diff --git a/crates/audio/src/lib.rs b/crates/audio/src/lib.rs index e0660d9c9..be0678303 100644 --- a/crates/audio/src/lib.rs +++ b/crates/audio/src/lib.rs @@ -58,6 +58,8 @@ mod read; mod track; #[cfg(any(feature = "decode-native", feature = "pcm-encode"))] mod wav; +/// Bounded WAV/RF64 cue, adtl and XMP marker metadata, without audio decoding. +pub mod wav_markers; #[cfg(any( feature = "pcm-encode", feature = "flac-encode", diff --git a/crates/audio/src/wav_markers.rs b/crates/audio/src/wav_markers.rs new file mode 100644 index 000000000..0bf4e7568 --- /dev/null +++ b/crates/audio/src/wav_markers.rs @@ -0,0 +1,721 @@ +//! Markers a WAV file carries, read from its RIFF chunks. +//! +//! Two encodings are read into separate lists for the caller to map and merge: +//! - **Cue points** (the standard): a `cue ` chunk with one point per marker, named by `labl` (or +//! `note`) sub-chunks of a `LIST` `adtl` chunk, and given a length by `ltxt`. +//! - **XMP** (Adobe's tools): an `_PMX` chunk holding an XMP packet whose `xmpDM:markers` list has +//! one entry per marker (`xmpDM:startTime`, `xmpDM:duration`, `xmpDM:name`), counted in frames +//! of `xmpDM:frameRate` (`f44100` for samples at 44.1 kHz), or with an explicit +//! FrameCount rate suffix (`15f24`, `15f30000s1001`). Durations with a distinct rate +//! are converted to the marker start rate without app-specific time quantization. +//! +//! RF64 files (WAVs over 4 GB) carry a `0xFFFFFFFF` size for a chunk too big for 32 bits and give its real size in a +//! `ds64` chunk; those sizes are resolved so the chunks after the audio are still found. +//! +//! The file is untrusted input: every read is bounds-checked, sizes use checked arithmetic, and +//! the number of markers, of `adtl` sub-chunks looked at and the size of the XMP packet are capped. + +use std::collections::{HashMap, HashSet}; + +/// Marker coordinates before app-specific time quantization. Explicit XMP duration +/// rates are normalized to the start coordinate rate; bare counts retain the legacy +/// track/file-rate interpretation. +#[derive(Clone, Debug, PartialEq)] +pub struct Marker { + /// Start in frames at `rate`. + pub start: f64, + /// Duration in frames at `rate`. + pub duration: f64, + /// Frames per second; consumers decide how invalid or absent rates map to their timeline. + pub rate: f64, + /// Label text. + pub name: String, +} +/// Separate source lists: rounding, deduplication, ordering and labels belong to the caller. +#[derive(Clone, Debug, Default, PartialEq)] +pub struct Markers { + /// RIFF cue/adtl markers in file order. + pub cue: Vec, + /// XMP markers in packet order. + pub xmp: Vec, + /// A chunk-inventory or cue-label retention limit omitted metadata. Marker coordinates + /// remain available when only their label was omitted. Existing marker/XMP prefix limits + /// are separate and do not set this flag. + pub limited: bool, +} + +/// Most markers read from one file. +pub const MAX_MARKERS: usize = 10_000; +/// Largest XMP packet read (bytes). +const MAX_XMP: usize = 4 << 20; +/// Most `adtl` sub-chunks (`labl`, `note`, `ltxt`) looked at in one file: a few per cue point is plenty. +const MAX_ADTL_RECORDS: usize = 4 * MAX_MARKERS; +/// Most entries read from an RF64 file's `ds64` size table. +const MAX_DS64_TABLE: usize = 1_024; +/// Maximum retained RIFF chunk descriptors (including empty chunks). +pub const MAX_CHUNKS: usize = 100_000; +/// Maximum UTF-8 bytes in one retained cue label. +pub const MAX_LABEL_BYTES: usize = 64 * 1024; +/// Aggregate cue label bytes, enforced separately for the name map and emitted markers. +pub const MAX_LABEL_TOTAL_BYTES: usize = 4 * 1024 * 1024; + +fn u32_at(b: &[u8], at: usize) -> Option { + let s = b.get(at..at.checked_add(4)?)