diff --git a/quest/m1/README.md b/quest/m1/README.md index 8ae524436b6..571e7cb90ba 100644 --- a/quest/m1/README.md +++ b/quest/m1/README.md @@ -108,6 +108,7 @@ lands before the cut. - [C++ and OBS package proof](/quest/m1/cpp-package-check.md) - the installed artifacts pass before their first release +- [Playwright browsers from Nix](/quest/m1/playwright-nix.md) - every Playwright check runs on Nix's Chromium locally and in CI, with npm pinned to match - [moq-net compiles once](/quest/m1/net-erase-transport.md) - the session drivers stop monomorphizing per transport and per crate, saving several MiB of the relay, FFI library, `libmoq` and CLI diff --git a/quest/m1/compress/migrate.md b/quest/m1/compress/migrate.md index 0f69d2576fb..1cb2d576385 100644 --- a/quest/m1/compress/migrate.md +++ b/quest/m1/compress/migrate.md @@ -21,6 +21,11 @@ See the [epic](/quest/m1/compress/README.md) for the decisions. `rs/hang/src/timeline.rs` set `compress` on their one track. - Consumers: `rs/moq-mux/src/catalog/consumer.rs`, `rs/moq-cli` (`subscribe.rs`, `archive.rs`), and `js/watch` lose their hangz paths. + moq-mux's hang catalog adapter over `hang::catalog::Consumer` + ([#6114](https://github.com/moq-dev/moq/pull/6114)) drops its `format` + field and `CatalogFormat::HangZ` goes away; check whether + `catalog::msf::Consumer` can reuse hang's escaping-path check instead of + its own (decided 2026-10-20 instead of a separate quest). - moq-stats deltas: the `.z` side uses merge-patch deltas and the plain side does not. Keep deltas on the single track unless measuring stats bytes with and without them on a compressed track shows the saving is small; then diff --git a/quest/m1/net-erase-transport.md b/quest/m1/net-erase-transport.md index c66f52879df..cdc800d255b 100644 --- a/quest/m1/net-erase-transport.md +++ b/quest/m1/net-erase-transport.md @@ -4,8 +4,9 @@ `moq-net`'s session drivers compile once per binary, not once per transport session type and per crate that drives them. The relay, `libmoq_ffi.so`, -`libmoq`, the CLI and `@moq/wasm` shrink by the duplicated copies, and the -relay fan-out benchmark shows no regression beyond its noise. +`libmoq`, the CLI and `@moq/wasm` shrink by the duplicated copies, they +compile faster for the same reason, and the relay fan-out benchmark shows no +regression beyond its noise. ## Plan @@ -34,3 +35,12 @@ session benchmark, before and after. The async bodies must end up non-generic inside `moq-net`, or each awaiting crate still instantiates them. Re-measure with the audit's method: `nm` symbol attribution by crate path and by instantiating crate on an unstripped `profiling` build. + +Compile time (2026-10-20 compile-time audit, +[#6106](https://github.com/moq-dev/moq/pull/6106)): moq-tokio's dev build is +1,661,503 lines of LLVM IR, 43.2% of it moq-net session code instantiated per +transport and again under `ControlStreamAdapter`. The PR reports +`cargo llvm-lines` for moq-tokio and moq-relay before and after, which is +deterministic, and a clean build of both crates before and after, run back to +back and reported as a ratio with the load average. Fewer IR lines alone do not +prove a faster build if the work moves to another compiler phase. diff --git a/quest/m1/playwright-nix.md b/quest/m1/playwright-nix.md new file mode 100644 index 00000000000..2e37fd2feab --- /dev/null +++ b/quest/m1/playwright-nix.md @@ -0,0 +1,57 @@ +# [M] Playwright browsers from Nix + +## Goal + +Every Playwright check runs in the root devShell, locally on NixOS and in CI, +against the Chromium build that Nix provides. No check needs a hand-set +`LD_LIBRARY_PATH` or a `playwright install`, and no contributor