Repository navigation
Desktop app crashes on a large local thread; recent history disappears from UI while rollout is intact #45162
Description
What version of the Codex App are you using (From “About Codex” dialog)?
26.903.9818.0
What subscription do you have?
Pro5x
What platform is your computer?
Microsoft Windows NT 10.0.26220.0 x64
What issue are you seeing?
After the Windows desktop app froze and crashed on 2026-09-13, a local thread showed an interrupted turn from 2026-09-11 as its latest content. Messages from 2026-09-12 and 2026-09-13 disappeared from the UI, although they are still present in the original local rollout JSONL.
The Windows error dialog showed ChatGPT.exe exception 0xe0000008. At 10:56:02 (UTC+8), the app log recorded: "memory allocation of 1613927764 bytes failed".
In-app feedback ID: 01a098bc-18cd-7a70-aca3-03e73443dccd
What steps can reproduce the bug?
I do not have a reliable way to reproduce the crash on demand.
- I used a long-running local thread over several days. Its rollout JSONL had grown to about 2.85 GB.
- I was still exchanging messages in it at about 10:41 on 2026-09-13 (UTC+8).
- The app froze and crashed around 10:56. I relaunched the app and reopened the same thread.
- The UI then stopped at an interrupted turn from 2026-09-11, rather than showing the later messages.
What is the expected behavior?
After relaunching the app, all messages already saved in the local rollout should remain visible, and the thread should resume from its actual latest turn.
Additional information
The original rollout is intact (2,854,511,278 bytes) and contains messages from 2026-09-13 10:41–10:42. The local history projection stops at about 1.24 GB / ordinal 10358 (2026-09-11), while the rollout continues to about 2.85 GB / ordinal 23035. The bundled app-server version was 0.153.4.
I made a hash-verified backup without modifying the original rollout or history database. Is there a supported, non-destructive way to rebuild the index and restore the messages in the UI?
I have not attached the complete rollout because it contains private conversation and tool data.
Activity
github-actions commentedon Sep 13, 2026
Potential duplicates detected. Please review them and close your issue if it is a duplicate.
Powered by Codex Action
LZhWi commentedon Sep 13, 2026
Thanks for the duplicate check. I rechecked the exact projection boundary and found a duplicate ordinal in my affected rollout:
This strongly suggests that the missing-history portion is related to #44511 and #44919. My incident also includes an app-server failure to allocate 1,613,927,764 bytes and a subsequent Windows desktop crash; I have not established whether that failure and the projection problem share a cause.
Please treat this as additional evidence for the projection issue. If maintainers prefer consolidation, I can close this issue and track the history-restoration problem in the existing report. I am keeping the full rollout and database backup private because they contain conversation data.