Re: Cygwin 1.7 wide char functions

[Bruno Haible]

Eric,

More details about this one:
> > checking whether mbrtowc has a correct return value... no
> 
> This is the first one the Cygwin developers should take care of.
> The test case is in m4/mbrtowc.m4 lines 260..296.
> 
> > ../../gltests/test-wcrtomb.c:51: assertion failed
> > ../../gltests/test-wcrtomb.sh: line 25:  9880 Aborted                 (core 
> > dumped) LC_ALL=$LOCALE_JA ./test-wcrtomb${EXEEXT} 3
> > FAIL: test-wcrtomb.sh
> 
> Looks like a problem with Cygwin's EUC-JP decoder.
> 
> > ../../gltests/test-mbrtowc.c:240: assertion failed
> > ../../gltests/test-mbrtowc3.sh: line 15: 19980 Aborted                 
> > (core 
> > dumped) LC_ALL=$LOCALE_JA ./test-mbrtowc${EXEEXT} 3
> > FAIL: test-mbrtowc3.sh
> 
> Likewise.

The setlocale function [1] installs function pointers to functions
__eucjp_wctomb and __eucjp_mbtowc, which are implemented in [2] and [3].

A second bug in Cygwin's EUC-JP decoder is that in [3] there is a comment
  "Cygwin defines its own doublebyte charset conversion functions 
   because the underlying OS requires wchar_t == UTF-16."
(and yes indeed, on Windows, the system calls expect UTF-16 encoded content
in wchar_t* strings), but the __eucjp_wctomb and __eucjp_mbtowc functions
return 16 bits combined from the two 8-bit values of the multibyte character -
without any conversion between JISX0208/JISX0212 and Unicode.

These functions need to be completely rewritten.

Bruno

[1] 
http://cygwin.com/cgi-bin/cvsweb.cgi/src/newlib/libc/locale/locale.c?rev=1.22&content-type=text/x-cvsweb-markup&cvsroot=src
[2] 
http://cygwin.com/cgi-bin/cvsweb.cgi/src/newlib/libc/stdlib/wctomb_r.c?rev=1.14&content-type=text/x-cvsweb-markup&cvsroot=src
[3] 
http://cygwin.com/cgi-bin/cvsweb.cgi/src/newlib/libc/stdlib/mbtowc_r.c?rev=1.13&content-type=text/x-cvsweb-markup&cvsroot=src