Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Path: utzoo!mnetor!seismo!umcp-cs!chris
From: chris@umcp-cs.UUCP (Chris Torek)
Newsgroups: net.lang
Subject: Re: null statements
Message-ID: <2565@umcp-cs.UUCP>
Date: Wed, 23-Jul-86 06:02:29 EDT
Article-I.D.: umcp-cs.2565
Posted: Wed Jul 23 06:02:29 1986
Date-Received: Thu, 24-Jul-86 01:31:11 EDT
References: <800015@ccvaxa> <463@opus.nbires.UUCP> <442@sunybcs.UUCP> <484@valid.UUCP>
Reply-To: chris@maryland.UUCP (Chris Torek)
Organization: University of Maryland, Dept. of Computer Sci.
Lines: 113

In article <484@valid.UUCP> sbs@valid.UUCP (Steven Brian McKechnie
Sargent) writes:
>Only a barbarian or Obfuscated C Contest winner would champion the use of
>	#define nothing	/* do nothing */
>but I figured I'd mention it anyway.

Actually, I came up with a use for something very like this recently.
In C one can define a macro that `looks' like a function or procedure:

	/* successor function, mod 5 */
	#define	succ(i)	(((i) + 1) % 5)

In many cases the function is more complex:

	#define	get_node() \
		(freenodes ? \
		    (temp = freenodes, freenodes = freenodes->next, temp) : \
		    allocate_more_nodes())

If the macro is emulating a procedure, it need not even be an expression:

	#define proc(exp) \
		if (exp) \
			a(); \
		else \
			b()

Note the lack of a semicolon.  Instances of the macro include the
semicolon, and it would not do to include it twice:

	v = e;
	proc(v);
	other(v);

Sometimes, however, there is nothing to do in the `else' clause:

	#define SaveB(b) \
		if ((b)->b_AfterGroupRst == NULL || \
		    (b)->b_AfterGroupRst->sv_level != CurrentGroup) \
			DoSave(b); \
		else \
			/* do nothing */

This is fine in normal contexts:

	v = e;
	SaveB(v);
	other(v);

However, now a missing semicolon can wreak silent havoc:

	v = e;
	SaveB(v)
	other(v);

This expands to `if (v->b_...) DoSave(v); else other(v);'---not at
all what was meant, but syntactically valid.

One solution is to invert the `if':

	#define SaveB(b) \
		if ((b)->b_AfterGroupRst != NULL && \
		    (b)->b_AfterGroupRst->sv_level == CurrentGroup) \
			/* do nothing */ ; \
		else \
			DoSave(b)

This is, however, confusing, at best.  What is needed is an
explicit `nothing'.  In fact, this works:

	#define SaveB(b) \
		if ((b)->b_AfterGroupRst == NULL || \
		    (b)->b_AfterGroupRst->sv_level != CurrentGroup) \
			DoSave(b); \
		else \
			0

since a constant is an expression, and an expression is a statement.
This will make the case of the missing semicolon produce a syntax
error.  Unfortunately, it also makes `lint' complain (at least when
used with `-h').  The solution below avoids both of these problems,
but is hardly pretty:

	#ifdef lint
	#define do_nothing	rand()
	#else
	#define do_nothing	0
	#endif

	/*
	 * Save a binding.
	 */
	#define SaveB(b) \
		if ((b)->b_AfterGroupRst == NULL || \
		    (b)->b_AfterGroupRst->sv_level != CurrentGroup) \
			DoSave(b); \
		else \
			do_nothing

	/*
	 * Undo a save.
	 */
	#define UnSaveB(b) \
		if ((b)->b_AfterGroupRst != NULL) \
			DoUnsave(b); \
		else \
			do_nothing

I think this also argues for an explicit `null statement'.
-- 
In-Real-Life: Chris Torek, Univ of MD Comp Sci Dept (+1 301 454 1516)
UUCP:	seismo!umcp-cs!chris
CSNet:	chris@umcp-cs		ARPA:	chris@mimsy.umd.edu