Thunk error in specification

[Mark Mitchell]

Section 3.2.3 of the ABI specification indicates what thunks should be
emitted with a particular function.

While constructing our ABI testsuite, we have noticed a problem
with this language:

  For each direct or indirect virtual base V of C such that either V
  declares f, or is derived from a class that declares f, the compiler
  must emit, in the same object file as the code for C::f, a
  V-adjusting entry point for C::f.

This sentence should be revised to remove the "or is derived from a class
that declares f" clause.  An example of where this matters is:

  struct A { virtual void f(); };

  struct B { virtual void g(); };

  struct C : public A, public B { }:

  struct D : virtual public C { virtual void g(); };

There is no point in creating a C::g thunk to convert from a C* to a
D* since C's vtable will not contain an entry for g.  (You do, of
course, need a thunk that converts from a B* to a C*, and then
adjusts by using the appropriate vcall offset in C's virtual table.)

(This language is a hold-over from an earlier draft of the ABI where
C's vtable *did* have an entry for g.)

As a result of removing the clause mentioned above, this language in the
next paragraph should also be removed:

  (Note that one implementation of the M-adjusting entry point is to
  convert from M* to V* and then transfer control to the V-adjusting
  entry point.)

There is no longer any guarantee that the V-adjusting entry point exists.

I will make this change in a couple of days, unless anyone objects.

-- 
Mark Mitchell                mark at codesourcery.com
CodeSourcery, LLC            http://www.codesourcery.com