Does it make sense to create a valgrind CI job for wgpu?

[Luni-4]

I do not know if this idea makes sense, because reading around it seems that valgrind can produce a certain amount of false-positives for GPU-wise contexts. I'm asking to you, who are more experts in the field, in order to know if a new GitHub action in that direction could be useful. Thank you!

[cwfitzgerald]

I wouldn't expect false positives - I can run the test suite under valgrind fine. The main problem is that valgrind is slow and running the whole test suite can take many minutes on a 10 core, let alone a 2 core CI machine.

[cwfitzgerald]

In short - running locally makes sense, in CI, not so much.

[Luni-4]

I wouldn't expect false positives - I can run the test suite under valgrind fine. The main problem is that valgrind is slow and running the whole test suite can take many minutes on a 10 core, let alone a 2 core CI machine.

This is true, but one can run valgrind either on specific tests or on the entire test suite before releasing a new version. Do you think that could be a good alternative?

[cwfitzgerald]

We don't really have enough bugs that valgrind to solve to bother with the maintenance of such a thing.