From 1290603b9b4d9066faec45c0477d755425f5b66a Mon Sep 17 00:00:00 2001 From: Luke Curley Date: Wed, 28 Oct 2026 21:27:28 -0700 Subject: [PATCH 1/3] quest: plan follow-ups from the 2026-10-28 m0/m1 spawn New quests for the regressions, gaps and diagnostics the spawn's PRs found, and folds of smaller findings into the quests that own them. Co-Authored-By: Claude Opus 5.5 --- quest/m0/ci/README.md | 1 + quest/m0/ci/cargo-alias-strings.md | 25 ++++++++++++++ quest/m1/README.md | 5 +++ quest/m1/baseline-page-logs.md | 21 ++++++++++++ quest/m1/bench-ci.md | 4 +++ quest/m1/c/README.md | 1 + quest/m1/c/api-gaps.md | 29 ++++++++++++++++ quest/m1/drop-set.md | 7 ++++ quest/m1/eager-encode/README.md | 3 ++ quest/m1/idle-poll-exit.md | 32 +++++++++++++++++ quest/m1/ietf-publisher-unknown-range.md | 36 ++++++++++++++++++++ quest/m1/ietf22-fetch-poll-cost.md | 34 ++++++++++++++++++ quest/m1/moxygen-pin-bump.md | 10 +++--- quest/m1/publish-controls.md | 22 ++++++++++++ quest/m1/push-to-talk/README.md | 8 +++++ quest/m1/push-to-talk/audio-support-cache.md | 17 +++++++++ quest/m1/ts-reliability/README.md | 1 + quest/m1/ts-reliability/follow-cancel.md | 21 ++++++++++++ quest/m2/README.md | 3 ++ quest/m2/audio-bufferbloat-av.md | 4 +++ quest/m2/audit-compile-time.md | 8 +++++ quest/m2/audit-size.md | 7 ++++ quest/m2/crossbeam-pin-fence.md | 32 +++++++++++++++++ quest/m2/lint-examples.md | 21 ++++++++++++ quest/m2/malformed-track-status.md | 27 +++++++++++++++ quest/m2/perf/README.md | 2 ++ quest/m2/perf/js-publisher-frame-cost.md | 27 +++++++++++++++ quest/m2/perf/single-delivery-write.md | 25 ++++++++++++++ quest/m2/release/c-retire.md | 5 +++ 29 files changed, 433 insertions(+), 5 deletions(-) create mode 100644 quest/m0/ci/cargo-alias-strings.md create mode 100644 quest/m1/baseline-page-logs.md create mode 100644 quest/m1/c/api-gaps.md create mode 100644 quest/m1/idle-poll-exit.md create mode 100644 quest/m1/ietf-publisher-unknown-range.md create mode 100644 quest/m1/ietf22-fetch-poll-cost.md create mode 100644 quest/m1/publish-controls.md create mode 100644 quest/m1/push-to-talk/audio-support-cache.md create mode 100644 quest/m1/ts-reliability/follow-cancel.md create mode 100644 quest/m2/crossbeam-pin-fence.md create mode 100644 quest/m2/lint-examples.md create mode 100644 quest/m2/malformed-track-status.md create mode 100644 quest/m2/perf/js-publisher-frame-cost.md create mode 100644 quest/m2/perf/single-delivery-write.md diff --git a/quest/m0/ci/README.md b/quest/m0/ci/README.md index c1666404b6d..d1bb60fdefb 100644 --- a/quest/m0/ci/README.md +++ b/quest/m0/ci/README.md @@ -139,4 +139,5 @@ The historical samples locate candidates; they are not controlled benchmarks. - [Padding runs](/quest/m0/ci/quic-padding.md) - moq-quic parses a PADDING run as one frame, keeping the padding stats' meaning - [tracing-test filter](/quest/m0/ci/tracing-filter.md) - moq-tokio's tests stop forcing every crate to format at TRACE - [Stats oversize](/quest/m0/ci/stats-oversize.md) - moq-stats refuses an oversized plain snapshot before parsing it back +- [Cargo aliases in nested checkouts](/quest/m0/ci/cargo-alias-strings.md) - `cargo build-ffi` and `cargo moq` work from a worktree nested inside another checkout - [Out-of-crate targets](/quest/m0/ci/out-of-crate-targets.md) - branch checks format and lint a crate whose target file lives outside it, such as the Rust interop clients diff --git a/quest/m0/ci/cargo-alias-strings.md b/quest/m0/ci/cargo-alias-strings.md new file mode 100644 index 00000000000..8ae24f36ff0 --- /dev/null +++ b/quest/m0/ci/cargo-alias-strings.md @@ -0,0 +1,25 @@ +# [XS] Cargo aliases work in a nested checkout + +## Goal + +`cargo build-ffi` and `cargo moq` run correctly from a worktree nested inside +another moq checkout, such as an agent worktree under +`~/work/moq/.claude/worktrees/`. + +## Plan + +Found 2026-10-28 by [moq-dev/moq#7156](https://github.com/moq-dev/moq/pull/7156): +cargo reads every `.cargo/config.toml` from the working directory up, and +concatenates list values across files instead of overriding. In a nested +worktree both checkouts' list aliases join, so `build-ffi` becomes +`build --locked … build --locked …` and fails with "unexpected argument +'build'". Every C, C++, OBS, Kotlin, Go and Dart check fails locally for +agents. `cargo moq` duplicates the same way, but the copy lands after `--`. + +Decided 2026-10-28 (maintainer): write the aliases in `.cargo/config.toml` as +strings, which override rather than concatenate. Check whether any other list +value there, such as the wasm32 `rustflags`, also duplicates harmfully. Add a +check that runs `cargo build-ffi --help` from a checkout nested inside +another. Agent worktrees stay where they are. + +Public API: none. Wire: none. diff --git a/quest/m1/README.md b/quest/m1/README.md index 4c9a1480afc..79d3ba9784a 100644 --- a/quest/m1/README.md +++ b/quest/m1/README.md @@ -160,6 +160,7 @@ removed). - [SUBSCRIBE honors the properties opt-out](/quest/m1/subscribe-timestamps.md) - the Rust publisher sends untimed objects to a SUBSCRIBE that declines Track Properties, and the Rust subscriber asks for them when it needs timestamps - [JS FETCH timeout test](/quest/m1/js-fetch-timeout-test.md) - a JS joining FETCH still waiting on its SUBSCRIBE at the deadline is refused with TIMEOUT - [JS joins at the current group](/quest/m1/ietf-js-join.md) - a late JS moq-transport subscriber gets the group in progress whole through a joining FETCH or fill, as Rust does +- [Rust marks a lost listed group unknown](/quest/m1/ietf-publisher-unknown-range.md) - the Rust IETF publisher sends End of Unknown Range for a listed group it can't serve, after the publisher's open quests land - [IETF fill timeout](/quest/m1/ietf-fill-timeout.md) - Rust honors native FILL_TIMEOUT to bound upstream waiting on drafts 18 and 22, alongside local content-age enforcement - [IETF takeover](/quest/m1/ietf-takeover.md) - carry the shared takeover contract through the negotiated IETF epoch extension - [Identity and recovery guide](/quest/m1/identity-guide.md) - explain epochs, pinned catalogs, lazy jobs, restart and version compatibility as they exist; takeover adds its own section @@ -168,6 +169,7 @@ removed). - [A/V clock](/quest/m1/av-clock.md) - the audio playhead drives Sync.reference while audio plays, through per-track sync handles - [Container reads drop backfill](/quest/m1/read-backfill.md) - the JS container consumer drops groups below its read cursor like Rust - [Buffered playback subscribes in ascending order](/quest/m1/buffered-ascending.md) - a buffered `` asks for the oldest group first, so a voice reply's opening arrives first, and audio's head hold shrinks to the floor +- [The publish controls stay visible without a preview](/quest/m1/publish-controls.md) - ``'s controls no longer collapse with a live camera and no preview - [Push-to-talk loses no speech](/quest/m1/push-to-talk/README.md) - consumers hold a disabled rendition, and an unmute encodes at once with the mic still open - [Always-on encoding](/quest/m1/eager-encode/README.md) - publishers encode while enabled whatever the demand, so the