Use int = 0 SFINAE in <memory> to improve compiler throughput

[fsb4000]

Fixes #248

Similar to https://github.com/microsoft/STL/pull/244/files

and transition now applies only to CUDA:

STL/stl/inc/functional

Lines 847 to 851 in a9321cf

	#ifdef __CUDACC__ // TRANSITION, CUDA
	#define _USE_FUNCTION_INT_0_SFINAE 0
	#else
	#define _USE_FUNCTION_INT_0_SFINAE 1
	#endif // __CUDACC__

[fsb4000]

I tested the constructor:

template <class _Fx, class _Alloc, typename _Mybase::template _Enable_if_callable_t<_Fx, function> = 0>
function(_Fx&& _Func) {
    this->_Reset(_STD forward<_Fx>(_Func));
}

#include <cstdio>
#include <functional>
using namespace std;

void test() {
    puts("hi");
}

int main() {
    function<void()> f = test;
    f();
}

$ nvcc -I"..\out\build\x64\out\inc" main.cpp
main.cpp
$ a.exe
hi

If someone knows what was the original repro it would be nice to test it.

[StephanTLavavej]

Regarding CUDA, I believe we retained the workaround out of caution, not because we had an actual repro. Now that we've updated to CUDA 11.6, removing the workaround seems like the right thing to do (we can always put it back if necessary).

[StephanTLavavej]

Reported VSO-1595465 "/clr /std:c++17 ICE with int = 0 SFINAE", pushed a conflict-free merge with main, then pushed a workaround to switch between the old code and new code depending on _M_CEE. (The type change from void to int is unconditional, that part is fine for /clr.)

[StephanTLavavej]

I'm mirroring this to the MSVC-internal repo - please notify me if any further changes are pushed.

[StephanTLavavej]

Thanks for modernizing these last remaining traces of old-style SFINAE! 🚀 ✨ 😺