This will affect only Darwin, as the rest alredy do that.
Details
Diff Detail
- Repository
- rG LLVM Github Monorepo
Event Timeline
Any concern?
Reproducing crashes without signal blocking is hard, we have tests which crashes 1 in 1M times.
So the question to reviewers is rather: does this regress anything on Darwin?
@kubamracek , do you have any historical context why this has been avoided so far in Darwin?
It was not avoided. It was used by msan, which is Linux only. And I just added this to Linux Asan as well. Without testing, I am not confident enabling this on Darwin.
If @vitalybuka is reasonably comfortable that testing in our CI will expose likely Darwin issues then I'm comfortable approving and just keeping an eye on it as it comes down into the automerger. If we need to do more testing than that, we should do what you suggest @thetruestblue and bring the diff down out-of-channel and do some more thorough testing first.
I applied this diff internally and at least as far as all of the tests run by
ninja check-runtimes
goes, there are no differences. I don't feel necessarily that I have any deep insight into why there should or should not be any subtle difference that would surface on macOS, but I also no longer want to be in the way of this one @vitalybuka. I'll keep an eye on it though and let you know if I see anything surprising.