This is the mail archive of the cygwin mailing list for the Cygwin project.

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: SOLVED: Removed 1.5.25 and installed 1.7.0, but still cannot access filenames containing Unicode

2009/11/2 Corinna Vinschen:
>> >> You must not use characters
>> >> in this range from U+f000 up to U+f0ff. ÂThere's no solution to this
>> >> except for "don't use these characters in filenames if they are not
>> >> explicitely written there by either Cygwin or Microsoft's SUA".
>> Actually there is a possible solution: when translating a U+F0xx
>> character, first check whether the xx byte really is illegal in the
>> target charset. If it's not, it won't roundtrip correctly, so encode
>> the U+F0xx as a ^X sequence instead. Doesn't seem worth the effort
>> though.
> I was contemplating this over the weekend. ÂI just applied a patch to
> do this. ÂI tested this with various filenames containing all sorts
> of characters, including f000, which would represent an ASCII NUL, if
> used wrongly.

I've had a look at the patch. It improves roundtrip transparency for
Windows filenames at the cost of reduced transparency for POSIX

Single U+F0xx's are now fine, but sequences of them still will not
necessarily roundtrip correctly, e.g., with a UTF-8 locale:

U+F0C3 U+F084 -> 0xC3 0x84 -> U+00C4  ('Ã')

And U+F0xx's on the POSIX side now won't roundtrip if they get mapped
to single bytes on the way back, e.g:

0xEF 0x80 0x8A -> U+F00A -> 0x0A  (newline)
0xEF 0x81 0xBC -> U+F07C -> 0x7C (pipe)
0xEF 0x82 0x80 -> U+F080 -> 0x80 (invalid UTF-8)


Problem reports:
Unsubscribe info:

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]