Seems the test for MultiByteToWideChar() == 0xFFFD can be dropped.

The MultiByteToWideChar doc mentions U+FFFD under Return Value. But when I test bad UTF-8 characters, it inserts 0xFFFD for the bad character(s) in the output string. It doesn't return 0xFFFD. Or if we pass the MB_ERR_INVALID_CHARS flag, it returns 0 and sets GetLastError() as usual.

(I wonder if I had a hand in this code - looks like my kind of thing to write a draft from the Windows doc but leave the Windows testing to someone else:-)