WriteFile UTF-8 console path uses ACP conversion and misreports partial writes

[739C1AE2]

In support/thunk/kernel32/u/WriteFile.cc, the console code path converts the input buffer using:

mingw-lite/support/thunk/kernel32/u/WriteFile.cc

Lines 34 to 35 in 1e051ec

	if (!w_buffer.from_a(static_cast<const char *>(lpBuffer),
	nNumberOfBytesToWrite)) {

which decodes the input using CP_ACP.

However, when handling a partial write, the code converts the wide string back using:

mingw-lite/support/thunk/kernel32/u/WriteFile.cc

Line 49 in 1e051ec

if (!u_written.from_w(w_buffer.c_str())) {

and u_str::from_w() encodes using UTF-8.

This appears inconsistent: the input buffer is decoded using ACP (from_a()), while the byte count is calculated using UTF-8 (u_str::from_w()).

Additionally, in the partial-write path:

mingw-lite/support/thunk/kernel32/u/WriteFile.cc

Lines 48 to 53 in 1e051ec

	d::u_str u_written;
	if (!u_written.from_w(w_buffer.c_str())) {
	SetLastError(ERROR_OUTOFMEMORY);
	return FALSE;
	}
	*lpNumberOfBytesWritten = u_written.size();

converts the entire buffer instead of only the first w_written characters reported by WriteConsoleW(), so lpNumberOfBytesWritten may not reflect the actual number of bytes written.