Repository navigation
Codex App on macOS crashes with SIGABRT after sending a message; earlier missing rollout path #21663
Description
What version of the Codex App are you using (From “About Codex” dialog)?
Codex 26.506.21252
What subscription do you have?
chatgpt pro
What platform is your computer?
Darwin 25.1.0 arm64 arm
What issue are you seeing?
## Summary
Codex App on macOS crashes after sending a message. Initially it failed to resume a local conversation because the expected rollout JSONL file did not exist. After reinstalling Codex App and clearing local Codex state, the app still crashes with `SIGABRT` / `thread caused non-unwinding panic`.
## Environment
- OS: macOS
- Architecture: [Apple Silicon / Intel]
- Codex App version: [paste version from About Codex]
- Install source: Codex App official download page
- Login method: ChatGPT account
- Workspace tested:
- `/Users/<user>/Documents/Codex`
- `/Users/<user>/codex-empty-test`
- Time of occurrence: 2026-05-08 around 12:30–12:41 local time
## Error shown in Codex App
```text
恢复对话失败
failed to resolve rollout path `/Users/<user>/.codex/sessions/2026/05/08/rollout-2026-05-08T12-30-06-<thread-id>.jsonl`: No such file or directory (os error 2)
(code=null, signal=SIGABRT).
Most recent error: thread caused non-unwinding panic. aborting.Later, after clearing state, the missing rollout path message no longer appeared, but the app still crashed with:
(code=null, signal=SIGABRT).
Most recent error:
0: 0x1087e53f0 - __mh_execute_header
1: 0x107d4151c - __mh_execute_header
2: 0x1087b4958 - __mh_execute_header
...
26: 0x19b699c08 - __pthread_cond_wait
thread caused non-unwinding panic. aborting.
Reproduction steps
- Install or reinstall Codex App on macOS.
- Open Codex App.
- Log in with ChatGPT account.
- Open a local workspace, for example
/Users/<user>/Documents/Codexor/Users/<user>/codex-empty-test. - Start a new conversation.
- Send a simple message such as
hi. - Codex App crashes with
SIGABRT.
What I tried
I backed up and moved the existing Codex local data:
cp -a ~/.codex ~/.codex.backup.$(date +%Y%m%d-%H%M%S)
mv ~/.codex ~/.codex.broken.$(date +%Y%m%d-%H%M%S)Then I reinstalled Codex App.
I recreated the expected sessions directory and verified it was writable:
mkdir -p ~/.codex/sessions/2026/05/08
chmod -R u+rwX ~/.codexI also removed GUI/global state files:
pkill -f Codex
pkill -f codex
BACKUP_DIR="$HOME/.codex/state-broken-$(date +%Y%m%d-%H%M%S)"
mkdir -p "$BACKUP_DIR"
mv ~/.codex/.codex-global-state.json "$BACKUP_DIR/" 2>/dev/null
mv ~/.codex/.codex-global-state.json.bak "$BACKUP_DIR/" 2>/dev/nullI then tried a deeper local state cleanup:
pkill -f Codex
pkill -f codex
BACKUP_DIR="$HOME/.codex/runtime-broken-$(date +%Y%m%d-%H%M%S)"
mkdir -p "$BACKUP_DIR"
mv ~/.codex/.codex-global-state.json* "$BACKUP_DIR/" 2>/dev/null
mv ~/.codex/state_*.sqlite* "$BACKUP_DIR/" 2>/dev/null
mv ~/.codex/logs_*.sqlite* "$BACKUP_DIR/" 2>/dev/null
mv ~/.codex/sqlite "$BACKUP_DIR/sqlite" 2>/dev/null
mv ~/.codex/cache "$BACKUP_DIR/cache" 2>/dev/null
mv ~/.codex/.tmp "$BACKUP_DIR/dot-tmp" 2>/dev/null
mv ~/.codex/tmp "$BACKUP_DIR/tmp" 2>/dev/null
mv ~/.codex/plugins "$BACKUP_DIR/plugins" 2>/dev/null
mv ~/.codex/vendor_imports "$BACKUP_DIR/vendor_imports" 2>/dev/null
mv ~/.codex/computer-use "$BACKUP_DIR/computer-use" 2>/dev/null
mkdir -p ~/.codex/sessions
chmod -R u+rwX ~/.codexThe crash still occurs after these steps.
