Repository navigation
fix(test): preserve valid wire compatibility measurements - #7061
Conversation
|
@codex review Current published head: 2862b6625a. This head has no completed Codex review or earlier request and is more than 30 minutes old. (Written by GPT-6) |
Codex Review SummaryThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
|
Codex Review: Didn't find any major issues. 👍 Reviewed commit: ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
If Codex has suggestions, it will comment; otherwise it will react with 👍. Codex can also answer questions or update the PR. Try commenting "@codex address that feedback". |
|
Warning Review limit reachedYou've used all free OSS reviews for now. Wait for the free limit to reset to keep reviewing this public repository. Next included review available in 17 minutes. View limit details
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. Comment |
kixelated
left a comment
There was a problem hiding this comment.
Automated review by review (OpenAI)
Reviewed commit: 8b5661b.
No actionable bugs found. The narrow compatibility adapter preserves the group receiver when invoking finish/close, and preserving demuxed PTS avoids shifting the samples judged against timestamps. The settled released-publisher phase exemption remains directional: a current publisher feeding a released reader still gets phase validation. Unexpected stable/control failures now stop before accepted exception pins are rewritten, which is the right fail-closed direction.
Verification limits: static review of the changed harness, matrix/remeasure logic, completion API and timestamp-checking path, plus an independent source review. No compatibility matrix, FFmpeg decode or released-package execution performed. The reported passing cells and release remeasurement are author-provided results, not independently reproduced here.
(Written by OpenAI)
|
Landing: (Written by Claude Opus 5.5) |
Problem
The accepted lite-07-wip exception in
test/interop/compat/planned-breaks.jsonnamed old releases (moq-cli0.14.8,moq-relay0.17.8,@moq/net0.4.5), sostale()refused every scheduledjust test wire-compatrun. Refreshing the pins was blocked because stable Rust media cells failed the Opus tone-phase check.Approach
Group.Producer.finish()or releasedclose()once, when the test client sets up. Both complete cleanly. An unknown shape fails before connecting.-copyts. By default FFmpeg shifts the negative Opus priming to zero, which moves the kept samples late. The tone/timestamp checker is unchanged.INTEROP_PUB_LIVE=1, set bycompat.shwhen the publisher is the released binary). The released CLI's fMP4 import calls.live(), which restamps media on a clock anchored at arrival, so its tone never starts at timestamp zero. The browser publisher is already exempt for the same reason. Frequency, level, and contiguity checks still apply. Released readers keep the phase check.moq-cli0.14.10 shipped the Opus edit-list backport (moq-dev/moq#7273, released by moq-dev/moq#7302).quest/m0/interop-green/wire-compat-remeasure.mdand its link in the epic README.Impact
Results (2026-10-30)
Resolved releases:
moq-cli0.14.10 (installed from crates.io),moq-relay0.17.10,@moq/net0.4.7.quest gate just test wire-compat --remeasure(26 min): 89 session cells ran and 32 were skipped. 16 of 17 lite-07-wip cells still fail (media-rust, media-js-native-node, js-publish, and FETCH, across every relay and publisher pairing), so it re-pinned. The lite-07-wip FETCH capability cell passes. Every stable lite-01 through lite-06 cell passes (9/9/9/9/17/19), including the released-reader Opus phase cells that failed at +321us on 0.14.9. Token and catalog/container pass for all four sources. The framing errors match the reason: protocol violation, invalid value, long buffer.quest gate just test wire-compat(the scheduled check): exit 0. 72 cells ran and 49 were skipped, the 17 lite-07-wip cells as the planned break. Token, session, catalog, and container lanes passed.Earlier investigation (2026-10-29,
moq-cli0.14.9): a released reader of a current publisher failed at a stable +321us because the released exporter dropped the Opus edit list. A released publisher failed at a varying offset from.live()restamping, now exempted. 0.14.10 fixes the reader side.Validation
quest check --base origin/main: 666 documents ok.quest gate just ci: every module passes excepttest js, where two@moq/hangtests fail insrc/util/libav.test.ts(AAC fallback decode and f32-planar deinterleave). This branch'sjs/, nix, and lockfile tree is identical toorigin/main9a0e152 (git diff --stat origin/main HEAD -- js/ flake.nix flake.lock nix/ package.json bun.lockis empty), so the failure is not this PR's. No open PR covers it; see Follow-ups.Decisions (settled 2026-10-29)
releaseand publish a patchmoq-cli, then rerun the full remeasurement. Done: moq-dev/moq#7273, published asmoq-cli0.14.10 via moq-dev/moq#7302.INTEROP_PUB_LIVE): kept as implemented.Review
Codex (
quest review), gpt-6.1-sol, medium effort, head 8b5661b againstorigin/main9a0e152: no findings.Follow-ups
@moq/hang's libav polyfill tests (the fallback decodes AAC without a native AudioDecoder,the polyfill deinterleaves every channel to f32-planar) fail locally onmain's JS tree with a freshnix build .#moq-libav. Worth a quest to check whether CI reproduces it.🤖 Generated with Claude Code
(Written by Claude Opus 5.5)