TITLE: Virtual inheritance and the open-closed principle

PROBLEM: 

              A       <-- base class
             / \
            /   \
           B     C
            \   /
             \ /
              C       <-- derived class


RESPONSE: fjh@munta.cs.mu.OZ.AU (Fergus James HENDERSON), 18 Jun 93

[...] although any D is-a C, and any C is-an A, it is not the case that
any D is-an A: in fact, any D is actually *two* As. Thus non-virtual
inheritance doesn't correctly model sub-typing. If sub-typing is
incorrectly modelled, then there is no way that identity can be correctly
modelled.


RESPONSE: maxtal@physics.su.OZ.AU (John Max Skaller)

	That was one of the major conclusions reached by Markku Sakkinen.
(Sakkinen proposed that the 'default' modes of inhertance be
virtual for public inheritance, and non-virtual for private inheritance)

It is why I have been questioning use of 'ordinary' public inheritance:
it doesn't model 'isA' relations, and doesnt represent subclassing
(in general). When it does, it usually does so by breaking the
open-ness part of the open/closed principle (it only works
if the programmer does not derive further classes).

	It is why I recommend that, almost always for abstract
classes, derivation should be virtual.


[ The open/closed principle was espoused by Meyer in "Object-Oriented
  Software Construction". It basically states that a class should not
  have to be modified (it is closed) in order to extend it by
  subclassing (openness). This is a worthwhile rule-of-thumb to use
  in designing classes. -adc ]