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

[via libcxx-commits]

EricWF wrote:

@philnik777 Nobody is overruling you. I appreciate you taking the time to spell out your concerns.

> I don't see any reason fur us to make our implementation work super well with cude if there doesn't seem to be much incentive from nvidia to make things work well for other libraries.

I would say the incentive for us is to help our users, who have asked us for this. 

> I'm also not convinced that this parallels with other macros we push and pop, since this isn't about conformance.

Could you say more? For me the parallel is "Expansion of this macro breaks libc++".

Fundamentally this change simply adds a name to a list. I don't share your concern regarding comparative complexity.

My concern is shipping an implementation where `_LIBCPP_NOINLINE` is a lie when it doesn't need to be. I believe having source code behave as written is a better outcome than having the compiler silently swallow the diagnostic. 

You've requested changes on this CL. Could you please reiterate the actionable items you would like taken?



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