chore: merge main into dev

[kixelated]

Merges main into dev. 29 files conflicted textually; several more conflicted semantically and merged clean. Both branches moved during the work, so this also carries a second round of integration against the newer tips.

Summary

Textual conflicts worth review

• js/net/src/origin.ts was a genuine collision: dev put the broadcast routing table there (feat(js/net)!: route publish and consume through Origins, share one connection per relay #2705), main moved the origin id module there from lite/ (feat(js/net): implement the MoQ Cluster extension for moq-transport #2910). The table keeps origin.ts; the id module became the internal js/net/src/hop.ts, shared by both wire protocols as main intended. Nothing new is exported from the package entrypoint.
• js/watch/src/broadcast.ts had opposed intents: dev rejects a whole catalog when a rendition's broadcast reference escapes the root (fix!: reject a catalog whose broadcast reference escapes the root #2630); main hides individual renditions whose broadcast is not announced (fix(js/watch): hide renditions whose broadcast isn't announced #2918). Both are kept. The announcement gate moved into #relativeTarget so playback and rendition selection cannot disagree about what is reachable, and filterCatalog now covers text renditions, which dev added.
• rs/moq-ffi/src/ffi.rs: dev's fix(ffi): release the state when a handle is cancelled #2935 rewrote Task<T> to hold Option<T> and release the state on cancel; main's feat(moq-ffi): compile for wasm32 #2911 had split the same type across native and wasm32. Resolved as dev's semantics under main's layering: T: kio::MaybeSend rather than Send, the two run bodies sharing dev's new drive, and cancel's state-drop routed through the spawn helper so it works on both targets. is_cancelled is native-only, since only the QUIC server distinguishes cancelled from busy.
• rs/moq-ffi/src/session.rs: main's wasm32 browser client alongside dev's reconnecting moq_tokio::Connection. Inner::Connection, MoqBackoff, and the moq_tokio::Status conversion are gated to native.
• js/net/src/ietf/publisher.ts: main's options-object constructor, plus dev's origin-backed broadcasts and main's cluster advert.
• rs/moq-net/src/ietf/subscriber.rs: main's Arrival parameter plus dev's GOAWAY drain cost.

Silent semantic conflicts (merged clean because neither side touched the other's lines)

• test/wasm/'s harness published through Established.publish, which feat(js/net)!: route publish and consume through Origins, share one connection per relay #2705 removed. It now publishes into an Origin and passes publish: origin.consume().
• test/wasm/, rs/justfile, rs/moq-bench/config/hd.toml, and main's new iroh page invoked --server-bind / --server-version / --client-connect, which fix(cli)!: refuse the renamed flags with a migration instead of ignoring them #2915 refuses outright. Reconciled against --help.
• Dev's timeline.rs needed the arc_with_non_send_sync allow the rest of moq-mux already carries, now that main compiles the crate for wasm32.
• doc/concept/layer/iroh.md and the VitePress sidebar pointed at /lib/rs/crate/moq-native, dead since dev's rename. (The sidebar link was already dead on dev; VitePress only validates in-content links.)
• Main's feat(moq-native): size the UDP socket buffers, warning when the kernel clamps #2940 added 215 lines to rs/moq-native/src/bind.rs, which rename detection correctly routed into rs/moq-tokio/src/bind.rs.

wasm32

web-transport-wasm 0.6 implements the poll traits moq-net requires (moq-dev/web-transport#369), so the hand-written adapters in moq-wasm and moq-ffi are deleted; both files are now just the dial (794 lines to 38, and 163 to 25). That is what lets main's moq-ffi wasm32 build (#2911) coexist with dev's poll-only transport (#2736). Pin bumped to "0.6"; web-transport-quinn picks up 0.12.1 under the existing "0.12" pin and web-transport-quiche is unchanged.

Three pre-existing dev bugs, each in its own commit

All three were invisible on dev because the CI that catches them is path-filtered, and this merge touches the paths.