?; + Some(u32::from_le_bytes(s.try_into().ok()?)) +} + +fn u64_at(b: &[u8], at: usize) -> Option { + let s = b.get(at..at.checked_add(8)?)?; + Some(u64::from_le_bytes(s.try_into().ok()?)) +} + +/// The text of a NUL-terminated label (the rest of the chunk), lossily decoded. +fn label(b: &[u8]) -> Option { + // Look only as far as one label can be retained; never copy an arbitrary chunk body. + let end = b.iter().take(MAX_LABEL_BYTES + 1).position(|&c| c == 0).unwrap_or(b.len()); + if end > MAX_LABEL_BYTES { + return None; + } + let text = String::from_utf8_lossy(b.get(..end)?); + let text = text.trim(); + (text.len() <= MAX_LABEL_BYTES).then(|| text.to_string()) +} + +/// Whether `bytes` start a RIFF/WAVE file. +pub fn is_wav(bytes: &[u8]) -> bool { + matches!(bytes.get(0..4), Some(b"RIFF" | b"RF64")) && bytes.get(8..12) == Some(b"WAVE") +} + +/// The sizes an RF64 file's `ds64` chunk gives for chunks too big for the 32-bit size field. +struct Ds64 { + /// The `data` chunk's size. + data: u64, + /// Other chunks: `(id, size)`. + table: Vec<([u8; 4], u64)>, +} + +impl Ds64 { + /// The real size of chunk `id`, when the file says. + fn size_of(&self, id: &[u8; 4]) -> Option { + if id == b"data" { Some(self.data) } else { self.table.iter().find(|(t, _)| t == id).map(|(_, size)| *size) } + } +} + +/// The `ds64` chunk of an RF64 file (`RF64`, size, `WAVE`, then `ds64`); `None` for any other file or one whose +/// `ds64` doesn't fit. +fn ds64(bytes: &[u8]) -> Option { + if bytes.get(0..4) != Some(b"RF64") || bytes.get(12..16) != Some(b"ds64") { + return None; + } + let size = usize::try_from(u32_at(bytes, 16)?).ok()?; + let body = bytes.get(20..20usize.checked_add(size)?)?; + // riff size (8), data size (8), sample count (8), table length (4), then `(id, size)` entries of 12 bytes + let data = u64_at(body, 8)?; + let entries = usize::try_from(u32_at(body, 24)?).ok()?; + let mut table = Vec::new(); + for i in 0..entries.min(MAX_DS64_TABLE) { + let at = 28usize.checked_add(i.checked_mul(12)?)?; + let id = <[u8; 4]>::try_from(body.get(at..at.checked_add(4)?)?).ok()?; + table.push((id, u64_at(body, at + 4)?)); + } + Some(Ds64 { data, table }) +} + +/// The chunks of a WAVE file as `(id, body)`, stopping at the first one that doesn't fit. +fn chunks<'a>(bytes: &'a [u8], limited: &mut bool) -> Vec<([u8; 4], &'a [u8])> { + let big = ds64(bytes); + let mut out = Vec::new(); + let mut at = 12usize; + while let Some(head) = bytes.get(at..at.saturating_add(8)) { + if out.len() == MAX_CHUNKS { + *limited = true; + break; + } + let (Ok(id), Some(field)) = (<[u8; 4]>::try_from(&head[..4]), u32_at(head, 4)) else { break }; + // In an RF64 file a size field of 0xFFFFFFFF stands for the size `ds64` gives. + let size = match &big { + Some(d) if field == u32::MAX => d.size_of(&id).unwrap_or(u64::from(field)), + _ => u64::from(field), + }; + let start = at.saturating_add(8); + let