past exists to fetch; on-demand is opt-in, and audio returns to it through a raw backbuffer - [TS fidelity and recovery](/quest/m1/ts-reliability/README.md) - both modes have explicit profiles, automated validation, loss tests, and a dedicated guide @@ -178,7 +180,10 @@ removed). - [moq-clock's tests pass under cargo test](/quest/m1/clock-test-process.md) - `paused_runtimes_own_their_anchors` stops depending on being first in its process - [kio's log tests pass in a workspace build](/quest/m1/kio-workspace-tests.md) - the eviction tests pass under the workspace's unified features, or eviction is fixed - [The JS stalled-consumer eviction test ages its groups](/quest/m1/js-cache-age-test.md) - the test drives the clock the cache reads, so it can fail +- [moq-transport-22 FETCH cost](/quest/m1/ietf22-fetch-poll-cost.md) - #7042's +12.8% instructions on draft-22 FETCH is fixed or attributed +- [An idle track poll exits cheaply](/quest/m1/idle-poll-exit.md) - a reader's poll on a track with nothing new costs a small constant, not ~3,800 instructions - [Main is no slower than release](/quest/m1/release-perf/README.md) - every benchmark where main is slower than release is fixed or accepted with a reason, checked by a final paired run before the cut +- [The browser baseline keeps its page logs](/quest/m1/baseline-page-logs.md) - a failing wasm-baseline cell is diagnosable from its artifact - [Interop coverage](/quest/m1/interop-coverage/README.md) - the riskiest moq-transport features moq-net exposes get cross-implementation tests over drafts 18 and 22 - [Calibrate the paired nightly benchmarks](/quest/m1/bench-ci.md) - A/A and deliberately slowed Nightly runs set the paired comparison's threshold, so its regression alerts can be trusted - [HLS and DASH segment conformance](/quest/m1/hls-conformance.md) - nightly ffprobe checks that advertised timing, sync flags, and bytes match the media served diff --git a/quest/m1/baseline-page-logs.md b/quest/m1/baseline-page-logs.md new file mode 100644 index 00000000000..6eac49ce5f6 --- /dev/null +++ b/quest/m1/baseline-page-logs.md @@ -0,0 +1,21 @@ +# [XS] The browser baseline keeps its page logs + +## Goal + +`just test wasm-baseline` keeps the publisher and consumer page logs and the +binding's error reason in its artifact, so a failing cell can be diagnosed +from the artifact alone. + +## Plan + +Found 2026-10-28 by [moq-dev/moq#7135](https://github.com/moq-dev/moq/pull/7135): +the harness (`test/justfile`'s `wasm-baseline`) kept neither page's log, and +the binding's error reason was lost, so diagnosing an `evaluate: +MoqError.Internal` failure meant rebuilding that context locally. + +Decided 2026-10-28 (maintainer): m1, since the baseline is a Nightly lane. + +Write each page's console and the error reason beside the existing +diagnostics, and check the Nightly job uploads them. + +Public API: none. Wire: none. diff --git a/quest/m1/bench-ci.md b/quest/m1/bench-ci.md index 9ffea2695bb..f48d36ce9cc 100644 --- a/quest/m1/bench-ci.md +++ b/quest/m1/bench-ci.md @@ -28,6 +28,10 @@ lands; agents never dispatch Nightly. - Check the paired legs' durations against their hour; the base build relies on the mbx cache for the previous night's artifacts, and a cold one roughly doubles each leg's build time. +- Give each revision in a paired comparison its own `CARGO_TARGET_DIR`. Two + source trees sharing one produced stale artifacts while attributing + [moq-dev/moq#7141](https://github.com/moq-dev/moq/pull/7141), such as a bench + linked against the other tree's library (decided 