Updated files

[Coleen Phillimore]

I attached my comments on construction vtables.  I won't be able to
listen to the con-call tomorrow (and definitely not fly out).

Have fun,
Coleen
-------------- next part --------------


>From the ctorvtbls writeup:

> c.Ed. Note: Don't we need the equivalent of secondary vptrs here, too? 

Yes, maybe it would be better to eliminate a,b,c and say in 2:

>2.Secondary VTTs: for each non-virtual subobject B of D, that requires a
> construction vtable by virtue of having virtual bases, a sub-VTT for
> B-in-D, which contains: <insert change color here> address in #1, #2, #3.

Note that step #2 is recursive on #1, #2 and #3.  This might simplify
the description somewhat.

Then for #3 put the better sentence from 2b "Subobject A of B needs a
construction ..."

Note this changes the definition slightly.  There is only a need for a
secondary VTT if the nonvirtual subobject has a virtual base (not if it
has an overridden virtual function).  There is a need for the secondary ptr
in the VTT (step #3) if either the base has virtual base or overridden virtual
functions.

The example:

I changed the example slightly to add 'int' members to all the classes, so 
there isn't so much sharing to confuse the issue.  I also added 'virtual'
for member function 'f' which I think was intended.  The example's hard to
follow, but I think it's mostly right, see my comments below.
>
> // 1. Primary vptr:
>  D * has virtual bases
>
>  // 2. Secondary VTTs:
>    C1 * has virtual base
>      V1-in-C1 in D secondary vptr
V1 is a virtual base class, so it's not in VTT
>        A1-in-...-C2 in D secondary vptr
nonvirtual base in virtual base class V1, not in VTT
>
>    C2 * has virtual bases
>      V2-in-C2 in D secondary vptr *
>        V1-in-V2-in-C2 in D secondary vptr *
Yes, this is right. V1 and V2 secondary vptrs for C2 need
construction vtables, because V2 has virtual bases and V1 has
virtual calls overridden on virtual path.
I think the nesting should be the same.
>          A1-in-...-C2 in D secondary vptr *
This one is shared by V1, so not in VTT.
>
>  // 3. Secondary vptrs:
>    C1-in-D secondary vptr
This is shared with D, so not in VTT.
>      V1-in-D secondary vptr *
>        A1-in-V1-in-D secondary vptr (shared with V1)
>    C2-in-D secondary vptr
>      V2-in-D secondary vptr
>
>  // 4. Virtual VTTs:
>    (V1 can use complete object constructor).
>    V2 * has virtual base, has overridden f()
>      V1-in-V2 in D secondary VTT
>        A1-in-V1 * has overridden f()
>      V1-in-V2 in D secondary vptr
>        A1-in-V1-in-V2 in D secondary vptr
>
I think this is right.

So, I have 10 entries in the VTT for D (ignore external names):

    static const long *__VTT_1D[8] =
    {__7__vtbl_1D,  // primary complete vptr for D
// Secondary VTT for C1
     __7__vtbl_2C1_2C1__1D,  // construction vtbl for C1 in D
// Secondary VTT for C2
     __7__vtbl_2C2_2C2__1D,  // construction vtbl for C2 in D
     __7__vtbl_2V2_2C2__1D,  // construction vtbl for V2 in C in D
     __7__vtbl_2V1_2C2__1D,  // construction vtbl for V1 in C in D
// secondary vptrs for D
     __7__vtbl_2C21D,
     __7__vtbl_2V21D, 
     __7__vtbl_2V11D, 
// virtual part
     __7__vtbl_2V2_2V2__1D, 
     __7__vtbl_2V1_2V2__1D};

The red in the 'Subobject Construction and Destruction' section looks
fine.

Coleen