John,

First off, thank you for the very insightful analysis. It certainly has
helped me clarify my thinking on the whole problem.

> For example, have you seen similar crashes on several different
> machines?

No, but haven't really tried that. A colleague of mine did see similar
crashes, but he had copied an image from me.

I've have been rebuilding the image from the distribution periodically.
Pretty much have to the way new versions come out. :-)  I suppose it's
possible that it may be a problem with the machine. I'll rearrange how I
work on Squeak to use a different machine, and see if I still run into
troubles that way.

> Have you had problems with other apps crashing mysteriously?

In general my system has been pretty stable. But Microsoft has probably
warped my perspective: I don't always have the greatest expectations of
Windows applications on that score. I did have problems with an older
version of Netscape, but that seems to have cleared up with 4.05.

> Is there any corrolation between Squeak crashing and having some other
> application running?

Not that I've managed to notice. Might be hard to correlate if there is a
big lag between the image getting corrupted and a problem showing up.

Also I may be living in a dream world, but aren't applications pretty well
protected from each other when running in NT?


> Have you been using Squeak in unusual ways, or using primitives that
> might not be too heavily exercised?

Not in any obvious way that I know of. But I do seem to find cracks in any
software that I use. Must be a personality thing. Guess I just have an
optimistic view of UI's where I think that any sequence of clicks, drags
and changes should be OK (at least until I learn better).

> (b) it doesn't seem reproducible (your current image is already damaged,
> but we don't know a sequence of steps that causes it to become damaged)

I agree: It does seem though that the image is already damaged in some
way. Is there any kind of a utility that can walk the image in a fairly
thorough fashion looking for damage? Periodic usage of such a tool would
be helpful in reducing the time between damage occurring to an image and
finding out about it. (Given that is what's occurring). That could really
help in determining the actual cause.

In the image that I got a snapshot of, the problem apparently manifests
itself during a #become: operation. But a garbage collection at the same
point doesn't seem to have a problem. If there is a wild pointer involved,
why doesn't GC have a problem? Any insights as to how that might be
explained? Hopefully Andreas' and/or Dan's postmortems can help answer
some of these questions.

Speaking of Andreas and Dan, I really appreciate their willingness to take
a look at my problem. Thanks. Even though it is starting to sound more
like *my* problem, given no one else is getting crashes like this. And
when the autopsy report comes back with "cause of death: damaged image",
but without being able to finger the perpetrator, I'll just need to learn
how to setup some VM surveillance so that I can catch 'im in the act
myself.

-Bill

-------------------------------------------
Bill Dargel            wdargel at shoshana.com
Shoshana Technologies
100 West Joy Road, Ann Arbor, MI 48105  USA