From adama@doppio Tue Sep 23 09:04:13 1997
Date: Tue, 23 Sep 1997 09:04:04 -0700
From: adama@doppio (Adam Abramski)
To: David.Bowen@Eng
Subject: Re: dual processor VM bug (?)


Hi David,

Here's more details regarding the VM bug.

If you need more information just let me know or you can work
through Tom Gullion if you wish.

Thanks again,
Adam

----- Begin Included Message -----

From ###@###.### Tue Sep 23 08:25:09 1997
Date: Tue, 23 Sep 1997 09:51:05 -0500
From: Tom Gullion <###@###.###>
Mime-Version: 1.0
To: Adam Abramski <Adam.Abramski@Eng>
Subject: Re: dual processor VM bug (?)
X-Priority: 3 (Normal)

Ok, here goes...

The bug is that Java windows (either application or applet) cease to
redraw after a few successful redraws on dual processor NT
machines. If you make the window redraw itself (by dragging parts
of it offscreen or by dragging another window over the top of it) it
can only do this successfully for a short time. Then its contents
get erased and never repaints itself. Resizing the window doesn't
refresh it either. The only way I found was to restart the app. Of
course you'll run into the same situation if the window gets redrawn
x number of times again. If you disable one of the cpu's, the
problem disappears.

I've been working on this problem for two weeks solid and I'll try
to list all of the diagnostic stuff I've done. I've also been in
contact with Microsoft tech support because I originally thought it
was a problem with NT.

I ruled out video cards/drivers because I swapped video cards (the
original Matrox Millenium for a Diamond Stealth 3D 3000) and the
problem persisted. After a long uphill battle I finally got
Microsoft to take the problem seriously and an engineer there found
a thread deadlock in their VM (JView) implementation of
Thread.start(). They wouldn't tell me the exact problem but I was
able to pry a bit. They told me one thread was looking through a
list of threads while another was trying to add the redraw thread
and they deadlocked. Makes sense - it only happens on dual
processor machines because those threads are actually happening
simultaeously.

This is still a little fuzzy because the two VM's repaint
differently. JView, the Microsoft VM, waits to repaint the screen
until you let go of the mouse while your VM repaints as soon as it
can. Somehow Microsoft VM is harder to kill in this way but it is
possible to make it happen. I've attached two simple java
applications using only AWT calls. The first one, DrawTestApp.java,
dies in the JavaSoft VM but not in the Microsoft VM. The second
one, AWTtest.java, dies in both. When I say die I mean they cease
to redraw. The application is still running underneath (as you can
tell by the System.out.println()'s). I've been using JDK 1.1.3 and
1.1.4 (problem is the same across both).

I think that's it. I'll add my phone number in case you have
further questions. Sorry, I've been through this scenario so many
times its hard to remember who I've told what!

Thanks A LOT for your help!!!!!
Tom Gullion
(312) 554-5928



Adam Abramski wrote:

