You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Compile queue: close the test gaps the #1728 reviews deferred #1731
Follow-up to #1728. The per-task reviews deferred these test gaps. None of them is a known defect, and the code paths read correct, but each is untested:
Settles on a demoted node.SettleAdmission with Stored, NotStored or Failed on a node that has been demoted must still release the charge and the lease. Only WorkerRefused is tested.
NoteDepartureCounted. The call in FrameEndpoint's AbandonIfPeerGone has no case driving it. Deleting it would double-count AdmissionAbandoned and nothing would go red. The gap is noted at the call site.
The launcher's pendingRefusal / granted wiring in src/apps/pace-cc/main.cpp.main.cpp is in no test target. RunMiss's seams are tested, but the lambda that records a held plan refusal only after a grant is not.
RelaysByVerb attribution (Responders.hpp). A non-relayed ADMIT on a follower, from a remote launcher or our own key, gets the forwarder's PeerWatchCounter. A departure is then filed as pace_admission_forward_abandoned_total.
Refactor.PresenceRound's bool* reachedScheduler out-parameter should become a returned {accepted, reached} value, the house idiom.
One more gap from the final review of #1728: a8092d3 (a connection's admission hold is released before FrameEndpoint stops counting it as open) has no test. The candidate test was removed because it stayed green with the fix reverted. A test that discriminates: give a hold a release callback that reads the server's open count and asserts it is still 1 or more at release time. That needs OpenCount exposed to tests; today it lives on the internal State.
Activity
christianparpart commentedon Oct 28, 2026
One more gap from the final review of #1728: a8092d3 (a connection's admission hold is released before
FrameEndpointstops counting it as open) has no test. The candidate test was removed because it stayed green with the fix reverted. A test that discriminates: give a hold a release callback that reads the server's open count and asserts it is still 1 or more at release time. That needsOpenCountexposed to tests; today it lives on the internalState.