Missing include on windows for using _wgetenv

[h-vetinari]

nvtx uses _wgetenv on windows

NVTX/c/include/nvtx3/nvtxDetail/nvtxInit.h

Lines 17 to 20 in 09e0d23

	#if defined(_WIN32)
	#define NVTX_PATHCHAR wchar_t
	#define NVTX_STR(x) L##x
	#define NVTX_GETENV _wgetenv

which requires either <stdlib.h> or <wchar.h>. From looking around the code under https://github.com/NVIDIA/NVTX/blob/v3.1.0/c/include/nvtx3/ a bit, all the includes happen in c/include/nvtx3/nvtxDetail/nvtxImpl.h, which is missing those headers for windows

NVTX/c/include/nvtx3/nvtxDetail/nvtxImpl.h

Lines 15 to 19 in 09e0d23

	#if defined(_WIN32)
	#include <Windows.h>
	#else

they do get included further down, so there may be some mix-up with the placement of the respective #else / #endif. For packages that need nvtx in conda-forge on windows (with CUDA 12 currently), this predictably fails:

%LIBRARY_INC%\nvtx3\nvtxDetail\nvtxInit.h(164): error C3861: '_wgetenv': identifier not found

Xref conda-forge/cuda-nvtx-feedstock#23

[jcohen-nvidia]

This issue is fixed in the dev branch by commit 909007c. It will be merged into the release-v3 branch in the next release, which will be in the next few weeks. If you are using the release-v3 branch and need a fix now, but are not comfortable switching the dev branch, you can temporarily take the contents of c\include\nvtx3\nvtxDetail\nvtxImpl.h from the dev branch replace your local copy with that. The only code change (i.e. not whitespace/comments) to that file since the last merge was the fix for this issue, by moving some #includes like stdlib.h and wchar.h out from a Linux-only section to be used on all platforms.

[h-vetinari]

Thank you!