From: wjb AT burl DOT att DOT com
Date: Mon, 24 May 93 15:11 EDT
To: att!att!ctron.com!dj AT omnigate DOT clarkson DOT edu, wjb AT burl DOT att DOT com
Subject: Re: 286 FAQ
Cc: att!sun.soe.clarkson.edu!djgpp AT omnigate DOT clarkson DOT edu

DJ,

I have a follow-on question to your reply.  Besides the questionable
worth of redesigning something that works fine on a 386 board to backport
it to an obsolete technology, is there a physical limitation to porting
DJGPP to a 286 (conceding that you probably wouldn't want your initials in
the name of the result)?

What I am aware of is:
	- the machine instructions taking advantage of 32-bit modes would
	  have to be replaced
	- machine constraints would keep code and data sizes way down,
	  unless explicit memory model support is provided ala commercial
	  compilers

What I am not sure about is:
	- how much (real) memory would such an implementation occupy while
	  running
	- how much core memory needs to be available to compile an
	  application (or, assuming it depends on the application size,
	  what is a good rough-order-of-magnitude approxiamtion for
	  programs of various sizes)?

Here's the bottom line: I know that DJGPP requires the 386.  But I'm
asking *why* it requires a 386.  If there is a technical reason inherent
to the GNU front end you started with, I can accept that.  But if it's a
matter of it just not being worth your time (which I also can accept;
don't take this as a criticism by any means), it might be worth my while.
I'm only fooling around with this as a hobby, and I have more time than
money to spend on it (and 286 CPU).  A back port might be fun.

		--Bill