Re: [bug #17647] glibc: GCC 4.1

[Thomas Schwinge]

Hello!

On Sun, Dec 17, 2006 at 03:53:56PM +0100, I wrote:
> On Sun, Nov 26, 2006 at 11:27:58PM +0100, I wrote:
> > [Building glibc with GCC 4.1.]
> > 
> > And in fact also glibc-proper code is affected:
> > 
> > #v+
> > thomas@leibniz:/var/tmp/glibc-GCC4.1-without_Roland_patches/glibc-2.3.6.ds1/build-tree/hurd-i386-libc
> >  $ LD_LIBRARY_PATH=. gdb --args locale/localedef --help
> > GNU gdb 6.5-debian
> > [...]
> > This GDB was configured as "i486-gnu"...
> > (gdb) r
> > Starting program: 
> > /var/tmp/glibc-GCC4.1-without_Roland_patches/glibc-2.3.6.ds1/build-tree/hurd-i386-libc/locale/localedef
> >  --help
> > Usage: localedef [OPTION...] NAME
> >   or:  localedef [OPTION...] [--add-to-archive|--delete-from-archive] 
> > FILE...
> >   or:  localedef [OPTION...] --list-archive [FILE]
> > Compile locale specification
> > 
> >  Input Files:
> >   -f, --charmap=FILE         Symbolic character names defined in FILE
> >   -i, --inputfile=FILE       Source definitions are found in FILE
> > 
> > Program received signal SIGSEGV, Segmentation fault.
> > 0x011389a8 in __argp_fmtstream_update (fs=0x8098158) at argp-fmtstream.c:232
> > 232               while (p >= buf && !isblank (*p))
> > (gdb) print p
> > $1 = 0x80981e9 "values\n\n Input Files:\n  -f, e] FILE...\n", '\223' 
> > <repeats 88 times>, "???"
> > (gdb) print *p
> > $2 = 118 'v'
> > (gdb) bt
> > #0  0x011389a8 in __argp_fmtstream_update (fs=0x8098158) at 
> > argp-fmtstream.c:232
> > #1  0x0113cc74 in _help (argp=0x1019ab0, state=0x1019ba4, stream=0x1188be0, 
> > flags=<value optimized out>, name=0x101a05e "localedef")
> >     at argp-fmtstream.h:255
> > #2  0x0113d31e in __argp_state_help (state=0x1019ba4, stream=0x80981e9, 
> > flags=634) at argp-help.c:1730
> > #3  0x0113d7a8 in argp_default_parser (key=63, arg=0x0, state=0x1019ba4) at 
> > argp-parse.c:119
> > #4  0x0113d595 in group_parse (group=0x8097efc, state=0x1019ba4, key=63, 
> > arg=0x0) at argp-parse.c:257
> > #5  0x0113df93 in __argp_parse (argp=0x1019ab0, argc=2, argv=0x1019db4, 
> > flags=0, end_index=0x1019d58, input=0x0) at argp-parse.c:755
> > #6  0x0804a854 in main (argc=2, argv=0x1019db4) at programs/localedef.c:206
> > #v-
> > 
> > Any offhand guesses what's going on there?  I would expect nothing in the
> > argp code to be Hurd specific and I'm not aware of any Linux system
> > having such a problem.
> 
> I can add another bit of information here: pay attention to `LC_ALL'.
> 
> #v+
> [...]
> thomas@leibniz:/var/tmp/glibc-GCC4.1-without_Roland_patches/glibc-2.3.6.ds1/build-tree/hurd-i386-libc
>  $ LC_ALL=C LD_LIBRARY_PATH=. locale/localedef --help
> Usage: localedef [OPTION...] NAME
>   or:  localedef [OPTION...] [--add-to-archive|--delete-from-archive] FILE...
>   or:  localedef [OPTION...] --list-archive [FILE]
> Compile locale specification
> 
>  Input Files:
>   -f, --charmap=FILE         Symbolic character names defined in FILE
>   -i, --inputfile=FILE       Source definitions are found in FILE
> Segmentation fault
> #v-
> 
> I.e. invalid values for `LC_ALL' work, as well as localized ones, only
> the `C' one makes it segfault.  Due to the situation with glibc HEAD
> being unbuildable at the moment, I can't test that.  Any ideas?

Rebuilding the `argp/' files with `-O' instead of `-O2' makes the problem
go away.  Further tests revealed that the `-O2'-built `argp-fmtstream.c'
is faulty.


Regards,
 Thomas

signature.asc
Description: Digital signature