Sender:	nate AT cartsys DOT com
Message-ID:	<35A2F203.8687D81D@cartsys.com>
Date:	Tue, 07 Jul 1998 21:13:55 -0700
From:	Nate Eldredge <nate AT cartsys DOT com>
MIME-Version:	1.0
To:	malfer AT teleline DOT es
CC:	djgpp AT delorie DOT com
Subject:	Re: About DJGPP v2.02
References:	<35A2AA9E DOT 7CE4 AT teleline DOT es>

Mariano Alvarez Fernandez wrote:
> 
> Nate wrote:
> 
> >> Is there any date or plan to release DJGPP v2.02?
> >
> >Real Soon Now :)
> 
> It's sound great!!!

That's a joke; it's the standard response anytime someone asks when
something will be ready, and nobody knows.  However, it has now been
frozen, so no new features are going to be added to 2.02; only bug
fixes.  That probably means release is not too far away.  (But I still
don't know just how far.)

> >> However I tried the alpha 2.02 (980101) and my application run fast
> >> and (apparently) reliable, I don't find bugs at the moment.
> >
> >You don't want to release any code linked with that libc; it has a bug
> >whereby all DJGPP programs will instantly crash under Windows.  There
> >are several others as well.  The latest alpha (980628) is better, but
> >still not perfect.
> 
> Yes, it crashed under Windows, I didn't try it. Oops!
> I will try the latest alpha.

It's not recommended to use for production code, but there's nothing to
stop you from doing so.  Use it at your own risk!

> >Perhaps you can profile the code and find just what was so slow in the
> >previous version, then ask specifically.
> 
> Yes, I profiled it and all the time is wasted whit __dpmi_int, as the
> 13.5 in the FAQ says.

If you get a profiling libc (2.01 with bugs fixed) from

http://www.cartsys.com/eldredge/djgpp-patches.html

you will be able to tell what is calling `__dpmi_int' for so much time.

Do you read a lot from a file, or seek a lot?  2.02 uses a different
algorithm which may speed that up, but I wouldn't expect it to be that
drastic.

> This is a comparation among versions of a simple test of my program
> (recalculating the cost of 400 pieces of furniture):
> 
> 16-bit version (BC 3.0) 20"
> 32-bit version (DJGPP 2.01) 32"
> 32-bit version (DJGPP 2.02 980101) 11"
> 
> The 16 and 32 bit are identical, except I use (in the 32 bit version) a
> big memory hash table to remember register key position, the 32 bit
> version is very slow without that.
> 
> Regards, M.Alvarez (sorry, again, my bad english)

-- 

Nate Eldredge
nate AT cartsys DOT com

- Raw text -