fix(moq-mux): carry all distinct SPS/PPS/VPS through transmux, not just the last seen

[kixelated]

Summary

Fixes moq-dev/moq#1798 (work item 1).

Publishing a real broadcast MPEG-TS feed and subscribing back with --format ts produced undecodable H.264: ffplay/ffprobe reported non-existing PPS 0 referenced / no frame! continuously. The source defines two PPS (ids 0 and 1) and ~97% of slices reference PPS 0, but the H.264/H.265 transmux held a single parameter set each, so only the last-seen PPS (id 1) survived the round-trip.

avcC/hvcC store SPS/PPS (and VPS for HEVC) as count-prefixed arrays, and slices reference them by id, so a single-slot cache is wrong by construction. This keeps an ordered set of parameter-set NALs and emits/re-injects all of them.

What changed

• Avc1 / Hvc1 transforms (h264/mod.rs, h265/mod.rs): keep the parameter sets carried by a keyframe and emit them all. build_avcc / build_hvcc now emit numOf{Sequence,Picture}ParameterSets > 1 (and multi-NAL hvcC arrays). The egress avcc_params / hvcc_params already loop over the counts, so TS re-injection needed no further change.
• avc3 / hev1 importers (h264/import.rs, h265/import.rs): re-inject every parameter set the keyframe references for mid-stream tune-in, with per-AU "seen" tracking.
• Shared push_distinct / reconcile_keyframe_params helpers added to annexb.rs.

Reinit: dropping superseded parameter sets

The active parameter set is scoped to the latest keyframe rather than accumulated for the life of the stream:

• In the transforms, a frame that carries parameter sets (a keyframe) redefines the active set per type (replace, not accumulate); the avcC/hvcC is rebuilt only when the set actually changes.
• In the importers, each keyframe reconciles: if it presents its own set, that set is adopted (dropping any the new GOP no longer uses); if it carries none, the retained set is re-injected for tune-in.
• ExportSource::refresh_description now tracks the transform's record after it is first set, so a reinit's new avcC/hvcC reaches the muxer's per-keyframe re-injection instead of a stale one.

So a mid-stream reconfiguration (resolution change, a PPS dropped from the rotation) drops the superseded entries instead of letting them linger. A single keyframe carrying multiple PPS (the original #1798 case) still keeps all of them.

Why main

This changes inline keyframe bytes and the synthesized avcC/hvcC only, not the wire or catalog format (avcC/hvcC carrying multiples is standard), so it is additive and non-breaking.

Out of scope

Issue items 2 (proper DTS / B-frame timing) and 3 (opaque MPEG-TS carriage) are larger, dev-targeted (wire / Frame / catalog changes), and independent of this blocker. Left for separate PRs as the issue recommends.

Test plan

• cargo test -p moq-mux (240 passing) including:
  • build_avcc_carries_multiple_pps
  • avc3_keyframe_with_two_pps_keeps_both
  • avc3_reinit_drops_superseded_pps (transform drops on reinit)
  • avc3_reinjects_all_cached_pps_on_keyframe (importer re-injection)
  • avc3_reinit_drops_superseded_pps_on_keyframe (importer drops on reinit)
  • export_avc3_preserves_multiple_pps (end-to-end TS export round-trip)
• cargo clippy -p moq-mux --all-targets clean
• cargo fmt
• moq-cli and hang build against the change

Cross-package sync

No js/net or js/hang mirror needed: this is internal moq-mux transmux, not a moq-net wire change or a hang catalog/container format change.

(Written by Claude)

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro

Run ID: 697d8468-be50-4833-ba8f-c95dbc61cdc1

📥 Commits

Reviewing files that changed from the base of the PR and between 3b0cacc and a85a412.

📒 Files selected for processing (1)

• rs/moq-mux/src/codec/h265/import.rs

🚧 Files skipped from review as they are similar to previous changes (1)

• rs/moq-mux/src/codec/h265/import.rs

---

Walkthrough

This change extends H.264 (avc1/avc3) and H.265 (hvc1) codec parameter-set handling from single cached values (Option<Bytes>) to ordered collections (Vec<Bytes>). Two new Annex-B helpers are added: push_distinct for deduplication and reconcile_keyframe_params for cross-GOP parameter-set injection and retention. The build_avcc and build_hvcc box serializers are updated to emit all collected SPS/PPS/VPS NALs instead of single values. The avc3 and H.265 import decoders now track parameter sets per access unit using seen-vectors and reconcile them against retained sets on keyframes. The container ExportSource::refresh_description is updated to allow mid-stream codec reconfiguration updates instead of freezing the description after first initialization.

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title accurately summarizes the primary change: fixing the transmux to carry all distinct SPS/PPS/VPS instead of just the last one.
Description check	✅ Passed	The description is well-detailed and related to the changeset, explaining the root cause, solution approach, and scope of changes.
Linked Issues check	✅ Passed	The PR addresses all coding requirements from issue #1798 work item 1: multi-parameter-set handling in Avc1/Hvc1 transforms, avc3/hev1 importers, and avcC/hvcC encoding with proper array counts.
Out of Scope Changes check	✅ Passed	All changes are within the scope of fixing multi-PPS collapse. The PR properly defers DTS/B-frame timing (#2) and opaque MPEG-TS carriage (#3) as independent follow-ups.
Docstring Coverage	✅ Passed	Docstring coverage is 100.00% which is sufficient. The required threshold is 80.00%.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

✨ Simplify code

• Create PR with simplified code
• Commit simplified code in branch claude/practical-tu-248392

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[coderabbitai]

Actionable comments posted: 1

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@rs/moq-mux/src/codec/h265/import.rs`:
- Around line 233-241: When H.265 reconfiguration is detected at the SPS stage
in the reconfigured block, the code incorrectly clears current.vps_seen which
loses the new VPS that was already collected in the current AU before the SPS
arrived. Remove the line that clears current.vps_seen (line 237) while keeping
the other clear() calls, since VPS precedes SPS in H.265 NAL order unlike H.264
where SPS is the first parameter set. This preserves the newly seen VPS so it
can be properly injected by reconcile_keyframe_params when the IDR frame is
processed.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro

Run ID: 8c03ff66-68c1-444e-800d-cf64c39dfb50

📥 Commits

Reviewing files that changed from the base of the PR and between 6ce3708 and 3b0cacc.

📒 Files selected for processing (7)

• rs/moq-mux/src/codec/annexb.rs
• rs/moq-mux/src/codec/h264/import.rs
• rs/moq-mux/src/codec/h264/mod.rs
• rs/moq-mux/src/codec/h265/import.rs
• rs/moq-mux/src/codec/h265/mod.rs
• rs/moq-mux/src/container/source.rs
• rs/moq-mux/src/container/ts/export_test.rs