Some(end) = usize::try_from(size).ok().and_then(|n| start.checked_add(n)) else { break }; + let Some(body) = bytes.get(start..end) else { break }; + out.push((id, body)); + // Chunks are padded to an even size. + at = end.saturating_add(end & 1); + } + out +} + +/// Read bounded cue/adtl and XMP metadata without interpreting application timeline policy. +pub fn read(bytes: &[u8]) -> Markers { + if !is_wav(bytes) { + return Markers::default(); + } + let mut limited = false; + let chunks = chunks(bytes, &mut limited); + let rate = chunks.iter().find(|(id, _)| id == b"fmt ").and_then(|(_, b)| u32_at(b, 4)).map_or(0.0, f64::from); + let cue = cue_markers(&chunks, rate, &mut limited); + let xmp = chunks.iter().find(|(id, _)| id == b"_PMX").map_or_else(Vec::new, |(_, xmp)| xmp_markers(xmp, rate)); + Markers { cue, xmp, limited } +} + +/// Cue points named and sized by the `adtl` list. +fn cue_markers(chunks: &[([u8; 4], &[u8])], rate: f64, limited: &mut bool) -> Vec { + let Some((_, cue)) = chunks.iter().find(|(id, _)| id == b"cue ") else { return Vec::new() }; + let count = (u32_at(cue, 0).unwrap_or(0) as usize).min(MAX_MARKERS); + // The cue points first: `(id, offset in samples)`. + let mut points = Vec::new(); + for i in 0..count { + let Some(point) = cue.get(4usize.saturating_add(i.saturating_mul(24))..) else { break }; + let (Some(id), Some(offset)) = (u32_at(point, 0), u32_at(point, 20)) else { break }; + points.push((id, offset)); + } + // Only the names and lengths of those ids are kept, so a hostile `adtl` list cannot grow them past the cue count, + // and each record is a lookup, not a scan. + let wanted: HashSet = points.iter().map(|(id, _)| *id).collect(); + let mut names: HashMap = HashMap::new(); + let mut lengths: HashMap = HashMap::new(); + let mut name_bytes = 0usize; + let mut records = 0usize; + 'lists: for (_, list) in chunks.iter().filter(|(id, b)| id == b"LIST" && b.get(0..4) == Some(b"adtl")) { + let mut at = 4usize; + while let Some(head) = list.get(at..at.saturating_add(8)) { + let (Ok(id), Some(size)) = (<[u8; 4]>::try_from(&head[..4]), u32_at(head, 4)) else { break }; + let start = at.saturating_add(8); + let Some(end) = start.checked_add(size as usize) else { break }; + let Some(body) = list.get(start..end) else { break }; + records += 1; + if records > MAX_ADTL_RECORDS { + break 'lists; + } + if let Some(cue_id) = u32_at(body, 0).filter(|c| wanted.contains(c)) { + match &id { + // A `labl` is the name; a `note` only when there is none. + b"labl" | b"note" if id == *b"labl" || !names.contains_key(&cue_id) => { + let without_old = name_bytes.saturating_sub(names.get(&cue_id).map_or(0, String::len)); + let text = label(body.get(4..).unwrap_or_default()) + .filter(|text| text.len() <= MAX_LABEL_TOTAL_BYTES.saturating_sub(without_old)) + .unwrap_or_else(|| { + *limited = true; + String::new() + }); + name_bytes = without_old + text.len(); + names.insert(cue_id, text); + } + // The first `ltxt` of a cue point gives its length. + b"ltxt" => { + lengths.entry(cue_id).or_insert_with(|| u32_at(body, 4).unwrap_or(0)); + } + _ => {} + } + } + at = end.saturating_add(end & 1); + } + } + let mut emitted_bytes = 0usize; + points + .into_iter() + .map(|(id, offset)| { + let comment = match names.get(&id) { + Some(text) if text.len() <= MAX_LABEL_TOTAL_BYTES.saturating_sub(emitted_bytes) => { + emitted_bytes += text.len(); + text.clone() + } + Some(_) => { + *limited = true; + String::new() + } + None => String::new(), + }; + let length = lengths.get(&id).copied().unwrap_or(0); + Marker { start: f64::from(offset), duration: f64::from(length), rate, name: comment } + }) + .collect() +} + +/// The text of `name` in an XMP entry, as an attribute (`name="v"`) or an element +/// (`v`). +fn xmp_value(entry: &str, name: &str) -> Option { + // Each entry starts just after ` in_tag = true, + b'>' => in_tag = false, + b'\'' | b'"' if in_tag => { + let quote = bytes[at]; + at += 1; + while at < bytes.len() && bytes[at] != quote { + at += 1; + } + } + _ if in_tag && (at == 0 || xml_space(bytes[at - 1])) && bytes[at..].starts_with(name.as_bytes()) => { + let mut value = at + name.len(); + while value < bytes.len() && xml_space(bytes[value]) { + value += 1; + } + if bytes.get(value) == Some(&b'=') { + value += 1; + while value < bytes.len() && xml_space(bytes[value]) { + value += 1; + } + if let Some("e @ (b'\'' | b'"')) = bytes.get(value) { + let start = value + 1; + let end = bytes.get(start..)?.iter().position(|&b| b == quote)? + start; + return Some(unescape(entry.get(start..end)?)); + } + } + } + _ => {} + } + at += 1; + } + let open = format!("<{name}>"); + let i = entry.find(&open)?; + let rest = entry.get(i + open.len()..)?; + Some(unescape(rest.get(..rest.find('<')?)?)) +} + +fn xml_space(b: u8) -> bool { + matches!(b, b' ' | b'\t' | b'\r' | b'\n') +} + +fn unescape(s: &str) -> String { + let mut out = String::with_capacity(s.len()); + let mut rest = s; + while let Some(i) = rest.find('&') { + out.push_str(rest.get(..i).unwrap_or_default()); + let tail = rest.get(i..).unwrap_or_default(); + let Some(semi) = tail.as_bytes().iter().take(11).position(|b| *b == b';') else { + out.push('&'); + rest = tail.get(1..).unwrap_or_default(); + continue; + }; + let ent = tail.get(1..semi).unwrap_or_default(); + match ent { + "amp" => out.push('&'), + "lt" => out.push('<'), + "gt" => out.push('>'), + "quot" => out.push('"'), + "apos" => out.push('\''), + _ => { + let code = ent + .strip_prefix("#x") + .and_then(|h| u32::from_str_radix(h, 16).ok()) + .or_else(|| ent.strip_prefix('#').and_then(|d| d.parse().ok())); + match code.and_then(char::from_u32) { + Some(c) => out.push(c), + None => out.push_str(tail.get(..=semi).unwrap_or_default()), + } + } + } + rest = tail.get(semi + 1..).unwrap_or_default(); + } + out.push_str(rest); + out.trim().to_string() +} + +/// A frame rate as XMP writes it: `f44100`, or `f30000s1001` (rate over scale). +fn xmp_rate(text: &str) -> Option { + let t = text.trim().strip_prefix('f')?; + let (num, den) = t.split_once('s').unwrap_or((t, "1")); + let (n, d): (f64, f64) = (num.parse().ok()?, den.parse().ok()?); + (n.is_finite() && d.is_finite() && n > 0.0 && d > 0.0 && (n / d).is_finite() && n / d > 0.0).then_some(n / d) +} + +/// Adobe XMP FrameCount: a count optionally followed by `fRATE[sSCALE]`. +/// Bare counts retain the caller's legacy track/file-rate interpretation. +fn xmp_count(text: &str, default_rate: f64) -> Option<(f64, f64)> { + let text = text.trim(); + let (count, rate) = match text.find('f') { + Some(at) => (text.get(..at)?, xmp_rate(text.get(at..)?)