fix(moq-mux): author DTS for B-frame MPEG-TS export (decode timeline not emitted)

[t0ms]

Problem

moq-cli TS export emits PTS only (no DTS), so B-frame streams carry no decode
timeline. Players see out-of-order DTS and need ffplay -fflags +igndts. Validated on a
real CNN 1080i contribution feed through moq pub | relay | moq sub. Follow-up to #1798
(multi-PPS, fixed); part of #1799.

Evidence (real feed)

• Source has 12480 B-frames (ffprobe -select_streams v -show_frames -show_entries frame=pict_type ... | grep -c ',B$'), so decode order != presentation order.
• Subscriber output: first packet dts=N/A, then ffmpeg's reconstructed DTS is
  non-monotonic and repeats (many frames share one DTS):
  pts, dts 25.277000, N/A 25.557000, 25.297000 25.437000, 25.557000 25.357000, 25.557000 (DTS 25.557 repeated ~12x)
• ffplay logs DTS 2289330 < 2301930 out of order; plays only with -fflags +igndts.

Root cause

rs/moq-mux/src/container/ts/export.rs write_pes sets dts: None (~line 652).
container::Frame carries a single timestamp (PTS); frames are delivered in decode
order with valid PTS, but no decode time is authored on export.

Recommended fix (exporter-side, non-breaking, main)

Frames already arrive in decode order, so the exporter can author a valid DTS with a
bounded reorder buffer (depth = max consecutive B-frames): within the window, assign DTS
= the PTS values in sorted order (exact for CFR broadcast). Emit dts: Some(..) when it
differs from PTS. No wire/format change; cost is a few frames of added latency. Removes
the out-of-order DTS and the +igndts requirement.

Optional follow-up (faithful, dev)

For byte-exact DTS (strict TR 101 290 / contribution), carry the original composition
offset (PTS-DTS) end to end: capture on TS import, add a field to container::Frame + the
hang container frame format, emit on TS/fMP4/FLV export. Breaking wire/format change ->
dev; mirror js/hang + doc/concept. Related: #1828's review found moq_mux::Export
send-timestamps use wall-clock, not PTS, and "needs a PTS-exposing API change" - the same
Export timing surface, worth solving together.

Branch

Exporter-side fix: main (additive). Faithful composition-offset carriage: dev.
(Written by Claude)

[coderabbitai]

🔗 Related PRs

#1781 - fix(hang): omit empty HEVC constraint component in codec string [merged]
#1808 - moq-gst: fix plugin license + broadcast-aligned timestamps [merged]
#1809 - fix(moq-gst): deterministic moqsrc pad names so CMAF playback works [merged]
#1812 - fix(moq-mux): carry all distinct SPS/PPS/VPS through transmux, not just the last seen [merged]
#1815 - feat(moq-mux): generic verbatim MPEG-TS carriage (mpegts catalog section) [merged]

---

📝 Issue Planner

_{Check the box below or use the @coderabbitai plan command to generate an implementation plan and prompts that you can use with your favorite coding assistant.}

• Create Plan

---

🧪 Issue enrichment is currently in open beta.

You can configure auto-planning by selecting labels in the issue_enrichment configuration.

To disable automatic issue enrichment, add the following to your .coderabbit.yaml:

issue_enrichment:
  auto_enrich:
    enabled: false

💬 Have feedback or questions? Drop into our discord!

[t0ms]

Verified end-to-end on main @ 6893b1c4 (#1815 + #1842 + #1843)

Re-ran a real broadcast capture through publish -> relay -> subscribe and compared the subscriber output to the source. Source is a live CNN International EMEA HD recording: 1080i H.264 + MPEG-1 audio + AC-3 + DVB teletext + 3x SCTE-35.

Test rig (matters): the feed is pushed byte-faithfully with TSDuck so every PID/PSI/timestamp reaches the publisher unaltered:

tsp -I file seg.ts -P regulate | moq-cli publish --url … --broadcast … ts
moq-relay localhost.toml
moq-cli subscribe --url … --broadcast … --format ts > sub.ts

Note for anyone reproducing: do not stage the feed through ffmpeg -c copy. ffmpeg's default stream selection silently drops the ancillary PIDs (AC-3/teletext/SCTE-35) and its mpegts muxer rewrites PIDs/timestamps, which makes it look like MoQ dropped them. Use raw TS (tsp, cat, or a real SRT/UDP source).

Verbatim carriage (#1835 / #1842 / #1815): PASS

All 7 elementary streams survive with their original PIDs:

Stream	Source PID	Subscriber PID
H.264 video	0x6f	0x6f
MPEG audio	0x79	0x79
AC-3	0x7b	0x7b
DVB teletext	0x83	0x83
SCTE-35	0x8d/0x8e/0x8f	0x8d/0x8e/0x8f

PCR PID preserved (0x6f). Publisher log shows the verbatim tracks 0.ts–4.ts created alongside 0.avc3 / 0.mp2, with no "unsupported / dropping" warnings. (Before the fix the same feed collapsed to 2 streams renumbered to 0x1001/0x1002.)

DTS / B-frames (#1836 / #1843): PASS

Over the captured GOPs: non_monotonic_dts = 0 and dts_gt_pts = 0 (DTS strictly increasing and <= PTS), a fully clean decode (ffmpeg -v error -i sub.ts -f null - prints nothing), and playback no longer needs -fflags +igndts. The reorder-depth reserve (PES DTS folded into the catalog jitter, import.rs) sizes correctly for this feed's ~5-frame reorder.

Both issues are confirmed fixed for native broadcast passthrough. Residual fidelity gaps that this lane does not yet preserve (SI/SDT service identity, constant mux-rate / PCR-accurate pacing, PMT PID, EIT/TDT) are written up separately so they don't muddy these two.

(Written by Claude)