TITLE: Q & A on the value of MI


IC = ichudov@wiltel.com (Igor Chudov)
RM = rmartin@rcmcon.com (Robert Martin), 2 Dec 94


IC: Do you use multiple inheritance in C++? 

RM: Yes, often.

IC: Do you have any benchmarks on how much slower is access to elements 
    of classes declared virtual?

RM: I have never noticed any efficiency issues.  My reading indicates that
    MI is no slower than normal inheritance unless there is a virtual base
    class, and then the difference is one pointer dereference.  (i.e. not
    big)

IC: Do you have any horror stories on how MI lead you to bad designs/
    maintenance frustrations/compiler bugs?

RM: No.

IC: Do you have any examples when MI was absolutely necessary and useful?

RM: Indeed.  It is often the case where one object must respond to more
    than one interface, and yet those two interfaces have clients that
    should not know about each other.  MI is the only solution for such
    problems.  And they are very frequent.

    Consider, I have my own exception model.  Owl has its own exception
    model.  Owl forces you to use OWL exceptions.  i.e. you must derive
    your exceptions from TxOwl.  I don't want my modules to depend upon
    OWL (i.e. I don't want them in #include any owl headers).  So I create
    my own abstract excecption class:  AbstractException.  I invoke an
    abstract factory to create AbstractException objects.  

    Then I derive the implemented exception both from AbstractException
    and TxOwl.  The result is that the bulk of my code remains independent
    of OWL even though it is using exceptions that OWL can deal with.

    This is not an isolated case.  Such connundrums come up very
    frequently in OOD and MI provides a very clean solution to them.