quest: plan live import clock and tracing capture audit

[kixelated]

Plans two follow-ups in the test-flakes-2 line, from #4687 and #4690.

• mux-live-import-clock [S]: crate::Clock uses web_async::time::Instant, so the moq-mux live import restart tests advance a paused clock instead of sleeping.
• trace-capture-audit [XS]: move any scoped tracing captures to moq-net: per-thread WARN capture for the drop tests #4690's process-wide helper, so a parallel test can't silence them through the call-site cache.

Decisions

C goal: the moq-mux live_import tests sleep no real time

• ✅ Yes
• Drop it

C shape

• ✅ Make crate::Clock use web_async::time::Instant
• Inject a now into the importers

E: plan an audit of tracing-capture tests

• ✅ Add to the plan, in the test-flakes-2 line
• Skip it

Review fixes: the #4687 citation no longer implies Clock already moved; the quest lists the public std::time::Instant boundary (Clock::at, json/binary timed writes) and leaves one open question for the maintainer before starting: convert at the boundary on main (recommended) or change the types on dev. The tracing audit waits on #4690 and notes its helper is WARN-specific.

Public API: none from this PR. Wire: none.

(Written by Claude Opus 5.5)

🤖 Generated with Claude Code

[kixelated]

ITERATE (head 1265ae09f385a991e0d35dd7836f9bada5bd2a3f)

Plans two test-flakes-2 children from the #4687 / #4690 line: drive moq-mux live-import idle gaps on a paused clock, and audit scoped tracing captures for the callsite-interest flake. The live-import facts check out on main (ts/fmp4 live_import and flv import still std::thread::sleep(idle) with a 300 ms case; gap goes through Anchor → Clock::now() → std::time::Instant::elapsed(); late_clock pins std::time::Instant).

Blocking

1. mux-live-import-clock.md mis-cites moq-mux: run the TS SI debounce test on the paused clock #4687. The Decided paragraph says crate::Clock's monotonic epoch becomes web_async::time::Instant “as moq-mux: run the TS SI debounce test on the paused clock #4687 did for the TS SI debounce.” On moq-mux: run the TS SI debounce test on the paused clock #4687’s head (d965302e), SI Capture dropped its crate::Clock and stores a local web_async::time::Instant for the debounce only; rs/moq-mux/src/clock.rs on main still uses std::time::Instant. This plan is a broader Instant-type change than moq-mux: run the TS SI debounce test on the paused clock #4687 shipped. Fix the citation (e.g. “same Instant type moq-mux: run the TS SI debounce test on the paused clock #4687 used for the SI debounce,” not “moq-mux: run the TS SI debounce test on the paused clock #4687 changed Clock”), so the implementer doesn’t treat Clock-wide Instant swap as already-landed precedent.

Non-blocking

2. Public API is already clear — don’t “check whether.” Clock / Clock::at(epoch: Instant, wall: SystemTime) are pub and re-exported from moq-mux. Swapping Instant to web_async::time::Instant (native: tokio::time::Instant) is a typed break. Call sites already pass std::time::Instant from test_util::late_clock, moq-cli publish tests, moq-hls export tests, moq-audio / moq-video capture fixtures, and catalog tests. Spell the child work as “update callers; Instant change is public API,” not “check whether Clock::at is exported.”

3. capture / stamp share the Instant surface. Clock::capture / stamp take Instant / Timed<_, Instant>, and moq_net::Timed’s docs say moq-mux uses std::time::Instant. Fold those into the wasm/build checklist with Clock::at, not only the constructor.

4. trace-capture-audit helper fit. moq-net: per-thread WARN capture for the drop tests #4690’s process-wide helper is WARN/message-specific (count_drop_warnings in rs/moq-net/src/model/test_tracing.rs). Moving “any” scoped capture there may need a more general process-wide + per-thread helper (or a shared copy), not a straight move of every capture onto that WARN counter. (On main the file still uses with_default; the process-wide shape is on moq-net: per-thread WARN capture for the drop tests #4690’s head — fine as a dependency.)

5. CI still queued at review time (Check / Test / Quest); no failure signal yet.

No “Alternative worth exploring”: Decisions already rejected translate_at injection; the Clock Instant swap is a coherent way to make the production now() path mockable, once the #4687 citation and API blast radius are honest.

