C++ PATCH: PR 16405

[Richard Kenner]

    Are the two versions of the type type_compatible_p?

No and they shouldn't be because that (as I understand it) is a symmetric
relation and it wouldn't be appropriate here.

I was previously taking the position that we should weaken the GIMPLE type
system to allow "compatible" types to be used interchangably, but I've since
been convinced that I was wrong and that the equivalence of types at the
GIMPLE level should be deterimined in a language-independent fashion by
seeing if two types would result in identical code, since that's what matters
for GIMPLE.  But this is also an issue for 4.1.

In the meantime, I've backed off and removed that langhook entirely for Ada
since it was causing problems.