Merge main into dev

[kixelated]

Merges the 11 commits main gained since #2852. 27 conflicts; the non-mechanical decisions are below.

Catalog broadcast references

Two independent changes, both kept:

• main fix(path): resolve catalog references like URLs #2855 changed the resolution algorithm to relative-URL semantics: a non-empty reference replaces the base's last segment, then ./.. apply.
• dev fix!: reject a catalog whose broadcast reference escapes the root #2630 changed escape handling to reject the whole catalog.

main already ships both resolve (clamping) and try_resolve (Option), so its path.rs is taken wholesale and dev's callers point at try_resolve. main itself had zero callers of the clamping resolve, so this moves no behavior.

This shifts what counts as escaping by one segment, which is why a number of tests move from ../x to ./x. From a two-segment base like a/pub, .. now lands on the root and ../.. escapes.

The escape policy is deliberately two-sided, not one choice:

• Publisher side (main): Source::retain_valid / resolve_reference drop the one bad rendition. 9 call sites across moq-mux, moq-hls, moq-rtc.
• Consumer side (dev): Error::EscapingBroadcast rejects the whole catalog, which is what the hang draft now mandates.

main's duplicate InvalidBroadcastReference variant is dropped for dev's better-documented EscapingBroadcast. Source grows a request/try_request pair mirroring resolve/try_resolve, so exporters keep skipping one bad rendition while consumers keep reporting the fault.

js/watch is a pure consumer, so it takes the reject policy alone; main's filterCatalog is dropped. Its findEscaping had to move to Path.tryResolve — main's resolve clamps and never returns undefined, so leaving it would have compiled, type-checked, and silently disabled the containment check.

Connecting removal

main #2856 deletes the mechanism; dev had rewritten the same thing into its poll driver (SubscriberDriver, AnnouncePrefix, 27 references). main's removal wins, applied on top of dev's driver: the per-prefix producer, the initial_count/initial_remaining bookkeeping that existed only to release it, and the client.rs wait sites all go, keeping dev's goaway plumbing.

Fixes carried across a rewrite

Two main fixes landed in code dev had deleted, so they were reapplied by hand rather than lost:

• fix(moq-net): serve an IETF subscribe from the live edge #2862 live-edge cursor fix moved from the async run_track into dev's poll-based TrackServe::new. Same lifecycle point: after SUBSCRIBE_OK, once per subscribe, before any group is consumed. Its regression test now drives the poll path.
• fix(relay): redial reconfigured cluster peers #2874 target-carrying DialSources absorbed dev's mDNS source, with run_mdns moved onto upsert/release (dev's 2-arg release was superseded). mDNS gains the fallback-to-another-source behavior the other discovery sources already had.

main #2841's WebSocket SessionInputs refactor gains dev's shutdown field.

Docs

The hang draft and /doc state main's URL resolution semantics and dev's reject-the-catalog rule.

Follow-up fix in this PR

retain_valid came from main, which has no text catalog section — text is dev's captions work. The merge left the publisher-side filter covering video and audio only, while the consumer side already covered all three: exactly the drift dev's comment warns about. Latent today (no exporter consumes text renditions, and the consumer rejects such a catalog anyway), fixed with a regression test that fails without it.

Verification

just check and just test pass: 3186 tests, 0 failures.

Adversarially reviewed by Codex against the mechanical auto-merge as the base, so the diff under review was purely the human conflict resolutions rather than main's already-reviewed commits. Verdict: approve, no material findings. The text gap above came from a separate manual pass.

(written by Opus 5)

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 4ce12d5d9a

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Preserve rendition jitter outside the decoder identity

When a rendition declares jitter or only framerate, identity.decoder contains neither field, but DecoderTrack still calls renditionJitter(props.config). It therefore always publishes undefined for video jitter, so video-only streams with grouped or delayed frames lose the catalog-derived buffer allowance and can repeatedly underrun. Keep the reload identity narrow, but pass or derive the full rendition's jitter metadata separately.

Useful? React with 👍 / 👎.

[chatgpt-codex-connector]

Point catalog consumers at the rejecting resolver

When an external consumer follows this broadcast field's adjacent Path::resolve documentation, excessive .. segments no longer produce no broadcast: this commit changed resolve to clamp and introduced try_resolve for containment checks. That can silently select an unrelated root broadcast instead of rejecting the catalog, so both VideoConfig and AudioConfig should direct consumers to Path::try_resolve.

AGENTS.md reference: AGENTS.md:L97-L99

Useful? React with 👍 / 👎.