"); + match adjust_if_changed(&e, host, "S", edit) { + None => assert!(want == e, "{src} on {host} after {edit:?}: reported unchanged, old gives {}", print(&want)), + Some(x) => { + assert!(want != e, "{src} on {host} after {edit:?}: reported changed but the old code keeps it"); + assert!(x == want, "{src} on {host} after {edit:?}: {} vs {}", print(&x), print(&want)); + } + } + n += 1; + } + } + for (dr, dc) in [(0, 0), (1, 0), (0, 1), (-1, -1), (5, 3), (-3, 2), (MAX_ROWS as i64, 0), (0, -(MAX_COLS as i64)), (1_048_575, 16_383)] { + let want = old::shift_relative(e.clone(), dr, dc); + let got = shift_relative(e.clone(), dr, dc); + assert!(got == want, "{src} shifted by ({dr}, {dc}): {} vs {}", print(&got), print(&want)); + } + n +} + +#[test] +fn adjust_matches_the_previous_implementation_for_every_edit_kind() { + let edits = edits(); + let mut cases = 0; + for src in FORMULAS { + let n = check(src, &edits); + assert!(n > 0, "{src} does not parse"); + cases += n; + } + for i in 0..48 { + // Rows 1-2 and 1,048,565-1,048,566 of the mix (its longest range reaches 9 rows further + // down, to the last row): near both edges of the sheet. + let row = if i < 24 { (i / 12) as u32 } else { MAX_ROWS - 12 + (i / 12 - 2) as u32 }; + let src = mix(i, row); + let n = check(&src, &edits); + assert!(n > 0, "{src} does not parse"); + cases += n; + } + assert!(cases > 100_000, "{cases}"); +} + +/// A small deterministic generator: a failing case is reproduced by rerunning the test, and +/// its message names the case's index in the stream. +struct Rng(u64); + +impl Rng { + fn next(&mut self) -> u64 { + self.0 = self.0.wrapping_mul(6364136223846793005).wrapping_add(1442695040888963407); + self.0 >> 33 + } + fn below(&mut self, n: u64) -> u64 { + self.next() % n + } + fn line(&mut self, max: u32) -> u32 { + match self.below(3) { + 0 => self.below(12) as u32, + 1 => max - 1 - self.below(12) as u32, + _ => self.below(max as u64) as u32, + } + } +} + +fn random_ref(rng: &mut Rng) -> String { + let sheet = ["", "", "S!", "s!", "Other!", "'My Sheet'!", "S:Other!"][rng.below(7) as usize]; + let dollar = |rng: &mut Rng| if rng.below(3) == 0 { "$" } else { "" }; + let cell = |rng: &mut Rng| format!("{}{}{}{}", dollar(rng), col(rng.line(MAX_COLS)), dollar(rng), rng.line(MAX_ROWS) + 1); + let body = match rng.below(4) { + 0 => cell(rng), + 1 => format!("{}:{}", cell(rng), cell(rng)), + 2 => { + let (a, b) = (rng.line(MAX_ROWS) + 1, rng.line(MAX_ROWS) + 1); + format!("{}{a}:{}{b}", dollar(rng), dollar(rng)) + } + _ => { + let (a, b) = (rng.line(MAX_COLS), rng.line(MAX_COLS)); + format!("{}{}:{}{}", dollar(rng), col(a), dollar(rng), col(b)) + } + }; + format!("{sheet}{body}") +} + +fn random_edit(rng: &mut Rng) -> Edit { + let axis = if rng.below(2) == 0 { Axis::Rows } else { Axis::Cols }; + let max = if axis == Axis::Rows { MAX_ROWS } else { MAX_COLS }; + let block = |rng: &mut Rng| { + let (r0, r1) = (rng.line(MAX_ROWS), rng.line(MAX_ROWS)); + let (c0, c1) = (rng.line(MAX_COLS), rng.line(MAX_COLS)); + range(r0.min(r1), c0.min(c1), r0.max(r1), c0.max(c1)) + }; + let count = if rng.below(8) == 0 { rng.below(u32::MAX as u64) as u32 } else { rng.below(20) as u32 }; + match rng.below(7) { + 0 => Edit::Insert { axis, at: rng.line(max), count }, + 1 => Edit::Delete { axis, at: rng.line(max), count }, + 2 => Edit::InsertCells { axis, range: block(rng) }, + 3 => Edit::DeleteCells { axis, range: block(rng) }, + 4 => Edit::Move { from: block(rng), to_row: rng.line(MAX_ROWS), to_col: rng.line(MAX_COLS) }, + _ => Edit::MoveToSheet { + from: block(rng), + to_sheet: ["Dest", "S", "other"][rng.below(3) as usize].to_string(), + to_row: rng.line(MAX_ROWS), + to_col: rng.line(MAX_COLS), + }, + } +} + +/// Random references near the sheet's edges and random edits: "unchanged" is reported exactly +/// when the previous implementation leaves the formula as it was, and a reported change is the +/// old result. +#[test] +fn adjust_if_changed_reports_unchanged_exactly_when_nothing_moves() { + let mut rng = Rng(0x5eed); + let (mut unchanged, mut changed) = (0, 0); + for case in 0..20_000 { + let refs: Vec = (0..1 + rng.below(4)).map(|_| random_ref(&mut rng)).collect(); + let src = format!("SUM({})", refs.join(",")); + let Ok(e) = parse(&src) else { panic!