[clang] MSan: poison the default-init allocation before calling constructors (PR #188001)

[Eli Friedman via cfe-commits]

efriedma-quic wrote:

I think explicit poisoning will blow up valid code if bitfields are involved, at least in theory.  (Not sure if anything can actually propagate the poison out of a memset of poison.)

Maybe it makes sense to have an msan-specific intrinsic for the exact behavior it wants.

https://github.com/llvm/llvm-project/pull/188001