2026-10-28). - Decide whether `bench/run.sh` pins glibc's malloc thresholds for every Criterion target, as [moq-dev/moq#6727](https://github.com/moq-dev/moq/pull/6727) diff --git a/quest/m1/c/README.md b/quest/m1/c/README.md index a8a62969b3d..c5223b8f164 100644 --- a/quest/m1/c/README.md +++ b/quest/m1/c/README.md @@ -32,3 +32,4 @@ Public API: moq-c gains each m1 binding feature. Wire: none. - [moq-c decode start](/quest/m1/c/decode-start.md) - a moq-c decoder can read a finished track from its start - [JSON windows](/quest/m1/c/json-window.md) - moq-c publishes and reads moq-json windows, as moq-ffi does - [Announce route](/quest/m1/c/announce-route.md) - a C publisher announces with a cost and restart epoch through its own struct, and cannot claim relay hops +- [API gaps](/quest/m1/c/api-gaps.md) - one malformed-epoch error code, an audio enabled query, and a relay hops test diff --git a/quest/m1/c/api-gaps.md b/quest/m1/c/api-gaps.md new file mode 100644 index 00000000000..4dccb08f922 --- /dev/null +++ b/quest/m1/c/api-gaps.md @@ -0,0 +1,29 @@ +# [S] moq-c closes its small API gaps + +## Goal + +moq-c reports a malformed epoch the same way everywhere, lets a C consumer see +that audio is disabled as it can for video, and has a test of the hops on a +route received through a relay. + +## Plan + +Found 2026-10-28 by [moq-dev/moq#7151](https://github.com/moq-dev/moq/pull/7151) +and [moq-dev/moq#7156](https://github.com/moq-dev/moq/pull/7156). + +Decided 2026-10-28 (maintainer): one PR for all three. moq-c already breaks +at this cut (0.6 to 0.7), so the error-code change rides that break. + +- `moq_broadcast_request_prepare` reports a malformed epoch as + `MOQ_ERROR_INVALID_CODE`; `moq_publication_announce` uses + `MOQ_ERROR_INVALID_CONFIG`. Use `MOQ_ERROR_INVALID_CONFIG` for both, with an + Impact migration note. +- Add an audio counterpart to `moq_consume_video_enabled`, so a C app can tell + that a held audio rendition is disabled. +- Add a moq-c test that announces through a relay session and checks the + received route's `hops`. + +Update `moq.h`, `rs/moq-c/README.md` and `doc/lib/c` as needed. + +Public API: moq-c gains an audio enabled query and changes one error code. +Wire: none. diff --git a/quest/m1/drop-set.md b/quest/m1/drop-set.md index 32792531040..c2a7bf5b90f 100644 --- a/quest/m1/drop-set.md +++ b/quest/m1/drop-set.md @@ -288,6 +288,13 @@ case needs an update to arrive between START and the first group. Cover it deterministically, through the track hooks rather than wire timing, when the drop runs are reworked here. +### Cancelled fetches hold the end + +Found 2026-10-28 by [moq-dev/moq#7149](https://github.com/moq-dev/moq/pull/7149): +a cancelled fetch below a track's end still holds the end open until the cache +expiry scan removes it, so how fast a subscription ends depends on expiry +timing. Settle it here, where cancelled and expired runs get a reason. + ## Related - [Subscribe ranges](/quest/m1/subscribe-ranges/README.md) - fills the cache miss an evicted run leaves diff --git a/quest/m1/eager-encode/README.md b/quest/m1/eager-encode/README.md index 07f3f1ececf..6119e6b5a09 100644 --- a/quest/m1/eager-encode/README.md +++ b/quest/m1/eager-encode/README.md @@ -40,6 +40,9 @@ The epic's own work, once the children land: watching"), `doc/concept/use-case/contribution.md` (on-demand becomes an option), the moq-c doc comment at `rs/moq-c/src/audio.rs:433`, and the Swift/Python/Go/Dart wrapper docs. +- `@moq/room`'s docs: a joined room publishes through `Publish.Broadcast`, so + it now encodes camera and