("case {case}: {src} does not parse") }; + let edit = random_edit(&mut rng); + let host = ["S", "Other"][rng.below(2) as usize]; + let want = old::adjust(e.clone(), host, "S", &edit); + assert!(adjust(e.clone(), host, "S", &edit) == want, "case {case}: {src} on {host} after {edit:?}"); + match adjust_if_changed(&e, host, "S", &edit) { + None => { + assert!(want == e, "case {case}: {src} on {host} after {edit:?}: reported unchanged, old gives {}", print(&want)); + unchanged += 1; + } + Some(x) => { + assert!(want != e && x == want, "case {case}: {src} on {host} after {edit:?}"); + changed += 1; + } + } + } + // Both outcomes are exercised in quantity. + assert!(unchanged > 2_000 && changed > 2_000, "unchanged {unchanged}, changed {changed}"); +} + +/// The whole formula corpus (one formula per line, with or without `=`) when +/// `GRIDCRAFT_ADJUST_CORPUS` names a file: `cargo test -p gridcraft-formula --test it -- --ignored adjust_corpus`. +#[test] +#[ignore = "needs GRIDCRAFT_ADJUST_CORPUS"] +fn adjust_corpus_matches_the_previous_implementation() { + let Ok(path) = std::env::var("GRIDCRAFT_ADJUST_CORPUS") else { return }; + let text = std::fs::read_to_string(&path).unwrap_or_else(|e| panic!("{path}: {e}")); + let edits = edits(); + let (mut parsed, mut skipped, mut cases) = (0, 0, 0); + for line in text.lines().filter(|l| !l.trim().is_empty()) { + match check(line, &edits) { + 0 => skipped += 1, + n => { + parsed += 1; + cases += n; + } + } + } + eprintln!("corpus: {parsed} formulas parsed, {skipped} skipped (do not parse), {cases} cases"); +} + +#[test] +fn unaffected_formula_is_returned_as_is() { + let e: Expr = parse("Other!B2+SUM(A1:A3)").unwrap(); + let ins = Edit::Insert { axis: Axis::Rows, at: 5, count: 1 }; + // Row 6 inserted: A1:A3 lies above it and Other!B2 is not on S, so nothing moves, whichever + // sheet the formula lives on (on Other, A1:A3 is on Other too). + assert!(adjust_if_changed(&e, "S", "S", &ins).is_none()); + assert!(adjust_if_changed(&e, "Other", "S", &ins).is_none()); + // Row 1 inserted: A1:A3 on S moves, Other!B2 does not. + let top = Edit::Insert { axis: Axis::Rows, at: 0, count: 1 }; + assert_eq!(adjust_if_changed(&e, "S", "S", &top).map(|x| print(&x)).as_deref(), Some("Other!B2+SUM(A2:A4)")); + // A cut that moves a block onto the same place still qualifies its references with the sheet. + let same_place = Edit::MoveToSheet { from: range(0, 0, 2, 0), to_sheet: "S".into(), to_row: 0, to_col: 0 }; + assert_eq!(adjust_if_changed(&e, "S", "S", &same_place).map(|x| print(&x)).as_deref(), Some("Other!B2+SUM(S!A1:A3)")); +} diff --git a/crates/formula/tests/it/main.rs b/crates/formula/tests/it/main.rs index 01bff98c1..0f0292479 100644 --- a/crates/formula/tests/it/main.rs +++ b/crates/formula/tests/it/main.rs @@ -1,4 +1,5 @@ //! The crate's integration tests, one module per file, built as one test binary. +mod adjust_oracle; mod printer_roundtrip; mod spanish_input; From 2ee261c0e8ff389be25ac494d121fc7bde4a622c Mon Sep 17 00:00:00 2001 From: vangel76 Date: Sun, 8 Nov 2026 03:51:57 +0100 Subject: [PATCH 23/24] perf(model): move whole rows when inserting or deleting rows and columns (#3575) CellStore::shift_rows, shift_cols, shift_cols_in_rows and take_range removed and re-set every affected cell one by one. They now take whole row maps (or the split-off part of one) out of the bands and put them back at their new position, so a row shift costs a few map operations per row instead of a band lookup, a remove and a set per cell. take_range does this when the block reaches the last column. Behaviour is unchanged, including blank cells left behind by get_mut, cells off the grid (they are still dropped, and still clear the cell they land on), cells shifted off the sheet, and copy-on-write: bands without moved cells stay shared with undo snapshots. A differential test runs 12,000 random operations against the old per-cell code and compares contents, length, iteration order and band sharing. shift_rows_in_cols, and take_range on a block with columns to its right, keep moving cell by cell: the row-map version measured no faster there. Co-Authored-By: Claude Opus 5.5 --- crates/model/src/store.rs | 402 ++++++++++++++++++++++++++++++++++---- 1 file changed, 367 insertions(+), 35 deletions(-) diff --git a/crates/model/src/store.rs b/crates/model/src/store.rs index be52bbe35..e2da31e43 