Merge pull request #498 from Silvertree2010/fix/wav-empty-data-chunk

WAV: an empty data chunk followed by metadata has no frames (#466)

ROADMAP.md

@@ -124,6 +124,7 @@ mutated and truncated movies. Field reports from the first public week are track
 
 - **2026-10-16:** Master fader volume automation uses each render block's actual last sample, matching track faders. Breakpoints in the next block no longer pull the preceding block's level toward them, and ramps reach their last-sample value. Regression tests cover nonzero starts, one-sample callbacks and several block sizes. No readiness percentage change.
 - **2026-10-14:** MCP `batch` rejects more than 1000 calls before executing any command, instead of silently running only the first 1000 (#460). No readiness percentage change.
+- **2026-10-13:** WAV: an empty data chunk followed by metadata (LIST, bext) reads as no audio; a recording that was never finalised (sizes left at 0) still gives what was written (#466)
 - **2026-10-13:** `mix.instrument` takes the built-in instrument's parameters in `params` (validated, clamped) and refuses any other field, so a synth setting written beside `plugin` is an error instead of being accepted and ignored (#265)
 - **2026-10-13:** A bounce of a bus that an aux returns onto lines up with the timeline: the aux insert's lookahead is compensated like the main mix's (it left the bus late and cut the end off), and tracks routed to that bus wait for it too (#246)
 - **2026-10-13:** Delay compensation covers the inserts of the master faders: a bounce of the main mix with a Maximizer on the master lines up with the same mix without it (it came out 71 samples late) (#351)
@@ -159,6 +160,7 @@ mutated and truncated movies. Field reports from the first public week are track
 |---|---|---|
 | 2026-10-16 | trivial | Master volume automation block boundaries corrected; readiness estimates unchanged |
 | 2026-10-14 | trivial | Progress log: oversized MCP batches rejected before execution (#460) |
+| 2026-10-13 | trivial | Progress log: empty WAV data chunk followed by metadata (#466) |
 | 2026-10-13 | trivial | Progress log: mix.instrument params, unknown fields refused (#265) |
 | 2026-10-13 | trivial | Progress log: bus bounce compensates aux returns (#246) |
 | 2026-10-13 | trivial | Progress log: master insert latency compensated (#351) |

crates/audio-io/src/wav.rs

@@ -140,6 +140,30 @@ pub(crate) fn is_wav(b: &[u8]) -> bool {
     matches!(tag(b, 0), Some(t) if &t == b"RIFF" || &t == b"RF64" || &t == b"BW64") && tag(b, 8).is_some_and(|t| &t == b"WAVE")
 }
 
+/// True when everything from `from` to the end of the file is whole RIFF chunks (four printable
+/// ASCII id bytes, a size that fits, the pad byte after an odd size; the last chunk's pad may be
+/// missing): the metadata of a finished file, not samples.
+fn only_chunks_follow(b: &[u8], from: usize) -> bool {
+    let mut pos = from;
+    while pos < b.len() {
+        let (Some(id), Some(size)) = (tag(b, pos), le_u32(b, pos + 4)) else { return false };
+        if !id.iter().all(|c| (0x20..0x7f).contains(c)) {
+            return false;
+        }
+        let size = size as usize;
+        let Some(end) = pos.checked_add(8).and_then(|body| body.checked_add(size)) else { return false };
+        if end > b.len() {
+            return false;
+        }
+        let next = end + (size & 1);
+        if end == b.len() || next == b.len() {
+            return true;
+        }
+        pos = next;
+    }
+    false
+}
+
 pub(crate) fn parse(b: &[u8]) -> Result<WavParse> {
     if !is_wav(b) {
         return Err(AudioError::Malformed("not a RIFF/RF64 WAVE file".into()));
@@ -168,7 +192,12 @@ pub(crate) fn parse(b: &[u8]) -> Result<WavParse> {
         if &id == b"data" {
             // Lenient: a truncated recording (or a streaming writer's 0 / 0xFFFFFFFF size) still
             // yields whatever audio is actually present.
-            let len = if size32 == 0 && !is_rf64 { remaining } else { declared.min(remaining) };
+            // A size of 0 in a file whose RIFF size is right and that has only chunks after it is an
+            // empty data chunk followed by metadata (LIST, bext, …); otherwise it is a recording
+            // that was never finalised, and what follows is audio.
+            let finished = u64::from(le_u32(b, 4).unwrap_or(0)) + 8 == b.len() as u64;
+            let empty = size32 == 0 && !is_rf64 && finished && only_chunks_follow(b, body);
+            let len = if size32 == 0 && !is_rf64 && !empty { remaining } else { declared.min(remaining) };
             if data.is_none() {
                 data = Some((body, len));
             }

