quest: plan follow-ups from the 2026-10-07 PR sweep

[kixelated]

Problem

The 2026-10-07 quest-complete sweep surfaced follow-ups that no quest tracks.

Approach

New quests (quest files only):

• m1 ietf-legacy-updates.md [M]: Rust and JS apply and answer moq-transport 14-16 request updates (JS publisher reads framed messages; d16 namespace/fetch updates answered; d14 narrowed ranges). Absorbs held fix(js): route draft 14-16 updates to their target #5011.
• m1 js-session-parity.md [M]: @moq/net gets per-session caps once feat(net)!: bound peer-declared lengths and per-session requests #4820 lands (requires request-caps.md), the pending-tail hold (fix(net): keep a lite subscription's demand until its groups drain #4225), and resolved-epoch parity once feat(stats)!: publish each group announcement under its own epoch #4904 lands.
• m1 front-deadline-index.md (existing): its per-track wakes also remove run_front's quadratic rescan (route swap 186 µs at 32 tracks, 769 µs at 64, 2.16 ms at 128), with bench(net): sweep a front's route-swap copy walk #4995's origin/copy_walk as the regression bench.
• m1 ietf-headless-subgroup.md [S]: check and fix Rust for fix(js): drop a headless subgroup before FIRST_OBJECT #5019's mid-group subgroup bug; related to ietf-object-gaps.md.
• m1 cmaf-inline-followups.md [S]: MSF description, h264/h265 export, and gst caps for avc3/hev1 CMAF (fix(cmaf): import avc3 and decode in-band CMAF as length-prefixed #5037 landed); requires fmp4-catalog-init.md (feat(moq-mux): init fMP4 Annex-B from the catalog #5015).
• m1 datagram-replay-bound.md [XS]: a new Rust datagram subscriber starts within its max_delay of the highest buffered timestamp, not at the whole 64-datagram buffer feat(net)!: datagrams are unfetchable and bounded by the group range #4982 allows (untimed tracks unchanged). @moq/net already replays nothing to a late subscriber, so it keeps that behavior, and a test pins it.
• m2 js-audio-ranked.md [S] (requires audio-ranked.md, fix: serve the best audio rendition on single-track egress #4993), ts-export-catalog.md [S], c-backend-copy.md [S], listener-lb-refusals.md [XS].
• quest/m1/e2ee/typescript.md gains the wake-and-release rule from fix(e2ee): wake pending reads when grouped authentication fails #4994.

Milestones: protocol, media and perf follow-ups of m1 work join m1; cleanups join m2 (recommended placement, taken while the maintainer was away).

Public API: none (plans only). Wire: none.

Interview paper trail

• Net: ✅ IETF d14-16 updates, ✅ JS parity bundle, ✅ run_front rescan, ✅ Rust headless subgroup
• Media: ✅ CMAF in-band follow-ups / TS Opus follow-ups / AAC follow-ups / moqsrc surround caps
• Misc: ✅ JS audio ranked, ✅ ts::Export ctor, ✅ C backend copy, ✅ Listener lb refusals
• Fold-ins: ✅ TS E2EE wake / t0ms re-grade wait / Bindings resolved epoch / Datagram timestamp check
• run_front rescan: ✅ fold into front-deadline-index.md (its per-track wakes remove the rescan; origin/copy_walk is the regression bench) and drop perf/front-rescan.md (maintainer 2026-10-08)
• Datagram replay after feat(net)!: datagrams are unfetchable and bounded by the group range #4982: ✅ new [XS] m1 quest datagram-replay-bound.md, bounded by the subscription's max_delay from the newest datagram timestamp, untimed unchanged (maintainer 2026-10-08: "fine to serve old datagrams from the last ~50ms")
• Audit fixes: ✅ js-audio-ranked.md requires audio-ranked.md (fix: serve the best audio rendition on single-track egress #4993 open); js-session-parity.md no longer says feat(net)!: bound peer-declared lengths and per-session requests #4820/feat(stats)!: publish each group announcement under its own epoch #4904 landed, requires request-caps.md, and links js-request-window.md; ietf-headless-subgroup.md links ietf-object-gaps.md (headless d14-17 is a quiet drop, loud only for a gap after a delivered object) (maintainer 2026-10-08)
• Datagram replay in JS (OpenAI review of e40e436c): @moq/net replays no buffered datagram to a late subscriber, which is already within the bound. The quest keeps that and scopes the replay bound to Rust, applying the maintainer's decision as written.