100644 --- a/crates/model/src/store.rs +++ b/crates/model/src/store.rs @@ -4,6 +4,8 @@ //! copies only the band pointers; editing a cell copies just its band. use std::collections::BTreeMap; +use std::collections::btree_map::Entry; +use std::ops::RangeInclusive; use std::sync::Arc; use gridcraft_core::{CellRef, RangeRef}; @@ -266,64 +268,187 @@ impl CellStore { } bytes } - /// Removes every cell in `r` and returns them. + /// Removes every cell in `r` and returns them, row-major. pub fn take_range(&mut self, r: RangeRef) -> Vec<(CellRef, Cell)> { - let keys: Vec = self.iter_range(r).map(|(c, _)| c).collect(); - keys.into_iter().filter_map(|c| self.remove(c).map(|cell| (c, cell))).collect() + if r.end.col < gridcraft_core::MAX_COLS - 1 { + // A block with columns to its right: cutting it out of each row map measured slower + // than removing its cells one by one (#3575). + let keys: Vec = self.iter_range(r).map(|(c, _)| c).collect(); + return keys.into_iter().filter_map(|c| self.remove(c).map(|cell| (c, cell))).collect(); + } + let rows = self.take_rows(r.start.row..=r.end.row, r.start.col..=r.end.col); + rows.into_iter().flat_map(|(row, cols)| cols.into_iter().map(move |(col, cell)| (CellRef::new(row, col), cell))).collect() } /// Rows strictly greater than or equal to `from` moved by `delta` (positive = down). Rows that /// would land before `from`... callers delete the vacated block first. pub fn shift_rows(&mut self, from: u32, delta: i64) { - let moved: Vec<(CellRef, Cell)> = { - let keys: Vec = self.iter().filter(|(c, _)| c.row >= from).map(|(c, _)| c).collect(); - keys.into_iter().filter_map(|c| self.remove(c).map(|cell| (c, cell))).collect() - }; - for (c, cell) in moved { - let nr = c.row as i64 + delta; - if (0..gridcraft_core::MAX_ROWS as i64).contains(&nr) { - self.set(CellRef::new(nr as u32, c.col), cell); - } - } + let moved = self.take_rows(from..=u32::MAX, 0..=u32::MAX); + self.put_rows(moved.into_iter().filter_map(|(row, cols)| Some((shifted(row, delta, gridcraft_core::MAX_ROWS)?, cols)))); } pub fn shift_cols(&mut self, from: u32, delta: i64) { - let keys: Vec = self.iter().filter(|(c, _)| c.col >= from).map(|(c, _)| c).collect(); - let moved: Vec<(CellRef, Cell)> = keys.into_iter().filter_map(|c| self.remove(c).map(|cell| (c, cell))).collect(); - for (c, cell) in moved { - let nc = c.col as i64 + delta; - if (0..gridcraft_core::MAX_COLS as i64).contains(&nc) { - self.set(CellRef::new(c.row, nc as u32), cell); - } - } + let moved = self.take_rows(0..=u32::MAX, from..=u32::MAX); + self.put_rows(moved.into_iter().map(|(row, cols)| (row, shift_keys(cols, delta)))); } /// Shifts cells within rows `r0..=r1` and columns `>= from` horizontally (Insert/Delete Cells /// with shift right/left). pub fn shift_cols_in_rows(&mut self, r0: u32, r1: u32, from: u32, delta: i64) { - let range = RangeRef::new(CellRef::new(r0, from), CellRef::new(r1, gridcraft_core::MAX_COLS - 1)); - let moved = self.take_range(range); - for (c, cell) in moved { - let nc = c.col as i64 + delta; - if (0..gridcraft_core::MAX_COLS as i64).contains(&nc) { - self.set(CellRef::new(c.row, nc as u32), cell); - } - } + let r = RangeRef::new(CellRef::new(r0, from), CellRef::new(r1, gridcraft_core::MAX_COLS - 1)); + let moved = self.take_rows(r.start.row..=r.end.row, r.start.col..=r.end.col); + self.put_rows(moved.into_iter().map(|(row, cols)| (row, shift_keys(cols, delta)))); } /// Shifts cells within columns `c0..=c1` and rows `>= from` vertically. pub fn shift_rows_in_cols(&mut self, c0: u32, c1: u32, from: u32, delta: i64) { + // Cell by cell: the moved cells sit between columns that stay, so moving row maps + // measured no faster here (#3575). let range = RangeRef::new(CellRef::new(from, c0), CellRef::new(gridcraft_core::MAX_ROWS - 1, c1)); - let moved = self.take_range(range); - for (c, cell) in moved { - let nr = c.row as i64 + delta; - if (0..gridcraft_core::MAX_ROWS as i64).contains(&nr) { - self.set(CellRef::new(nr as u32, c.col), cell); + for (c, cell) in self.take_range(range) { + if let Some(nr) = shifted(c.row, delta, gridcraft_core::MAX_ROWS) { + self.set(CellRef::new(nr, c.col), cell); + } + } + } + /// Removes the cells in `rows` x `cols` and returns them grouped by row, in row order. Moves + /// whole column maps (or the split-off part of one) instead of removing cell by cell, and only + /// copies bands that hold a cell in the block, so snapshots keep sharing the others. + fn take_rows(&mut self, rows: RangeInclusive, cols: RangeInclusive) -> Vec<(u32, BTreeMap)> { + let (r0, r1, c0, c1) = (*rows.start(), *rows.end(), *cols.start(), *cols.end()); + let mut out = Vec::new(); + if r0 > r1 || c0 > c1 { + return out; + } + let whole_rows = c0 == 0 && c1 == u32::MAX; + let keys: Vec = self + .bands + .range(r0 / BAND..