crates/audio-io/tests/wav_empty_data.rs

@@ -0,0 +1,87 @@
+//! A WAV whose data chunk is empty and followed by metadata decodes to no audio; a recording
+//! that was never finalised (RIFF and data sizes left at 0) still yields what was written (#466).
+
+use soundcraft_audio_io::{decode, probe};
+
+fn fmt16(channels: u16, sr: u32) -> Vec<u8> {
+    let align = channels * 2;
+    let mut f = Vec::new();
+    f.extend_from_slice(&1u16.to_le_bytes());
+    f.extend_from_slice(&channels.to_le_bytes());
+    f.extend_from_slice(&sr.to_le_bytes());
+    f.extend_from_slice(&(sr * u32::from(align)).to_le_bytes());
+    f.extend_from_slice(&align.to_le_bytes());
+    f.extend_from_slice(&16u16.to_le_bytes());
+    f
+}
+
+fn chunk(id: &[u8; 4], data: &[u8]) -> Vec<u8> {
+    let mut c = id.to_vec();
+    c.extend_from_slice(&(data.len() as u32).to_le_bytes());
+    c.extend_from_slice(data);
+    if data.len() % 2 == 1 {
+        c.push(0);
+    }
+    c
+}
+
+/// A finished file: correct RIFF size, chunks as given.
+fn riff(chunks: &[Vec<u8>]) -> Vec<u8> {
+    let mut body = b"WAVE".to_vec();
+    for c in chunks {
+        body.extend_from_slice(c);
+    }
+    let mut out = b"RIFF".to_vec();
+    out.extend_from_slice(&(body.len() as u32).to_le_bytes());
+    out.extend_from_slice(&body);
+    out
+}
+
+fn list_info() -> Vec<u8> {
+    let mut l = b"INFO".to_vec();
+    l.extend_from_slice(&chunk(b"INAM", b"Silence"));
+    chunk(b"LIST", &l)
+}
+
+#[test]
+fn an_empty_data_chunk_followed_by_metadata_has_no_frames() {
+    let file = riff(&[chunk(b"fmt ", &fmt16(2, 48_000)), chunk(b"data", &[]), list_info(), chunk(b"bext", &[0u8; 602])]);
+    let info = probe(&file, Some("wav")).unwrap();
+    assert_eq!(info.frames, 0, "the LIST and bext chunks are not samples");
+    let (info, buf) = decode(&file, Some("wav")).unwrap();
+    assert_eq!((info.frames, buf.frames()), (0, 0));
+    // The same with the metadata first.
+    let file = riff(&[chunk(b"fmt ", &fmt16(1, 44_100)), list_info(), chunk(b"data", &[])]);
+    assert_eq!(probe(&file, Some("wav")).unwrap().frames, 0);
+    // And with a single trailing chunk of odd size and no pad byte at the very end.
+    let mut file = riff(&[chunk(b"fmt ", &fmt16(1, 44_100)), chunk(b"data", &[]), chunk(b"note", b"abc")]);
+    file.pop();
+    let riff_size = (file.len() - 8) as u32;
+    file[4..8].copy_from_slice(&riff_size.to_le_bytes());
+    assert_eq!(probe(&file, Some("wav")).unwrap().frames, 0);
+}
+
+#[test]
+fn a_recording_that_was_never_finalised_still_gives_its_audio() {
+    let samples: Vec<u8> = (0..400i16).flat_map(|i| (i * 40).to_le_bytes()).collect();
+    // RIFF size and data size both left at 0, as a recorder writes them until it finishes.
+    let mut file = riff(&[chunk(b"fmt ", &fmt16(2, 48_000)), chunk(b"data", &[])]);
+    file.extend_from_slice(&samples);
+    file[4..8].copy_from_slice(&0u32.to_le_bytes());
+    let (info, buf) = decode(&file, Some("wav")).unwrap();
+    assert_eq!(info.frames, 200);
+    assert_eq!(buf.frames(), 200);
+    // Even if the RIFF size is the file's, audio that is not a run of chunks is still audio.
+    let mut file = riff(&[chunk(b"fmt ", &fmt16(2, 48_000)), chunk(b"data", &[])]);
+    file.extend_from_slice(&samples);
+    let riff_size = (file.len() - 8) as u32;
+    file[4..8].copy_from_slice(&riff_size.to_le_bytes());
+    assert_eq!(probe(&file, Some("wav")).unwrap().frames, 200);
+}
+
+#[test]
+fn a_data_chunk_with_audio_is_unchanged() {
+    let samples: Vec<u8> = (0..100i16).flat_map(|i| i.to_le_bytes()).collect();
+    let file = riff(&[chunk(b"fmt ", &fmt16(1, 44_100)), chunk(b"data", &samples), list_info()]);
+    assert_eq!(probe(&file, Some("wav")).unwrap().frames, 100);
+}

