[libcxx-commits] [libcxx] [libc++][test][msan] Fix bots after #67799 (PR #73152)

[via libcxx-commits]

huixie90 wrote:

> @huixie90 This smells awfully like the padding bytes issue for `long double`, no? What else would make the compare exchange weak fail all the time? If so, we could consider disabling these tests for `long double` entirely until the Clang support isn't completely broken. This is actually kinda dangerous right now, people might be writing code using `atomic<long double>` and expecting it to work, when in reality it doesn't work properly and hasn't for probably 10 years.

Yea the stacktrace clearly shows that it is long double. Yes we did have a big report from 2020 as we seen in the code review. the things I don’t understand is
1: we have a workaround for fetch_add and it is working on our CI envs but not here?
2. aarch64 long double has no padding

could we clarify in this build:
1. The compiler is clang and OS is Linux
2. does the non-msan build have the same issue?


In your debugger, is that is_fp80_long_double return true?

https://github.com/llvm/llvm-project/pull/73152