Repository navigation
feat(hang)!: video warmup leaves hang-04, and the players stop reading it - #8960
Conversation
Quest-Claim: 4b5c0bc890ac43a1
…g it moq-hang-04 defines `warmup` for audio renditions only. A video rendition's `warmup` is an unknown field, every video group starts with a keyframe again, and the renamed `track` is the only version-4 trigger. The one-decoder video warmup path in @moq/watch, moq_video::decode, moq play, moq-gst, moq-ffi, and moq-c is removed; audio warmup and the container consumer's reach stay. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
kixelated
left a comment
There was a problem hiding this comment.
Automated review by review (OpenAI)
Reviewed 59e48e1cbf515115ed43ea8324e66aac1e190a6a. No blocker found in the removal. Ordinary video keeps keyframe and leading-picture recovery, native Start::Latest still waits for the live edge, and audio pre-roll stays wired. I’d defer video warmup for 0.22 until a producer and the browser matrix are ready; a later return should mark the entry as needing a newer reader so older players can skip it. When resolving the overlap with #8950, keep renamed track as the only hang-04 entry-version trigger. Static review only; I did not rerun playback or interop.
There was a problem hiding this comment.
ℹ️ No critical issues — one stray marker in the draft, inline.
Reviewed changes
Reviewed the full removal of video warmup from moq-hang-04 across the catalog, players, bindings, and draft, with audio warmup left intact.
- Catalog.
VideoConfig::warmupleaves Rusthangand@moq/hang, and both version computations now raiseversionto 4 only for a renamedtrack. A videowarmupparses as an unknown field, and the parity fixture covers it in both languages. - Browser player.
Recoverykeeps only its leading-picture refusal.Renditiondrops the widened subscription, the warmup hold, and the H.264 recovery-point rewrite. DroppingRecovery.restart()is safe: a fresh encoder's first keyframe already clears#neededand the floor. - Rust players.
moq_video::decode::Consumerstill waits for the live edge onStart::Latest.moq play's buffer dropspresent,joined,fresh(), andwarmed(). moq-gst passes no warmup for video. - Bindings and C.
MoqVideo.warmupis gone from moq-ffi, along with the Dart bindings and Go'sbridge.Video. moq-c'smoq_video_configloseswarmup_usandhas_warmup, andMedia::warmupnow defaults to zero, which only audio overrides. - Draft. The
warmupsection moves under Audio and intoAudioDecoderConfigExtensions. The group rule goes back to keyframe-only, and the Version section and changelog are updated to match.
claude-opus-5-5 | 𝕏
|
No actionable comments were generated in the recent review. WalkthroughThis PR removes video Merge Risk:
|
…as cut Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
There was a problem hiding this comment.
✅ No new issues found.
Reviewed changes
Reviewed the delta since the prior Pullfrog review at 59e48e1.
- Draft cleanup. Removed the stray
>>>left where the oldwarmupsection was cut, sojitterflows straight intodelay. This addresses the prior inline finding.
claude-sonnet-5-5 | 𝕏

Claims quest/m2/hang04/video-warmup.md and finishes it (the PR deletes the quest).
Problem
moq-hang-04 promised video
warmup(intra-refresh groups that start without a keyframe) with no producer that writes it, Firefox and Safari never run, and an H.265 refresh rendition that Chromium refuses. Worse, a 0.22 player that claims to read it would accept a later producer's rendition and mis-decode it instead of skipping it.Approach
warmupleaveshang::catalog::VideoConfigand@moq/hang's video schema, so a videowarmupis an unknown field, parsed past and never written back. Both version computations lose the warmup trigger: a renamedtrackis hang-04's only reason forversion4. The MSF mappings only ever derived Opus warmup, so they are unchanged.@moq/watch). The warmup hold inRecoverygoes (its leading-picture refusal stays, back to its pre-feat(watch): run a min-delay and a max-delay decoder for warmup renditions #7555 shape), and so do the widened subscription, the consumer'swarmupreach for video, thewarmupdecoder-config field, andinsertRecoveryPoint, the H.264 recovery-point SEI rewrite Chromium needed. Thedoc/lib/js/watch.mdparagraph goes.moq_video::decode::Consumerno longer widens its subscription or hides pictures after a join;Start::Lateststill waits for the live edge, as documented.moq playdrops the buffer's warmup hold,Buffer::fresh, and the live-edge wait it only did for warmup. moq-gst passes no warmup for video.MoqVideo.warmupleaves moq-ffi (regenerated Dart bindings; Go'sbridge.Videovalidator had nothing left to check, so it and its test go). moq-c'smoq_video_configloseswarmup_usandhas_warmup; its media consumer applies a warmup for audio only.warmupsection moves from Common Rendition Fields to Audio (and intoAudioDecoderConfigExtensions), describing Opus pre-roll only, without the dual-decoder sentences. The group rule returns to "each group MUST start with a keyframe". The Version section and the hang-04 changelog drop videowarmup.moq_mux::container::Playout { delay, warmup }, the JS consumer'swarmupreach, and every audio use. Done by hand, not withgit revert, so the dual decoder feat(watch,moq-video,moq-cli): a warmup rendition plays through one decoder #8857 deleted (rs/moq-video/src/decode/select.rs) stays deleted.shape.md, andentry-version.mddrop their links to this quest; the intra-refresh epic and catalog warmup link this PR as the inverse to restore from.Decisions
Taken while the maintainer was away; each is reversible in review.
