Path: utzoo!attcan!uunet!wuarchive!psuvax1!swatsun!jackiw
From: jackiw@cs.swarthmore.edu (Nick Jackiw)
Newsgroups: comp.sys.mac.programmer
Subject: Re: Two suggestions for THINK C
Message-ID: <39FN4Y@cs.swarthmore.edu>
Date: 2 Aug 90 13:41:16 GMT
References: <1990Aug1.220231.18581@ucselx.sdsu.edu> <1990Aug2.065801.5767@d.cs.okstate.edu>
Reply-To: jackiw@cs.swarthmore.edu (Nick Jackiw)
Organization: Visual Geometry Project, Swarthmore College, PA
Lines: 38

minich@d.cs.okstate.edu (Robert Minich) writes:
>   I agree an OS should deal with CPU scheduling, but I don't think that
> it would be prohibitively difficult to support background compilation.
> Judging how often to yield CPU time might entail a little bit of work,
> but it seems obvious to me that they are already checking for events,
> since they catch cmd-'.' to interrupt compiling.
>   I s'pose the main difficulty would be ensuring the foreground process
> regular CPU bursts. However, I think it would be relatively simple to
> add some code around the are that provides status updates ("xxx lines")
> that looks at the current number of ticks and calls WNE() after a
> reasonable amount have gone by.

My project (THINK Pascal 3.0, but the argument's equivalent) takes a good
10 minutes to compile on an SE/30, locking up my machine entirely. Thus I
frequently have time to ruminate about these issues.  In that apple's not
going to come up with user-based OS task scheduling soon, I frequently
dream about my more sophisticated and time-consuming utilities (e. g. my
compilers) allowing me at least to schedule their individual behavior.
Some sort of meter in the Compile Options dialog could represent a percent
of CPU time I desire to allocate to the compilation.  It would default to
100--never switch out, never call WNE.  (This is what it does now, I 
assume--it checks for cmd-. by consulting the keymap, not the event queue.)
However, if I set it to 75%, I would expect my compile to take 13 minutes,
but I'd have enough spare cycles to watch me tap the space bar in netnews.
Alternately, I could allot it only 20%, have it take an hour, and work on
my documentation with a fairly-good response time in the word processor.

This sort of self-scheduling can never be better than approximative, in that
you can't force tasks to return to you at a given time, but 5 minutes worth
of scattered timeouts is still far more desirable than the 10 minute total
lockup now confronting me every time I change a trans-project header.


-- 

------------------------
Nick Jackiw		jackiw@cs.swarthmore.edu  "We are what we think,
Visual Geometry Project	jackiw@swarthmr.bitnet     having become what we