Closed rshariffdeen closed 1 year ago
I think that we had an unresolved discussion about this. Right now, we don't explicitly set any UBSAN_OPTIONS during patch evaluation, so we roll with whatever the defaults are. I could try to update UBSAN_OPTIONS to exit with a specific code (do you have a reference for UBSAN_OPTIONS settings?), but that would also conflict with our desire to keep halt_on_error=0
(i.e., don't exit when we encounter a failure).
Edit: From this blog post, it looks like exitcode
is the name of the UBSAN variable that we would want to change (if we halt on error).
To get rid of this blocker and allow us to workaround the issue, I'm adding support for custom ASAN_OPTIONS and UBSAN_OPTIONS overrides in bug.json now (#131). We can discuss how to deal with the default case after the internal deadline.
@rshariffdeen Did you address this issue using the features added above?
I updated the bad-output from runtime error to UndefinedBehaviorSanitizer
This is addressed
Incorrectly classify patch as successful when UBSan fails Didn't we enable UBSAN_OPTIONS with errorexitcode? Even if our bad_output does not capture this, errorexitcode can be set to detect these bugs.