Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Posting-Version: $Revision: 1.6.2.16 $; site ima.UUCP
Path: utzoo!watmath!clyde!bonnie!akgua!gatech!seismo!harvard!bbnccv!ima!johnl
From: johnl@ima.UUCP
Newsgroups: net.micro
Subject: Re: Re: 386 Family Products (8086 sucks
Message-ID: <97800015@ima.UUCP>
Date: Tue, 10-Dec-85 16:14:00 EST
Article-I.D.: ima.97800015
Posted: Tue Dec 10 16:14:00 1985
Date-Received: Fri, 13-Dec-85 07:30:19 EST
References: <133@daisy.UUCP>
Lines: 38
Nf-ID: #R:daisy:-13300:ima:97800015:000:2051
Nf-From: ima!johnl    Dec 10 16:14:00 1985


/* Written  2:16 pm  Dec  8, 1985 by brad@looking in ima:net.micro */
> In article <97800013@ima.UUCP> johnl@ima.UUCP writes:
> >Oh, stop it.  It's just not true that existing compilers can hide the 
> >808x's segmentation.  I have just spent the last year and a half writing a 
> I'll correct you, because you are wrong. ...
> 
> You appear (correct *me* if I'm wrong) to be confusing "dealing with
> segmentation" with "writing good C programs."
> 
> Now I will tell you that we did have lots of segment headaches later,
> but that was only because of a conscious decision to switch to a
> "hybrid" model using the Microsoft C compiler.  We did this because we
> required the program to run on a 256K IBM-PC.  If we had been content
> to run on a 320K machine, there would have been no troubles.

No, I'm right, but I think we actually agree.  Our program is a lot bigger 
than yours -- it's over 400K of code, mostly medium model but about 10% 
large model.  The problems I'm concerned about are not debugging problems, 
but algorithmic problems.  You have to use a hybrid model to make the 
program fit (and we overlay that) and you have to write lots of extra code 
to deal with mixed models.  If your objects are bigger than 64K, then you 
have to decide whether to use huge model code which is huge and slow, or 
large model code which needs fiddling to deal with large objects.  

Then we had to write even more code to deal with the Lotus/Intel bank 
switching scheme which, after all, is only needed because of the addressing 
complications.  There is simply a lot of code in our program that is there 
solely because of the segmented architecture.  I suspect that if we redid 
it for a 386, the total size would be smaller because even though some of 
the instructions would be bigger, the ones that do the segment munging 
wouldn't be there at all.

Sure, if your program fits into your computer compiled huge model and runs
fast enough for you, segmentation is no problem.  For the rest of us, it is
a royal pain.

John Levine, ima!johnl