Itanium C++ ABI primary virtual bases - c++

Question

I was reading here about how primary bases are chosen:
"...2. If C is a dynamic class type:
a. Identify all virtual base classes, direct or indirect, that are primary base classes for some other direct or indirect base class. Call these indirect primary base classes.
b. If C has a dynamic base class, attempt to choose a primary base class B. It is the first (in direct base class order) non-virtual dynamic base class, if one exists. Otherwise, it is a nearly empty virtual base class, the first one in (preorder) inheritance graph order which is not an indirect primary base class if any exist, or just the first one if they are all indirect primaries..."
And after there is this correction:
"Case (2b) above is now considered to be an error in the design. The use of the first indirect primary base class as the derived class' primary base does not save any space in the object, and will cause some duplication of virtual function pointers in the additional copy of the base classes virtual table.
The benefit is that using the derived class virtual pointer as the base class virtual pointer will often save a load, and no adjustment to the this pointer will be required for calls to its virtual functions.
It was thought that 2b would allow the compiler to avoid adjusting this in some cases, but this was incorrect, as the virtual function call algorithm requires that the function be looked up through a pointer to a class that defines the function, not one that just inherits it. Removing that requirement would not be a good idea, as there would then no longer be a way to emit all thunks with the functions they jump to. For instance, consider this example:
struct A { virtual void f(); };
struct B : virtual public A { int i; };
struct C : virtual public A { int j; };
struct D : public B, public C {};
When B and C are declared, A is a primary base in each case, so although vcall offsets are allocated in the A-in-B and A-in-C vtables, no this adjustment is required and no thunk is generated. However, inside D objects, A is no longer a primary base of C, so if we allowed calls to C::f() to use the copy of A's vtable in the C subobject, we would need to adjust this from C* to B::A*, which would require a third-party thunk. Since we require that a call to C::f() first convert to A*, C-in-D's copy of A's vtable is never referenced, so this is not necessary."
Could you please explain with an example what this refers to: "Removing that requirement would not be a good idea, as there would then no longer be a way to emit all thunks with the functions they jump to"?
Also, what are third-party thunks?
I do not understand either what the quoted example tries to show.

Answer 1

A is a nearly empty class, one that contains only a vptr and no visible data members:
struct A { virtual void f(); };
The layout of A is:
A_vtable *vptr
B has a single nearly empty base class used as a "primary":
struct B : virtual public A { int i; };
It means that the layout of B begins with the layout of an A, so that a pointer to a B is a pointer to an A (in assembly language). Layout of B subobject:
B_vtable *A_vptr
int i
A_vptr will point to a B vtable obviously, which is binary compatible with A vtable.
The B_vtable extends the A_vtable, adding all necessary information to navigate to the virtual base class A.
Layout of B complete object:
A base_subobject
int i
And same for C:
C_vtable *A_vptr
int j
Layout of C complete object:
A base_subobject
int j
In D obviously there is only an A subobject, so the layout of a complete object is:
A base_subobject
int i
not(A) not(base_subobject) aka (C::A)_vptr
int j
not(A) is the representation of an A nearly empty base class, that is, a vptr for A, but not a true A subobject: it looks like an A but the visible A is two words above. It's a ghost A!
(C::A)_vptr is the vptr to vtable with layout vtable for C (so also one with layout vtable for A), but for a C subobject where A is not finally a primary base: the C subobject has lost the privilege to host the A base class. So obviously the virtual calls through (C::A)_vptr to virtual functions defined A (there is only one: A::f()) need a this ajustement, with a thunk "C::A::f()" that receives a pointer to not(base_subobject) and adjusts it to the real base_subobject of type A that is the (two words above in the example). (Or if there is an overrider in D, so the D object that is at the exact same address, two words above in the example.)
So given these definitions:
struct A { virtual void f(); };
struct B : virtual public A { int i; };
struct C : virtual public A { int j; };
struct D : public B, public C {};
should the use of a ghost lvalue of a non existant A primary base work?
D d;
C *volatile cp = &d;
A *volatile ghost_ap = reinterpret_cast<A*> (cp);
ghost_ap->f(); // use the vptr of C::A: safe?
(volatile used here to avoid propagation of type knowledge by the compiler)
If for lvalues of C, for a virtual functions that is inherited from A, the call is done via the the C vptr that is also a C::A vptr because A is a "static" primary base of C, then the code should work, because a thunk has been generated that goes from C to D.
In practice it doesn't seem to work with GCC, but if you add an overrider in C:
struct C : virtual public A {
int j;
virtual void f()
{
std::cout << "C:f() \n";
}
};
it works because such concrete function is in the vtable of vtable of C::A.
Even with just a pure virtual overrider:
struct C : virtual public A {
int j;
virtual void f() = 0;
};
and a concrete overrider in D, it also works: the pure virtual override is enough to have the proper entry in the vtable of C::A.
Test code: http://codepad.org/AzmN2Xeh