mux: reconstruct display-order PTS for sparse-PTS program streams

HD-DVD EVO (and DVD VOB) program streams timestamp video at GOP
granularity: only one access unit per GOP carries a PES PTS. The H.264 /
HEVC / VC-1 parsers collapsed a missing PTS to 0, so on such a source
every non-anchor frame landed on the same block timestamp and a decoder
reported "non monotonically increasing dts".

Add a shared SparsePtsReorder that rebuilds a display-order PTS per frame
from the coded picture type (I/P/B) plus the sparse anchor PTS, with a
per-frame duration self-calibrated from the spacing between consecutive
GOP anchors (no external frame-rate needed). Display order is derived via
the classic single-anchor-delay rule (an anchor displays only after the
previously-held anchor; a B displays immediately), exact for the
non-hierarchical GOP structures HD-DVD H.264/VC-1 use. It mirrors the
MPEG-2 parser's GOP-buffered origin-locking.

Gated to the program-stream path only: the three parsers enable it via
with_ps_reorder(is_dvd_ps), so the BD/UHD transport path (per-frame PTS)
is byte-identical and untouched.
This commit is contained in:
Matthew Jackson
2026-07-08 21:54:09 -07:00
parent f8bea78db5
commit 5fdff5664f
5 changed files with 497 additions and 12 deletions
+100 -3
View File
@@ -52,6 +52,10 @@ pub struct H264Parser {
// the stale avcC copy after a mid-title redefinition.
cur_sps: Option<Vec<u8>>,
cur_pps: Option<Vec<u8>>,
/// Display-order PTS reconstruction, enabled only on the program-stream
/// (HD-DVD EVO) path where the source stamps a PTS once per GOP. `None` on
/// the BD/UHD transport path, which carries a per-frame PTS.
reorder: Option<super::reorder::SparsePtsReorder>,
}
impl Default for H264Parser {
@@ -68,6 +72,25 @@ impl H264Parser {
pps: None,
cur_sps: None,
cur_pps: None,
reorder: None,
}
}
/// Enable display-order PTS reconstruction for a program-stream source.
/// No-op (leaves timestamps as parsed) for a transport-stream source.
pub(crate) fn with_ps_reorder(mut self, enabled: bool) -> Self {
if enabled {
self.reorder = Some(super::reorder::SparsePtsReorder::new());
}
self
}
/// Route a finished frame through the PTS reorderer when enabled, else emit
/// it directly (unchanged transport-stream behaviour).
fn finish(&mut self, explicit: Option<i64>, frame: Frame) -> Vec<Frame> {
match self.reorder.as_mut() {
Some(r) => r.push(explicit, frame),
None => vec![frame],
}
}
}
@@ -152,7 +175,8 @@ impl CodecParser for H264Parser {
// decode order and the player reorders by timecode. Use PTS, not DTS —
// DTS presents B-frames in decode order (visible judder) and breaks
// PTS-based seeking. Fall back to DTS only if PTS is absent.
let pts_ns = pes.pts.or(pes.dts).map(pts_to_ns).unwrap_or(0);
let explicit_pts = pes.pts.or(pes.dts).map(pts_to_ns);
let pts_ns = explicit_pts.unwrap_or(0);
// Single pass: detect IDR keyframes, seed/strip param sets, and convert
// Annex B (start-code prefixed) NALUs to length-prefixed NALUs (MKV with
@@ -236,7 +260,7 @@ impl CodecParser for H264Parser {
}
}
vec![Frame {
let frame = Frame {
// Coding-type only: H.264 field order is not decoded here, so
// `field_order()` stays `None` — honestly absent, never guessed.
coding: coding_type.map(PictureInfo::coding_type_only),
@@ -248,7 +272,15 @@ impl CodecParser for H264Parser {
discontinuity: pes.discontinuity,
data: frame_data,
duration_ns: None,
}]
};
self.finish(explicit_pts, frame)
}
fn flush(&mut self) -> Vec<Frame> {
match self.reorder.as_mut() {
Some(r) => r.flush(),
None => Vec::new(),
}
}
fn codec_private(&self) -> Option<Vec<u8>> {
@@ -667,6 +699,71 @@ mod tests {
);
}
/// End-to-end sparse-PTS reconstruction through the REAL parser + reorder:
/// a program-stream source (`with_ps_reorder(true)`) that stamps a PTS only
/// on each GOP's I-frame must yield distinct, display-ordered PTS for every
/// frame — the property the mkv muxer needs so a decoder derives monotonic
/// DTS. Without the reorder the non-anchor frames all collapse to one PTS.
#[test]
fn h264_ps_reorder_reconstructs_distinct_display_pts() {
use super::super::coding::CodingType;
// slice bodies: 0x88 → I (IDR), 0x98 → P, 0x9C → B (non-IDR).
// Decode order of a classic single-B GOP: I P B P B.
let gop = |anchor_pts: Option<i64>| {
vec![
(NAL_SLICE_IDR, 0x88u8, anchor_pts),
(NAL_SLICE_NON_IDR, 0x98, None),
(NAL_SLICE_NON_IDR, 0x9C, None),
(NAL_SLICE_NON_IDR, 0x98, None),
(NAL_SLICE_NON_IDR, 0x9C, None),
]
};
let feed = |reorder: bool| -> Vec<super::super::Frame> {
let mut p = H264Parser::new().with_ps_reorder(reorder);
let mut out = Vec::new();
// Two GOPs; the second I carries an anchor 5 frames later (90 kHz:
// 5 * 3750 = 18750 ticks) so the reorder can calibrate a duration.
for (nal, body, pts) in gop(Some(0)).into_iter().chain(gop(Some(18750))) {
out.extend(p.parse(&make_pes(h264_nal(nal, &[body]), pts)));
}
out.extend(p.flush());
out
};
// With reorder ON: all 10 frames emitted, every PTS distinct.
let recon = feed(true);
assert_eq!(recon.len(), 10, "no frame dropped");
let mut pts: Vec<i64> = recon.iter().map(|f| f.pts_ns).collect();
let n = pts.len();
pts.sort_unstable();
pts.dedup();
assert_eq!(
pts.len(),
n,
"reconstructed PTS are all distinct (no DTS collision)"
);
// The GOP's first-displayed frame is the I; the B in decode position 2
// must display BEFORE the P in decode position 1 (classic reorder).
let g1 = &recon[0..5];
assert_eq!(g1[0].coding.unwrap().coding_type(), CodingType::I);
assert!(
g1[2].pts_ns < g1[1].pts_ns,
"B (decode idx 2) displays before its forward-anchor P (decode idx 1)"
);
assert_eq!(g1[0].pts_ns, 0, "GOP anchor locks the I to its true PTS");
// With reorder OFF (transport-stream behaviour): the non-anchor frames
// collapse to a single colliding PTS — the bug this fix removes.
let raw = feed(false);
let collisions = raw.iter().filter(|f| f.pts_ns == 0).count();
assert!(
collisions >= 8,
"without reorder the sparse-PTS frames collide on 0 (got {collisions})"
);
}
/// Regression (Fight Club bug, H.264 variant): PPS id 0 = body A (→ avcC),
/// redefined to B, then switched BACK to A. A streaming decoder is on B; the
/// revert to A == avcC must still be emitted in-band or the A-segment