[kixelated]

Implements quest/m1/compress/migrate.md, and deletes that quest.

Problem

moq-net now compresses tracks marked compress: true. Even so, first-party compressed data still went out as an uncompressed track plus an application-compressed .z sibling (catalog.json.z, the moq-stats .z tracks). That meant duplicate tracks, duplicate codec state, and a flate library (moq-flate, @moq/flate, and the bindings' flate namespaces) in every stack.

Approach

• One compressed track per name. catalog.json (moq-mux and js/publish), each moq-stats .json track, the <track>.hls record tracks, the CLI's stats.json and feedback report tracks, and the moq-c JSON tracks are created with compress: true. hang::Catalog::default_track_info() sets it.

• Consumers read a compressed track like a plain one: the hang/moq-mux catalog consumers, js/watch, the moq-json and @moq/json decoders, and every binding's JSON consumer lose their compression options.

• moq-stats: TrackPair became one Output per track and keeps merge-patch deltas with refresh(8). Measured over 600 ticks with moq-net's codec (raw DEFLATE level 6, sync flush per frame, fresh window per group), deltas save 35–82% over compressed snapshots:

  track	refresh	compressed snapshot	compressed delta	delta saves
  totals, 1 tier	every 8 frames	148,595 B	44,170 B	70.3%
  totals, 4 tiers	every 8 frames	251,627 B	97,790 B	61.1%
  prefix, 20 children	every 8 frames	313,545 B	155,409 B	50.4%
  prefix, 200 children	every 8 frames	1,050,155 B	678,139 B	35.4%

• Drafts: draft-lcurley-moq-hang drops the compression field and catalog.json.z, and its Compression section now calls compression a track property that the transport carries. draft-lcurley-moq-e2ee drops the .z sibling but keeps compress-before-AEAD.

• Quests: quest/m1/compress/README.md is sized [S] for its remaining end-to-end proof and guide.

Choices made without asking:

• The MSF catalog track stays uncompressed, because third-party moq-transport readers consume it.
• Empty #[non_exhaustive] config structs stay where the Rust API takes one (moq_json::stream::Config, window::ConsumerConfig, moq_stats::consume::Config). Empty FFI records, Go option structs, and C structs are removed instead, along with stream::Decoder::reset.
• The hang binary catalog section stays. Only moq-mux's binary_* producer helpers go.
• catalog::msf::Consumer doesn't reuse hang's escape check. MSF namespaces are absolute, so the check could never fire there.
• Tests that measured compressed frame sizes through a local reader are deleted. Local readers now see inflated frames.

Impact

All breaking; each package's MIGRATE.md has the migration.

• Wire / catalog: catalog.json.z and the moq-stats .z tracks are gone. Readers subscribe to catalog.json and the .json stats tracks, which moq-net compresses. hang's catalog compression field is removed (Rust, JS schema, draft).
• hang: Catalog::COMPRESSED_NAME, TRACK_COMPRESSED, and the compression field are removed. default_track_info() sets compress: true.
• moq-mux: CatalogFormat::HangZ, the moq_mux::binary module, and catalog::Producer::binary_* are removed.
• moq-json / @moq/json: Config::compression, Compression, Desync, and stream::Decoder::reset are removed. @moq/json now depends on @moq/signals.
• moq-stats: Encoding and TrackPair are removed. totals_track, presence_track, and prefix_track take no encoding. consume::Config is empty, and aggregate::Config loses encoding.
• moq-flate, @moq/flate: deleted.
• CLI / @moq/watch: the hangz catalog format is removed.
• moq-ffi: the MoqFlate* objects and MoqJsonStreamConfig are removed.
• moq-c: moq_binary_config and moq_publish_binary_* are removed.
• Python, Kotlin, Swift, Go, Dart: the flate namespace and the JSON compression options are removed. The Dart bindings and the JS API reports are regenerated.

Alternatives

• Snapshot-only moq-stats tracks. Simpler, but compressed deltas are 35–82% smaller (table above), so the deltas stay.
• Compress the MSF catalog too. Rejected because third-party readers consume it.

Checks

• cargo check --workspace --all-targets --all-features, cargo clippy -D warnings, and cargo test --all-features on the touched crates pass. JS biome, build with API reports, tests for the changed packages, and tsc for @moq/json pass. sh/dart/generate.sh, just drafts check, and quest check --base origin/main pass.
• Not run locally: the Swift, Kotlin, Go, Python, and Dart wrapper suites and the interop lanes.

Follow-ups

• Encoder heap on long-lived groups. A compressed group's encoder holds about 256 KiB (flate::ENCODER_HEAP) while the group is open. HLS record and stats tracks keep groups open, so each rendition's few-byte record track now costs 256 KiB. Options: a smaller window or memLevel for small groups, or drop the encoder between writes and re-prime it.
• E2EE and compressed catalogs. default_track_info() now compresses catalog.json, so an encrypted catalog would be compressed as ciphertext, which the e2ee draft forbids. The m2 e2ee quests (quest/m2/net-wrap/e2ee.md, quest/m2/e2ee/rust-publish.md) must compress before AEAD.
• Wire-size visibility. The moq-json telemetry example no longer reports wire bytes, because a local reader sees inflated frames. A moq-net helper or stats hook that exposes a group's stored size would restore it.

🤖 Generated with Claude Code

(Written by Claude Opus 5.5)

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-28T21:01:35.461170Z	313c819	New commits

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" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[kixelated]

Automated review by review (OpenAI)

Reviewed commit: d7e43b4.

No actionable bugs found in the inspected changes. Moving compression into track metadata is a sound simplification: it removes duplicate codec state and sibling tracks while preserving catalog retention/priority, stream single-group failure semantics, and stats demand/refresh handling. Keeping merge-patch deltas and leaving MSF uncompressed are justified distinctions rather than unnecessary parallel implementations.

Verification limits: static review of the changed Rust/JS runtime paths, relevant surrounding code, tests, and C/FFI/wrapper API wiring, including an independent Rust pass. I did not run builds, tests, wrapper suites, or network interoperability checks; this is not a claim that those pass.

[kixelated]

Merged main into the branch (313c819) after the queue dropped it for a merge conflict:

• go/wrapper/README.md and demo/carla/sim/tests/test_runtime.py: main's binding publication wrapper (CreateBroadcast returns a publication; the broadcast is publication.Broadcast()/.broadcast) combined with this PR's compressed TrackInfo and the info=None fake.
• rs/moq-c/MIGRATE.md, rs/moq-mux/MIGRATE.md: kept both Unreleased entries (this PR's compression section above main's #7110 identity section).

just check origin/main passed locally, CI passed, and Codex reviewed the new head without findings. quest check is clean. Landing again.

(Written by Claude Opus 5.5)