Re: Binutils 2.20 - Windres problems building pthreads/libiconv on Open Solaris

[Dave Korn <dave dot korn dot cygwin at googlemail dot com>]

On 10/02/2010 11:17, Tony Theodore wrote:

> Program received signal SIGSEGV, Segmentation fault.
> 0xcea46832 in iconv () from /lib/libc.so.1
> #0  0xcea46832 in iconv () from /lib/libc.so.1
> #1  0x08071dbd in iconv_onechar (cd=0xffffffff, s=0x81053d0 "FileDescription",
>     d=0x8046940 "\022", d_len=32, n_s=0xffffffff, n_d=0xffffffff)
>     at winduni.c:633
> #2  0x08071fc3 in wind_MultiByteToWideChar (cp=4294967295,
>     mb=0x81053d0 "FileDescription", u=0x0, u_len=0) at winduni.c:694
> #3  0x08072052 in unicode_from_codepage (length=0x0, u=0x81053d0,
>     src=0x81053d0 "FileDescription", cp=0) at winduni.c:584
> #4  0x080720ae in unicode_from_ascii (length=0x0, unicode=0x8046a28,
>     ascii=0x81053d0 "FileDescription") at winduni.c:194
> #5  0x0806e74f in yyparse () at rcparse.y:1703
> #6  0x08064c2e in read_rc_file (filename=0x8047ad7 "version.rc",
>     preprocessor=0x80e002c "gcc -E -xc -DRC_INVOKED",
>     preprocargs=0x80dfd80 "", language=1033, use_temp_file=0) at resrc.c:562
> #7  0x08063aee in main (argc=3, argv=0x8047970) at windres.c:1025

  I've reproduced this on osol0906jeos in a VM.  What happens is that here:

> static rc_uint_type
> wind_MultiByteToWideChar (rc_uint_type cp, const char *mb,
> 			  unichar *u, rc_uint_type u_len)
> {
>   rc_uint_type ret = 0;
> 
> #if defined (_WIN32) || defined (__CYGWIN__)
   [ ... elided ... ]
> #elif defined (HAVE_ICONV)
>   int first = 1;
>   char tmp[32];
>   char *p_tmp;
>   const char *iconv_name = wind_iconv_cp (cp);
> 
>   if (!mb || !iconv_name)
>     return 0;
>   iconv_t cd = iconv_open ("UTF-16", iconv_name);

... this call to iconv_open fails (iconv_name is "MS-ANSI", returning
(iconv_t)-1 ...

>   while (1)
>     {
>       int iret;
>       const char *n_mb = "";
>       char *n_tmp = "";
> 
>       p_tmp = tmp;
>       iret = iconv_onechar (cd, (ICONV_CONST char *) mb, p_tmp, 32, & n_mb, & n_tmp);

... that we promptly go and use as if it were valid.  Hmm, that's obviously
not good.  Aborting with a "invalid code page" error of some kind is one thing
we could do.  But the other thing I'm wondering is if MS-ANSI isn't just a
slightly non-standard name for CP1252, so maybe we could change the default to
use that one (or perhaps attempt it as a secondary fallback).  I'm not very
familiar with iconv stuff, so I'll have to do some rtfming to make sure I've
got the right end of the stick, but one reference I found(*) suggests that my
be the case.

    cheers,
      DaveK
-- 
(*) - http://www.firstobject.com/character-set-name-alias-code-page.htm