=r1 / BAND) + .filter(|(_, band)| band.rows.range(r0..=r1).any(|(_, row)| row.range(c0..=c1).next().is_some())) + .map(|(k, _)| *k) + .collect(); + for key in keys { + let Some(band) = self.bands.get_mut(&key) else { continue }; + let inside = |row: Option<&u32>| row.is_some_and(|r| (r0..=r1).contains(r)); + if whole_rows && inside(band.rows.keys().next()) && inside(band.rows.keys().next_back()) { + // The whole band moves: no splitting. + let Some(band) = self.bands.remove(&key) else { continue }; + for (row, taken) in Arc::unwrap_or_clone(band).rows { + self.len -= taken.len(); + out.push((row, taken)); + } + continue; + } + let band = Arc::make_mut(band); + let mut block = band.rows.split_off(&r0); + let mut tail = match r1.checked_add(1) { + Some(next) => block.split_off(&next), + None => BTreeMap::new(), + }; + for (row, mut kept) in block { + let taken = if whole_rows { + std::mem::take(&mut kept) + } else if c1.checked_add(1).is_none_or(|next| kept.range(next..).next().is_none()) { + // Nothing right of the block: one split. + kept.split_off(&c0) + } else { + let mut taken = BTreeMap::new(); + for (col, cell) in kept.extract_if(c0..=c1, |_, _| true) { + taken.insert(col, cell); + } + taken + }; + if !kept.is_empty() { + band.rows.insert(row, kept); + } + if !taken.is_empty() { + self.len -= taken.len(); + out.push((row, taken)); + } + } + band.rows.append(&mut tail); + if band.rows.is_empty() { + self.bands.remove(&key); + } + } + out + } + /// Stores rows of cells (each row at most once) the way `set` stores each cell: a cell + /// replaces what is there, and a blank cell clears its destination instead of being stored. + /// A row with nothing stored yet takes the whole map. Rows in order share one band lookup. + fn put_rows(&mut self, rows: impl Iterator)>) { + let mut rows = rows.peekable(); + let mut group = Vec::new(); + while let Some((first, cols)) = rows.next() { + let key = first / BAND; + group.clear(); + group.push((first, cols)); + while let Some(next) = rows.next_if(|(row, _)| row / BAND == key) { + group.push(next); + } + // Leave the band shared when only blank cells land on empty positions. + let existing = self.bands.get(&key); + let writes = group.iter().any(|(row, cols)| { + let dest = existing.and_then(|band| band.rows.get(row)); + cols.iter().any(|(col, cell)| !cell.is_blank() || dest.is_some_and(|d| d.contains_key(col))) + }); + if !writes { + continue; + } + let band = Arc::make_mut(self.bands.entry(key).or_default()); + for (row, mut cols) in group.drain(..) { + match band.rows.entry(row) { + Entry::Vacant(slot) => { + if cols.values().any(Cell::is_blank) { + cols.retain(|_, cell| !cell.is_blank()); + } + if !cols.is_empty() { + self.len += cols.len(); + slot.insert(cols); + } + } + Entry::Occupied(mut slot) => { + let dest = slot.get_mut(); + for (col, cell) in cols { + if cell.is_blank() { + if dest.remove(&col).is_some() { + self.len -= 1; + } + } else if dest.insert(col, cell).is_none() { + self.len += 1; + } + } + if dest.is_empty() { + slot.remove(); + } + } + } + } + if band.rows.is_empty() { + self.bands.remove(&key); } } } } +/// `at + delta` when it lies in `0..limit`. +fn shifted(at: u32, delta: i64, limit: u32) -> Option { + let to = (at as i64).checked_add(delta)?; + if (0..limit as i64).contains(&to) { Some(to as u32) } else { None } +} + +/// The cells of one row with their columns moved by `delta`; cells that would leave the sheet are +/// dropped. +fn shift_keys(cols: BTreeMap, delta: i64) -> BTreeMap { + let mut out = BTreeMap::new(); + for (col, cell) in cols { + if let Some(c) = shifted(col, delta, gridcraft_core::MAX_COLS) { + out.insert(c, cell); + } + } + out +} + #[cfg(test)] mod tests { use super::*; - use gridcraft_core::Value; + use gridcraft_core::{MAX_COLS, MAX_ROWS, Value}; fn c(r: u32, col: u32) -> CellRef { CellRef::new(r, col) @@ -396,4 +521,211 @@ mod tests { s.shift_cols(0, 1); assert!