Repository navigation
fix: prevent release lock cycles and agent tool crashes - #870
Merged
Merged
Conversation
RetroGameDeveloper
force-pushed
the
fix/release-lock-order
branch
from
November 8, 2026 16:16
9f94fbf to
d1fd0c2
Compare
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Fixes
log.parallel_claims=Falsefrom bounded background CPU handoffs. Its normal default remains enabled. A regression verifies an existing agent lease is untouched.Evidence and limits
Recent local tool transcripts contain a release taking 299 s and another taking 113 s despite almost no reported project-lock or synchronization wait. Inspection found release tool entry retaining the project lock before its receipt acquires the candidate lock, reversing compile publication order. A bounded synthetic reproduction through both actual transports makes the compile unable to acquire its project lock on the old implementation; both finish in the correct order with this fix. This establishes the cycle, not that it explains every observed delay.
Three unexpected MCP tool exceptions in the reviewed logs identify the missing-source and empty-flow crashes above. Broader verification also exposed the ignored CPU handoff claim policy; it is corrected rather than excluded.
Long cross-project verification-lane waits remain unresolved. The newer phase fields now show some waits concentrated in verification admission, rather than the compiler or SQL. This PR does not increase timeouts, skip imports, weaken matching or claim all long waits are solved.
Verification
src tests projects) andgit diff --checkpass.After #869 merged, rebased onto
b10d1bce; retained both performance-document sections. All four changed implementation files are byte-identical to their tested pre-rebase versions. 129 integration tests passed on the combined tree, including new m2c session/autopilot pipeline coverage; full Ruff and diff checks passed again.The release entry also rechecks function identity after acquiring both locks: a function replaced while queued must not make the handler acquire a different candidate lock under the retained project lock. The new rejection regression and 218 targeted tests passed (45.22 s) on the final guard change; full Ruff/diff checks passed again. Claim/source state is not changed by this rejection.
Remote CI status at merge inspection: Linux and Windows full-suite jobs are running; macOS and label jobs remain queued. No remote failures are reported, and the main branch has no required remote checks (
gh pr checks --requiredreports none; branch protection API reports unprotected). The local regression/lint checks above are complete. Hosted CI remains pending and is not represented as passed; its eventual result must be followed up.