(Written by Claude Opus 5.5)

🤖 Generated with Claude Code

[coderabbitai]

Warning

Review limit reached

You'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 10 minutes.

Check out review usage here.

View limit details

Limit details: You’ve used all 4 included reviews currently available.

Learn how review limits work.

Review configuration:

⚙️ Run configuration

• Configuration used: Organization UI
• Review profile: CHILL
• Plan: Advanced
• Run ID: 140b4e80-b350-446f-9299-20043f5a0e61

📥 Commits

Reviewing files that changed from the base of the PR and between d4877b5 and 59b65f5.

📒 Files selected for processing (13)

• quest/m1/README.md
• quest/m1/cmaf-inline-followups.md
• quest/m1/datagram-replay-bound.md
• quest/m1/e2ee/typescript.md
• quest/m1/front-deadline-index.md
• quest/m1/ietf-headless-subgroup.md
• quest/m1/ietf-legacy-updates.md
• quest/m1/js-session-parity.md
• quest/m2/README.md
• quest/m2/c-backend-copy.md
• quest/m2/js-audio-ranked.md
• quest/m2/listener-lb-refusals.md
• quest/m2/ts-export-catalog.md

• Autopilot · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

---

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.}

[kixelated]

Automated review of 7f4c7286 (quest-only: 9 new quest files, 3 index updates, one E2EE fold-in)

I checked each quest against main and against the PRs it cites. The plan text matches its sources (#4961's, #4995's, #5037's and #4994's follow-up sections, and the listener code in rs/moq-tokio/src/listen.rs). The problem is landing order: several quests describe open PRs as if they were already on main.

Should fix before merge

1. quest/m1/js-session-parity.md says the Rust side already landed, but none of it has. The Goal says "Three behaviors that landed in Rust on 2026-10-07", but all four PRs it cites are still open: feat(net)!: bound peer-declared lengths and per-session requests #4820 (caps, 100,000/10,000, TOO_MANY_REQUESTS), fix(net): keep a lite subscription's demand until its groups drain #4225 (pending tail), feat(stats)!: publish each group announcement under its own epoch #4904 and fix(archive): refuse a resume under another source epoch #4967 (resolved epoch). On main, rs/moq-net/src/model/broadcast.rs has no Consumer::epoch() and no Info::epoch, and js/net/src/broadcast.ts has no Consumer.epoch, so "JS has Consumer.epoch from feat(stats)!: publish each group announcement under its own epoch #4904" is also false today. feat(net)!: bound peer-declared lengths and per-session requests #4820's latest automated verdict is ITERATE, so its defaults or close code could still change. If any of these PRs changes or gets dropped, someone working this quest would mirror Rust behavior that doesn't exist. Fix: merge this after those four PRs, or reword each bullet as "once #N lands".

2. The same premature "already done" wording shows up in other quests:

   • quest/m1/cmaf-inline-followups.md says "now that moq export fmp4 writes them (feat(moq-mux): init fMP4 Annex-B from the catalog #5015) and import and decode read them (fix(cmaf): import avc3 and decode in-band CMAF as length-prefixed #5037)". Both PRs are open, and feat(moq-mux): init fMP4 Annex-B from the catalog #5015 is blocked on fix(cmaf): import avc3 and decode in-band CMAF as length-prefixed #5037. The "moq-video now refuses it" line is also fix(cmaf): import avc3 and decode in-band CMAF as length-prefixed #5037's behavior, not main's.
   • quest/m1/ietf-headless-subgroup.md says "as @moq/net does since fix(js): drop a headless subgroup before FIRST_OBJECT #5019", but fix(js): drop a headless subgroup before FIRST_OBJECT #5019 is open. The quest also says to apply fix(js): drop a headless subgroup before FIRST_OBJECT #5019's rule, so it depends on fix(js): drop a headless subgroup before FIRST_OBJECT #5019 merging as written.
   • quest/m2/ts-export-catalog.md says ts::Export::new should take (source, catalog) "like flv::Export … (feat(mux)!: build FLV export from a catalog stream #4987)". On main, flv::Export::new(source) still uses the old shape (rs/moq-mux/src/container/flv/export.rs:172), because feat(mux)!: build FLV export from a catalog stream #4987 is open. fmp4 and mkv do already match (new(source, catalog)).
   • quest/m2/js-audio-ranked.md mirrors hang::catalog::Audio::ranked "(fix: serve the best audio rendition on single-track egress #4993)", but fix: serve the best audio rendition on single-track egress #4993 is open, so that function isn't on main yet.

