Re: [STUMP] stumpwm "crashing" X server

[Lionel Flandrin]

On Sat, Mar 05, 2011 at 07:14:21PM -0500, Mehul Sanghvi wrote:
> I have talked about this in the IRC channel a few times but so far I
> had not been able to solve this.
> What happens is the following:
> 
>      I start my X session using startx.  I have got stumpwm from the
> git repository and built it with clisp.
>      At times it will be a few hours, at other times it is pretty much
> with a few minutes, but opening/closing
>      some dialog boxes will almost certainly crash everything.  When I
> say "crash" what happens is that the entire window
>      manager crashes, taking down the X session and X server with it,
> putting me back at the console prompt.
>      There is a backtrace/dump that I can see on the console, but so
> far had not been able to
>       capture it because you can't scroll back on the console (well
> not that far back).
> 
> I used VNC today and ran stumpwm under there.  I was able to duplicate
> what happens _and_ able
> to capture all the output of the backtrace.  I have included the
> backtrace below.  If anyone has any thoughts
> and/or suggestions, please let me know.  Any help would be
> appreciated, and if I can narrow down the issue
> let me know.  Applications that I have used which cause this crash to
> occur are DBVisualiser (a Java app), Pidgin,
> TOra,  PCManFm/Thunar (one or the other or both, can't remember at this 
> point).
> 
> 
> ----------  BEGIN ERROR REPORT ----------
> *** glibc detected *** stumpwm: munmap_chunk(): invalid pointer: 0xb703f1ac 
> ***
> ======= Backtrace: =========
> /lib/i686/cmov/libc.so.6(+0x6b281)[0xb743a281]
> /lib/i686/cmov/libc.so.6(+0x6c4fe)[0xb743b4fe]
> /usr/lib/libX11.so.6(XFree+0x1d)[0xb7623d0d]
> stumpwm(C_subr_xlib_wm_hints+0xc0)[0x809116c]
  [snip remaining backtrace]

This is a glibc backtrace caused, it seems, by some memory
corruption. Since stumpwm is 100% lisp, there's no way it could have
caused that directly: it's either a bug in clisp or in libX11.so.6
(either clisp's clx passes a bogus pointer to the libX11 or the libX11
itself does something non-kosher with it). Besides, 0xb703f1ac looks
like a pointer to the stack, I can't imagine any scenario where such a
pointer would make sense for an munmap.

Maybe you could ask the clisp folks what they think about it, and/or
rebuild clisp and/or your libX11 with debuging symbols and run stumpwm
in gdb. If you know some C you could probably pinpoint the origin of
the problem that way.

At any rate, I don't this kind of corruption can not be directly
imputed on stumpwm, the underlying lisp process should never allow
this kind of corruption to happen.

Cheers,
-- 
Lionel Flandrin

pgpY6UGSTJstu.pgp
Description: PGP signature