fix c++ gc problems following t_r_o_c

[Geoff Keating]

Richard Henderson <rth@redhat.com> writes:

> Hopefully, anyway.  I never saw the problem in the first place, and I'm
> not patient enough to wait for a gcac build.  The analysis seems plausible,
> however.  Thanks to geoffk for identifying the failing call stack.
> 
> 
> r~
> 
> 
>         * optimize.c (optimize_function): Push/pop ggc context around
>         the call to optimize_inline_calls.

I have a better fix undergoing a gcac build now, I'll back this out as
part of it, it turns out that we have deeper problems in the C++
frontend and in fact the first call stack I provided is not the one
triggering the problem on Darwin (although it still shows a bug).

I would strongly encourage people to run gcac builds occasionally.
With the new GC heuristics, GC hardly ever runs in a normal bootstrap
and testsuite run on a big machine, so any GC bugs introduced won't be
noticed until some poor user with a 32Mb box tries to build something.

Maybe we should have a 'gcsc', GC Sometimes Collect, which can be on
by default in enable-checking builds and doesn't cost quite so much as
gcac.

-- 
- Geoffrey Keating <geoffk@geoffk.org>