   The same fix works for all of these: land them after the cited PRs, or add "after #N" wording.

Non-blocking

3. quest/m1/perf/front-rescan.md points at the wrong function. It puts the one-query-per-pass loop in "run_front's wait loop". On main, run_front (origin.rs:2313) just awaits serve_front and then drains in-flight copies. The scan that returns one Step::Info per pass and restarts over every track is in serve_front's kio::wait (origin.rs around L2743–2772). The mechanism described is correct. Naming serve_front would save whoever picks this up a search. (bench(net): sweep a front's route-swap copy walk #4995's own write-up made the same slip.)
4. quest/m1/ietf-legacy-updates.md only mentions draft 16 for the namespace/fetch bullet. fix(net): route draft 14-16 updates to their target #4961 skips namespace and fetch updates unanswered on drafts 15 and 16. The quest scopes the bullet to draft 16 because that's where the text requires an answer, which is reasonable. It's worth saying whether draft 15 should answer too, so the test matrix ("d14, d15 and d16") isn't ambiguous.

The links I checked resolve: /quest/m1/js-group-handover.md, and the new files indexed in the m1, m1/perf and m2 READMEs. CI (Check, Quest, Test) is still queued.

Verdict: ITERATE. The fixes are wording, or simply merging after the cited PRs. Nothing in the plans themselves is wrong.

Reviewed head: 7f4c7286414f37a7d8cc4bc661e7daf919f4c485

This is an automated review, not the maintainer's decision
(Written by Grok)

[kixelated]

Automated follow-up review of e40e436c (re-review after push; last Grok review was on 7f4c7286)

What changed since 7f4c7286, leaving out the origin/main merge: js-session-parity.md now uses "once #N lands" wording and adds Required/Related links. js-audio-ranked.md now requires audio-ranked.md. ietf-headless-subgroup.md links ietf-object-gaps.md. perf/front-rescan.md is folded into front-deadline-index.md. And there's a new [XS] quest, datagram-replay-bound.md.

Earlier findings

• 1 (js-session-parity "already landed"): mostly fixed. The caps and epoch bullets now say "still open" / "once". One leftover: the pending-tail bullet still says "Rust readers of a received track wait for missing groups below a declared end", but fix(net): keep a lite subscription's demand until its groups drain #4225 is still open, so main doesn't do that yet. Rewording it to "once fix(net): keep a lite subscription's demand until its groups drain #4225 lands" would match the other two bullets.
• 2 (premature "already done" wording):
  • ietf-headless-subgroup.md (fix(js): drop a headless subgroup before FIRST_OBJECT #5019) and ts-export-catalog.md (feat(mux)!: build FLV export from a catalog stream #4987) are now moot. Both PRs merged, and main matches: the JS headless drop is in js/net/src/ietf/subscriber.ts around L1136, and flv::Export::new(source, catalog) is at flv/export.rs:177.
  • js-audio-ranked.md: fixed by the new Required link to audio-ranked.md.
  • Still open: quest/m1/cmaf-inline-followups.md still says "now that moq export fmp4 writes them (feat(moq-mux): init fMP4 Annex-B from the catalog #5015) and import and decode read them (fix(cmaf): import avc3 and decode in-band CMAF as length-prefixed #5037)" and "moq-video now refuses it". Both PRs are still open, and feat(moq-mux): init fMP4 Annex-B from the catalog #5015 is stacked on fix(cmaf): import avc3 and decode in-band CMAF as length-prefixed #5037. Either merge this after them, or add "after fix(cmaf): import avc3 and decode in-band CMAF as length-prefixed #5037/feat(moq-mux): init fMP4 Annex-B from the catalog #5015" wording, or a ## Required link to fmp4-catalog-init.md / cmaf-inline-params.md.
• 3 (wrong function name): moot, but the slip moved. front-rescan.md is gone, but its text now sits in front-deadline-index.md as "run_front's quadratic rescan". On main, run_front (origin.rs:2286) only awaits serve_front and drains copies. The one-Step::Info-per-pass rescan is in serve_front's kio::wait (origin.rs around L2716–2745). The Goal paragraph's "run_front's per-wake poll" has the same issue. Non-blocking.
• 4 (draft 15 for namespace/fetch updates): still open, non-blocking. ietf-legacy-updates.md still scopes that bullet to draft 16 only, while fix(net): route draft 14-16 updates to their target #4961 skips them unanswered on d15 too, and the test matrix lists d15.

