On Tue, 26 Oct 1999 14:01:24   John Duncan wrote:

>>Is that right?  I was under the impression that
>>the object isn't purged until the gc occurs.
>>[Indeed, what is described is something more like
>>reference counting than gc].  Further, can't you
>>register objects for finalization, thus giving
>>each object a chance to say "I want to stay?" or
>>to copy itself somewhere -- or would that action
>>be beyond the scope of a valid finalization
>>action?

>The GC occurs in short, high-frequency bursts so
>that it's not a dog on response time. So, the
>effect of GC is almost immediate. But, if you
>imagine my example, you'll see that there are
>response-time benefits for not having to rerender
>the glyphs very often, so normal GC doesn't seem
>like the best way. Nor does traditional caching,
>because the data have properties that reflect
>temporal locality, and traditional caching will
>have a higher miss rate than some sort of more
>intelligent method. The best cache is a 1-1
>correspondence between cache locations and
>references to be used before the program switches
>to a different locale. This is the goal that I'm
>shooting for, but I'm trying to figure out where
>to start. I began prototyping, but then I came to
>this reference counting idea, and I think it has
>potential.

John, I recommend a careful look at the extensive work
done for Eiffel's memory reclamation and the issues
involved, in
<a href="http://www.eiffel.com/doc/oosc/page.html">OOSC-2</a>,
 about pages 304-314.  Also, we're fortunate that a
lot of this stuff
<a href="http://www.eiffel.com/doc/manuals/technology/internal/gc/page.html">is
 online</a>, too.


[snip]


>:)

>-John.


---
________________________________________________
Jan Theodore Galkowski  demiourgos at smalltalk.org
home.stny.rr.com/algebraist/         squeak.org/
www.smalltalk.org/


Get your FREE, private e-mail account at http://www.salonmember.com.