Repository navigation
feat!: socket endpoints leave every package for Kixel - #9029
Conversation
Quest-Claim: 003da3a7b62bc08b
Deletes moq_net::socket, @moq/net's Socket, the moq-ffi and moq-c socket handles, the Python, Swift, Kotlin, Dart, and Go wrappers, the interop sockets lane, and the concept page. What existed only for sockets goes with them (origin hidden_withheld and its pre-lite-07 session hold, the JS hiddenWithheld hook). reserve_unlisted_track becomes public, and track::Producer::abort_handle adds a non-owning AbortHandle, both for Kixel's port. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
|
Maintainer decision (2026-11-17, asked in the
(Written by Claude Opus 5.5 with high) |
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
|
Applied the maintainer decision on 4c5d6b7: all three answers are recorded under (Written by Claude Opus 5.5 with low) |
|
Warning Review limit reachedYou've used all free OSS reviews for now. Wait for the free limit to reset to keep reviewing this public repository. Next included review available in 13 minutes. View limit details
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. |
There was a problem hiding this comment.
ℹ️ No critical issues. One cross-package gap is noted inline.
Approach: sound. The PR deletes sockets everywhere, removes the hidden-withholding machinery that only socket dials read, and keeps a small, weak-only track::AbortHandle plus a public reserve_unlisted_track for Kixel's port.
Reviewed changes
This run read the full diff. It focused on the non-deletion hunks and used git grep to check that no removed symbol is still referenced.
track::AbortHandle: holds aWeak<Alive>, so it never keeps a track open. Itsabortunlists the track and then runscommit_abort, which matchesProducer::abort. The new listing tests cover the unlist, freeing the name, and the dropped-track case.broadcast::Producer::reserve_unlisted_track: nowpub. Its doc and the newdoc/lib/rs/moq-net.mdentry say unlisted controls discovery, not authorization.- Hidden withholding removed:
HiddenWithheld/HiddenWithholding, the lite session'swithholdingguard, andVersion::has_hiddenare gone in Rust, along withwithholdHidden/hiddenWithheld/hasHiddenand the origin'swithheldcounter in JS. Nothing else read them. - Bindings: moq-ffi
TrackPhase::Open.familyand moq-cFamily::Open.keptare no longerOption, since only sockets used the empty variant. The Go bridge hooks are gone, andtrackInfoFFIis still used bypublish.go. quic_simport: the bulk-transfer pacing test now sends a plain announced broadcast through the relay instead of a socket session.- Leftovers:
git grepfinds no reference to the removed Rust, JS, or binding symbols, interop examples, or bench entries outside historical quest prose and theMIGRATE.mdentries the cut will drop.
claude-opus-5-5 | 𝕏
# Conflicts: # quest/m2/README.md
There was a problem hiding this comment.
✅ No new issues found.
Reviewed changes
This run reviewed the one commit since the last Pullfrog review: a merge of origin/main into the branch.
- Merged
main: the only change to the PR's own diff is how the merge resolvedquest/m2/README.md.mainhad removedSession codesfrom themoq-net APIline, and the PR's link to #9029 is kept. No other file in the PR's diff changed. - Checked the merged code for leftovers:
git grepfinds no reference to the removed symbols (has_hidden,HiddenWithheld,hiddenWithheld,withholdHidden,hasHidden,moq_net::socket,from_socket,MoqSocket) in the code merged frommain. That includeslite/session.rs,model/track.rs, andfuzz.rs, which the merge and this PR both touched.
The earlier finding about JS hooks.reserveUnlistedTrack is settled by decision 4: Kixel ports only Rust, so the JS hook stays internal.
claude-opus-5-5 | 𝕏

Socket endpoints leave this repository for Kixel, as the sockets epic decided on 2026-11-16. No package this repository publishes has a socket endpoint after this PR.
