Looks good at first view. I can't think of a better place to add the suppression files.

This is awesome! I think this is a good idea and I don't see any issues in the new github actions configuration. After this is merged we should probably open a tracking issue for any followups for any suppressions we think are real bugs that should be looked at closer.

One idea: maybe we can consolidate the two UBSAN jobs and one ASAN job we'd have if this PR is merged into two jobs that both simultaneously run ASAN and UBSAN. It looks like meson supports this: https://mesonbuild.com/Builtin-options.html#base-options

IMO given the different suppressions you had to write on linux and macos, I think it is worth testing both but that does mean adding at least one new CI job.

I asked Ralf to look at this if he has time since he's good at spotting issues in CI configuration.

@ngoldbaum went ahead and consolidated job clang_ubsan and clang_asan into clang_asan_ubsan. Can back out into separate jobs if requested again.

OK, I think this is looking good now.

My only reservation is that the new CI job takes 36 minutes to complete. Compare that with the less than 10 minutes the MacOS sanitizer job needs.

I’m not sure if a slow CI job that is mostly redundant with the Mac sanitizer job is worth adding. It does test code that doesn’t run on macs, but I’m not sure how much of that there is.

Before @mclare does anything else here I’d appreciate it if one or more of @seberg @charris or @rgommers could weigh in about how to handle the new CI job.

I'd be happy to run these sorts of lengthy tests on a weekly basis if there were a good way to insure someone notices if there is a problem. Not sure how well the tests identify the bad code. Running on PRs does have the advantage of narrowing down the problematic code. Apart from that, it would be nice to know precisely what this offers in the way of code coverage and completeness that TSAN doesn't. They are both clang based, correct?

Apart from that, it would be nice to know precisely what this offers in the way of code coverage and completeness that TSAN doesn't. They are both clang based, correct?

TSAN, ASAN, and UBSAN are all sanitizers in the clang and gcc compiler sanitizer suites. Each suite of sanitizers implements different checks. TSAN is for thread safety checks, ASAN is for memory safety, and UBSAN is for C undefined behavior checks.

All three are useful to run on PRs because they have disjoint sets of checks.

What's not clear on this PR is if it's worth adding checks using clang on x86_64 ubuntu as well as on ARM64 MacOS. Since we're doing both ASAN and UBSAN on both sets of checks, the only reason for running both is for checking OS-specific code for memory safety and undefined behavior.

See the clang docs for mire details. There are lots more sanitizers that we haven't even tried to use yet.

@ngoldbaum yeah I can work on that tonight, I'll set up an issue with a checklist and mirror some of the language in the suppressions themselves.

@ngoldbaum see here: #29524

Thanks so much! You've left things in a position where people can work on these issues one at a time, which is perfect for sprints and other drive-by contributions.