Eliot & Linda
elcm at pacbell.net
Sat Feb 28 17:15:00 UTC 1998
Tim Olson wrote:
> Hi, all:
> Here's a cute trick / crufty hack I recently posted to
> comp.lang.smalltalk. It was in regards to someone complaining that they
> had to recreate a bunch of methods when using a delegate to implement
> multiple-inheritance-like operation. Instead of manually doing this
> (painful) or dynamically doing this by implementing doesNotUnderstand:
> (slow), why not combine the two and have the doesNotUnderstand: method
> install the custom-created method?
The technique was the basis of Dan Ingalls' and Allan Borning's multiple
inheritance system for Smalltalk-80 v2.0. An instance of a class with
multiple superclasses checked its extra superclasses for methods in
doesNotUnderstand: and recompiled them in its own class so that the next
time it was sent there would be no doesNotUnderstand:. The scheme
maintained a dependency graph so that if one changed a method (either
modified or deleted) that was copied all the copies would be removed.
When a class with multiple superclasses was first created for all
selectors that had more than one reachable implementation, i.e.
ambiguous methods, the class would create a method that did ^self
conflictingInheritanceError. The programmer could then resolve the
ambiguities as required.
The implementation was buggy; self was not renamed when copying down;
conflicting inheritance methods overrode the resolutions if you
redefined the class; redefining superclasses could cause bugs due to the
traversal machinery expecting a tree and not a graph, etc.
Evewn with these bugs fixed the general feeling was that it was too
complex. The best alternative I've seen is Gilad Bracha's mixin-based
multiple inheritance scheme.
More information about the Squeak-dev