Kixel ports from this PR's parent commit,
2dcb5f8994, which still hasrs/moq-net/src/model/socket.rs.Summary
Deleted:
moq_net::socket(model/socket.rs, 1,150 lines), its integration tests, and moq-relay'ssocket_authtests.@moq/net'sSocket(socket.ts, its tests, its bench), the export, and the API report entry.socket.rsandMoqSocketEndpoint, moq-c'smoq_socket_*, and the Python, Swift, Kotlin, Dart, and Go wrappers, with their tests and README sections.test/interop/socket.sh, the--socketslane,run_sockets, the JS sockets plan, four clients, and theinterop-socketandinterop-socket-jsexamples.doc/concept/socket.md, its sidebar and index entries, and socket sections indoc/lib/{rs,js,py},py/moq-rs/docs,bench/README.md, andrs/moq-net/CHANGELOG.md.socketbench insh/gh/nightly.sh.Removed because nothing else used them:
origin::Consumer::hidden_withheld,HiddenWithheld/HiddenWithholding, the hold a pre-lite-07 client session took on them,lite::Version::has_hidden,broadcast::Producer::poll_closed,TrackWeak::abort, and thewithheldfield of the fuzzFootprint.hooks.hiddenWithheld,hooks.withholdHidden, the origin'swithheldcounter, andhasHidden().hooks.requestWantedstays, sinceutil/demand.tsand the lite subscriber use it.MoqTrackProducer::from_socket. A track'sfamilyis now a plainChildren.Consume::socket_read,Publish::socket_track/socket_close,Children::unkept, andChildren::counts.bridge.OriginProducer,WrapTrackProducer,WrapTrackConsumer, andTrackInfohooks.Added for Kixel's port:
broadcast::Producer::reserve_unlisted_trackis nowpub, as decided on 2026-11-16. Its doc says unlisted controls discovery, not authorization. A new test shows an accepted reservation can be read by name and never appears in a track listing.track::Producer::abort_handle()returns atrack::AbortHandlewithabort(err)andis_closed(). It holds only astd::sync::Weakto the track'sAliveguard, so it never keeps a track open. Itsabortunlists the track first, asProducer::abortdoes. Tests pin that it unlists a listed track and frees its name, and that a held handle keeps no dropped track alive. To support it,kio::Producergainsis_closed().TrackWeakstayspub(super).Ported rather than deleted: moq-relay's
quic_simbulk-transfer test (a_socket_session_carries_a_bulk_transfer_losslessly) tests relay pacing, not sockets. It is nowa_group_carries_a_bulk_transfer_losslessly: a plain broadcast's 40 MB group crosses a relay to a slow reader without loss. The lite-07-onlyserve_versionshelper went with the socket version of the test.Impact
Public API: breaking, but only for API that exists on
mainalone.origin/releasenever had sockets, andjust api-diffagainst the published releases shows no socket item anywhere. Migration: none for released users. Amainuser ofmoq_net::socket,@moq/net'sSocket, or a binding's socket handle must move to Kixel's copy.Additive:
broadcast::Producer::reserve_unlisted_track,track::Producer::abort_handle,track::AbortHandle, andkio::Producer::is_closed.Behavior: a pre-lite-07 moq-lite client session no longer registers with its publish origin while it withholds hidden broadcasts. Only socket dials read that registration. Hidden broadcasts themselves are unchanged.
Wire: none. Sockets were only a path and track naming convention.
