Virtual Calls: Make the ABI match the implementations

[Jason Merrill]

I have two problems with this proposal:

1) In addition to the problem I mentioned before with weak followed by
   strong, your proposal would end up with two copies of the thunk in
   the executable: one weak copy in a COMDAT, and one strong outside.
   I've encountered this problem when I tried to use the same trick in
   gcc; I ended up giving up and not trying to generate strong
   definitions.

   One way to avoid this would be garbage collection, but that's not
   going to happen.

   Another way would be a 'choose me' bit in the COMDAT, but I believe
   that idea was rejected when I brought it up long ago.

2) I just don't see the point.  It doesn't seem hard to emit the
   appropriate thunks with the function; they don't need to be written
   to fall through, they just have to be in the same object file.

Jason