(s.get(c(7, 1)).is_some()); } + + /// The per-cell algorithms the shifts used before they moved whole rows (#3575), kept as the + /// reference the structural versions must match cell for cell. + mod per_cell { + use super::*; + + fn reset(s: &mut CellStore, moved: Vec<(CellRef, Cell)>, to: impl Fn(CellRef) -> Option) { + for (at, cell) in moved { + if let Some(at) = to(at) { + s.set(at, cell); + } + } + } + fn row_by(delta: i64) -> impl Fn(CellRef) -> Option { + move |at| { + let r = at.row as i64 + delta; + (0..MAX_ROWS as i64).contains(&r).then(|| CellRef::new(r as u32, at.col)) + } + } + fn col_by(delta: i64) -> impl Fn(CellRef) -> Option { + move |at| { + let c = at.col as i64 + delta; + (0..MAX_COLS as i64).contains(&c).then(|| CellRef::new(at.row, c as u32)) + } + } + pub fn take_range(s: &mut CellStore, r: RangeRef) -> Vec<(CellRef, Cell)> { + let keys: Vec = s.iter_range(r).map(|(c, _)| c).collect(); + keys.into_iter().filter_map(|c| s.remove(c).map(|cell| (c, cell))).collect() + } + pub fn shift_rows(s: &mut CellStore, from: u32, delta: i64) { + let keys: Vec = s.iter().filter(|(c, _)| c.row >= from).map(|(c, _)| c).collect(); + let moved = keys.into_iter().filter_map(|c| s.remove(c).map(|cell| (c, cell))).collect(); + reset(s, moved, row_by(delta)); + } + pub fn shift_cols(s: &mut CellStore, from: u32, delta: i64) { + let keys: Vec = s.iter().filter(|(c, _)| c.col >= from).map(|(c, _)| c).collect(); + let moved = keys.into_iter().filter_map(|c| s.remove(c).map(|cell| (c, cell))).collect(); + reset(s, moved, col_by(delta)); + } + pub fn shift_cols_in_rows(s: &mut CellStore, r0: u32, r1: u32, from: u32, delta: i64) { + let moved = take_range(s, RangeRef::new(CellRef::new(r0, from), CellRef::new(r1, MAX_COLS - 1))); + reset(s, moved, col_by(delta)); + } + pub fn shift_rows_in_cols(s: &mut CellStore, c0: u32, c1: u32, from: u32, delta: i64) { + let moved = take_range(s, RangeRef::new(CellRef::new(from, c0), CellRef::new(MAX_ROWS - 1, c1))); + reset(s, moved, row_by(delta)); + } + } + + struct Lcg(u64); + impl Lcg { + fn below(&mut self, n: u64) -> u64 { + self.0 = self.0.wrapping_mul(6364136223846793005).wrapping_add(1442695040888963407); + (self.0 >> 33) % n.max(1) + } + fn pick(&mut self, xs: &[T]) -> T { + xs[self.below(xs.len() as u64) as usize] + } + } + + /// A position near a band edge, the sheet edge, or anywhere below `max`. + fn coord(rng: &mut Lcg, max: u32, edges: &[u32]) -> u32 { + match rng.below(3) { + 0 => rng.pick(edges).min(max - 1), + 1 => max - 1 - rng.below(70.min(max as u64)) as u32, + _ => rng.below(max as u64) as u32, + } + } + + /// Random store with values, styled-only cells, a few cells off the grid, and blank cells left + /// behind by `get_mut` (which `set` never stores, but the old shifts dropped and let clear + /// their destination). + fn random_store(rng: &mut Lcg, max_r: u32, max_c: u32) -> CellStore { + let row_edges = [0, 1, 62, 63, 64, 65, 127, 128, 129, 191, 192, MAX_ROWS - 64, MAX_ROWS - 1]; + let col_edges = [0, 1, 2, 63, 64, 65, MAX_COLS - 2, MAX_COLS - 1]; + let mut s = CellStore::new(); + for _ in 0..rng.below(300) { + let mut at = CellRef::new(coord(rng, max_r, &row_edges), coord(rng, max_c, &col_edges)); + // Now and then a cell outside the grid (`set` does not check): the shifts must treat + // it as the old code did. + match rng.below(60) { + 0 => at.row = MAX_ROWS + rng.below(3) as u32, + 1 => at.col = MAX_COLS + rng.below(3) as u32, + _ => {} + } + let cell = match rng.below(5) { + 0 => Cell { style: crate::style::StyleId(1 + rng.below(3) as u32), ..Cell::default() }, + 1 => Cell::value(Value::from("t")), + _ => Cell::value(Value::Number(rng.below(1000) as f64)), + }; + s.set(at, cell); + if rng.below(8) == 0 + && let Some(cell) = s.get_mut(at) + { + *cell = Cell::default(); + } + } + s + } + + fn assert_same(new: &CellStore, old: &CellStore, base: &CellStore, base_copy: &CellStore, what: &str) { + assert_eq!(new, old, "{what}"); + assert_eq!(new.len(), new.iter().count(), "{what}: len"); + assert!(new.iter().map(|(c, x)| (c, x.clone())).eq(old.iter().map(|(c, x)| (c, x.clone()))), "{what}: iteration"); + assert!