Re: msgfmt -cv aborts with no message at all

[Bruno Haible]

Jochen Hein wrote:
> I'm doing a "msgfmt -cv " on the attached file.  This file probably
> contains broken UTF-8 encoding, which I tried to catch that way.
> ...
> #1  0x40075872 in raise () from /lib/libc.so.6
> #2  0x40076986 in abort () from /lib/libc.so.6
> #3  0x40023986 in po_callback_comment () from

Thanks for reporting this bug. The appended patch fixes it.

> msgfmt should at least give an informative message and point to the
> file (and character) where the encoding is broken.

When a program calls abort(), it's an indicator of a bug in the program.
This was the case here. The encoding of the file is not broken, it
"just" contains an UTF-8 encoded character > 0x10FFFF; this is valid
according to the system's (glibc's) iconv module for UTF-8.

Bruno


diff -c -3 -r1.7 -r1.8
*** gettext-tools/src/po-lex.c  9 Sep 2003 13:40:36 -0000       1.7
--- gettext-tools/src/po-lex.c  11 Sep 2003 16:40:32 -0000      1.8
***************
*** 511,518 ****
                abort ();
              /* Convert it from UTF-8 to UCS-4.  */
              mbc->uc_valid = true;
!             if (u8_mbtouc (&mbc->uc, scratchbuf, outbytes) != outbytes)
!               abort ();
              break;
            }
        }
--- 511,521 ----
                abort ();
              /* Convert it from UTF-8 to UCS-4.  */
              mbc->uc_valid = true;
!             /* We ignore the return value of u8_mbtouc(): Usually it returns
!                outbytes, but if scratchbuf contains an out-of-range Unicode
!                character (> 0x10ffff), it can also return 1 and set mbc->uc
!                to 0xfffd.  This is precisely what we need.  */
!             u8_mbtouc (&mbc->uc, scratchbuf, outbytes);
              break;
            }
        }