2044025 - Import script failing to auto ignore gkrust leaks

Status: ASSIGNED

(Testing :: web-platform-tests, defect)

Description

[Andrew McCreight [:mccr8]]

As seen in bug 2034815 and bug 2044021, two directories of WPTs are experiencing frequent leaks like this:
INFO - TEST-UNEXPECTED-FAIL | LeakSanitizer | leak at gkrust.ebd94ecb2afbeb2-cgu.0, gkrust.ebd94ecb2afbeb2-cgu.0, gkrust.ebd94ecb2afbeb2-cgu.0, gkrust.ebd94ecb2afbeb2-cgu.0

The problem is that the sequence of weird characters keeps changing causing large spikes in leaks until the import script runs again.

James, I don't remember exactly how the lsan-allowed matching works. If gkrust gets added, will it match frames like gkrust.ebd94ecb2afbeb2-cgu.0? These doesn't seem like a very frequent problem so I'm glad to go through and clean up the 2 directories for this specific issue but I can't tell offhand if it'll work or not. Thanks.

Comment 1

[Andrew McCreight [:mccr8]]

Attachment: Bug 2044025 - Ignore gkrust more generally in WPT lsan-allowed.

Comment 2

[James Graham [:jgraham]]

Looks like mozleak interprets the allow list as a set of regexp: https://searchfox.org/firefox-main/source/testing/mozbase/mozleak/mozleak/lsan.py#94-96. The sync bot won't ever create anything other than static strings, but I think something like gkrust\..*-cgu\.0 would work.

Comment 3

[Andrew McCreight [:mccr8]]

(In reply to James Graham [:jgraham] from comment #2)

Looks like mozleak interprets the allow list as a set of regexp: https://searchfox.org/firefox-main/source/testing/mozbase/mozleak/mozleak/lsan.py#94-96. The sync bot won't ever create anything other than static strings, but I think something like gkrust\..*-cgu\.0 would work.

I think the re.escape will prevent you from passing in an actual regexp. But it looks like it uses self.allowedRegexp.match(frame) eventually, so I think it'll match the prefix just fine, fortunately.