Path: utzoo!mnetor!uunet!lll-winken!lll-lcc!mordor!sri-spam!ames!pasteur!ucbvax!denali.UUCP!bill
From: bill@denali.UUCP (Bill Meine [Sun Rocky Mtn TSE])
Newsgroups: comp.windows.news
Subject: Re:  Problems with window/resource loops..
Message-ID: <8804202342.AA04039@denali.sunpeaks.uucp>
Date: 21 Apr 88 08:03:15 GMT
Sender: daemon@ucbvax.BERKELEY.EDU
Organization: The Internet
Lines: 26

I've had this problem with my code too.  I ended up putting "destroy"
methods in every object that keeps a handle on another object in it's
own class or instance slots.  These methods set any object-handle
values in their object (dictionary) to nulls.  They also usually send
destroy messages to any other objects that they link to (setting up
entire graphs for scavenging).

Another way I found to avoid the problem is to keep only the name of
other objects instead of their handle.  Then, when a method needs to
reference that object, it looks it up first.  The technique is more
computationally intensive, but always works.

I have also wondered if expressing interest in an event will create a
back-reference in the event handler that would keep an object around,
but I haven't devised a test for that case yet.

Sorry to say, I don't know of any way to take a direct look at the heap
to try to track down the offending objects that have cyclic references.
Avoidance is the only policy I have found to work so far.

What would be nice is a garbage collector that can detect
disconnected, self-referencing graphs and collect the whole mass.  Many
Lisp systems have (mostly) solved this problem already.
 
-Bill

(sorry about not including the original question)