Say the string is \377\000, basename will use mbrlen() to check whether
it's a start of a multibyte sequence. While on Linux it'll return -1 for
any char in the extended ASCII, on Windows it's returning 1. From what I
see the reason is that Windows doesn't implement UTF-8 in the CRT lib,
it's rather 16-bit Unicode or DBCS. Since extended ASCII is convertable
to Unicode directly - thus the behavior. On Linux however, it's a true
UTF-8 locale and implementation, for it \377\000 is invalid.
Maybe mbrlen needs an independent implementation for Windows supporting
UTF-8. For now I just split out this case so the most of the big basename
test doesn't fail on this one case.
Anatol Belski