?), + None => (text, default_rate), + }; + let count: f64 = count.parse().ok()?; + (count.is_finite() && rate.is_finite() && rate > 0.0).then_some((count, rate)) +} + +/// The markers of an XMP packet's `xmpDM:markers` entries. +fn xmp_markers(packet: &[u8], file_rate: f64) -> Vec { + let text = String::from_utf8_lossy(packet.get(..packet.len().min(MAX_XMP)).unwrap_or_default()); + let mut out = Vec::new(); + // The frame rate belongs to the track an entry follows. + let mut rate = file_rate; + for entry in text.split("= MAX_MARKERS { + break; + } + } + out +} + +#[cfg(test)] +mod tests { + use super::*; + + struct TimedMarker { + time: f64, + duration: f64, + comment: String, + } + fn read(bytes: &[u8]) -> Vec { + let metadata = super::read(bytes); + let time = |frames: f64, rate: f64| if rate > 0.0 && frames.is_finite() { (frames / rate).max(0.0) } else { 0.0 }; + let map = |m: Marker| TimedMarker { time: time(m.start, m.rate), duration: time(m.duration, m.rate), comment: m.name }; + let mut out: Vec<_> = metadata.cue.into_iter().map(map).collect(); + let mut seen: HashSet<_> = out.iter().map(|m| (m.comment.clone(), m.time.to_bits())).collect(); + for m in metadata.xmp.into_iter().map(map) { + if seen.insert((m.comment.clone(), m.time.to_bits())) { + out.push(m); + } + } + out.truncate(MAX_MARKERS); + out.sort_by(|a, b| a.time.total_cmp(&b.time)); + out + } + + fn chunk(id: &[u8; 4], body: &[u8]) -> Vec { + let mut v = id.to_vec(); + v.extend((body.len() as u32).to_le_bytes()); + v.extend(body); + if body.len() % 2 == 1 { + v.push(0); + } + v + } + + fn fmt(rate: u32) -> Vec { + let mut b = Vec::new(); + b.extend(1u16.to_le_bytes()); // PCM + b.extend(1u16.to_le_bytes()); // mono + b.extend(rate.to_le_bytes()); + b.extend((rate * 2).to_le_bytes()); + b.extend(2u16.to_le_bytes()); + b.extend(16u16.to_le_bytes()); + chunk(b"fmt ", &b) + } + + fn wav(chunks: &[Vec]) -> Vec { + let body: Vec = chunks.iter().flatten().copied().collect(); + let mut v = b"RIFF".to_vec(); + v.extend((body.len() as u32 + 4).to_le_bytes()); + v.extend(b"WAVE"); + v.extend(body); + v + } + + fn cue(points: &[(u32, u32)]) -> Vec { + let mut b = (points.len() as u32).to_le_bytes().to_vec(); + for (id, offset) in points { + b.extend(id.to_le_bytes()); + b.extend(0u32.to_le_bytes()); + b.extend(b"data"); + b.extend(0u32.to_le_bytes()); + b.extend(0u32.to_le_bytes()); + b.extend(offset.to_le_bytes()); + } + chunk(b"cue ", &b) + } + + fn labl(id: &[u8; 4], cue_id: u32, text: &str) -> Vec { + let mut b = cue_id.to_le_bytes().to_vec(); + b.extend(text.as_bytes()); + b.push(0); + chunk(id, &b) + } + + fn adtl(subs: &[Vec]) -> Vec { + let mut b = b"adtl".to_vec(); + b.extend(subs.iter().flatten()); + chunk(b"LIST", &b) + } + + #[test] + fn cue_points_are_named_by_the_adtl_list_and_timed_in_samples() { + let file = wav(&[ + fmt(44100), + chunk(b"data", &[0; 8]), + cue(&[(2, 88200), (1, 22050)]), + adtl(&[labl(b"labl", 1, "SEQ01_Prologue"), labl(b"labl", 2, "SEQ02_Vallee_P1_Agriculture")]), + ]); + let m = read(&file); + assert_eq!(m.len(), 2); + assert_eq!((m[0].comment.as_str(), m[0].time), ("SEQ01_Prologue", 0.5)); + assert_eq!((m[1].comment.as_str(), m[1].time), ("SEQ02_Vallee_P1_Agriculture", 2.0)); + } + + #[test] + fn a_note_names_a_cue_only_when_it_has_no_label_and_ltxt_gives_a_length() { + let mut ltxt = 1u32.to_le_bytes().to_vec(); + ltxt.extend(44100u32.to_le_bytes()); + ltxt.extend([0; 16]); + let file = wav(&[ + fmt(44100), + cue(&[(1, 0), (2, 44100)]), + adtl(&[labl(b"note", 1, "from note"), chunk(b"ltxt", <xt), labl(b"labl", 2, "named"), labl(b"note", 2, "ignored")]), + ]); + let m = read(&file); + assert_eq!