For the release cut: the socket entries in these
MIGRATE.mdfiles describe API that never shipped, so the cut should drop them rather than carry them into a version section:rs/moq-ffi,rs/moq-net("Socket readers start at the beginning"; keep its IETF explicit-start paragraph),js/net,py/moq-rs,kt/moq,swift,dart/moq,go/wrapper,c/moq, andcpp/moq. Several of those sections also describe the non-socketGroupProducer.ready(size)change, which stays.Decisions
Recorded from the maintainer decision of 2026-11-17:
track::AbortHandle(abort,is_closed) is Kixel's cleanup handle (rejected: publishingTrackWeak, whose methods are the broadcast cache's own invariants; no handle).TrackWeakstays private; a method such asis_wantedis added only when a consumer needs it.MIGRATE.mdstay for the release cut to drop, as listed in Impact (rejected: a deletion exemption insh/migrate.sh). The quest asked to delete them, butsh/migrate.sh(moq-dev/moq#8424) failsjust checkon anyMIGRATE.mdedit outside the cut.quest/m0/test-flakes-3/py-client-body-error.md(rejected: drop it), recorded in its Plan.reserve_unlisted_trackis public in Rust only; JShooks.reserveUnlistedTrackstays internal (rejected: publishing it in JS too). The sockets epic's plan of 2026-11-16 has Kixel port only the Rust endpoint and no JS, so no JS consumer exists. Answers Pullfrog's finding.Choices made without asking
Validation
just checkon0a10b1c70d: passes. An earlier run failed in three places. TheMIGRATE.mdedits and the nightlysocketbench entry are fixed. The third was two load-dependent flakes in tests this PR does not touch: moq-cserver_accepts_a_sessionpassed 5 of 5 alone, is known to fail onmain, and is owned byquest/m0/test-flakes-3/c-server-accepts-session.mdfrom moq-dev/moq#8964. Pythontest_client_context_keeps_the_body_errorpassed 6 of 6 alone and gets a new follow-up below.just test interop --all: all checks pass, 204 PASS and 1 SKIP. The skip is the existing Chromium route-key case.cargo test -p moq-relay --features test-support --test integration quic_sim:a_group_carries_a_bulk_transfer_losslessly,one_relay, andtwo_relayspass.cargo test -p moq-net --lib listing::: 35 pass, includingan_unlisted_track_is_read_by_name_but_never_listed,an_abort_handle_unlists_the_track_it_ends, andan_abort_handle_keeps_no_dropped_track_alive. As an independent check, the unlist test only passes becauseAbortHandle::abortunlists first: without that,-aarrives only when the producer drops.cargo clippy -p moq-net -p kio -p moq-relay --all-targets --all-features: clean after removing the dead code above.bun run checkandbun run testinjs/net(3347 pass), and the API report was regenerated.just api-diff: no socket items against any published release.quest review --harness codex(gpt-6.1-sol high), 0 findings.just check's own checks, andgo vet, since the generatedmoq.dev/moq-ffimodule was unavailable.moq.hand the Dart bindings are generated and untracked, so the next generation drops their socket entries.Follow-ups
just check,test_client_context_keeps_the_body_errorgetsMoqError.Cancelledinstead of the body's error. Kept at the pick; no open decisions.Related: moq-dev/moq#8941 renames
rs/moq-ctors/libmoq. Whichever lands second resolves the delete/rename conflict onrs/moq-c/src/socket.rs.Models
Size: M
Implementer: Claude, claude-opus-5-5, unknown effort (arm A)
Implementer: Claude Code, claude-opus-5-5, low effort (iterate, S, arm A)
Implementer: Claude Code, claude-opus-5-5, low effort (iterate, XS, arm A)
(Written by Claude Opus 5.5)
🤖 Generated with Claude Code
Review
Rewritten by
quest reviewafter each round.Head: 128fce9
Base: origin/main at 9a4ed22
Risk: high,
Removing socket APIs across language bindings can break downstream callers until the removals are reverted or callers migrate.Classified: medium by default
Reviewer: Codex, gpt-6.1-sol, high effort
Session:
01a1eb30-9094-7b22-af9f-4ff0e8cac9b4, round 3Score: 10/10,
The branch’s patch is unchanged apart from a correctly resolved planning-document merge, with no actionable findings; builds and tests were not run, as instructed.Raised the risk to high:
Removing socket APIs across language bindings can break downstream callers until the removals are reverted or callers migrate.Rounds: 3; first round 0; findings 0: fixed 0, refuted 0, open 0
Findings: none