(new.bands.values().all(|b| !b.rows.is_empty() && b.rows.values().all(|r| !r.is_empty())), "{what}: empty band or row"); + // Copy-on-write: every band the old code left shared with the snapshot stays shared. + for (k, b) in &base.bands { + if old.bands.get(k).is_some_and(|o| Arc::ptr_eq(o, b)) { + assert!(new.bands.get(k).is_some_and(|n| Arc::ptr_eq(n, b)), "{what}: band {k} copied"); + } + } + assert_eq!(base, base_copy, "{what}: snapshot changed"); + } + + #[test] + fn structural_shifts_match_per_cell_shifts() { + let mut rng = Lcg(3575); + let deltas = [0i64, 1, -1, 2, -2, 63, -63, 64, -64, 65, -65, 128, -128, 1000, -1000, MAX_ROWS as i64, -(MAX_ROWS as i64)]; + let row_edges = [0, 1, 63, 64, 65, 127, 128, MAX_ROWS - 1, MAX_ROWS]; + let col_edges = [0, 1, 63, 64, 65, MAX_COLS - 1, MAX_COLS]; + for trial in 0..3000 { + let max_r = [200, 5000, MAX_ROWS][trial % 3]; + let max_c = [30, MAX_COLS][trial / 3 % 2]; + let mut base = random_store(&mut rng, max_r, max_c); + for step in 0..4 { + let snapshot = base.clone(); + let (mut new, mut old) = (base.clone(), base.clone()); + let delta = if rng.below(2) == 0 { rng.pick(&deltas) } else { rng.below(301) as i64 - 150 }; + let row = if rng.below(2) == 0 { rng.pick(&row_edges) } else { rng.below(max_r as u64 + 1) as u32 }; + let col = if rng.below(2) == 0 { rng.pick(&col_edges) } else { rng.below(max_c as u64 + 1) as u32 }; + let (row2, col2) = (rng.below(max_r as u64 + 1) as u32, rng.below(max_c as u64 + 1) as u32); + let op = rng.below(5); + match op { + 0 => { + new.shift_rows(row, delta); + per_cell::shift_rows(&mut old, row, delta); + } + 1 => { + new.shift_cols(col, delta); + per_cell::shift_cols(&mut old, col, delta); + } + 2 => { + new.shift_cols_in_rows(row, row2, col, delta); + per_cell::shift_cols_in_rows(&mut old, row, row2, col, delta); + } + 3 => { + new.shift_rows_in_cols(col, col2, row, delta); + per_cell::shift_rows_in_cols(&mut old, col, col2, row, delta); + } + _ => { + let r = RangeRef::new(CellRef::new(row, col), CellRef::new(row2, col2)); + assert_eq!(new.take_range(r), per_cell::take_range(&mut old, r), "trial {trial} step {step}: taken cells"); + } + } + let what = format!("trial {trial} step {step} op {op} row {row}..{row2} col {col}..{col2} delta {delta}"); + assert_same(&new, &old, &base, &snapshot, &what); + base = new; + } + } + } + + #[test] + fn off_grid_cells_right_of_a_block_stay_put() { + let mut s = CellStore::new(); + for col in [0, 5, MAX_COLS - 1, MAX_COLS + 1] { + s.set(c(3, col), Cell::value((col as f64).into())); + } + let mut old = s.clone(); + s.shift_cols_in_rows(3, 3, 0, 1); + per_cell::shift_cols_in_rows(&mut old, 3, 3, 0, 1); + assert_eq!(s, old); + assert_eq!(s.get(c(3, MAX_COLS + 1)).map(|x| x.value.clone()), Some(Value::Number((MAX_COLS + 1) as f64))); + let r = RangeRef::rows(3, 3); + assert_eq!(s.take_range(r), per_cell::take_range(&mut old, r)); + assert_eq!(s, old); + } + + #[test] + fn moved_blank_cell_clears_its_destination() { + let mut s = CellStore::new(); + s.set(c(3, 0), Cell::value(1.0.into())); + s.set(c(5, 0), Cell::value(2.0.into())); + s.set(c(5, 1), Cell::value(3.0.into())); + if let Some(cell) = s.get_mut(c(5, 0)) { + *cell = Cell::default(); + } + s.shift_rows(5, -2); + assert!(s.get(c(3, 0)).is_none()); + assert_eq!(s.get(c(3, 1)).map(|x| x.value.clone()), Some(Value::Number(3.0))); + assert_eq!(s.len(), 1); + } + + #[test] + fn shifting_keeps_untouched_bands_shared() { + let mut s = CellStore::new(); + for r in 0..256 { + s.set(c(r, 0), Cell::value((r as f64).into())); + } + s.set(c(200, 5), Cell::value("x".into())); + let snap = s.clone(); + s.shift_rows(130, 1); + assert!(Arc::ptr_eq(&s.bands[&0], &snap.bands[&0])); + assert!(Arc::ptr_eq(&s.bands[&1], &snap.bands[&1])); + s.shift_cols(5, 1); + assert!(Arc::ptr_eq(&s.bands[&0], &snap.bands[&0])); + assert_eq!