Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Path: utzoo!watmath!clyde!burl!ulysses!allegra!mit-eddie!genrad!panda!husc6!harvard!seismo!umcp-cs!chris
From: chris@umcp-cs.UUCP (Chris Torek)
Newsgroups: net.lang.c
Subject: Re:  questions from using lint
Message-ID: <1503@umcp-cs.UUCP>
Date: Thu, 15-May-86 04:31:30 EDT
Article-I.D.: umcp-cs.1503
Posted: Thu May 15 04:31:30 1986
Date-Received: Sat, 17-May-86 03:34:29 EDT
References: <501@brl-smoke.ARPA> <797@bentley.UUCP> <640@bunkerb.UUCP>
Reply-To: chris@maryland.UUCP (Chris Torek)
Organization: University of Maryland, Dept. of Computer Sci.
Lines: 32

In article <640@bunkerb.UUCP> garys@bunkerb.UUCP (Gary M. Samuelson) writes:
>If I say "a = b", I have clearly used the value of the expression.

Which expression?  `b', certainly, but what about the value of `a = b'?

The problem is that there really are several `kinds' of expressions
in C.  All expressions have values; but some are more useful than
others.  The following are grouped into `error' and `non-error'
classes:

	Error:			Non-Error:

	open("foo", 1);		strcpy(hold, s);
	a + 1;			a++;

My best guess at the moment at the `true' distinction between the
`error' and `non-error' classes is that the latter `end' in side
effects.  The return value from `open' is usually as important as
its side effect; but that from strcpy() is often not.  `a + 1'
has no side effects in most cases (one where it might is if integer
overflow traps are enabled); `a++' has a clear side effect.

So perhaps lint needs, in addition to /*VARARGS*/ and /*ARGSUSED*/,
and the System V style /*PRINTFLIKE*/, another pragma: /*FORSIDEEFFECTS*/.
Routines like strcpy(), strcat(), and perhaps even close(), would be
so declared in llib-lc.  This would tell lint that the return value,
though available, was purely for convenience: that the function is
normally called simply for its side effects.
-- 
In-Real-Life: Chris Torek, Univ of MD Comp Sci Dept (+1 301 454 1415)
UUCP:	seismo!umcp-cs!chris
CSNet:	chris@umcp-cs		ARPA:	chris@mimsy.umd.edu