Repository navigation
Conversation
|
Codex corrected shutdown-test event ordering and retry of transient status-probe disconnects. A reset injection fails the old test and passes the correction; destroying active work still fails it. Passed 759 base tests with coverage, build, typecheck, lint and formatting, plus 901 tests and a clean build/typecheck on the restacked tip. Production code is unchanged. Current-head Node 22/24 CI and Devin pass; no unresolved review threads remain. Earlier human reviews predate these test-only changes. — Codex |
Codecov Report❌ Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## main #4791 +/- ##
==========================================
+ Coverage 76.64% 76.85% +0.22%
==========================================
Files 1102 1108 +6
Lines 55495 56124 +629
Branches 12236 12424 +188
==========================================
+ Hits 42530 43130 +600
- Misses 11976 12004 +28
- Partials 989 990 +1 ☔ View full report in Codecov by Harness. 🚀 New features to boost your workflow:
|
7b3bbf9 to
04b5495
Compare
This comment has been minimized.
This comment has been minimized.
This comment has been minimized.
This comment has been minimized.
04b5495 to
5261e34
Compare
This comment has been minimized.
This comment has been minimized.
This comment has been minimized.
This comment has been minimized.
5261e34 to
f2c3b5c
Compare
This comment has been minimized.
This comment has been minimized.
This comment has been minimized.
This comment has been minimized.
f2c3b5c to
0de7ec0
Compare
This comment has been minimized.
This comment has been minimized.
This comment has been minimized.
This comment has been minimized.
0de7ec0 to
1899090
Compare
This comment has been minimized.
This comment has been minimized.
This comment has been minimized.
This comment has been minimized.
|
Subscribed to pull request
Generated by CodeMention Warning: The preamble and epilogue options in commentConfiguration are deprecated. Use template instead. |
1899090 to
011dfd3
Compare
This comment has been minimized.
This comment has been minimized.
This comment has been minimized.
This comment has been minimized.
011dfd3 to
ff2aea7
Compare
This comment has been minimized.
This comment has been minimized.
This comment has been minimized.
This comment has been minimized.
Co-authored-by: Codex <codex@openai.com>
ff2aea7 to
ee19c5d
Compare
|
✅ Thank you for adding a changeset! |
Why
Repeated control commands reconnect to the preview server. Let agents keep the connection ready between commands.
How
Add an optional local helper and share action preparation, uploads, downloads and execution. Ordinary control commands reuse the helper when available. The idle timeout defaults to 120 seconds.
--idle-timeouttakes seconds. Closing the connection leaves the simulator session running. Keep-alive supports macOS and Linux; ordinary commands still work on Windows. Control import (#4686) and start import (#4687) use this shared executor.Security and cleanup
The helper uses an owned, private Unix socket directory. Processes running as the same OS user share this authority. Discovery is scoped to EAS credentials, project, session, API environment and CLI installation/version. Preview credentials pass over parent-child IPC; no new credential file is written.
Callers run one at a time. A dropped reply never triggers an automatic replay. Messages, queues and waits are bounded. Close finishes the in-flight action, rejects later actions and permits bounded upload cleanup. Immediate reopen waits for the drain.
Saved-login changes, idle expiry and remote disconnect close the helper. Server-side access-token revocation is not polled. An already-dispatched action can finish during shutdown.
Test Plan
Codex reproduced the shutdown-test races under coverage. Tests now allow close events to settle and retry dropped status probes until the helper confirms closing. The old test fails a deterministic reset injection; the correction passes. Destroying active work still fails the corrected test. The base passed 759 simulator/help tests with coverage, clean build, typecheck, lint and formatting; the complete stack passed 901 tests. Production code is unchanged. Current-head CI passes on Node 22 and Node 24. Devin completed with no unresolved findings.
Current-head Codecov patch coverage is 94.57%. Worker bootstrap/exit paths and some error branches remain uncovered; no coverage exclusions were added. Hosted EAS session testing and latency measurements remain pending.
Authorship and LLM use
Gabe chose the connection commands and timeout, then asked to share the control code with imports. Codex implemented the helper, moved the action executor into this layer and restacked the imports. Gabe reviewed the changes at
1899090. Codex then rebased onto current main and later corrected shutdown-test races exposed by Node 22 CI. The existing poll delay now lets the close event abort the connection before checking refusal replies. The close-caller test also retries transient probe disconnects and still checks the close acknowledgement and active-work drain. These test-only adjustments have not had a new human review; production code is unchanged.Astra and Opus reviews led to removing redundant authentication and validation, then fixing admission and shutdown races. Devin identified the close/reopen race. Coverage reviews corrected a masked shutdown assertion and a fake-timer race. Independent rechecks were clean on the earlier combined implementation; Codex checked the reorganized layers and preserved feature tree. Those independent reviews and hosted testing were not repeated for the split.
— Codex