X-Authentication-Warning:	delorie.com: mail set sender to djgpp-bounces using -f
From:	Charles Sandmann <sandmann AT clio DOT rice DOT edu>
Newsgroups:	comp.os.msdos.djgpp
Subject:	Re: CWSDPMI r5 update test
Date:	Sat, 16 Jun 2007 22:05:39 CDT
Organization:	Rice University, Houston, TX
Lines:	26
Message-ID:	<4674a503.sandmann@clio.rice.edu>
References:	<4670269e$0$6399$9b4e6d93 AT newsspool2 DOT arcor-online DOT net>
NNTP-Posting-Host:	clio.rice.edu
X-Trace:	joe.rice.edu 1182050296 14440 128.42.105.3 (17 Jun 2007 03:18:16 GMT)
X-Complaints-To:	abuse AT rice DOT edu
NNTP-Posting-Date:	Sun, 17 Jun 2007 03:18:16 +0000 (UTC)
X-NewsEditor:	ED-1.5.9
To:	djgpp AT delorie DOT com
DJ-Gateway:	from newsgroup comp.os.msdos.djgpp
Reply-To:	djgpp AT delorie DOT com

> > r5 (or the update) is fine for 99.99% of the other usages.
> 
> What about `bigstack.zip', `csdpmi6t.zip', `psefix.zip', or
> `stacksiz.zip' available from <http://clio.rice.edu/djgpp/>?
> 
> When do I need these? I used `bigstack.zip' to get a game (DJGPP +
> Allegro) running in Bochs 2.3. Otherwise CWSDPMI loops forever, because
> _oldint2f was overwritten and points to _dpmiint2f?

The DPMI specification is very clear about stack usage - if the the 
interrupt requires more than the specification provides for the application
should allocate a separate stack.  This is more of a programming pain
than many developers want to go to, and Windows provides a stack which 
is much larger than the default.

CWSDPMI provides about 2X what the specification specifies.  If I expand
the stack size to what windows provides then the DOS memory footprint 
needs to increase another 4KB to 8KB (depending on how many nested
programs you decide to support).  This breaks some environments.

So there is no good fix here - it's been discussed and tested before.
We looked at making it run-time configurable - but that code was less
forgiving if you overran the stack (typically a machine hang).  I provided
other versions with bigger stacks - some people were happy, other people
had problems and wanted it put back.  In the end the DPMI specification
won the argument.

- Raw text -