Update OS minimum version and remove XP runtime detection

[AlexGuteniev]

Currently STL has to target Vista as minimum OS version for x86 and x64

Update filesystem.cpp

STL/stl/src/filesystem.cpp

Lines 56 to 99 in 87dc1d3

	#if _STL_WIN32_WINNT < _WIN32_WINNT_VISTA
	using _GetFinalPathNameByHandleW_t = decltype(&GetFinalPathNameByHandleW);
	_GetFinalPathNameByHandleW_t GetFinalPathNameByHandleW_p{};
	unsigned long __stdcall _Not_supported_GetFinalPathNameByHandleW(HANDLE, wchar_t*, unsigned long, unsigned long) {
	SetLastError(ERROR_NOT_SUPPORTED);
	return 0;
	}
	#define __vcrt_GetFinalPathNameByHandleW \
	(_Runtime_dynamic_link(&GetFinalPathNameByHandleW_p, L"kernel32.dll", "GetFinalPathNameByHandleW", \
	&_Not_supported_GetFinalPathNameByHandleW))
	#else // ^^^ _STL_WIN32_WINNT < _WIN32_WINNT_VISTA ^^^ // vvv _STL_WIN32_WINNT >= _WIN32_WINNT_VISTA vvv
	#define __vcrt_GetFinalPathNameByHandleW GetFinalPathNameByHandleW
	#endif // _STL_WIN32_WINNT < _WIN32_WINNT_VISTA
	// MACRO __vcrt_GetFileInformationByHandleEx
	#if _STL_WIN32_WINNT < _WIN32_WINNT_VISTA
	using _GetFileInformationByHandleEx_t = decltype(&GetFileInformationByHandleEx);
	_GetFileInformationByHandleEx_t _GetFileInformationByHandleEx_p{};
	BOOL __stdcall _Not_supported_GetFileInformationByHandleEx(
	HANDLE, FILE_INFO_BY_HANDLE_CLASS, void*, unsigned long) {
	SetLastError(ERROR_NOT_SUPPORTED);
	return 0;
	}
	#define __vcrt_GetFileInformationByHandleEx \
	(_Runtime_dynamic_link(&_GetFileInformationByHandleEx_p, L"kernel32.dll", "GetFileInformationByHandleEx", \
	&_Not_supported_GetFileInformationByHandleEx))
	struct _GetFileInfoByHandleEx {
	const _GetFileInformationByHandleEx_t _Fn = __vcrt_GetFileInformationByHandleEx;
	[[nodiscard]] bool _Supported() const noexcept {
	return _Fn != _Not_supported_GetFileInformationByHandleEx;
	}
	[[nodiscard]] auto operator()(const HANDLE _Handle, const FILE_INFO_BY_HANDLE_CLASS _Class, void* const _Info,
	const unsigned long _Size) const noexcept {
	return _Fn(_Handle, _Class, _Info, _Size);
	}
	};
	#else // ^^^ _STL_WIN32_WINNT < _WIN32_WINNT_VISTA ^^^ // vvv _STL_WIN32_WINNT >= _WIN32_WINNT_VISTA vvv

Update awint.hpp / winapisupp.cpp

Note: the ABI compatibility for XP toolset should stay. The XP toolset should build; it just doesn't need to support Windows XP anymore.

[StephanTLavavej]

We definitely have to be extra-careful about preserving ABI compatibility here. Our object files, static libraries, import libraries, and DLLs don't have to worry about running on XP anymore, but we can't change any data structure representations that would be visible across boundaries, can't remove DLL exports, and can't alter DLL export behavior. Changing the version defined in CMakeLists.txt is the scariest in this sense - we would need to verify every possible effect it has. Altering individual functions (e.g. in winapisupp.cpp) is much easier to reason about.

There is also an argument that this isn't worth the effort and risk for the remaining life of v19.

[AlexGuteniev]

Removed CMakeFile.txt reference.
On the contrary, it is little performance thing: call by stored pointer invokes control flow guard overhead, whereas call by import address table does not.

[StephanTLavavej]

That's a good point - both perf and code size (customers love code size improvements and that is worth some level of effort / risk).

[MikeGitb]

Any chance Vista support will be dropped for VNEXT considering that market share for vista seems to be even lower than XP (not sure how reliable those online market share statistics at statcounter of netmarketshare are)?

[AlexGuteniev]

Dropping Vista enables only one static import, though an important one: TryAcquireSRWLockExclusive

[sylveon]

Might as well go for supported operating systems only and target 8.1 as minimum ;)

In more serious talk, I don't think the STL team can make that decision - it's the VS team which will decide the minimum supported OS.

[MikeGitb]

In more serious talk, I don't think the STL team can make that decision - it's the VS team which will decide the minimum supported OS.

This wasn't meant as a request from my side but asking for a prediction of what the upper echolons may decide.

[StephanTLavavej]

I have argued very strongly, and will continue to do so, that vNext should drop Vista support, because the ability to use SRWLOCK unconditionally is too important. (It is also long out of support, and never loved, making the situation very different from XP.) Given how long it's been since we started vNext planning, I don't think this is controversial anymore.