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

[Mark de Wever via libcxx-commits]

https://github.com/mordante requested changes to this pull request.

I'm also against merging.
@ldionne asked for an issue filed at the CUDA project. This has not been done.

I feel it sets a bad precedent when we make changes when other members in the ecosystem don't want to play nicely. Especially, since they were aware it clashed with existing GCC practice. Instead of using `__cuda_noinline__` they decided to introduce a macro that breaks existing code.

The fix is quite simple, but it has a maintenance burden for us. Every time we use our noinline macro we have to remember to push and pop our macros. Which most likely is only flagged by the CI so needs another CI run. Our CI resources are limited and they slow down development, at least for me it does regularly.

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