Bugzilla

It definitely looks like CheckedUnsafePtr caught a case where a CheckedUnsafePtr outlived the worker dropping its self-ref.  The fact that this happened in xpcom-shutdown-threads suggests this might not be a content global but instead something like the OS.File global and that ASAN performance characteristics slowed things down to shuffle thread scheduling.  We definitely need to investigate and/or add more instrumentation before relanding, and should probably target a bunch of ASAN jobs like this as good candidates to be introducing a chaos mode factor.

It definitely looks like CheckedUnsafePtr caught a case where a CheckedUnsafePtr outlived the worker dropping its self-ref.  The fact that this happened in xpcom-shutdown-threads suggests this might not be a content global but instead something like the OS.File global and that ASAN performance characteristics slowed things down to shuffle thread scheduling.  (edit: Although this was a content process which inherently has different shutdown characteristics than the parent, so maybe that's also related and maybe it was a content global that just didn't have a lot of time to shutdown before XPCOM shutdown.)  We definitely need to investigate and/or add more instrumentation before relanding, and should probably target a bunch of ASAN jobs like this as good candidates to be introducing a chaos mode factor.