Vbase offsets

[Mark Mitchell]

>>>>> "Jim" == Jim Dehnert <dehnert at baalbek.engr.sgi.com> writes:

    Jim> I don't know about you, but we are.  This paragraph didn't
    Jim> get fixed when we decided to always include vbase offsets for
    Jim> all virtual bases.  I've fixed it now (with the example) --
    Jim> everyone please verify that it matches our decisions.

OK.  But, given:

  struct S {
    virtual void f();
  };
  struct T : virtual public S {};
  struct U : public T {};
  
Are there two vbase offsets for `S' in the vtable for `U'?  Here `T'
is the primary base for `U', so the vtable for `U' already has an `S'
entry.  I assume we don't want to add an extra one.  Maybe that
example would be helpful, too?  What if `T' is inherited virtually?
Then, `T' is still the primary base of `U', so I don't think we need
an extra offset here either?  If I'm right, that's probably worth
mentioning too.  So, I think the rule is:

  If there is no primary base class, then there is a vbase offset for
  every direct or indirect virtual base.  Otherwise, a vbase offset is
  present in the derived class for every virtual base that does not
  already have a vbase offset in the primary base class vtable.  In
  either case, the vbase offsets are allocated in reverse inheritance
  graph order.

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