((m[0].comment.as_str(), m[0].duration), ("from note", 1.0)); + assert_eq!(m[1].comment, "named"); + } + + const XMP_ELEMENTS: &str = r#"MarkersCuef44100441000One & Two13230022050Three"#; + + #[test] + fn xmp_markers_are_read_as_elements_in_frames_of_the_track_rate() { + let file = wav(&[fmt(48000), chunk(b"_PMX", XMP_ELEMENTS.as_bytes())]); + let m = read(&file); + assert_eq!(m.len(), 2); + // 44100 frames at f44100 is one second, whatever the file's own rate. + assert_eq!((m[0].comment.as_str(), m[0].time), ("One & Two", 1.0)); + assert_eq!((m[1].comment.as_str(), m[1].time, m[1].duration), ("Three", 3.0, 0.5)); + } + + #[test] + fn xmp_markers_are_read_as_attributes_and_use_the_file_rate_without_a_track_rate() { + let xmp = r#""#; + let m = read(&wav(&[fmt(48000), chunk(b"_PMX", xmp.as_bytes())])); + assert_eq!((m[0].comment.as_str(), m[0].time), ("Attr é", 0.5)); + } + + #[test] + fn a_marker_in_both_encodings_is_listed_once() { + let xmp = r#""#; + let file = wav(&[fmt(44100), cue(&[(1, 44100)]), adtl(&[labl(b"labl", 1, "Same")]), chunk(b"_PMX", xmp.as_bytes())]); + let names: Vec = read(&file).into_iter().map(|m| m.comment).collect(); + assert_eq!(names, ["Same", "Only XMP"]); + } + + #[test] + fn damaged_or_hostile_files_read_as_far_as_they_go_without_panicking() { + assert!(read(b"").is_empty()); + assert!(read(b"RIFF").is_empty()); + assert!(read(b"not a wav file at all").is_empty()); + let good = wav(&[fmt(44100), cue(&[(1, 44100)]), adtl(&[labl(b"labl", 1, "Kept")])]); + // Every truncation is safe. + for n in 0..good.len() { + let _ = read(&good[..n]); + } + // A cue count far beyond the chunk, and chunk sizes past the end of the file. + let mut huge = 0xFFFF_FFFFu32.to_le_bytes().to_vec(); + huge.extend([0; 24]); + let m = read(&wav(&[fmt(44100), chunk(b"cue ", &huge)])); + assert!(m.len() <= 1); + let mut lying = good.clone(); + lying.extend(b"LIST\xff\xff\xff\xffadtl"); + assert_eq!(read(&lying).len(), 1); + // No sample rate: times are zero, not a division by zero. + let m = read(&wav(&[cue(&[(1, 44100)])])); + assert_eq!(m[0].time, 0.0); + } + + /// A chunk whose size field says `field` (an RF64 file's `0xFFFFFFFF`) while its body is `body`. + fn chunk_with_size_field(id: &[u8; 4], field: u32, body: &[u8]) -> Vec { + let mut v = id.to_vec(); + v.extend(field.to_le_bytes()); + v.extend(body); + if body.len() % 2 == 1 { + v.push(0); + } + v + } + + /// An RF64 `ds64` chunk: the `data` chunk's size and a table of other chunks' sizes. + fn ds64_chunk(data_size: u64, table: &[([u8; 4], u64)]) -> Vec { + let mut b = 0u64.to_le_bytes().to_vec(); // riff size (not used) + b.extend(data_size.to_le_bytes()); + b.extend(0u64.to_le_bytes()); // sample count + b.extend((table.len() as u32).to_le_bytes()); + for (id, size) in table { + b.extend(id); + b.extend(size.to_le_bytes()); + } + chunk(b"ds64", &b) + } + + fn rf64(chunks: &[Vec]) -> Vec { + let body: Vec = chunks.iter().flatten().copied().collect(); + let mut v = b"RF64".to_vec(); + v.extend(u32::MAX.to_le_bytes()); + v.extend(b"WAVE"); + v.extend(body); + v + } + + #[test] + fn rf64_markers_after_a_data_chunk_of_sentinel_size_are_read() { + let audio = [0u8; 8]; + let file = rf64(&[ + ds64_chunk(audio.len() as u64, &[]), + fmt(44100), + chunk_with_size_field(b"data", u32::MAX, &audio), + cue(&[(1, 44100)]), + adtl(&[labl(b"labl", 1, "After the audio")]), + ]); + let m = read(&file); + assert_eq!