Path: utzoo!censor!geac!torsqnt!news-server.csri.toronto.edu!cs.utexas.edu!milano!cadillac!lynx!franke
From: franke@lynx.cad.mcc.com (David Franke)
Newsgroups: comp.object
Subject: Re: Examples of Multiple Inheritance?
Message-ID: <13688@cadillac.CAD.MCC.COM>
Date: 5 Dec 90 15:22:29 GMT
References: <60700005@inmet>
Sender: news@cadillac.CAD.MCC.COM
Reply-To: franke%cad@MCC.COM (David Franke)
Organization: MCC CAD Program, Austin, TX
Lines: 59

In article <60700005@inmet> stt@inmet.inmet.com writes:
>
>We are looking for a few good examples of multiple inheritance
>in the programming language (especially C++) arena (as opposed to OODB,
>OOD, OOA arenas)
> ...
>We are looking for an example of a C++ class (or Eiffel, Objective C,
>etc.) which is a "true" sub-class of two (or more) parent classes;
>that is, it bears the "is-a" relationship to multiple parents.

I believe that I can provide such an example.  I have been interested
in imbedding rule inferencing in CAD tools, have developed an
objected-oriented approach, and have implemented this approach in C++.
(An article, "Embedding Rule Inferencing in Applications" will appear
in Dec. 90 IEEE Expert).

The basic idea is that an inference-object class is defined that a tool
developer can include (derive from) when creating classes needed for the
tool.  While the tool views instances of the derived class for its
purposes (e.g. schematic editing, timing analysis), the inference engine 
can view these same instances for its purpose.  As you state, this
derived class "is-a" tool object and "is-a" inference engine object.

>Our hypothesis is that multiple inheritance can rarely be useful
>unless planned for in advance, which seems to defeat the purpose.

It is certainly true in this case that the multiple inheritance was
planned in advance.  The primary motivation has been the ability to
integrate the inferencing capability into an existing tool (developed
in C++) and tools under development with the least possible impact
on tools defined classes.  Multiple inheritance provides such a
mechanism to accomplish this integration.

>Multiple inheritance seems to add significant overhead to an OOPL even
>when not being used in a system, so we are looking for convincing
>examples that it's benefits outweigh the cost.

IMO, multiple inheritance solves a significant problem in the application
I have described.  Specifically, the problem is that not all classes
defined by a tool will be of interest in inferencing, and hence need not
derive from the inference-object class.  If the class hierarchy defined
by the tool is very rich, one does not want to reference the 
inference-object class from one of the most general classes in the
hierarchy (i.e. use only single inheritance), since it (the inference-
object class) will probably participate in derived classes for which
there is no interest with respect to inferencing.  Multiple inheritance 
allows the tool developer to define an inferencing capability for only
those objects of interest, avoiding overhead in those objects for which
inferencing is not needed.

>S. Tucker Taft    
>Intermetrics, Inc.
>733 Concord Avenue
>Cambridge, MA  02138
>
>stt@inmet.inmet.com     ...!uunet!inmet!stt

 David Franke, MCC CAD Program | ARPA: franke@mcc.com | Phone: [512] 338-3641
 UUCP: {uunet,harvard,gatech,pyramid}!cs.utexas.edu!milano!cadillac!franke


Brought to you by Super Global Mega Corp .com