Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!csd4.milw.wisc.edu!uxc!uxc.cso.uiuc.edu!uxd.cso.uiuc.edu!uxe.cso.uiuc.edu!mcdonald
From: mcdonald@uxe.cso.uiuc.edu
Newsgroups: comp.lang.c
Subject: Re: Book on Microsoft C
Message-ID: <225800147@uxe.cso.uiuc.edu>
Date: 29 Mar 89 15:04:00 GMT
References: <754@oravax.UUCP>
Lines: 47
Nf-ID: #R:oravax.UUCP:754:uxe.cso.uiuc.edu:225800147:000:2316
Nf-From: uxe.cso.uiuc.edu!mcdonald    Mar 29 09:04:00 1989


mcdonald@uxe.cso.uiuc.edu:
	"Portability" is a word seldom heard outside the academic discussions
	of Usenet.

gwyn@smoke.brl.mil (Doug Gwyn):
	That's utter nonsense.[....]

ejd@caen.engin.umich.edu (Edward J Driscoll):
	Sure, but there are also numerous developers who use all kinds of
	non-portable code in order to get the best performance they can.
	The majority of users probably only use a particular application
	on one particular system, so they're not going to be attracted
	by portability....

Fs@ernie.berkeley.edu (Jim Shankland):
       Second, it's a canard that there is a direct tradeoff between portability
       and performance.  In most cases, the performance impact is unnoticeable,
       and the (very) few cases where it makes a difference can be carefully
       isolated.  Learning to code portably can take a little extra programmer time
       and discipline up front; even once the skill is learned, portable coding
       *may* take a little extra development time.  It's worth it.

      Those are the facts.  Flout them at your own (or your employer's) risk.

Me again:
       It is silly to assume that there is no tradeoff between portability
       and performance. It is true that in most cases it can be carefully
       isloated. Portability things like not assuming sizeof(a) == sizeof(b),
       coding in real ANSI C with full prototypes, are simply good 
       coding practice. And codeing in pure ANSI C guarantees a good
       bit of portability ( :-( except to certain benighted systems with
       only old-fashioned compilers.).

       But, doing certain things portably can cost dearly. Graphics is
       one area. Not using the full functionality of a machine's
       special features is another. I was, and am, talking not so much
       about the "core" functionality of a program as I am about 
       the user interface. And, I must emphasize most emphatically,
       that a portable user interface MUST either be machine-specific
       or be crippled by using the lowest common denominator (i.e.
       a tty emulator text interface.) People who think that they can 
       get away with a (codewise) portable user interface are either
       deluded or working in a sheltered market segment (accounting
       programs for cookie shops?)

Doug McDonald