doc tells +anyone to run one. + +Inventory on `main` (2026-10-21). Recheck with `git grep -n -i chromium` +outside `quest/` and `test/smoke/`, keeping the hits that install, download, or +say where Chromium comes from (comments included), and with every +`package.json` that depends on `playwright`: + +- Packages: `test/interop/clients/js`, `test/video`, `test/p2p`, `test/wasm`, + `test/audio-quality/clients/js`, `bench/browser`. +- Scripts: `test/interop/interop.sh`, `test/portrait/run.sh` (run in CI as + `just test portrait`), `test/wasm/run.sh`, + `test/audio-quality/{run,record}.sh`, `bench/browser/run.sh`, and the + `test/justfile` recipes. +- Workflows: `interop.yml`, `video.yml`, `wasm.yml`, `nightly.yml`, including + their comments that the devShell lacks Chromium. +- Docs and comments: `test/interop/README.md` (says the devShell lacks + Chromium), `test/wasm/README.md` (says to run + `bunx playwright install chromium`), `rs/justfile`'s browser clock recipe + (says `wasm.yml` installs it), `js/justfile`'s fairness sweep ("requires + installed Chromium"), and `test/video/renderer.browser.ts` (says + `just test video` installs it). + +Out of scope: the Firefox player row, whose `media-firefox` job in +`nightly.yml` keeps installing its engines with `--with-deps`, and +`test/smoke`, which keeps its own flake on purpose. + +## Plan + +Decided 2026-10-20 with the maintainer: + +- **Nix browsers, with npm pinned to match.** Today each package pins npm + `playwright ^1.64`, and the Chromium it downloads cannot load + `libglib-2.0.so.0` on NixOS (found while landing the browser rename check). + The root devShell exports `PLAYWRIGHT_BROWSERS_PATH` from + `playwright-driver.browsers`, `PLAYWRIGHT_VERSION` from + `playwright-driver.version`, and `PLAYWRIGHT_SKIP_VALIDATE_HOST_REQUIREMENTS`, + as `test/smoke/flake.nix` already does; without the last, Playwright's host + check rejects the Nix browser on NixOS. Every + package pins npm `playwright` to exactly that version, even when this + downgrades it (1.64 to nixpkgs' 1.63 today). The alternatives were rejected: + supplying host libraries to a downloaded Chromium (nix-ld or an FHS env) + depends on the host's library compatibility, and building playwright 1.64 + in the repo's overlay is maintenance for a gap nixpkgs closes on its own. +- **CI switches too.** Every workflow, script, and recipe in the inventory + drops its Chromium install, so CI and local runs share one Chromium. +- **The pin is enforced.** A check in `just check` fails when any package's + npm `playwright` differs from `PLAYWRIGHT_VERSION`, like + `test/smoke/freshness.sh`, so a nixpkgs bump cannot silently drift. +- The docs and comments in the inventory say the devShell provides Chromium. + No new user-facing docs: this changes only contributor tooling. diff --git a/quest/m1/time/crates.md b/quest/m1/time/crates.md index c84e16ca454..da2c6197b07 100644 --- a/quest/m1/time/crates.md +++ b/quest/m1/time/crates.md @@ -42,6 +42,10 @@ rest of the workspace and public `SystemTime` fields still need migration. In pa SRT and WebRTC hand instants to third-party APIs on std's clock, and audio/video device threads block outside an async runtime; those boundaries still need clock support rather than a second host-clock read at each call site. +moq-srt's `a_late_payload_keeps_its_past_host_instant` compares a direct +`StdInstant::now()` with `host_instant(moq_net::time::now())`, so it can flake +under load; the SRT boundary migration rewrites it on controlled time +(decided 2026-10-20 instead of a separate flake quest). Public API: breaking wherever a crate exposes an instant or a time source. Wire: none.