From: Ville Hallik <ville@tartu.cyber.ee>
  To  : ggi-develop@eskimo.com
  Date: Tue, 13 Jul 1999 09:57:12 +0300 (EEST)

Re: hacking methods to find out bugs

On Tue, 13 Jul 1999, Andrew Apted wrote:

> Christoph Egger writes:
> 
> >  On Mon, 12 Jul 1999, Andrew Apted wrote:
> >  
> >  > Or run in X and use gdb (hacking libgg so that it doesn't catch SIGSEGV
> >  > and prevent a core dump).
> >  
> >  With "hacking a lib so that it doesn't catch SIGSEGV" you mean I should
> >  fix the bug to find the bug?
>  
> Nah.  Normally on UNIX, a SIGSEGV signal will produce a core dump. 
> The LibGG library (in the LibGII tarball) catches that signal (etc) so
> that graphics can be restored sanely.  But this means no core dump.

But You can raise SIGSEGV again after cleanup and voila: coredump is here.
This works at least for the following simple program under glibc2.1
(core shows exactly the right crashpoint: *p = 'a'; ):

#include <stdio.h>
#include <signal.h>

void crash( int sig ) {
  fprintf( stderr, "SIGSEGV received\n" );
  signal( SIGSEGV, SIG_DFL );
  raise( sig );
}

int main() {
  char *p = NULL;  
  fprintf( stderr, "Establishing signal handler\n" );
  signal( SIGSEGV, crash );
  fprintf( stderr, "Crashing program\n" );
  *p = 'a';
  fprintf( stderr, "This message will never be seen\n" );
  return 0;
}

-- 

Ville Hallik   +372-52-77891
http://home.cyber.ee/~ville/