Re: SIGSEGV in msgmerge

[Bruno Haible]

Jan-Marek Glogowski writes:

> When running msgmerge on the attached PO files I get a SIGSEGV.
> 
> Version: 0.11.2
> 
> Command: msgmerge de.old de.ref -o de.po
> 
> I don't have time to investigate this problem, so I just attached
> a backtrace and the POs

Thanks a lot for the report. This patch fixes it. Changing the charset
field from iso9959-1 to the correct iso-8859-1 also avoids the problem.

Bruno


*** gettext-0/src/write-po.c    2002-04-27 12:36:39.000000000 +0200
--- gettext-1/src/write-po.c    2002-05-14 12:02:58.000000000 +0200
***************
*** 247,252 ****
--- 247,253 ----
       enum is_wrap do_wrap;
       const char *charset;
  {
+   const char *canon_charset;
    const char *s;
    bool first_line;
  #if HAVE_ICONV
***************
*** 255,260 ****
--- 256,263 ----
  #endif
    bool weird_cjk;
  
+   canon_charset = po_charset_canonicalize (charset);
+ 
  #if HAVE_ICONV
    /* The old Solaris/openwin msgfmt and GNU msgfmt <= 0.10.35 don't know
       about multibyte encodings, and require a spurious backslash after
***************
*** 267,286 ****
      /* Write a PO file in old format, with extraneous backslashes.  */
      conv = (iconv_t)(-1);
    else
!     /* Avoid glibc-2.1 bug with EUC-KR.  */
! # if (__GLIBC__ - 0 == 2 && __GLIBC_MINOR__ - 0 <= 1) && !defined 
_LIBICONV_VERSION
!     if (strcmp (charset, "EUC-KR") == 0)
        conv = (iconv_t)(-1);
      else
  # endif
!     /* Use iconv() to parse multibyte characters.  */
!     conv = iconv_open ("UTF-8", charset);
  
    if (conv != (iconv_t)(-1))
      weird_cjk = false;
    else
  #endif
!     weird_cjk = po_is_charset_weird_cjk (po_charset_canonicalize (charset));
  
    /* Loop over the '\n' delimited portions of value.  */
    s = value;
--- 270,299 ----
      /* Write a PO file in old format, with extraneous backslashes.  */
      conv = (iconv_t)(-1);
    else
!     if (canon_charset == NULL)
!       /* Invalid PO file encoding.  */
        conv = (iconv_t)(-1);
      else
+       /* Avoid glibc-2.1 bug with EUC-KR.  */
+ # if (__GLIBC__ - 0 == 2 && __GLIBC_MINOR__ - 0 <= 1) && !defined 
_LIBICONV_VERSION
+       if (strcmp (canon_charset, "EUC-KR") == 0)
+       conv = (iconv_t)(-1);
+       else
  # endif
!       /* Use iconv() to parse multibyte characters.  */
!       conv = iconv_open ("UTF-8", canon_charset);
  
    if (conv != (iconv_t)(-1))
      weird_cjk = false;
    else
  #endif
!     if (canon_charset == NULL)
!       weird_cjk = false;
!     else
!       weird_cjk = po_is_charset_weird_cjk (canon_charset);
! 
!   if (canon_charset == NULL)
!     canon_charset = po_charset_ascii;
  
    /* Loop over the '\n' delimited portions of value.  */
    s = value;
***************
*** 528,534 ****
  
        /* Do line breaking on the portion.  */
        mbs_width_linebreaks (portion, portion_len, width, startcol, 0,
!                           overrides, charset, linebreaks);
  
        /* If this is the first line, and we are not using the indented
         style, and the line would wrap, then use an empty first line
--- 541,547 ----
  
        /* Do line breaking on the portion.  */
        mbs_width_linebreaks (portion, portion_len, width, startcol, 0,
!                           overrides, canon_charset, linebreaks);
  
        /* If this is the first line, and we are not using the indented
         style, and the line would wrap, then use an empty first line