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
Windows arm64: the packaged app exited once mid-suite with no crash record (1 of 5 runs) #392
During platform validation of pull request 387, the packaged Windows arm64 app exited once in the middle of the default packaged UI suite (during the appearance-themes spec), in 1 of 5 full runs. The Profiler recorded crash.unclean-exit on the next start.
Evidence
The app's last log line is a routine page load; no panic, shutdown, quit or WebView failure line.
Windows recorded nothing in the Application or System logs for that minute, no error report and no dump.
The last memory sample was normal.
A new app process started about 150 ms after the old one's last log line.
The spec alone passed 5 of 5 twice, and three further full runs on the final build passed with no exit.
What it could be
The WebDriver service used by the suite restarted the app after a failed status check; on Windows that ends the process at once and leaves no record. The quick restart fits; the service's restart warning was not found.
A native crash that left no record: on Windows arm64 the Profiler's dump capture is disabled.
What would decide it
Route the WebDriver service launcher's own log to a file kept per run, and enable native dumps on arm64 (or a LocalDumps key in the lab). Then the order of "service restarts the app" and "the app's last line" tells the two apart.
Closing as not planned: not reproducible. It happened once in five full runs, did not recur in three runs on the final build, and the records cannot tell a test-tooling restart from a silent native crash. The issue body says what to capture; reopen with that evidence if it happens again.
Activity
mubeda commentedon Nov 19, 2026
Closing as not planned: not reproducible. It happened once in five full runs, did not recur in three runs on the final build, and the records cannot tell a test-tooling restart from a silent native crash. The issue body says what to capture; reopen with that evidence if it happens again.