To: djgpp AT sun DOT soe DOT clarkson DOT edu, sl25 AT cus DOT cam DOT ac DOT uk
Subject: Re: System32 or whatever
Date: Wed, 20 May 92 15:49:48 +0100
From: "S. Linton" <sl25 AT cus DOT cam DOT ac DOT uk>
Status: O


Eric Backus writes:

> Personnally, I'd rather not have a system32().  I'd like our
> environment to look as much like UNIX as possible.

> Seems like having these 32-bit programs calling each other should
> eventually be good, not bad.  Our problem today is we spend a lot of
> time swapping things to disk.  Perhaps things would work better if:

> 	1. We had a single TSR go32 (rather than one copy per
> 	   program).
	
This might help, but I don't think it make a huge difference	

> 	2. We had a pager somewhere that only swapped out a program's
> 	   pages when necessary (rather than every time the program
> 	   calls system()).

This would be nice, but how would you ensure that other programs 
(not using go32 at all) obeyed whatever conventions it needed. The 
reason go32 works as it does is that it can't make any assumptions
about the sanity or virtue of programs run from system calls.

> 	3. We had faster computers and disks.
> 	4. We all had more RAM.

Then the fatser disk would take just as long to swap the more RAM used
by our larger programs

> 	5. We all installed UNIX on our PCs.

Well, yes, but my PC certainly isn't up to that, nor likely to be for a while.


We need some way for go32 to recognise when the program it is swapping in
is virtuous (eg also a go32 application) and can be trusted to share a pager 
somehow. This could be done with a system32 call (or call this one system
and have a DOSsystem call that runs more general apps). Alternatively,
it could be done by looking at the command line and seeing if the program to be
run is a) a single program not a pipeline and b) starts with a copy of GO32 
(as recognised by a magic number and perhaps checked by CRC).

		Steve Linton