Re: PATCH for c++/15764 (adds new pre-lowering EH optimization pass)

[Jason Merrill <jason at redhat dot com>]

Richard Guenther wrote:
> This seems to break the gmp testsuite which now complains about non-freed
> memory.  In fact, for a reduced t-assign.cc
>
> void
> check_mpq (void)
> {
>   // operator=(const mpq_class &)
>   {
>     mpq_class a(1, 2), b;
>     b = a; ASSERT_ALWAYS(b == 0.5);
>   }
> }

Thanks for bringing this to my attention, but this isn't a useful 
testcase; there's no definition of mpq_class.  As it happens, I have GMP 
sources lying around, but next time please send a complete testcase.

Jason