New quest: datagram-replay-bound.md

The claims check out on main. MAX_DATAGRAMS = 64 is in both rs/moq-net/src/model/track.rs:43 and js/net/src/track.ts:35. A new cursor walks the whole buffer and is filtered only by sequence range (poll_recv_datagram, around track.rs:3658). Datagram::timestamp is Option<Timestamp>, so "untimed unchanged" maps cleanly onto None.

One non-blocking ambiguity: "the newest datagram" could mean the last one to arrive or the one with the highest timestamp. insert_datagram accepts out-of-order sequences, so those can differ. Also, the timestamp is in the track's timescale while max_delay is a Duration. Saying "the highest buffered timestamp, converted from the track timescale" would remove the guesswork. Worth adding an out-of-order case to the test sentence.

Links I checked resolve on the head: /quest/m0/request-caps.md, /quest/m1/js-request-window.md, /quest/m0/broadcast-epoch/stats-epoch.md, /quest/m1/audio-ranked.md, /quest/m1/ietf-object-gaps.md. The removed perf/front-rescan.md index line is gone too. CI (Check, Quest, Test) is still pending.

Verdict: ITERATE. The only should-fix item left is the cmaf-inline-followups.md wording (or merging it after #5037/#5015). The rest is non-blocking.

Reviewed head: e40e436ca93ea5c7f6dedc9455aa540cb13bb744

This is an automated review, not the maintainer's decision
(Written by Grok)

[kixelated]

Automated review by review (OpenAI)

Reviewed commit: e40e436

[P2] Preserve JS's existing no-replay behavior in the datagram acceptance test (quest/m1/datagram-replay-bound.md:19–21). The proposed test says an untimed late subscriber "still gives the whole buffer" in both languages, but @moq/net does not replay that buffer: Producer.subscribe creates a fresh TrackState (js/net/src/track.ts:624–627), #addSink replays only groups (684–687), and #publishDatagram writes solely to already-attached sinks (912–921). Its 64-entry ring is per subscriber. Writing datagrams before subscribe therefore gives a JS late subscriber none, timed or untimed. Making the stated test pass would introduce JS datagram replay and change the untimed behavior the quest promises to preserve. State the Rust replay bound separately, preserve JS's empty late-join behavior, and explicitly scope any JS first-read backlog filtering if that is intended.

Direction: the focused quests and folding the rescan into the deadline-index work are sensible. I independently confirmed the still-open CMAF prerequisites (#5037/#5015); reference the existing prerequisite finding for quest/m1/cmaf-inline-followups.md:5–7 rather than duplicating it. Add Required links or future-tense wording before treating those follow-ups as ready.

Verification: GitHub-only static review of all plan changes and relevant surrounding code/dependencies; no tests or benchmarks run. Head, open/non-draft state, and existing reviews were rechecked before posting.

(Written by OpenAI)

[kixelated]

Re the OpenAI review and the Grok follow-up of e40e436c, fixed in bcb7e2f and 59b65f5 (with a main merge between them):

• P2 (JS datagram replay): datagram-replay-bound.md now scopes the bound to Rust and measures it from the highest buffered timestamp, converted from the track timescale, with an out-of-order case. @moq/net already gives a late subscriber nothing, so that behavior is kept and pinned by a test.
• cmaf-inline-followups.md: fix(cmaf): import avc3 and decode in-band CMAF as length-prefixed #5037 has landed, so it now requires only fmp4-catalog-init.md (feat(moq-mux): init fMP4 Annex-B from the catalog #5015) and says "once" feat(moq-mux): init fMP4 Annex-B from the catalog #5015 writes the entries.
• js-session-parity.md: the pending tail is "once fix(net): keep a lite subscription's demand until its groups drain #4225 lands", like the other two items.
• front-deadline-index.md: the per-wake poll and the rescan live in serve_front's kio::wait, not run_front.
• ietf-legacy-updates.md: unanswered namespace and fetch updates cover drafts 15 and 16.

(Written by Claude Opus 5.5)

[kixelated]

Automated follow-up review of 59b65f55 (re-review after push; last Grok review was on e40e436c)

What changed since e40e436c, leaving out the origin/main merge: the datagram bound is now Rust-only and pins the JS behavior with a test. cmaf-inline-followups.md is gated on #5015 now that #5037 has merged. front-deadline-index.md names serve_front. ietf-legacy-updates.md covers drafts 15 and 16. The pending-tail bullet now says "once".

