Virtual Calls: Make the ABI match the implementations

[Jason Merrill]

>>>>> Mark Mitchell <mark at codesourcery.com> writes:

>>>>> "Jason" == Jason Merrill <jason at redhat.com> writes:
 Jason> Ah, I see.  I'm strongly opposed to requiring all compilers
 Jason> to do work to support their semi-compliant peers.

 > Even if that means that fewer compilers are compliant?  In pratice,
 > vendors have a unique opportunity here: because they are all shipping
 > new systems and new compilers, they can break the ABI.  They will not
 > want to do so subsequently.  They will not want to delay
 > time-to-market in order to provide ABI compliance -- most vendors will
 > not find the promise of interoperability that important.  So, the ABI
 > will become less valuable -- many systems will not comply.

 > This is a major philosophical question, and something that the ABI
 > committee should have a cohesive position on.

I agree, but it seems to me (judging from the amount of innovation in this
ABI) that we decided long ago not to give a lot of weight to difficulty of
implementation.  Perhaps that was the wrong decision, but it's a bit late
to reconsider now.

And I'm a bit puzzled why this issue is coming up with respect to this
particular detail; it seems pretty easy to determine what thunks to emit
with the function.  The fall-through implementation is hard, but we've
already agreed that isn't necessary for compliance (by the as-if rule).

Jason