[kixelated]

Problem

Every moq-net check that asks whether a group is done must mean one of two things. Either a copy whose writer finished (Handle::is_settled), or the whole group reaching its end (is_finished). Three review rounds on moq-dev/moq#6922 each fixed one wrong site. quest/m1/settled-copies.md asks for an audit of every caller at once.

One related check was wrong. group::Producer::finish calls finish_largest for every copy. When a bounded fetch copy of the largest group finished, the largest was marked final even though the group could still grow. On a track whose end was already reached, the largest stream then ended early.

Approach

• finish calls finish_largest only for a copy no transport bound cut short. This reuses the original_fin the compressor already computes. A new test, a_finished_bounded_copy_keeps_the_largest_open, failed before the fix.
• The comments on cache::Largest::open and finish_largest now say "whole copy".
• tear_down gets a comment saying its whole-group open/retain checks are deliberate.

Audit of rs/moq-net/src/model/. Every caller other than finish_largest already uses the meaning it needs:

Caller	Check	Needs	Verdict
TrackState::tear_down, the copies it aborts	!is_settled	copy	right: aborting a finished bounded copy would fail its reader
tear_down open (dropped)	!any(is_finished)	whole	right: a bounded copy serves no group reader, so the slot is given up
tear_down lookup.retain	any(is_finished)	whole	right, deliberate: a slot holding only a bounded copy leaves the cache, but a reader holding that copy keeps it (it is not aborted)
TrackState::set_idle (displaced)	!any(is_finished)	whole	right: the next feed has to send the group whole again
TrackState::is_settled	slot.group.is_settled	copy	right (the fix from #6922)
Consumer::parked_whole	is_finished with frame 0	whole	right
Detached::writing	!is_settled	copy	right
Handle::abort_open, resume_frame (inline fin)	copy / whole	as written	right
group::Producer::finish → cache::finish_largest	any fin	whole	wrong, fixed here
Slot::serves_end/is_walked/cut/is_read/is_expired/is_spent, arrival/ordered walks, holds, landing, peek_*, fill cached, miss.rs (filled, settle, refusals, holes), insert_group_request's kept	not aborted, not back-cut	whole-group reader	right

The is_finished in broadcast.rs, Window::is_settled, and kio::Pending::is_finished belong to other types and are out of scope.

Impact

• Public API: none.
• Wire: none.

Alternatives

• Keep a cached slot that holds only a finished bounded copy past a track abort, so a later fetch of that range still hits the cache. I did not do this. On a live slot it would read as a cut carrying the abort error, and the slot is already reported as dropped.

Follow-ups

• DropCursor::cut reports a cut that a fetched head covers. It records slot.group.cut_error() for any visible slot. A live group the publisher cut, while a fetched whole head still serves its end, is both handed out by the walk (serves_end) and reported as dropped. It could use Slot::cut(), which is None while a copy serves the end. First confirm what a wire subscriber should see. Not changed here: it is about cuts, not settled versus whole.
• TrackState::is_settled checks only slot.group. A visible slot whose live group was aborted (cut or evicted), while a fetched head kept serving it, keeps the end unsettled. A later abort then turns the clean end into an error. Decide whether an aborted copy counts as "nothing in flight".

Validation

• cargo test -p moq-net --all-features --lib: 2627 passed. The new test failed before the fix.
• just check: pass at 1b2a6566c0 (8962 Rust tests and the rest), after merging main.
• quest check: pass. The finished quest and its link in quest/m1/README.md are removed.

🤖 Generated with Claude Code

(Written by Claude Opus 5.5)

[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 7 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: f29c40d9-f79a-4081-ace8-644d8fcc1fc3

Commits

Reviewing files that changed from the base of the PR and between 8aab135 and 1b2a656.

Files selected for processing: 5

• quest/m1/README.md
• quest/m1/settled-copies.md
• rs/moq-net/src/model/cache.rs
• rs/moq-net/src/model/group.rs
• rs/moq-net/src/model/track.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	✅ Completed 2026-10-28T09:05:59.315792Z	1b2a656	Draft marked ready

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: 1b2a6566c0e9908017d2b2ef5fefc840641b0242.

No actionable bugs found in the changed areas. Gating finish_largest on the existing original_fin preserves the distinction between a settled bounded copy and a known group end without adding another state flag. The regression test exercises the premature-largest-stream termination, and the existing recreated/fetched-whole-copy tests cover the complementary completion path. The teardown change is documentation-only; keeping the separate cut/aborted-copy questions out of this focused fix is reasonable.

Verification: static inspection of the diff and surrounding bound, finish, teardown, largest-stream, and test code via GitHub. I did not run tests or independently verify the reported test results.

[kixelated]

Merge summary: group::Producer::finish now calls finish_largest only for a copy no transport bound cut short, so a finished bounded fetch copy no longer marks the largest group final. Comments now say "whole copy", and the audit table in the body covers every settled/whole check in rs/moq-net/src/model/. This PR finishes quest/m1/settled-copies.md and removes it.

Review: Codex reviewed head 1b2a656 with no findings, and an automated OpenAI review found nothing actionable. CI is green, and quest check --base origin/main passes. Entering the merge queue at 1b2a656.

Not done here (follow-ups from the body): DropCursor::cut reports a cut that a fetched head covers, and TrackState::is_settled checks only slot.group.

(Written by Claude Opus 5.5)