On Sat, Nov 05, 2005 at 10:44:00PM +0100, Pierre THIERRY wrote:
> Scribit Matthew Palmer dies 05/11/2005 hora 11:02:
> > Also, Pierre, you might like to try installing the libc6-dbg package,
>
> Here's the backtrace with LD_LIBRARY_PATH=/usr/lib/debug:
>
> (gdb) run
> Starting program: /usr/games/xconq
>
> Program received signal SIGSEGV, Segmentation fault.
> 0xb7c7a740 in *__GI___strcasecmp (s1=0x0, s2=0xb7ee0028 "symbol") at
> ../sysdeps/generic/strcasecmp.c:65
> 65 ../sysdeps/generic/strcasecmp.c: Aucun fichier ou répertoire de ce
> type.
> in ../sysdeps/generic/strcasecmp.c
And we have our bug. strcasecmp() is getting a NULL pointer sent to it, and
it is understandably not happy. I think it's time to start looking into
what the Tcl/Tk libraries are up to. Problem is, I can't find a call to
strcasecmp() in Tk_DrawChars(). Definitely time to build -dbg versions of
Tcl/Tk.
See #337747 and #337749 for patches to build {tcl,tk}8.3-dbg packages for
install.
> > It's a pity that the Tcl/Tk libraries don't provide debug symbols as
> > well; it's relatively straightforward to produce debug symbol
> > packages, though.
>
> Maybe one could file a bug for it...
*cough*
> Is there a policy on wether it is a Good Thing to include the unstripped
> version in the Debian archive? (maybe it's not when the program has no
> particular problem when running, but it always is for libraries...)
I reckon it should be serious-bug policy to fail to provide lib*-dbg
packages; it's just Too Damn Hard to debug C programs otherwise. As for
regular binaries, I'd say it's worthwhile for fairly alpha/beta quality
programs, or those with a known history of strange corner cases causing
segfaults and other painful phenomena; your ordinary run-of-the-mill
production quality binary isn't going to need debug symbols often enough to
make it worthwhile, and it's fairly easy to rebuild a single binary package
with debug support enabled anyway for the very, very rare occasion.
- Matt
