Repository navigation
Conversation
…the node queues Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
There was a problem hiding this comment.
✅ No new issues found.
Approach: sound — the bench assertions are updated to count what the lite-07 route node actually queues, rather than changing the node to match the old counts.
Reviewed changes — the single-file diff to rs/moq-net/benches/routes.rs.
routes/refollow— the assertion now expects2 + 2 * claims(twoAskedits, plus each claim withdrawn and served again), and the doc line says "stops asking for a prefix".routes/received— the assertion now expects2 * changes(each changed claim served into the origin and forwarded).
I did not run the bench. The PR description reports a reproduced panic on main and a passing full run after the fix.
claude-sonnet-5-5 | 𝕏
|
No actionable comments were generated in the recent review. WalkthroughThe refollow benchmark description now states that the node stops and resumes its request for a prefix. The Merge Risk:
|

Problem
Nightly's
moq-net/routesbench panics before measuring:routes/refollowasserts the node queues one output per claim (left: 4, right: 1on main), androutes/receivedasserts one forwarded batch per change (left: 200, right: 101). Both counts describe a router the lite-07 route node never was: the node sends eachAskedit as its own wire message, withdraws a dropped prefix's claims before serving them again, and sends oneAnnounceper claim changed (Router::exportinroute/relay.rs).Approach
The two assertions count what the node actually queues:
2 + 2 * claims, the remove and add edits, and each claim withdrawn and served again.2 * changes, each changed claim served into the origin and forwarded.The refollow doc line also says "stops asking for a prefix" rather than "closes its stream", which the node no longer does.
Impact
Alternatives
Announces into one message per peer would make the oldchanges + 1true, but the lite-07 publisher encodes each claim as its ownAnnounceBroadcastregardless, so it saves nothing on the wire.Validation
cargo bench -p moq-net --features fuzz --bench routes -- --testreproduced the panic on4f3ba3317f; with a non-panicking check every other routes case already matched, onlyreceived/*_100dandrefollow/*differed.--test) run of every routes case with short warm-up/measurement passes, so the counts also hold across repeated iterations on the kept fixture.refollow/16p_1candreceived/100c_100dto confirm the new counts are the messages described above, not something spurious.just checkpasses.Caused by
RouteFollower/RouteRelayonto the lite-07 route node but kept the old router's output counts in the bench assertions (found by check: Nightly).Follow-ups
moq-net/publisherwake/8192setup is O(drops x ranges) since moq-dev/moq#8638, which pushes the job past its timeout; tracked separately.Models
Size: S
Implementer: Claude Code, claude-opus-5-5, unknown effort
(Written by Claude Opus 5.5)
Review
Rewritten by
quest reviewafter each round.Head: 073e654
Base: origin/main at 4f3ba33
Risk: medium,
the base has no RISK.md to name what is always high or customer-facing, so every change is mediumClassified: medium by default
Reviewer: Codex, gpt-6.1-sol, medium effort
Session:
01a1efc4-9c01-77b0-9594-276196a143de, round 1Score: 10/10,
The updated assertions match the route fixtures, and source inspection identified no defects; builds and tests were not run.Rounds: 1; first round 0; findings 0: fixed 0, refuted 0, open 0
Findings: none