(s.get(c(201, 6)).map(|x| x.value.clone()), Some(Value::from("x"))); + } } From 9ab297ced1115c2088bb0065588a4500563476f2 Mon Sep 17 00:00:00 2001 From: vangel76 Date: Sun, 8 Nov 2026 04:06:21 +0100 Subject: [PATCH 24/24] perf(engine): rewrite only the formulas a structural edit changes rewrite_all_formulas borrows each tree with Formula::with_expr and calls adjust_if_changed, so a formula the edit leaves alone is neither copied nor printed, and a changed one is printed once. A legacy array whose range moves while its formula stays the same still gets the new range. Sheet rename and delete compare the adjusted tree with the old one instead of printed text, and replace only formulas that changed. Delete no longer rebuilds every formula in the workbook, so unrelated formulas keep their array range and dynamic-array mark. A stored non-canonical formula text is no longer re-printed by an edit that does not change the formula; a test pins this. Part of #3576 Co-Authored-By: Claude Opus 5.5 --- crates/engine/src/cmd/edit/paste_special.rs | 18 ++-- crates/engine/src/cmd/sheet/tabs.rs | 25 +++--- crates/engine/tests/it/main.rs | 1 + .../tests/it/structural_edit_rewrite.rs | 84 +++++++++++++++++++ 4 files changed, 106 insertions(+), 22 deletions(-) create mode 100644 crates/engine/tests/it/structural_edit_rewrite.rs diff --git a/crates/engine/src/cmd/edit/paste_special.rs b/crates/engine/src/cmd/edit/paste_special.rs index df2b0de61..c90dd9ec8 100644 --- a/crates/engine/src/cmd/edit/paste_special.rs +++ b/crates/engine/src/cmd/edit/paste_special.rs @@ -567,19 +567,19 @@ pub(crate) fn rewrite_all_formulas(wb: &mut gridcraft_model::Workbook, target: & let on_target = moves_arrays && host.eq_ignore_ascii_case(target); let Some(sheet) = wb.sheet_mut(si) else { continue }; for (c, f) in keys { - let Some(expr) = f.expr() else { continue }; - let new = gridcraft_formula::adjust::adjust(expr, &host, target, e); - let text = gridcraft_formula::print(&new); + // Borrows the tree and copies and prints only the formulas the edit changes. A formula + // the edit leaves alone keeps its stored text, even if that text is not canonical. + let Some(new) = f.with_expr(|x| gridcraft_formula::adjust::adjust_if_changed(x, &host, target, e)) else { continue }; let array = match f.array { Some(a) if on_target => crate::cmd::sheet::map_range(a, e), a => a, }; - if (text != f.text || array != f.array) - && let Some(cell) = sheet.cells.get_mut(c) - { - let mut nf = Formula::from_expr(new); - nf.array = array; - nf.dynamic = f.dynamic; + let nf = match new { + Some(new) => Formula { array, dynamic: f.dynamic, ..Formula::from_expr(new) }, + None if array != f.array => Formula { array, ..(*f).clone() }, + None => continue, + }; + if let Some(cell) = sheet.cells.get_mut(c) { cell.formula = Some(std::sync::Arc::new(nf)); } } diff --git a/crates/engine/src/cmd/sheet/tabs.rs b/crates/engine/src/cmd/sheet/tabs.rs index d21851553..2c1657ef9 100644 --- a/crates/engine/src/cmd/sheet/tabs.rs +++ b/crates/engine/src/cmd/sheet/tabs.rs @@ -55,11 +55,12 @@ pub(super) fn delete_sheet(s: &mut Session, p: &Json) -> Result { cx.wb.sheets.get(si).map(|s| s.cells.iter().filter_map(|(c, x)| x.formula.clone().map(|f| (c, f))).collect()).unwrap_or_default(); let Some(sh) = cx.wb.sheet_mut(si) else { continue }; for (c, f) in keys { - if let Some(e) = f.expr() { - let ne = gridcraft_formula::adjust::delete_sheet(e, &name, &order); - if let Some(cell) = sh.cells.get_mut(c) { - cell.formula = Some(Arc::new(gridcraft_model::Formula::from_expr(ne))); - } + // Only formulas whose tree changes are rewritten; the others keep their text. + let changed = f.with_expr(|e| Some(gridcraft_formula::adjust::delete_sheet(e.clone(), &name, &order)).filter(|ne| ne != e)); + if let Some(Some(ne)) = changed + && let Some(cell) = sh.cells.get_mut(c) + { + cell.formula = Some(Arc::new(gridcraft_model::Formula::from_expr(ne))); } } } @@ -109,14 +110,12 @@ pub(super) fn rename_sheet(s: &mut Session, p: &Json) -> Result { cx.wb.sheets.get(si).map(|s| s.cells.iter().filter_map(|(c, x)| x.formula.clone().map(|f| (c, f))).collect()).unwrap_or_default(); let Some(sh) = cx.wb.sheet_mut(si) else { continue }; for (c, f) in keys { - if let Some(e) = f.expr() { - let ne = gridcraft_formula::adjust::rename_sheet(e, &old, &name); - let text = gridcraft_formula::print(&ne); - if text != f.text - && let Some(cell) = sh.cells.get_mut(c) - { - cell.formula = Some(Arc::new(gridcraft_model::Formula::from_expr(ne))); - } + // Compares trees, so unchanged formulas are not printed; they keep their text. + let changed = f.with_expr(|e| Some(gridcraft_formula::adjust::rename_sheet(e.clone(), &old, &name)).filter(|ne| ne != e)); + if let Some(Some(ne)) = changed + && let Some(cell) = sh.cells.get_mut(c) + { + cell.formula = Some(Arc::new(gridcraft_model::Formula::from_expr(ne))); } } for ch in sh.charts.iter_mut() { diff --git a/crates/engine/tests/it/main.rs