[libcxx-commits] [libcxx] [libcxx] Poison memory in variant destroy (PR #101048)

Louis Dionne via libcxx-commits libcxx-commits at lists.llvm.org
Tue Oct 1 14:16:36 PDT 2024


ldionne wrote:

Gentle ping @vitalybuka. IMO the desired outcome for libc++ is pretty clear, it would be nice if MSAN's behavior could be aligned with it.

Said differently, our code would also be valid if we already called the destructor inside `variant::__destroy()` today, but MSAN would still have these false negatives. In that sense, I view this as a MSAN bug and not something we should work around in libc++ by calling the MSAN API directly.

https://github.com/llvm/llvm-project/pull/101048


More information about the libcxx-commits mailing list