Repository navigation
chore: bump moq to catalog updates at their timestamp - #208
Conversation
Pins moq at 81e43217c, the last main commit before moq-dev/moq#8931 made APIs the studio reads crate-private. Follows the group API rename (finish_at/finish), timestamped append_group, the catalog's sync section, and moq-video's ashpd async-io backend. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
|
No actionable comments were generated in the recent review. WalkthroughThe moq subproject reference now points to a new commit. The check job disables automatic Git maintenance. Scene and studio code and tests use updated timestamp-based group APIs and finish methods. Catalog mirroring and related tests use the Merge Risk:
|
There was a problem hiding this comment.
✅ No new issues found.
Approach: sound. Pinning to 81e43217c, the last moq main commit before has_instance went crate-private, gets #146 its timed catalog without the rest of the tip's API churn. I checked the pin against the GitHub compare API: it is on moq's main, 116 commits behind the tip.
Reviewed changes
I checked the submodule bump and each port against the new pin's moq-net, moq-mux, and hang source.
- Group ends: I read
group.rsat the new pin. The newfinish()(end unknown) behaves like the oldclose(), andfinish_at(end)behaves like the oldfinish(end). So the ends thatpublish.rscopyandreplay.rscopy_heldforward are unchanged. append_group(Timestamp):Info::rescaleconverts a scaled timestamp into the track's timescale and rejects a scale mismatch. That makesunscaledcorrect on the untimed uplink probe and thehealth.rstest track. Thets()/from_microsvalues on the hang (µs) tracks also equal the old raw keys.- Catalog
media→sync: This is a rename inListing::mirrorand the tests.moq_mux::catalog::Consumer::nextstill yields a plainCatalog<E>, notTimed, so the studio's catalog readers keep reading at the live edge. - ashpd
async-io: This is forced by moq-video's ashpd backend. zbus keepstokiothrough the keyring store'srt-tokio, so the portals' runtime does not change. - Test-only ports:
ClaimsAccess.pathinguests.rs, andnext_back()inhealth::echo.
ℹ️ CI's Check job failed on a git race, not on this change
The The moq pin is on moq's main step exited with fatal: shallow file has changed since we read it from its git fetch, after 18 s. Because of that, just ci never ran on this head. Re-run the job to get a real CI signal before merging. I confirmed the pin-on-main condition separately.
claude-opus-5-5 | 𝕏
|
Maintainer decision (2026-11-18,
(Written by Claude Opus 5.5) |
actions/checkout disables gc only in the superproject, so the moq submodule's checkout fetch leaves a detached git maintenance that rewrites moq's shallow file while sh/check-moq-main.sh unshallows it, failing with 'shallow file has changed since we read it'. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
|
Applied the maintainer's answers on 576698d: the pin stays at The Check failure was not a flake: it reproduced on every run locally with the runner's git 2.56.1. The submodule's checkout fetch leaves a detached (Written by Claude Opus 5.5 with medium effort) |
There was a problem hiding this comment.
✅ No new issues found.
Reviewed changes
Reviewed the two commits added since the prior pullfrog review (23149c0).
- Pin check race fix:
check.ymlsetsmaintenance.auto=falsethrough job-levelGIT_CONFIG_*env, so the checkout action's submodule fetch no longer leaves a detachedgit maintenancethat rewrites moq'sshallowfile during the unshallow. Job-level env reaches both the checkout step andsh/check-moq-main.sh, and nothing else in the repo usesGIT_CONFIG_*. - Follow-up quest: added
quest/m0/pin-moq-tip.md, a docs-only quest for catching up to moq's tip.
