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
Daemon suite flakes on ubuntu CI: GitRepo fixture state vanishes mid-test #707
The ubuntu Build and test leg intermittently fails one daemon-suite test whose fixture is a GitRepo, with git state that should be stable disappearing or mutating between two steps of the same test. Two recent occurrences, different tests, same shape:
Run 33260948585 (main, 53be3b75): Borrowed_launch_creates_no_worktree_and_runs_in_the_cwd — the before/after tree snapshots differ by exactly .git/objects/maintenance.lock, a transient artifact of git's background maintenance, which evidently runs inside the fixture repos on CI.
Run 33397578952 (PR Block chat messaging to a flow participant in the desktop app #706, App-only diff): Interactive_codex_launch_threads_the_posture_into_the_runtime_start_context — GitRepo.CreateWithCommit fails 10 ms in with git add -A → 128 fatal: not a git repository, immediately after a successful git init in a uniquely named owned TempDir (Directory.CreateTempSubdirectory, so a path collision is ruled out).
Both runs' suite stderr also carries UnixSpawnerThread terminated unexpectedly / Environment.FailFast — possibly the intentional crash-contract test's child process rather than part of the flake; not yet confirmed either way.
The maintenance.lock sighting suggests the assembly-wide git config pin (Guards/GitConfigGlobalSetup) does not disable background maintenance/auto-gc in fixture repos; a detached git maintenance process from an earlier test also outlives its TempDir. Whether that also explains the not a git repository case is open.
Activity
linear-code commentedon Aug 31, 2026
AI-2370
Turn off git's own maintenance in fixture repositories (#707)
Indent the pinned git config without tabs (#707)
Close the races behind the suite's intermittent failures (#941)