[kixelated]

Problem

quest/m0/ci/quic-padding.md, found by moq-dev/moq#6987's profile: moq-quic's frame::Iter yielded one Frame::Padding per PADDING byte. A 1200-byte padded Initial became about a thousand iterator steps, each moving a Frame through Option<Result<..>> and the receive loop. In debug builds those moves dominate packet receive.

Change

• frame::Iter consumes a whole run of PADDING (consecutive 0x00 bytes) in one step and yields a single Frame::Padding. next_with_size reports the run's full length. The remaining per-byte work is one zero check per byte.
• QMux draft-00 still allows only one frame per message: the mux counts a decoded PADDING run as one frame per byte, so a message of two PADDING bytes is still a FRAME_ENCODING_ERROR. Without this, the collapse would have silently started accepting it.
• wire_decode bench: a padded case (below).
• Tests: an iterator test for runs before, between, and after other frames, and a QMux draft-00 test that one PADDING byte is accepted and two are refused.

Padding stats

No stat counts PADDING. FrameStats::record ignores Frame::Padding (and every receive loop skips it before recording), so there is no field whose meaning could change. UDP byte counts are taken from datagrams, not frames. Nothing to keep or migrate.

Measurement

Debug build of the moq-quic lib tests, tests::characterize (handshake-heavy), single-threaded, frame pointers, perf record:

	Total cycles	Connection::handle_packet (inclusive)	frame::Iter (inclusive)	memmove (self)
main	2.47 G	0.95 G (38.4%)	~11.6%	8.0%
this PR	2.29 G	0.64 G (27.8%)	~1.4%	3.5%

Packet receive CPU falls by a third. User CPU of that test group: 0.60 to 0.63 s on main, 0.53 to 0.55 s here (3 runs each). The whole moq-quic lib suite: 10.8 to 11.2 s on main, 10.6 to 10.7 s here, since most of it is not handshakes.

It matters in release too. The wire_decode bench gains a padded case: one small DATAGRAM frame with PADDING filling the rest of a 1360-byte payload, as a padded Initial or MTU probe arrives.

wire_decode/padded	main	this PR
packets_64	712 µs	24.1 µs
packets_1024	12.70 ms	393 µs

That is about 11 µs of decode per padded packet on main, against about 0.4 µs now. For comparison, frames_1 (one DATAGRAM frame filling the packet) measures 12.7 µs and 212 µs.

Impact

• Public API: none. Frame and frame::Iter are pub(crate).
• Wire: none. Every input decodes to the same frames apart from PADDING grouping, and QMux draft-00 keeps refusing multi-frame messages.
• Stats: unchanged (see above).

Choices made without asking

• Kept QMux draft-00's one-frame-per-message rule exact for PADDING runs (fail loud on malformed input) instead of letting a multi-byte PADDING message through. It costs one match in the mux receive loop.
• Extended the existing wire_decode bench with a padded shape rather than adding a new bench target.
• The run is found with a plain position scan. It still touches every byte (about 0.13 ns a byte in release); a chunked zero check would be faster but is not worth the extra code at these numbers.

Follow-ups

None found.

(Written by Claude Opus 5.5)

🤖 Generated with Claude Code

[coderabbitai]

Warning

Review limit reached

You'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 23 minutes.

Check out review usage here.

View limit details

Limit details: You’ve used all 4 included reviews currently available.

Learn how review limits work.

Review configuration:

Run configuration

• Source mode: Auto
• Repository: Unavailable
• Branch: Unavailable
• File: No configuration file from this repository applied
• Commit: Unavailable
• Layers: Organization UI
• Review profile: CHILL
• Plan: Advanced
• Run ID: b3cba703-7010-4422-886f-1fb8fcec7d23

Commits

Reviewing files that changed from the base of the PR and between 66bdbb9 and 5fd50de.

Files selected for processing: 6

• quest/m0/ci/README.md
• quest/m0/ci/quic-padding.md
• rs/moq-quic/benches/wire.rs
• rs/moq-quic/src/connection/mux/mod.rs
• rs/moq-quic/src/connection/mux/tests.rs
• rs/moq-quic/src/frame.rs

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

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.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[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	🔄 Running since 2026-10-29T15:26:46.463637Z	5fd50de	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: aeff48a

No actionable bugs found in the changed code and relevant receive paths. Collapsing zero-byte runs is a focused optimization: the iterator preserves consumed-byte accounting, QUIC receive paths already skip padding, and QMux draft-00 still rejects multiple padding frames. The plain scan is simpler than chunked/vectorized parsing and avoids unnecessary complexity. The new iterator and draft-00 regression tests address the main semantic risks; the padded benchmark covers the intended workload.

Verification limits: static GitHub review only; I did not run tests or benchmarks or independently verify the reported performance numbers.

[kixelated]

Merge summary for head aeff48a547a6f408c491be1e5b637b07cc0644c0:

• Reviews: the automated OpenAI review on this head found nothing actionable, and Codex finished with a thumbs up (no findings). CodeRabbit was rate-limited, so its review doesn't count.
• CI: all 12 checks pass. The merge state is clean, with no conflicts against main.
• quest check --base origin/main passes. This PR deletes quest/m0/ci/quic-padding.md and its link in quest/m0/ci/README.md.
• No changes were made during merge and no decisions were left open. Landing through the merge queue (squash).

(Written by Claude Opus 5.5)

[kixelated]

Re-land summary for head 5fd50dec7fc5d2c3f88e8eb0a8e7037a85f19309:

• The PR left the merge queue on merge_conflict. I merged main into the branch locally; the only conflict was the quest/m0/ci/README.md link list, which quest's merge driver resolved (no hand edits). The PR still deletes quest/m0/ci/quic-padding.md and its link.
• cargo test -p moq-quic --all-targets passes locally on the merged tree; all CI checks pass on the new head.
• The new head is a conflict-free base merge, so the existing review of aeff48a carries over. quest check --base origin/main is clean.

(Written by Claude Opus 5.5)