> Hi Tom,
>
> I have an engineer who would like all the details and if you have
> a reproducible case he'd like that also. Any help you can give
> I'll be sure to pass along to him.
>
> Thanks again,
> Adam
>
> > From ###@###.### Mon Sep 22 13:26:40 1997
> > Date: Mon, 22 Sep 1997 15:06:40 -0500
> > From: Tom Gullion <###@###.###>
> > Mime-Version: 1.0
> > To: Adam Abramski <Adam.Abramski@Eng>
> > Subject: Re: dual processor VM bug (?)
> > X-Priority: 3 (Normal)
> >
> > Thanks! I just got in contact with our local Sun rep (Rich
> Miller -
> > ###@###.###) who is promising to put it in the right
>
> > hands. I sent him a rather detailed email complete with code
> > samples. So hopefully it will be smooth sailing from here on.
> >
> > Thanks again for your help,
> > Tom
> >
> >
> > Adam Abramski wrote:
> >
> > > Hi Tom,
> > >
> > > Great, actually submitting the bug is the best way so that
> > > it can be tracked and hopefully fixed.
> > >
> > > Thanks again for you help and assistance. I've forwarded your
>
> > > previous email to the AWT group and just forwarded your last
> > > email to the Product Manager for JDK asking him to forward it
> > > to the right VM person.
> > >
> > > Cheers,
> > > Adam
> > >
> > > > From ###@###.### Mon Sep 22 10:30:06
> 1997
> > > > Date: Mon, 22 Sep 1997 12:18:08 -0500
> > > > From: Tom Gullion <###@###.###>
> > > > Mime-Version: 1.0
> > > > To: Adam Abramski <Adam.Abramski@Eng>
> > > > Subject: Re: dual processor VM bug (?)
> > > > X-Priority: 3 (Normal)
> > > >
> > > > Yeah, I submitted one through the javasoft site. I didn't
> make
> > > > calls about it because I thought it was an NT bug and not a
> VM
> > > bug.
> > > > Is there something more official that I should be doing? I
> > > realize
> > > > you are not the guy I should be contacting about this but
> you're
> > > our
> > > > only contact there. Tell me what to do and/or who to
> contact
> > > and
> > > > I'll do it.
> > > >
> > > > Thanks,
> > > > Tom
> > > >
> > > > Adam Abramski wrote:
> > > >
> > > > > Hi Tom,
> > > > >
> > > > > Thanks for the heads up. I'll forward this to the VM
> group.
> > > Did
> > > > > you file a bug on this?
> > > > >
> > > > > Thanks again,
> > > > > Adam
> > > > >
> > > > > > From ###@###.### Mon Sep 22 08:16:38
>
> > > 1997
> > > > > > Date: Mon, 22 Sep 1997 09:33:42 -0500
> > > > > > From: Tom Gullion <###@###.###>
> > > > > > Mime-Version: 1.0
> > > > > > To: Adam Abramski <###@###.###>
> > > > > > Subject: dual processor VM bug (?)
> > > > > > X-Priority: 3 (Normal)
> > > > > >
> > > > > > Adam,
> > > > > >
> > > > > > I hope you remember - I'm the guy from Swissbank who was
>
> > > bugging
> > > > > you
> > > > > > about redraw problems on dual processor NT machines.
> I've
> > > been
> > > > > > hounding Microsoft because I thought it was an NT bug.
> > > Well, I
> > > > > just
> > > > > > got an email back from one of their VM engineers who
> said he
> > >
> > > > > just
> > > > > > found a thread deadlock in the VM code that was causing
> the
> > > > > problem!
> > > > > >
> > > > > > Could you please forward this on to your VM group? Its
> > > probably
> > > > > the
> > > > > > same problem for your VM. I'll be happy to provide with
>
> > > > > anything
> > > > > > and everything I have that could help out. If you don't
>
> > > have a
> > > > > dual
> > > > > > proc NT box, you're even welcome to come to Chicago and
> hang
> > > out
> > > > > and
> > > > > > debug the VM.
> > > > > >
> > > > > > Thanks,
> > > > > > Tom Gullion
> > > > > > Swissbank
> > > > > > 312 554-5928
> > > >
> > > >
> >
> >



----- End Included Message -----

From ###@###.### Tue Sep 23 11:35:17 1997
Date: Tue, 23 Sep 1997 13:21:21 -0500
From: Tom Gullion <###@###.###>
Mime-Version: 1.0
To: David Bowen <David.Bowen@Eng>
Subject: Re: dual processor VM bug (?)
X-Priority: 3 (Normal)

Hello Dave,

I'm attaching a cumbersome NT diagnostic report (isn't information
overload wonderful?).

The quick answer, though, is a Compaq Professional Workstation (with
dual Pentium Pro 200's, 128M RAM, etc.). NT 4.0, service pack 3,
without the Compaq SSD stuff. Compaq has a "Compaq-specific" NT
build. We're not using that - just plain, off-the-street NT. You
can see that in the report.

Thanks,
Tom

David Bowen wrote:

> Hi Tom,
>
> Just entering your bug into the database. What hardware version
> and
> NT version are you using?
>
> - Dave Bowen
> Manager, Java Language Group