microphone while nobody watches (found + 2026-10-28 by [moq-dev/moq#7143](https://github.com/moq-dev/moq/pull/7143)). - Rewrite the quests that assume demand-gated capture: [studio replay](/quest/m3/studio/replay.md) and its compose sources ("arm replay = local subscriber"), [iOS capture](/quest/m2/mobile-capture-ios.md), diff --git a/quest/m1/idle-poll-exit.md b/quest/m1/idle-poll-exit.md new file mode 100644 index 00000000000..ca32d748d05 --- /dev/null +++ b/quest/m1/idle-poll-exit.md @@ -0,0 +1,32 @@ +# [M] An idle track poll exits cheaply + +## Goal + +A reader's poll on a track with nothing new costs a small constant number of +instructions, not about 3,800, so catalog and media fanout scale with the +readers that have work rather than every pending reader. + +## Plan + +Found 2026-10-28 by [moq-dev/moq#7140](https://github.com/moq-dev/moq/pull/7140), +which moved the arrival-grace and clock checks behind the cheap "nothing +changed" checks in `Misses::judge`. Even at the old baseline, a pending catalog +poll costs about 3,800 instructions per reader, spread across the cursor's +poll path (`Cursor::probe` in `rs/moq-net/src/model/track.rs` and what it +calls) on a track with nothing new. + +Decided 2026-10-28 (maintainer): m1, for relay fanout scale. + +- Profile one idle poll with instruction counts (callgrind or + `perf stat --no-scale`) and find where the cost goes. +- Add one early exit the whole path shares, keyed on something every change + already bumps (for example a track revision the cursor last saw), rather + than a short-cut per function. +- Measure the moq-mux catalog fanout benches and the moq-net session benches + over tracks and readers, before and after. + +Public API: none. Wire: none. + +## Related + +- [Mesh session cost](/quest/m2/perf/mesh-session-cost.md) - per-session costs in the same model diff --git a/quest/m1/ietf-publisher-unknown-range.md b/quest/m1/ietf-publisher-unknown-range.md new file mode 100644 index 00000000000..b4956f05d6f --- /dev/null +++ b/quest/m1/ietf-publisher-unknown-range.md @@ -0,0 +1,36 @@ +# [S] The Rust IETF publisher marks a lost listed group unknown + +## Goal + +When the Rust moq-transport publisher has listed a group in a FETCH answer or +a draft-22 fill but can no longer serve it, it sends End of Unknown Range +(`0x10C`) for that group on drafts 18 and 22, as the JS publisher does, instead +of skipping it silently. + +## Plan + +Found 2026-10-28 by [moq-dev/moq#7150](https://github.com/moq-dev/moq/pull/7150), +which fixed the JS publisher. In `rs/moq-net/src/ietf/publisher.rs` Rust skips +such a group in two places: the FETCH answer loop in `serve_fetch_stream`, on +`NotFound | NotFetchable | Unsupported | Old`, and `run_fill`, when +`fetch_group` fails with the same errors. + +Rust can refetch from upstream, so a skip there means a failed refetch rather +than only an eviction. `NotFound` and `Old` map to `0x10C`. Decide in the PR +whether `Unsupported` and `NotFetchable` mean "unknown" too, or need their own +handling, and record the reason. + +Decided 2026-10-28 (maintainer): it starts after the three quests below land, +since each edits the same publisher. + +Tests: a listed group evicted before it is written, for FETCH on drafts 18 +and 22 in both group orders and for a draft-22 fill, each failing before the +fix. Run `just test interop --all` if the framing order changes. + +Public API: none. Wire: none; the marker exists in both drafts. + +## Required + +- [Rust FETCH parity](/quest/m1/ietf-fetch-parity.md) - edits the same FETCH answer path +- [IETF