From: sandmann AT clio DOT rice DOT edu (Charles Sandmann)
Message-Id: <10206141524.AA16081@clio.rice.edu>
Subject: Re: unixy sbrk and win2k
To: djgpp-workers AT delorie DOT com
Date: Fri, 14 Jun 2002 10:24:40 -0500 (CDT)
In-Reply-To: <3D0913C9.5419EBCE@phekda.freeserve.co.uk> from "Richard Dawe" at Jun 13, 2002 10:51:05 PM
X-Mailer: ELM [version 2.5 PL2]
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Reply-To: djgpp-workers AT delorie DOT com
Errors-To: nobody AT delorie DOT com
X-Mailing-List: djgpp-workers AT delorie DOT com
X-Unsubscribes-To: listserv AT delorie DOT com
Precedence: bulk

> Emacs run from cmd.exe does not crash on start-up, when I hold down Ctrl. I
> does bomb out after pressing Ctrl 20-30 times, say (maybe more, maybe less).
> NTVDM disappears from the Task Manager. I don't get a traceback.

Interesting information.  Since NTVDM dies, that really points to a
Windows bug.  From the types of bugs we've seen in the past, we know that
nesting confuses it at times; I also know it doesn't like hardware 
interrupts and real mode callbacks very much.  This test really points 
the finger at the enable/disable interrupt problem around the sbrk().
But I wouldn't expect EMACS to be calling sbrk() when CTRL is being
pressed?  

> Emacs run directly from a shortcut seems pretty stable. I spent a couple
> minutes doing operations involving Ctrl and it did not crash.

I would not expect a difference from launched from a command window and
from a shortcut.  (Launched from CMD is not nesting).  If it makes a
difference that points to memory layout?

> This is all with Unixy sbrk. If I disable Unixy sbrk, then Emacs works fine.

This points to the arena possibly being moved; since the crash happens
with keyboard presses (hardware interrupts) this seems to point to the
section of code we were test patching.  It's possible the get/set
DPMI calls themselves (since they don't work) could also cause the
crash.

> So, is this a nesting problem? What else can I do to help diagnose the
> problem?

The only other thing I can think of is to disable the Int 9 keyboard
hooking and see if the problem goes away; double checking the cli was
actually linked in the binary image on the patched version that didn't
help.