a_warmup_update_keeps_the_reader_and_reaches_the_next_onedrove the "a changed warmup never restarts the reader" rule through a video rendition, and moq-gst's sibling-instance fixtures (Publisher,Recorder) only publish and record video. Porting it to audio meant new fixtures for one rule, soa_warmup_is_not_part_of_the_shapechecks the rule where it lives,Shape's equality, anda_rendition_subscribes_with_its_catalog_warmupkeeps the audio subscription check.warmupbound cases (0, 2^53-1, 2^53, -1, 1.5) become one,video warmup is an unknown field, where-1no longer refuses the catalog in either language. The audio bound cases are untouched.Recovery.restart()(JS) and therecover = falseon an encoder end (Rust) came with feat(watch,moq-video,moq-cli): a warmup rendition plays through one decoder #8857 to keep a fresh encoder from warming up. Without warmup they only skip setting a leading-picture floor at a fresh encoder's first keyframe, where no leading picture exists, so both go.warmupdraft section moves under Audio. It sat under Common Rendition Fields, which text renditions share too, and no schema listed it; nowAudioDecoderConfigExtensionsdoes.Impact
Breaking, Rust, JS, bindings, and C:
hang:VideoConfig::warmupis removed. Drop any assignment; a catalog's videowarmupis ignored on read.moq-video:decode::Consumerno longer reads a warmup (there is none to read). Behavior only.moq-ffiand every binding:MoqVideo.warmupis removed (Gomedia.Video.Warmup, Python/Kotlin/Swift/DartVideo.warmup). Drop it from video rendition literals;MoqAudio.warmupis unchanged.moq-c:moq_video_configloseswarmup_usandhas_warmup. Remove them from initializers and rebuild against the new header;moq_audio_config's stay.@moq/hang:VideoConfigSchema(andVideoConfig) losewarmup; a parsed catalog drops a videowarmupas an unknown field.@moq/watch: no public API change.rs/moq-ffi,go/wrapper,kt/moq,py/moq-rs,swift, anddart/moqMIGRATE.mdsay the video record gainswarmuptoo. Only the cut editsMIGRATE.md(check migrate), so the cut should rewrite them to name the audio record alone.Wire: the hang-04 catalog JSON and container group rule, both unpublished. A video
warmupis no longer defined and never raisesversion.Validation
cargo check --workspace --all-targets.just check(nextest over every affected crate, JS, drafts, bindings, Go, Kotlin, Dart): 5994 Rust tests pass. The only failure in the last run waspy/moq-rs/tests/test_socket.py::test_abort_and_endpoint_close_are_errors, the known flake py-socket-abort that moq-dev/moq#8900 fixes. An earlier run timed out fourmoq-rtmppayload tests at load average 24; they pass alone in 30 ms and passed in the next run.just test interop --all: all checks pass.bun test:js/hang563 pass,js/watch626 pass, including the video decoder's join, gap, and leading-picture tests with the hold gone.cargo teston hang, moq-video, moq-cli, moq-ffi, moq-mux, moq-gst, moq-c: all pass under nextest. Under plaincargo test, moq-c'sraw_track_subscription_options_and_updateandroute_broadcast_survives_id_reusefail intermittently, the same on main at 13e593c (3 of 4 runs), since its tests share one process's global state; CI runs nextest, where they pass.warmup: 500and audiowarmup: 80, decoded and re-encoded by@moq/hang(scratch script), writes the video entry withoutwarmup, keeps the audio80, and writes noversion;hang's newa_video_warmup_is_ignoredwrites the same video entry, and the shared parity casevideo warmup is an unknown fieldis read identically by both languages.Follow-ups
None.
Caused by
None: video warmup was deliberate work (moq-dev/moq#5212, moq-dev/moq#7555, moq-dev/moq#8857) that the pre-cut audit chose to move past hang-04, not a defect.
Models
Size: M
Implementer: Claude, claude-opus-5-5, high effort (arm B)
Implementer: Claude Code, claude-sonnet-5-5, unknown effort (iterate, S, arm B)
🤖 Generated with Claude Code
(Written by Claude Opus 5.5 with high effort)
Review
Rewritten by
quest reviewafter each round.Head: 6dccb76
Base: origin/main at 38a1082
Risk: medium,
the base has no RISK.md to name what is always high or customer-facing, so every change is mediumClassified: medium by default
Reviewer: Codex, gpt-6.1-sol, high effort
Session:
01a1e573-acaa-7392-ad82-76bc123e4216, round 2Score: 9/10,
The only new change removes a stray draft marker and introduces no actionable regression; builds and tests were not run as instructed.Rounds: 2; first round 0; findings 0: fixed 0, refuted 0, open 0
Findings: none