claude-sonnet-5-5 | 𝕏
Main carries #208's timed catalogs, so the composition moves from its interim JSON snapshot track into a `live.kixel.studio.composition` section of the composition broadcast's hang catalog, each catalog stamped with the program frame it applies on. The catalog also lists the program's renditions with `"broadcast": "."`, so a player that knows no compositions plays the program from it. The watcher reads it with hang's catalog consumer and refuses an untimed one. Deletes quest/m2/compose/pin-timed-catalog.md, satisfied by #208, and the layout quest this PR completes. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Brings in #194 (keep out of the way), #207 (command palette), and #208 (moq pin bump). A layer keeping clear places only the landscape picture: the compositor's keepers read and place the landscape view alone, and BUILD's keep helpers take the picture edited, holding or grabbing nothing in portrait. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

Problem
moq-dev/studio#146 (the composition) waits on hang applying each catalog update at its timestamp: the quest
quest/m2/compose/pin-timed-catalog.md, which exists only on #146's branch. That landed upstream as moq-dev/moq#8907, but the studio's gitlink (moq6c726f455) predates it.Approach
just upgrade-moq 81e43217c, not moq's tip. The quest's condition holds at that pin:rs/hang/src/catalog/consumer.rsreturns each catalog as amoq_json::Timed<Catalog>with its timestamp, andquest/m2/catalog-at-timestamp.mdis gone.Moq's tip (
4c8af4c8f) can't build here yet. Its "no API without a caller" sweeps counted only callers inside moq, so they made private two APIs the studio uses:lite::Version::has_instancecrate-private.Stream::check_failover(rs/moq-studio/src/stream.rs) needs it.hang::healthprivate. The health grade (rs/moq-studio/src/health.rs,engine/air.rs,setup.rs,fx.rs) is built on it. Moq's own questquest/m3/stats/health-consumers.mdplans to publish it again.81e43217cis the last main commit before #8931, 94 commits past the old pin. It passessh/check-moq-main.sh.What the bump broke, and the fixes:
async-iobackend (moq-dev/moq#8891), and ashpd refuses to build with both backends. The studio's ashpd now takesasync-iotoo.zbus-secret-service-keyring-store'srt-tokiostill turns on zbus'stokiofeature, which zbus prefers, so the portals still run on tokio as before.finish(end)is nowfinish_at(end), andclose()is nowfinish().append_grouptakes aTimestamp:unscaledon untimed tracks, and the track's own unit on timed ones.mediasection is nowsync(moq-dev/moq#8977).Access(moq-dev/moq#8933), so the guest tests compare itspath.next_back()overlast()inhealth.rs.fatal: shallow file has changed since we read it, which was not a flake.actions/checkoutdisables gc only in the superproject, so the moq submodule's checkout fetch leaves a detachedgit maintenance run --autothat rewrites moq'sshallowfile whilesh/check-moq-main.shunshallows it. The Check job now setsmaintenance.auto=false(throughGIT_CONFIG_*), since the checkout is thrown away.Impact
Alternatives
hang::healthpublic, and a publicmoq_net::Versioninstance query), then this bump. It also crosses the rest of the tip's breaks, which I tried and set aside:encode::Producerbecameencode::Publisher(feat(moq-video)!: encode::Producer takes raw frames and follows the grant moq#8964), andInfo::with_refreshbecame aRefresh(feat(net)!: refresh a track by interval, and on request moq#9054), which means rewritingladder::JoinsonDemand::poll_refresh. RoutePreferencebecamePriority,Key::to_publicbecamepublic, andDemand::wantedbecame a bool. Too large for an S pin that feat(scene)!: the studio publishes a composition #146 waits on.has_instanceby matching the version string. That fails closed, but it hard-codesmoq-lite-07-wip.Decisions
Maintainer, 2026-11-18, in the
/quest-runpick round:81e43217cnow (rejected: hold for moq's tip). It unblocks moq-dev/studio#146.Follow-ups
quest/m0/pin-moq-tip.md[M]: the studio catches up to moq'smain, absorbing the tip's breaks listed in Alternatives. Its## Waitingnames the two moq changes it needs first:hang::healthpublic again (moq'squest/m3/stats/health-consumers.md), and a publiclite::Versioninstance query, whose quest is pending planning in moq's m1 so it does not gate moq's 0.22.0 cut. Answers recorded in its Plan; no open decisions.Caused by
Validation
just ci: lints, plus 423 tests passed.just render-swon Mesa lavapipe, which composites, encodes, publishes, and reaches a PipeWire camera: 67 + 18 + 14 + 1 tests passed.cargo tree -e features -i zbus --features moq-studio/pipewire: zbus hastokio(fromsecret-service/rt-tokio), and ashpd has onlyasync-io, so the portal runtime is unchanged.fatalon every run;pgrepshowedgit maintenance run --auto --detachalive inmoq/right after checkout, andmoq/.git/shallow's mtime moved ~3 s later. Withmaintenance.auto=falsein the environment, no maintenance ran and two replays passed.main(quest/m*/pin-*.md) still wait. Each upstream quest they name still exists at this pin, andmoq_audio::playback::Devicereports no output kind, so none is deleted.Models
Size: S
Implementer: Claude Code, claude-opus-5-5, medium effort
Implementer: Claude Code, claude-opus-5-5, medium effort (iterate, S)
🤖 Generated with Claude Code
Review
Rewritten by
quest reviewafter each round.Head: 576698d
Base: origin/main at fa87036
Risk: high,
it changes .github/workflows/check.yml, which is always high: workflows run with the repository's secrets and decide what merges and shipsClassified: high by the always-high paths
Reviewer: Codex, gpt-6.1-sol, medium effort
Session:
01a1f0e7-7756-7b81-ac52-51718bf9b224, round 2Score: 8/10,
The incremental source review found no actionable regressions; builds and tests were not run under the review constraints.Rounds: 2; first round 0; findings 0: fixed 0, refuted 0, open 0
Findings: none
Reviewer: Codex, gpt-6.1-sol, low effort, shadow
Session:
01a1f0e7-7754-7d31-b35e-2922927ada13Rounds: 1; first round 0; findings 0: fixed 0, refuted 0, open 0