Better to just convert WTF-16 (Windows filenames re not guaranteed to be valid UTF-16) to/from WTF-8 at the API boundary and then do the same processing internally on all platforms.