Fix filesystem's handling of nonexistent network paths emitting ERROR_INVALID_NAME

[StephanTLavavej]

On my machine, P0218R1_filesystem was failing whenever it tested nonexistent network paths. (I'm not sure why nobody else has observed this yet; I'm just running a normal Microsoft corpnet installation of Windows 10 Enterprise.)

I debugged into this, and observed that when __std_fs_get_stats() calls GetFileAttributesExW(), it fails with ERROR_INVALID_NAME:

STL/stl/src/filesystem.cpp

Lines 919 to 921 in 65d98ff

	if (!GetFileAttributesExW(_Path, GetFileExInfoStandard, &_Data)) {
	return __std_win_error{GetLastError()};
	}

This is returned unchanged by _File_size():

STL/stl/inc/filesystem

Lines 3377 to 3388 in 65d98ff

	_NODISCARD inline __std_win_error _File_size(const path& _Path, uintmax_t& _Result) noexcept {
	__std_fs_stats _Stats;
	const auto _Error = __std_fs_get_stats(
	_Path.c_str(), &_Stats, __std_fs_stats_flags::_Follow_symlinks | __std_fs_stats_flags::_File_size);
	if (_Error == __std_win_error::_Success) {
	_Result = _Stats._File_size;
	} else {
	_Result = static_cast<uintmax_t>(-1);
	}
	return _Error;
	}

The test expects file_size() to return bad_file_size, which it does, but it also expects ec == errc::no_such_file_or_directory, which fails:

STL/tests/std/tests/P0218R1_filesystem/test.cpp

Lines 2713 to 2716 in 65d98ff

	for (auto&& nonexistent : nonexistentPaths) {
	EXPECT(bad_file_size == file_size(nonexistent, ec));
	EXPECT(ec == errc::no_such_file_or_directory);
	}

This is because we map ERROR_INVALID_NAME to errc::invalid_argument:

STL/stl/src/syserror.cpp

Line 50 in 65d98ff

{ERROR_INVALID_NAME, errc::invalid_argument},

The Windows documentation for System Error Codes describes ERROR_INVALID_NAME as "The filename, directory name, or volume label syntax is incorrect." so this is specific to filesystem operations; it isn't a general error code for invalid arguments.

Additionally, __std_is_file_not_found() already considers __std_win_error::_Invalid_name (our synonym for ERROR_INVALID_NAME) to be a "file not found" error code:

STL/stl/inc/xfilesystem_abi.h

Lines 51 to 61 in 65d98ff

	_NODISCARD inline bool __std_is_file_not_found(const __std_win_error _Error) noexcept {
	switch (_Error) {
	case __std_win_error::_File_not_found:
	case __std_win_error::_Path_not_found:
	case __std_win_error::_Error_bad_netpath:
	case __std_win_error::_Invalid_name:
	return true;
	default:
	return false;
	}
	}

The other 3 error codes are already mapped to errc::no_such_file_or_directory:

STL/stl/src/syserror.cpp

Line 44 in 65d98ff

{ERROR_FILE_NOT_FOUND, errc::no_such_file_or_directory},

STL/stl/src/syserror.cpp

Line 64 in 65d98ff

{ERROR_PATH_NOT_FOUND, errc::no_such_file_or_directory},

STL/stl/src/syserror.cpp

Line 27 in 65d98ff

{ERROR_BAD_NETPATH, errc::no_such_file_or_directory},

Changing the mapping of ERROR_INVALID_NAME (and updating the test's direct inspection of ec.value()) fixes this bug.

This is a behavioral change (similar to #406 and #616, although they added new mappings). As this fixes the behavior of <filesystem>, I believe it is necessary.