A Python 3.11 backend on Windows has had access violations while faulthandler is enabled with all_threads=True. Some dumps appear to be triggered by COM exceptions that the application handles. Could the handler itself be involved, and what would be useful to check?
Could faulthandler be contributing to crashes on Windows?
Answers
It is a plausible lead, especially if the access violations happen while a dump is being written. Try faulthandler.enable(file=..., all_threads=False) and see whether the mid-dump crashes stop. That still gives you the current thread's traceback, but not the other threads.
There was a reported case where switching to current-thread-only dumps was followed by several days without a mid-dump crash, despite further COM exceptions. That's suggestive, not proof of the mechanism, and it doesn't rule out a separate crash.
Reply to this answer
Check the exception codes and whether the program actually exits after each dump. On Windows, a vectored exception handler can see a first-chance exception before another handler catches it. That means a dump can appear for an exception the application goes on to handle.
For example, 0x80010108 is a COM object-disconnected code. Repeated dumps with that code aren't necessarily repeated crashes, although the source of the exceptions may still be worth investigating.
Reply to this answer
If a current-thread traceback doesn't explain a later access violation, consider configuring Windows Error Reporting LocalDumps for python.exe. A native crash dump may give you more to investigate than the Python traceback alone. Keep the startup or import crashes separate from the mid-dump issue until there's evidence they share a cause.
Reply to this answer
Add your answer
Answers are reviewed before publication.
7x+canvasratchet_462:
Truncated output in the middle of another thread's traceback is a useful clue, but it doesn't by itself establish that walking that thread caused the access violation.