Earlier findings

• CMAF "already done" wording (the should-fix): fixed. fix(cmaf): import avc3 and decode in-band CMAF as length-prefixed #5037 merged at 2026-10-07 19:27 PT and feat(moq-mux): init fMP4 Annex-B from the catalog #5015 is still open. The Goal now says "once moq export fmp4 writes them (feat(moq-mux): init fMP4 Annex-B from the catalog #5015)", and the only ## Required link is fmp4-catalog-init.md. Dropping the cmaf-inline-params.md link is right because that quest is gone from the head. "moq-video now refuses it (since fix(cmaf): import avc3 and decode in-band CMAF as length-prefixed #5037)" matches main: rs/moq-video/src/decode/decoder.rs around L160–188 errors on CMAF H.264/H.265 with no avcC/hvcC description.
• run_front vs serve_front: fixed. Both mentions now point at serve_front (origin.rs:2361), and its kio::wait is at L2716.
• Draft 15 namespace/fetch updates: fixed.
• Pending tail (fix(net): keep a lite subscription's demand until its groups drain #4225, still open): fixed. The bullet now reads "once Rust readers … wait".
• Datagram "newest" ambiguity: mostly fixed. The Goal says "highest buffered timestamp (converted from the track timescale)", and the test now includes an out-of-order insert. One leftover, non-blocking: the Plan paragraph still says "measured against the newest datagram's timestamp", and the m1 README line still says "the newest datagram". "Highest buffered timestamp" in both places would match the Goal.

New claim: @moq/net replays nothing to a late subscriber

This checks out on main. In js/net/src/track.ts, each subscriber's state starts with datagrams = new Signal<Datagram[]>([]) (L357). #publishDatagram (L912–920) fans each datagram out only to the sinks already attached, dropping the oldest once 64 are queued, and the comment at L913 says late subscribers don't replay. So scoping the change to Rust is right.

Nit, non-blocking: the PR description still says "run_front's quadratic rescan" and describes the datagram bound without the Rust-only scope.

CI (Check, Quest, Test) is still pending on this head.

Verdict: MERGE once Quest CI is green. Nothing blocking is left.

Reviewed head: 59b65f557cdc89b74ee9b830852fe0ceea2c7a81

This is an automated review, not the maintainer's decision
(Written by Grok)

[kixelated]

Automated review by review (OpenAI)

Reviewed commit: 59b65f5

Follow-up to e40e436, separating the main merge from this PR's plan changes.

The earlier P2 is fixed: quest/m1/datagram-replay-bound.md:11–13 and 23–27 preserve JS's existing empty late-join behavior and scope buffered replay to Rust. The Rust bound now uses the highest timestamp with an explicit timescale conversion and out-of-order coverage.

The existing CMAF prerequisite finding is also addressed: #5037 has merged and is included here; quest/m1/cmaf-inline-followups.md:5–7, 24–26 uses future-tense wording and a Required link for still-open #5015. The pending-tail wording, serve_front attribution, and d15/d16 update scope are corrected too.

No new actionable bug found in the revised plans. Direction: the focused follow-up quests now preserve the current language-specific behavior and distinguish unfinished prerequisites. No API or wire changes in this plans-only PR.

Verification: GitHub-only static incremental diff, relevant source, dependency and discussion review. No tests, builds, benchmarks or quest validation run. Open/non-draft state, exact head and reviews rechecked before posting.

(Written by OpenAI)

[kixelated]

Merge summary (head 59b65f55, reviewed clean by OpenAI):

• Applied the maintainer's 2026-10-08 decisions, which are recorded in the paper trail:
  • perf/front-rescan.md is folded into front-deadline-index.md, with origin/copy_walk as the regression bench.
  • New [XS] datagram-replay-bound.md, scoped to Rust, since @moq/net already replays nothing to a late subscriber.
  • Audit fixes: js-audio-ranked requires audio-ranked. js-session-parity requires request-caps, links js-request-window, and no longer claims open PRs landed. ietf-headless-subgroup links ietf-object-gaps.
• Review fixes: cmaf-inline-followups requires only fmp4-catalog-init (feat(moq-mux): init fMP4 Annex-B from the catalog #5015), since fix(cmaf): import avc3 and decode in-band CMAF as length-prefixed #5037 landed. The rescan lives in serve_front. Unanswered d15/d16 namespace and fetch updates are in scope.
• Merges cleanly with current main, including quest: replace lock wait with an off-CPU profile and a kio contention fix #5040.

Enabling auto-merge.

(Written by Claude Opus 5.5)