(m.len(), 1); + assert_eq!((m[0].comment.as_str(), m[0].time), ("After the audio", 1.0)); + } + + #[test] + fn rf64_sizes_of_other_chunks_come_from_the_ds64_table() { + let junk = [7u8; 6]; + let tail = || [cue(&[(1, 8000)]), adtl(&[labl(b"labl", 1, "Found")])].concat(); + let with_entry = rf64(&[ds64_chunk(0, &[(*b"junk", junk.len() as u64)]), fmt(8000), chunk_with_size_field(b"junk", u32::MAX, &junk), tail()]); + assert_eq!(read(&with_entry)[0].comment, "Found"); + // Without the table entry the sentinel is all there is, and what follows is not reached. + let without = rf64(&[ds64_chunk(0, &[]), fmt(8000), chunk_with_size_field(b"junk", u32::MAX, &junk), tail()]); + assert!(read(&without).is_empty()); + } + + #[test] + fn hostile_rf64_size_tables_read_as_far_as_they_go_without_panicking() { + // sizes that overflow when added to an offset + for data in [u64::MAX, u64::MAX - 7, 1 << 63] { + let file = + rf64(&[ds64_chunk(data, &[(*b"junk", u64::MAX)]), fmt(8000), chunk_with_size_field(b"data", u32::MAX, &[0; 4]), cue(&[(1, 0)])]); + assert!(read(&file).is_empty()); + } + // a table that claims far more entries than its chunk holds + let mut liar = ds64_chunk(0, &[]); + liar[8 + 24..8 + 28].copy_from_slice(&u32::MAX.to_le_bytes()); + let _ = read(&rf64(&[liar, fmt(8000), cue(&[(1, 0)])])); + // every truncation of a good RF64 file is safe + let good = rf64(&[ + ds64_chunk(4, &[]), + fmt(44100), + chunk_with_size_field(b"data", u32::MAX, &[0; 4]), + cue(&[(1, 44100)]), + adtl(&[labl(b"labl", 1, "Kept")]), + ]); + assert_eq!(read(&good).len(), 1); + for n in 0..good.len() { + let _ = read(&good[..n]); + } + } + + #[test] + fn a_huge_adtl_list_is_read_in_linear_time_and_only_for_the_cue_points_that_exist() { + // Records for ids that have no cue point are skipped, not stored, and the number looked at is capped. + let junk = |n: u32| (1000..1000 + n).map(|id| labl(b"labl", id, "n")).collect::>(); + let started = std::time::Instant::now(); + // the real label comes first: read + let mut subs = vec![labl(b"labl", 1, "Real")]; + subs.extend(junk(100_000)); + let m = read(&wav(&[fmt(44100), cue(&[(1, 44100)]), adtl(&subs)])); + assert_eq!(m[0].comment, "Real"); + // it comes after more records than are looked at: not read, and the marker is still there + let mut subs = junk(MAX_ADTL_RECORDS as u32 + 10); + subs.push(labl(b"labl", 1, "Too late")); + let m = read(&wav(&[fmt(44100), cue(&[(1, 44100)]), adtl(&subs)])); + assert_eq!((m.len(), m[0].comment.as_str()), (1, "")); + assert!(started.elapsed() < std::time::Duration::from_secs(5), "took {:?}", started.elapsed()); + } + + #[test] + fn metadata_keeps_original_coordinates_and_separate_sources() { + let xmp = br#""#; + let file = wav(&[fmt(44100), cue(&[(1, 1)]), adtl(&[labl(b"labl", 1, "Same")]), chunk(b"_PMX", xmp)]); + let data = super::read(&file); + assert_eq!(data.cue[0].start, 1.0); + assert_eq!(data.cue[0].rate, 44100.0); + assert_eq!(data.xmp[0].start, 1.0); + assert_eq!(data.xmp[0].duration, 2.0); + assert_eq!(data.cue.len(), 1); + assert_eq!