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

[kixelated]

Automated review by review (OpenAI)

Reviewed 1265ae0 (full plan diff).

No new actionable findings beyond the existing API-impact clarification. The three import helpers really do sleep through the idle gap, and their anchor reads Clock::now() on std::time::Instant.

Independently confirming and extending the existing API note: replace the uncertain export check in quest/m1/test-flakes-2/mux-live-import-clock.md:24–28 with an explicit boundary decision. Clock::at is already public and re-exported (rs/moq-mux/src/clock.rs:89, lib.rs:45). The same std::time::Instant contract also reaches public binary/JSON timed writes (binary.rs:182,257; json.rs:225,309) through Clock::stamp. Preserve those input contracts deliberately or declare the typed API break and include their callers/docs in the child scope and branch choice. capture and stamp themselves are currently crate-private.

Direction: using the shared mockable monotonic clock is coherent and avoids importer-only time plumbing; retain the fixed wall mapping. #4687 is precedent for the Instant type in private SI debounce code, rather than a completed Clock-wide migration. For trace-capture-audit.md:17–21, adopt the helper after accounting for the open #4690 installation-window finding, rather than copying that unresolved race into other crates. No duplicate inline findings added.

Verification: static plan, dependency-PR, API/caller, and import-test inspection. No code changes, builds, or tests run locally; this does not independently verify the proposed implementations.

(Written by review (OpenAI))

[kixelated]

Automated review by review (OpenAI)

Delta reviewed: 1265ae0 → ac25d88, separating the inherited #4687/base merge from the two plan updates.

The prior planning concerns are addressed:

• mux-live-import-clock.md:25–40 now identifies the public constructor and binary/JSON timed-input contracts, covers callers/docs, and requires a maintainer decision between preserving those inputs and an explicit typed break on dev. The #4687 citation correctly limits its precedent to private SI debounce code.
• trace-capture-audit.md:17–27 now waits for #4690's installation-window finding to be resolved and accounts for the helper's WARN/message-specific scope. This fixes the dependency plan; it does not claim that #4690's implementation is already fixed.

No new actionable findings. Direction remains sound: make the existing clock controllable, preserve its wall mapping, and reuse only the tracing-capture behavior each test actually needs. Keep the API choice as an explicit prerequisite to the implementation.

Verification: static delta and prior-code-context review only. No builds or tests run; this documentation change does not verify either proposed implementation.

(Written by review (OpenAI))

[kixelated]

MERGE (re-review after push; head ac25d88ab58eb1aba8e064630148bb2d313d33b6)

Address-review commit clears the prior blockers: #4687 is cited as SI-debounce Instant only (Clock still std::time::Instant), and the Instant/API surface is spelled for implementers.

Prior findings

1. Fixed — #4687 mis-cite (mux-live-import-clock.md). Decided now says #4687 used the same Instant type only for the private SI debounce after dropping its crate::Clock; Clock itself still reads std::time::Instant. Matches #4687 / current clock.rs on this head.
2. Fixed — “check whether Clock::at” (mux-live-import-clock.md). Plan states Clock::at is re-exported, lists stamp/capture + json/binary Timed<_, Instant> callers (late_clock, moq-cli/hls/audio/video, catalog tests), and opens a maintainer choice: convert at the boundary (recommended) vs typed break on dev.
3. Fixed — helper fit (trace-capture-audit.md). Notes #4690’s helper is WARN/message-specific and needs generalization (or a shared copy); also starts only after #4690 lands with its install-window race closed.

Blocking

None.

Non-blocking

1. CI still pending on this head (Quest queued at review time); no failure signal for a docs-only quest PR.
2. Open choice is the right gate — not a defect. Boundary convert keeps public Instant on std (matches moq_net::Timed docs); implementers should convert in Clock::at and capture/stamp so “ahead of now” / elapsed ride the tokio clock under start_paused.

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

[kixelated]

Merging. Review fixes in ac25d88: the #4687 citation is limited to the private SI debounce, the live import quest lists the public std::time::Instant boundary and gates on one maintainer choice (convert at the boundary on main, recommended, or a typed break on dev), and the tracing audit waits on #4690 and notes its helper is WARN-specific. Settled decisions unchanged.

(Written by Claude Opus 5.5)