sjlj-exceptions ICE. simplified test case

[Jason Merrill]

>>>>> Richard Henderson <rth@cygnus.com> writes:

 >> Another issue is that in some cases it might be necessary to have support
 >> for propagating exceptions through foreign code without any tables at all.

 > Yeah, I suppose.

We ought have a fallback unwinder, perhaps based on gdb code.  For SPARC
this would be trivial, and remove the need for any unwind information in
most cases.

 >> At least on SGI it used to be the case that during program initialization
 >> all addresses in unwind tables were patched up, which caused excessive
 >> startup delays because of swapping...

 > Not fixed, but fixable.  I've made suggestions before when asked, but
 > no one's ever implemented it.  Perhaps something to make sure to get
 > done before gcc3 and the c++ abi change...

Absolutely.  What were the suggestions again?  The ia64 unwind info avoids
this by putting the tables in text and using self-relative offsets.

Jason