Path: utzoo!utgpu!water!watmath!clyde!att!osu-cis!tut.cis.ohio-state.edu!mailrus!ames!pasteur!ucbvax!decwrl!nsc!voder!apple!lsr
From: lsr@Apple.COM (Larry Rosenstein)
Newsgroups: comp.sys.mac.programmer
Subject: Re: Inheritance in The Crunched Shell
Keywords: Object-oriented programming
Message-ID: <14934@apple.Apple.COM>
Date: 29 Jul 88 16:09:21 GMT
References: <2328@pt.cs.cmu.edu> <12362@agate.BERKELEY.EDU> <60865@sun.uucp> <12405@agate.BERKELEY.EDU>
Reply-To: lsr@apple.apple.com.UUCP (Larry Rosenstein)
Organization: Advanced Technology Group, Apple Computer
Lines: 39

In article <12405@agate.BERKELEY.EDU> elcond@garnet.berkeley.edu (Gregory Dow) writes:
>
>The SuperClass ordering is one problem (depth-first or breadth-first search),
>the other is more mundane.  Message names are implemented as an enumerated
>type, which C automatically converts to integer values.  The root object
>class has 3 messages, which get numbered as 0, 1, and 2.  If Class A,
>a child of the root class, has 4 messages, they get numbered 3, 4, 5, and 6.

This is the dispatching scheme we used in the very first implementation of
Object Pascal (nee Clascal), and it is used in the single-inheritance C++.
It has the virtue of being very fast, because it pre-computes the address of
the code for each message understood by a class.  The time to call a method
is constant.

When you make a subclass, the subclass' table starts out as a copy of the
superclass' table.  You add any new methods onto the end, and replace any
entries that are overridden.  So even if you override 1 method, you still
have to copy the entire superclass table.

The current Object Pascal dispatcher minimizes the space by only putting the
methods a class actually implements into the table.  For info on this, see
the December 1986 MacTutor.

>Can anyone suggest an alternative?  I'd love it if I could easily implement
>the "set" data structure from Pascal in C.  Then I could just take the
>union of the messages of the superclasses of a class.  I'd really like
>to find a clean solution to this problem, because standard C++ doesn't
>support multiple inheritance either.

I think Stroustrup had a paper in the C++ Workshop Proceedings on multiple
inheritance in C++.  AT&T also had the paper in its booth at the last
OOPSLA.  It retains the constant-time method call feature of C++, with only
a little increase in space and complexity.

		 Larry Rosenstein,  Object Specialist
 Apple Computer, Inc.  20525 Mariani Ave, MS 46-B  Cupertino, CA 95014
	    AppleLink:Rosenstein1    domain:lsr@Apple.COM
		UUCP:{sun,voder,nsc,decwrl}!apple!lsr