b/crates/engine/tests/it/main.rs index 0ba7c0db8..ca960d759 100644 --- a/crates/engine/tests/it/main.rs +++ b/crates/engine/tests/it/main.rs @@ -29,6 +29,7 @@ mod sheet_copy_spill; mod sheet_name_quoting; mod sparse_clear_invalidation; mod stored_list_shape; +mod structural_edit_rewrite; mod table_resize_overlap; mod table_total_row_last_row; mod trace_error; diff --git a/crates/engine/tests/it/structural_edit_rewrite.rs b/crates/engine/tests/it/structural_edit_rewrite.rs new file mode 100644 index 000000000..bb87ba3c4 --- /dev/null +++ b/crates/engine/tests/it/structural_edit_rewrite.rs @@ -0,0 +1,84 @@ +//! Structural edits (insert and delete of lines, sheet rename and delete) rewrite only the formulas +//! whose tree they change. A formula they leave alone keeps its stored text and its flags. + +use std::sync::Arc; + +use gridcraft_core::{CellRef, RangeRef}; +use gridcraft_engine::Session; +use gridcraft_model::{Cell, Formula}; +use serde_json::{Value as Json, json}; + +fn run(s: &mut Session, command: &str, params: Json) { + s.execute(command, params).unwrap(); +} + +fn workbook() -> Session { + let mut s = Session::new(); + s.new_workbook(); + run(&mut s, "home.insertSheet", json!({"name": "Other"})); + run(&mut s, "home.insertSheet", json!({"name": "Spare"})); + run(&mut s, "sheet.activate", json!({"sheet": "Sheet1"})); + s +} + +fn formula(s: &Session, sheet: &str, cell: &str) -> Arc { + let wb = &s.doc().unwrap().wb; + wb.sheet(wb.sheet_index(sheet).unwrap()).unwrap().cell(CellRef::parse(cell).unwrap()).unwrap().formula.clone().unwrap() +} + +/// Stores a formula with this exact text and no cached tree, as a JSON workbook loads it. +fn put_raw(s: &mut Session, cell: &str, text: &str) { + let wb = Arc::make_mut(&mut s.doc_mut().unwrap().wb); + let i = wb.sheet_index("Sheet1").unwrap(); + let f = Formula { text: text.to_string(), expr: None, array: None, dynamic: false }; + wb.sheet_mut(i).unwrap().cells.set(CellRef::parse(cell).unwrap(), Cell::formula(f)); +} + +#[test] +fn an_edit_that_leaves_a_formula_alone_keeps_its_stored_text() { + // The rewrite compares trees, not printed text, so a stored non-canonical text is not + // re-printed by an edit that does not change the formula. An edit that changes it prints it. + let mut s = workbook(); + put_raw(&mut s, "C1", "sum(a1, Other!b2)"); + run(&mut s, "home.insertRows", json!({"rows": "5:5"})); + run(&mut s, "home.insertColumns", json!({"cols": "F:F"})); + run(&mut s, "sheet.rename", json!({"sheet": "Spare", "name": "Spare2"})); + run(&mut s, "home.deleteSheet", json!({"sheet": "Spare2"})); + assert_eq!(formula(&s, "Sheet1", "C1").text, "sum(a1, Other!b2)"); + + run(&mut s, "home.insertRows", json!({"rows": "1:1"})); + assert_eq!(formula(&s, "Sheet1", "C2").text, "SUM(A2,Other!B2)"); +} + +#[test] +fn renaming_a_referenced_sheet_rewrites_the_formula() { + let mut s = workbook(); + put_raw(&mut s, "C1", "sum(a1, Other!b2)"); + run(&mut s, "sheet.rename", json!({"sheet": "Other", "name": "Data"})); + assert_eq!(formula(&s, "Sheet1", "C1").text, "SUM(A1,Data!B2)"); + run(&mut s, "home.deleteSheet", json!({"sheet": "Data"})); + assert_eq!(formula(&s, "Sheet1", "C1").text, "SUM(A1,#REF!)"); +} + +#[test] +fn inserting_rows_above_a_legacy_array_moves_its_range_when_the_formula_is_unchanged() { + let mut s = workbook(); + run(&mut s, "range.setValues", json!({"range": "B1", "values": [[1], [2]]})); + run(&mut s, "selection.set", json!({"range": "C5:C6"})); + run(&mut s, "cell.set", json!({"cell": "C5", "input": "=B1:B2*2", "array": true})); + assert_eq!(formula(&s, "Sheet1", "C5").array, Some(RangeRef::parse("C5:C6").unwrap())); + run(&mut s, "home.insertRows", json!({"rows": "3:3"})); + let f = formula(&s, "Sheet1", "C6"); + assert_eq!(f.text, "B1:B2*2"); + assert_eq!(f.array, Some(RangeRef::parse("C6:C7").unwrap())); +} + +#[test] +fn deleting_an_unrelated_sheet_keeps_legacy_arrays() { + let mut s = workbook(); + run(&mut s, "range.setValues", json!({"range": "B1", "values": [[1], [2]]})); + run(&mut s, "selection.set", json!({"range": "C1:C2"})); + run(&mut s, "cell.set", json!({"cell": "C1", "input": "=B1:B2*2", "array": true})); + run(&mut s, "home.deleteSheet", json!({"sheet": "Spare"})); + assert_eq!(formula(&s, "Sheet1", "C1").array, Some(RangeRef::parse("C1:C2").unwrap())); +}