A SEGV when compiling OpenOrb

[minyard at acm dot org]

I had a SEGV in a piece of code that I couldn't give out (and couldn't
easily isolate the problem with), but I've manage to reproduce the
same problem with OpenOrb.  This is on Linux x86, Linux PPC didn't
have the same problems (but has some other problems, maybe related).

To reproduce, download OpenOrb from
  http://ftp.openorb.org/ftp/openorb/jars/1_1_0/OpenORB-1.1.0.tgz

tar xvzf OpenORB-1.1.0.tgz OpenORB-1.1.0/lib/openorb-1.1.0.jar
jar xf OpenORB-1.1.0/lib/openorb-1.1.0.jar
gcj -c -g -fPIC -O3 org/openorb/CORBA/dynany/DynAnyImpl.class

This is only with -O3, -O2 or less passes.

Running it in GDB reveals:

Program received signal SIGSEGV, Segmentation fault.
0x821d587 in ggc_set_mark (p=0x12400a8) at ../../gcc-3.0/gcc/ggc-page.c:439
439	  return base[L1][L2];
(gdb) l
434	
435	  /* Extract the level 1 and 2 indicies.  */
436	  L1 = LOOKUP_L1 (p);
437	  L2 = LOOKUP_L2 (p);
438	
439	  return base[L1][L2];
440	}
441	
442	/* Set the page table entry for a page.  */
443	

I'm seeing some wierd stuff with this (I'm up late because it's
driving me a little crazy :-).  I have two compile builds/installs,
one will compile it ok, the other won't.  They are both exactly the
same except for the file locations.  Plus, if I specify a -fclasspath
with the one that did compile it (even though it is the default
classpath), it has the same failure.  Since this seems to be in the
garbage collector, maybe that's not surprising.

Does anyone have any insight on this?

Also, should I be reporting these things here, or should I use GNATS?

Thanks,

-Corey