Observed local state issue
Before clearing state, ~/.codex/.codex-global-state.json contained projectless thread IDs that pointed to threads whose rollout JSONL files did not exist.
Example, redacted:
{
"projectless-thread-ids": [
"<thread-id-1>",
"<thread-id-2>",
"<thread-id-3>"
],
"thread-workspace-root-hints": {
"<thread-id-1>": "/Users/<user>/Documents/Codex",
"<thread-id-2>": "/Users/<user>/Documents/Codex",
"<thread-id-3>": "/Users/<user>/Documents/Codex"
}
}The expected path under ~/.codex/sessions/2026/05/08/ did not contain the corresponding rollout-*.jsonl file.
Relevant macOS log excerpts
From:
log show --last 10m --predicate 'process CONTAINS[c] "Codex"' --info --debugI saw repeated Codex / Codex Helper process starts and repeated FSEvents errors:
Codex Helper: (libsystem_info.dylib) Retrieve User by ID
Codex Helper: (TCC) TCCAccessRequest() IPC
codex: (libsystem_info.dylib) Retrieve User by ID
codex: (Network) networkd_settings_read_from_file initialized networkd settings by reading plist directly
codex: (Security) SecKeyCopyAttributes
codex: (Security) SecKeyCopyPublicBytes
codex: (Security) SecTrustEvaluateIfNecessary
Repeated FSEvents errors:
codex: (FSEvents) dev 0 () : purging events up to event id <event-id>
codex: (FSEvents) [com.apple.fsevents:client] FSEventsPurgeEventsForDeviceUpToEventId: f2d_purge_events_for_device_up_to_event_id_rpc() failed: 5
Expected behavior
Codex App should create a new local thread/session successfully, write the required rollout file if needed, and not crash when sending a message.
If a rollout file is missing, Codex App should handle it gracefully rather than aborting.
Actual behavior
Codex App crashes with:
(code=null, signal=SIGABRT)
thread caused non-unwinding panic. aborting.
Initially it also failed with:
failed to resolve rollout path `~/.codex/sessions/.../rollout-<timestamp>-<thread-id>.jsonl`: No such file or directory (os error 2)
Attachments
I can provide a redacted codex-gui-crash.log generated with:
log show --last 10m --predicate 'process CONTAINS[c] "Codex"' --info --debug \
| sed "s#/Users/$USER#/Users/<user>#g" \
| sed "s#$USER#<user>#g" \
> ~/codex-gui-crash-redacted.logI will not upload auth.json, config.toml, installation_id, or the full ~/.codex directory because they may contain sensitive local or authentication information.
What steps can reproduce the bug?
1.Install or reinstall Codex App on macOS.
2.Open Codex App.
3.Log in with ChatGPT account.
4.Open a local workspace, for example /Users//Documents/Codex or /Users//codex-empty-test.
5.Start a new conversation.
6.Send a simple message such as hi.
7.Codex App crashes with SIGABRT.
What is the expected behavior?
No response
Additional information
No response
Activity
github-actions commentedon May 8, 2026
Potential duplicates detected. Please review them and close your issue if it is a duplicate.
Powered by Codex Action
tyson-trader commentedon May 8, 2026
Workaround: running Codex with a clean HOME works:
HOME=/tmp/codex-home-test codex --cd /Users//codex-empty-test
But using the real HOME crashes even with /tmp workspace:
RUST_BACKTRACE=1 codex --cd /tmp/codex-work-test
This suggests Codex crashes when handling something in the real HOME environment, likely through macOS FSEvents / notify invalid UTF-8 path handling.
tyson-trader commentedon May 8, 2026
Update: After clearing Codex/OpenAI local Library state, Codex App can now send messages again. The earlier crash appears tied to local app/cache/state plus macOS FSEvents/notify invalid UTF-8 handling. I am still checking whether Codex CLI with the real HOME is fixed as well.