docs/file-format-parity.md

@@ -1,6 +1,6 @@
 # File-format parity
 
-> **Last reviewed:** 2026-10-13 · **Last updated:** 2026-10-13 · **Change:** minor (continuous fade curvature in SoundCraft sessions) · **Target:** Avid Pro Tools Ultimate 2026.4.1
+> **Last reviewed:** 2026-10-13 · **Last updated:** 2026-10-13 · **Change:** trivial (WAV: empty data chunk with trailing metadata (#466); previously minor: continuous fade curvature in SoundCraft sessions) · **Target:** Avid Pro Tools Ultimate 2026.4.1
 
 Every format Pro Tools reads or writes, and ours. The Pro Tools list comes from the document types
 its bundle declares (`CFBundleDocumentTypes` in Info.plist: names only, nothing else read) and its
@@ -33,7 +33,7 @@ Legend: ✅ supported and tested · 🟡 partial · ❌ missing · — not appli
 
 | Format | Pro Tools | Read | Write | Tests / notes |
 |---|---|---|---|---|
-| WAV / BWF (PCM 16/24/32, float 32), `bext` | native | ✅ | ✅ | Native reader/writer, TPDF dither option; also 8-bit and float 64 read |
+| WAV / BWF (PCM 16/24/32, float 32), `bext` | native | ✅ | ✅ | Native reader/writer, TPDF dither option; also 8-bit and float 64 read. An empty data chunk followed by metadata reads as no audio; an unfinalised recording's audio is still read (#466). |
 | RF64 / BW64 (> 4 GiB) | read/write | ✅ | ✅ | `encode_wav_rf64` |
 | WAV `WAVE_FORMAT_EXTENSIBLE` multichannel with channel mask | read/write | ✅ | ✅ | `encode_with_channel_mask`; default 5.1 import creates a stereo track (#89, PR #128) |
 | AIFF / AIFF-C | native | ✅ | ✅ (`fl32`) | Native |
@@ -80,5 +80,6 @@ Legend: ✅ supported and tested · 🟡 partial · ❌ missing · — not appli
 
 | Date | Change | Summary |
 |---|---|---|
+| 2026-10-13 | trivial | WAV: empty data chunk with trailing metadata (#466) |
 | 2026-10-13 | minor | Fade curvature round-trips in sessions with a zero default for older files. Format version and readiness estimate unchanged |
 | 2026-10-10 | major | Created: format inventory against Pro Tools 2026.4.1's declared document types and import/bounce options |