fill timeout](/quest/m1/ietf-fill-timeout.md) - edits the same fill path +- [SUBSCRIBE honors the properties opt-out](/quest/m1/subscribe-timestamps.md) - edits the same publisher diff --git a/quest/m1/ietf22-fetch-poll-cost.md b/quest/m1/ietf22-fetch-poll-cost.md new file mode 100644 index 00000000000..6adfd302ed5 --- /dev/null +++ b/quest/m1/ietf22-fetch-poll-cost.md @@ -0,0 +1,34 @@ +# [M] moq-transport-22 FETCH costs what it did before the waiter refactor + +## Goal + +A moq-transport-22 FETCH of 256-frame groups costs no more instructions than +it did before [moq-dev/moq#7042](https://github.com/moq-dev/moq/pull/7042), or +the remaining cost is attributed to work the refactor needs and recorded in +its PR. Other versions and group sizes don't regress. + +## Plan + +Found 2026-10-28 by the paired-relay measurements in +[moq-dev/moq#7141](https://github.com/moq-dev/moq/pull/7141): #7042 +(`6eff65506e`, "transport polls take kio waiters through &self") adds +12.8% +instructions on moq-transport-22 FETCH of 256-frame groups (8.02M to 9.04M per +round). The commit before it measures the same as the head. It landed after the +range the 2026-10-28 Nightly compared, so the next paired Nightly may flag it. + +Decided 2026-10-28 (maintainer): m1, not a Nightly blocker. + +- Attribute with instruction counts (`perf stat --no-scale` pinned to one core + type, or callgrind), not wall time: the hosted runners and a busy host are + too noisy for a 13% delta. Give each revision its own `CARGO_TARGET_DIR`. +- Find which poll path the refactor made more expensive on draft 22's FETCH + (per-call waiter registration, extra polls, or a lost fast path), and fix it + at the cause without reverting the `&self` waiter API. +- Sweep the relay session bench's FETCH cases across lite-07, draft 18 and + draft 22, at 4- and 256-frame groups, and record before and after. + +Public API: none. Wire: none. + +## Related + +- [Calibrate the paired nightly benchmarks](/quest/m1/bench-ci.md) - owns the gate's threshold if the Nightly flags this first diff --git a/quest/m1/moxygen-pin-bump.md b/quest/m1/moxygen-pin-bump.md index f229aae2654..6165b0c1c6a 100644 --- a/quest/m1/moxygen-pin-bump.md +++ b/quest/m1/moxygen-pin-bump.md @@ -8,14 +8,14 @@ merges, `nix/moxygen.nix` pins a Moxygen that includes it, `nix/moxygen-track-status.patch` is deleted, and the Moxygen interop and conformance lanes still pass. -Check moxygen#264's state. Delete this quest once the patch is gone, or if -moxygen#264 lands before the patch reaches `main`. +Check moxygen#264's state. Delete this quest once the patch is gone. ## Plan -The patch is not on `main` yet: -[moq-dev/moq#6729](https://github.com/moq-dev/moq/pull/6729), still open, adds -it to `nix/moxygen.nix`'s `patches` and carries it until upstream lands it. Upstream's client has moved to draft-20 Location +The patch is on `main`: +[moq-dev/moq#6729](https://github.com/moq-dev/moq/pull/6729) (merged +2026-10-25) added it to `nix/moxygen.nix`'s `patches`, carried until upstream +lands it. Upstream's client has moved to draft-20 Location Filters, so the profile patch the pin also applies needs rebasing in the same bump. Run `just test interop --all` and the Moxygen smoke lane. diff --git a/quest/m1/publish-controls.md b/quest/m1/publish-controls.md new file mode 100644 index 00000000000..d395e66373d --- /dev/null +++ b/quest/m1/publish-controls.md @@ -0,0 +1,22 @@ +# [XS] The publish controls stay visible without a preview + +## Goal + +``'s controls stay visible when a camera is live but the element +shows no `