Ville Syrjälä schrieb:
> On Sun, Jan 07, 2007 at 07:19:57PM +0100, Denis Oliver Kropp wrote:
>> Ville Syrjälä wrote:
>>> Not sure. A core dump might at least give me a better idea where to
>>> start looking...
>> I think I could run it remotely in a gdb.
>
> The core dump problem was me logging out and ulimit -c resetting itself
> to 0 :)
>
> I tried to look at this a bit but the real bug still eludes me.
Thanks for your investigations so far.
> This ugly hack eliminates the crash, but causes a double free on exit.
>
> --- o/xorg-server-1.1.1/xkb/xkb.c 2006-07-05 21:38:48.000000000 +0300
> +++ xorg-server-1.1.1/xkb/xkb.c 2007-01-07 21:18:55.000000000 +0200
> @@ -2910,6 +2910,9 @@
> if (!sli)
> return BadAlloc;
>
> + if (!sli->maps)//
> + sli->maps = xkbi->desc->indicators->maps;//
> +
> from = (xkbIndicatorMapWireDesc *)&stuff[1];
> for (i=0,bit=1;i<XkbNumIndicators;i++,bit<<=1) {
> if (stuff->which&bit) {
>
> Something manages to clear the maps pointer. It is allocated just fine
> and it is fine during some other xkb calls. I didn't figure out what
> clears the pointer. I added printfs to all obvious places but that
> didn't reveal anything. Xorg server built from the exact same source
> tree works fine. gdb + watchpoint could help here.
Some entries have six symbols instead of four. I think the old code
was assuming four or was it independent? Array overrun could explain
why the pointer are cleared.
> I also found another bug. The code uses XKB_BASE_DIRECTORY but configure
> doesn't turn that into an absolute path. That caused some xkb warnings
> during server startup. I fixed that by passing --with-xkb-path=... to
> configure.
Maybe I added a symbolic link to get around it...
Isn't it possible to build a server completely without XKB?
The option "--disable-xkb" didn't help.
--
Best regards,
Denis Oliver Kropp
.------------------------------------------.
| DirectFB - Hardware accelerated graphics |
| http://www.directfb.org/ |
"------------------------------------------"
_______________________________________________
directfb-dev mailing list
[email protected]
http://mail.directfb.org/cgi-bin/mailman/listinfo/directfb-dev
