[UI] Well, shall we do something then?

[Bill Schwab]

Dave,

This sounds like a problem with objects not being gc'd.  There are
subtle problems with one of the of the #when:* variants - IIRC, it's
#when:evaluate:, that will result in entire webs of objects not being
collected.  OA found and partially fixed, partially designed around it
in Dolphin, and we have had no such problems.  This is starting to sound
like an as-yet unidentified bug that is causing people to shun a very
nice design.

Bill




Wilhelm K. Schwab, Ph.D.
University of Florida
Department of Anesthesiology
PO Box 100254
Gainesville, FL 32610-0254

Email: bschwab at anest.ufl.edu
Tel: (352) 846-1285
FAX: (352) 392-7029

>>> lewis at mail.msen.com 09/19/07 11:24 PM >>>
I don't know what a "usual load" is. The effect would be be cumulative
for all of the event-driven thingamabobs in all of the projects in an
entire image. I'm guessing that the nature of the beast that that the
total number of these things could accumulate in an image over time, and
sooner or later (probably sooner) you would start fielding questions
from folks whose images had slowed to a crawl for "no reason at all".

I ran into this issue in an experimental version of my CommandShell,
in which I had tried to move the event driven stuff to use
#when:send:to:.
The net result was that when I created command pipelines with lots
of event-driven activity, I ended up with an image that moved like
molasses. I had to go back and change everything to use the old
update protocol, which turned out to be perfectly satisfactory for
my needs, and which did not have a performance problem.

HTH,
Dave

On Wed, Sep 19, 2007 at 08:29:22PM -0400, Bill Schwab wrote:
> Dave,
> 
> I take your word for it.  I would hope that with my usual load of
> browsers, inspectors, and a debugger or two, it would be ok.  Anything
> that gets into thousands of widgets, I would long-since have been
> emulating.  Is that realistic, or is worse than that?
> 
> Bill
> 
> 
> 
> 
> Wilhelm K. Schwab, Ph.D.
> University of Florida
> Department of Anesthesiology
> PO Box 100254
> Gainesville, FL 32610-0254
> 
> Email: bschwab at anest.ufl.edu
> Tel: (352) 846-1285
> FAX: (352) 392-7029
> 
> 
> >>> lewis at mail.msen.com 9/19/2007 4:03:29 PM >>>
> On Wed, Sep 19, 2007 at 08:47:18AM -0400, Bill Schwab wrote:
> > Is the weak dictionary problem unique to Squeak?  Dolphin seems to
> get
> > along just fine with MVP.  OTOH, what you describe could be a part
> of
> > why I emulate for large grids, etc., or it could be a "SquWeakness"
> that
> > needs attention.  I use events with the emulated framework, but
there
> is
> > also some 
> 
> I don't know about other Smalltalks, but it is an issue for Squeak.
> The original update mechanism is perfectly fine by the way.
> 
> > Years ago now, I made inquires about Squeak's process
> syncrhonization
> > objects, and scaling of threads, and was regaled with stories of
> great
> > victories over the dragons of software complexity ;)  More recently,
> I
> > am getting the sense that (e.g. "we almost had to rewrite in Java")
> some
> > of this stuff is just plain broke.  It is unclear whether it really
> is
> > sound, or was sound and has since gotten into trouble, etc.
> 
> It's not broken, it just does not scale with respect to performance,
> and
> that fact is not well known or documented.
> 
> Dave
> 
> _______________________________________________
> UI mailing list
> UI at lists.squeakfoundation.org
> http://lists.squeakfoundation.org/mailman/listinfo/ui
> _______________________________________________
> UI mailing list
> UI at lists.squeakfoundation.org
> http://lists.squeakfoundation.org/mailman/listinfo/ui
_______________________________________________
UI mailing list
UI at lists.squeakfoundation.org
http://lists.squeakfoundation.org/mailman/listinfo/ui