(data.xmp.len(), 1); + } + + #[test] + fn repeated_unterminated_entities_preserve_text_with_bounded_lookahead() { + let text = "&".repeat(100_000); + assert_eq!(unescape(&text), text); + assert_eq!(unescape("A A & &unknown;"), "A A & &unknown;"); + } + #[test] + fn repeated_cue_ids_cannot_multiply_label_storage() { + let text = "x".repeat(MAX_LABEL_BYTES); + let points: Vec<_> = (0..MAX_MARKERS).map(|i| (1, i as u32)).collect(); + let file = wav(&[fmt(44100), cue(&points), adtl(&[labl(b"labl", 1, &text)])]); + let metadata = super::read(&file); + assert_eq!(metadata.cue.len(), MAX_MARKERS); + assert_eq!(metadata.cue.iter().map(|m| m.name.len()).sum::(), MAX_LABEL_TOTAL_BYTES); + assert!(metadata.limited); + assert_eq!(metadata.cue[0].name, text); + assert!(metadata.cue[MAX_MARKERS - 1].name.is_empty()); + assert_eq!(metadata.cue[MAX_MARKERS - 1].start, (MAX_MARKERS - 1) as f64); + } + + #[test] + fn cue_name_map_and_single_label_are_bounded() { + let count = MAX_LABEL_TOTAL_BYTES / MAX_LABEL_BYTES + 2; + let text = "x".repeat(MAX_LABEL_BYTES); + let points: Vec<_> = (0..count as u32).map(|i| (i, i)).collect(); + let labels: Vec<_> = (0..count as u32).map(|i| labl(b"labl", i, &text)).collect(); + let metadata = super::read(&wav(&[fmt(44100), cue(&points), adtl(&labels)])); + assert!(metadata.limited); + assert_eq!(metadata.cue.iter().map(|m| m.name.len()).sum::(), MAX_LABEL_TOTAL_BYTES); + let oversized = "x".repeat(MAX_LABEL_BYTES + 1); + let metadata = super::read(&wav(&[fmt(44100), cue(&[(1, 3)]), adtl(&[labl(b"labl", 1, &oversized)])])); + assert!(metadata.limited); + assert_eq!(metadata.cue[0].start, 3.0); + assert!(metadata.cue[0].name.is_empty()); + } + + #[test] + fn empty_chunk_inventory_is_bounded() { + let empty = chunk(b"junk", &[]); + let many = empty.repeat(MAX_CHUNKS + 1); + let metadata = super::read(&wav(&[many, fmt(44100), cue(&[(1, 1)])])); + assert!(metadata.limited); + assert!(metadata.cue.is_empty()); + } + + #[test] + fn entity_lookahead_keeps_previous_semicolon_boundary() { + assert_eq!(unescape("A"), "A"); + assert_eq!(unescape("A"), "A"); + } + #[test] + fn xmp_quotes_spacing_and_attribute_boundaries() { + let packet = br#""#; + let markers = xmp_markers(packet, 48000.0); + assert_eq!(markers.len(), 1); + assert_eq!((markers[0].start, markers[0].duration, markers[0].rate), (24000.0, 12000.0, 48000.0)); + assert_eq!(markers[0].name, "single & quoted"); + let spoof = r#" label=" xmpDM:startTime='7'" otherxmpDM:startTime='8' xmpDM:startTime = '9'/>"#; + assert_eq!(xmp_value(spoof, "xmpDM:startTime").as_deref(), Some("9")); + assert!(xmp_value(" xmpDM:startTime='unterminated", "xmpDM:startTime").is_none()); + } + + #[test] + fn xmp_frame_counts_preserve_explicit_and_mixed_rates() { + let packet = br#""#; + let markers = xmp_markers(packet, 44100.0); + assert_eq!(markers.len(), 3); + assert_eq!((markers[0].start, markers[0].rate), (15.0, 24.0)); + assert!((markers[0].duration / markers[0].rate - 0.5005).abs() < 1e-12); + assert_eq!((markers[1].start, markers[1].duration, markers[1].rate), (48000.0, 24000.0, 48000.0)); + assert_eq!((markers[2].start, markers[2].duration, markers[2].rate), (15.0, 12.0, 24.0)); + for invalid in ["15f0", "15f24s0", "15fNaN", "15finf", "NaN", "1e999", "15f24junk"] { + assert!(xmp_count(invalid, 48000.0).is_none(), "{invalid}"); + } + } +}