COM compatibility

[Jim Dehnert]

Christophe de Dinechin wrote:
> 
> > A while ago Jason was worried about COM compatibility.
> > Part of that is to ensure that vtables can be expressed in C.
> > But the resolution of issue B-4 says that a vtable contains
> > function descriptors rather than function descriptor pointers.
> 
> Did we resolve that issue that way? I was quite convinced we had
> finally decided for "every pointer is a C-style pointer" precisely
> because of this kind of problems. Mmmh, I'd better check today with
> the rest of the team. Are there other cases where we have a function
> pointer that is not a C-style function pointer?
> 
> Christophe

I thought we had resolved it as:  If C uses function descriptors (i.e.
address/GP pairs), and C pointers point to them, we would put a function
descriptor in the Vtable; but if the C ABI were changed to not use
descriptors, we would put a simple pointer there.

(Since I think the latter is unlikely at this time, I tend to interpret
this as using descriptors.  I do think the savings of an indirection is
likely significant enough to make inability to directly express them in
C a worthwhile tradeoff.)