[libcxx-commits] [libcxx] [libc++] Protect the libc++ implementation from CUDA SDK's `__noinline__` macro (PR #73838)

[Louis Dionne via libcxx-commits]

ldionne wrote:

If the CUDA SDK truly dismisses this interoperation issue with Clang and GCC, they probably don't care that their customers are able to use those compilers / standard libraries alongside the CUDA SDK, which would be a reason to push back on doing anything here.

I agree we should strive to make our users lives better, but blindly working around this is harmful in the long term for everyone, since the badly-behaved vendor will just keep doing this sort of stuff and we'll only accumulate more and more workarounds. We need to give a good reason for that vendor to change things, and then we can make our users lives better in the meantime while they are transitioning away from the macro.

I will file a bug report against CCCL myself.

https://github.com/llvm/llvm-project/pull/73838