I had something similar a while back and the following additional flag
to gcc was suggested:

-fsjlj-exceptions

Cheers,

Patrick

crvich@us.ibm.com wrote:
> 
> >I'm by no means a c++ person, i just thought i'd do some more
> >investigation.  Sounds like there are some Issues, perhaps with EGCS
> 
> Thanks for confirming that it's reproducible elsewhere.  I was sort of
> hoping it was just that I needed some additional flag or cr*0.o object.
> We're sufficing with a static library for the time being.  The exes are
> huge, but they pass our full test suite and our NetBSD customers have their
> port.  We'll continue trying to get a shared lib version working as time
> permits, though...we have a lot of exes, so in addition to the size
> reduction we'd like to get, it bites to have to relink all of them every
> time we make a change in the lib.
> 
> >Have you tried other versions of the compiler?
> 
> gcc 2.7.x isn't up to the task (internal compiler errors galore), and 2.8.1
> gave us some unexplainable runtime problems.  gcc 2.9.1 (egcs 1.1.2) has
> been reliable, so we've stuck with that.  I tried building and using gcc
> 2.9.5 a few weeks ago, but it didn't cooperate well with the NetBSD 1.3.3
> assembler and linker, so I gave up...I'll probably try again at some point.