Repository navigation
fix(assistant): address sidecar review follow-ups - #19
Conversation
Fix-forward report
Verification:
|
|
CI update: corrected follow-up commit |
caba9e2 to
ddf1a09
Compare
|
AI sidecar follow-up run status: incomplete.
No commit, push, rebase, PR-body update, or merge was performed. CI run ID: not available because no new push was made. |
|
Follow-up after isolating B1 with stash SHA
CI run ID: unavailable; no push was made. |
bac0882 to
89f5d97
Compare
89f5d97 to
8ede0e8
Compare
Final review follow-up修正を反映し、マージせずに push と検証を完了しました。
検証結果:
|
|
Squash merge completed. The merge commit on |
Summary
Fix-forward for the independent review of PR #13.
state-lock; poll and worker paths use locked snapshots; I/O remains outside the lock.discards-a-stale-sidecar-pending-before-the-next-requestrecords fake-sidecar input and assertsstaleis absent andfreshis delivered.stops-cleanly-when-init-gate-rejects-a-live-reader.+assistant-sidecar-response-timeout-seconds+(90 seconds) once per poll. If no event arrives for 90 seconds after the last event, it stops the sidecar and publishes a Japanese terminal error; each received event extends the deadline.does-not-time-out-while-response-events-continue;stops-and-delivers-a-terminal-event-after-response-silence.state-lockinterval; process and cleanup operations stay outside the lock. Cleanup preservesstartingduring the startup transition and advances the generation consistently. If pre-thread handoff fails,starting-pis cleared under the lock so a later start retries.clears-starting-state-when-startup-handoff-fails.reports-version-probe-output-when-the-sidecar-exits-with-an-error.%assistant-test-await-stateuses a time deadline; added stale-payload and SIGTERM-ignoring child cleanup coverage.discards-a-stale-sidecar-pending-before-the-next-request;kills-a-sidecar-that-ignores-sigterm.Commit mapping
22d3e294c3b32f01efabe2cae77029a51891faa6b08ede0e8Verification
nix build .#checks.aarch64-darwin.default -L: passed,2293 passed, 34 skipped, 0 todo, 0 failed, 0 errored, 2327 total.e2e-main-wait-returns-background-statusdid not hang in this CI run; the earlier gate hang was an existing defect fixed by PR fix: reap completed background jobs before reading status #22 SIGCHLD handling.