• cpp/obs did not compile against moq.h. The codec enums cross the ABI as uint32_t so that an unknown discriminant from C is an error rather than UB, which leaves no signature reaching them; the ENUMS allowlist in rs/libmoq/build.rs is what forces cbindgen to emit them anyway. moq_video_format and moq_container_format were never added to it, so the header declared neither and every C caller of moq_publish_video / moq_publish_container had to hardcode the discriminants. Nothing compiled a C consumer of libmoq in PR CI until obs.yml (feat(obs): compile the plugin in CI, and type-check it without obs-deps #2867) arrived from main.
• Every python -> * smoke case failed. refactor(bindings)!: split media publishing by kind #2927 split publish_media_stream into publish_video_stream / publish_container_stream and updated py/moq-rs, but not the smoke client calling it, so the publisher died with 'BroadcastProducer' object has no attribute 'publish_media_stream'. smoke.yml runs on a PR only when it touches test/smoke, test/ts, test/justfile, package.json, or itself; refactor(bindings)!: split media publishing by kind #2927 touched none of those.
• The wasm harness failed ietf: reads a published broadcast with old. It subscribes with the default Latency::REAL_TIME budget, which drops a group the live edge has moved past rather than delivering it late. feat(moq-net): enforce the subscriber latency budget #2890 made that enforcement real, so a publisher writing every 50ms can retire a group the reader is still on. Demanding consecutive sequences and treating old as a failure asserted a guarantee real-time delivery does not make. Skips are now accepted and bounded by an attempt cap, so a subscription that never delivers a whole group still fails; order and byte-exactness of a delivered group are unchanged. (Contrast test(mux): read the SI history with a latency budget that keeps it #2948, which hit the same budget in a Rust test that wanted history, where a wider budget was the right answer.)

One test changed deliberately

moq_net::model::resume::consecutive_updates_wake asserted an absolute wake count of 2 and saw 3. The extra wake happens during the poll, not at the update: main's kio::Park now reuses a still-registered waiter (#2905), so applying the subscription change notifies a list the poll is itself parked on and self-wakes once. That costs a redundant poll and nothing else, while a lost wakeup parks the task forever, so the assertion measures the delta now. Reason is recorded in the test.

Public API changes

moq.h gains the moq_video_format and moq_container_format declarations described above. That is additive: the discriminants are unchanged, so an existing caller that hardcoded them keeps working.

Otherwise no new or changed public surface beyond what each parent already shipped. Internal only: js/net/src/hop.ts is a new internal module (not re-exported from the entrypoint); moq_wasm::transport::Session is now a re-export of web_transport_wasm::Session rather than a local newtype (moq-wasm is publish = false); moq_ffi::transport's newtypes are gone in favour of the upstream types.

Wire behavior changes

None. No encoding, message, or framing changed in either direction, so no drafts/ update applies.

Testing

• just rs check, just rs wasm -- pass
• cargo nextest run --workspace -- 3295/3295 (dev's test(mux): read the SI history with a latency budget that keeps it #2948 fixed the moq-mux TS clock test that was failing on dev before this merge)
• bun run --filter='*' check (type-check + docs build) -- pass
• just obs ci -- all 11 targets build and link
• ./smoke/smoke.sh --publishers python --subscribers rust,python -- pass
• ruff + pyright on the changed Python -- clean
• just test wasm -- 9/9 browser cases across all three protocol flavours, which is the real check on the transport swap: compiling says nothing about whether the bindings still open a session
• bun biome check -- 3 errors, all pre-existing and identical on a pristine main checkout (cpp/obs/{CMakePresets,buildspec}.json formatting, doc/.vitepress/theme/index.js import order). Left alone rather than mixed into a merge.

🤖 Generated with Claude Code

https://claude.ai/code/session_017wFQ5wqKbvWwET3G5MJXY9

(Written by Claude Opus 5 (1M context))

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.
To continue using code reviews, add credits to your account and enable them for code reviews in your settings.