Path: utzoo!utgpu!jarvis.csri.toronto.edu!rutgers!cs.utexas.edu!usc!brutus.cs.uiuc.edu!psuvax1!husc6!cmcl2!stealth.acf.nyu.edu!brnstnd
From: brnstnd@stealth.acf.nyu.edu
Newsgroups: comp.lang.c
Subject: Summary: Is this swap() macro correct?
Message-ID: <790@stealth.acf.nyu.edu>
Date: 21 Jan 90 03:30:09 GMT
References: <1990Jan18.002842.441@aqdata.uucp> <4514@rtech.rtech.com>
Reply-To: brnstnd@stealth.acf.nyu.edu (Dan Bernstein)
Distribution: usa
Organization: IR
Lines: 26

The macro is correct, except for namespace problems. I depend upon the
user not doing anything like swap(x,_MV_x,real)---after all, the user
isn't supposed to use such names. Then again, the macro shouldn't use
them either! Without better namespace management there's no solution.

do { ... } while(0) is the usual way to keep the macro working before an
else. (Any language that doesn't terminate all its control structures
should throw away its syntax and commit suicide, but that's a side issue.)
Sorry for not explaining this. By the way, an empty do { } while(0) fails;
define block as ``do { ;'' for general use.

Several people asked what was wrong with

  #define swap(x,y,typ) { typ tmp; tmp = y; y = x; x = tmp; }

There are two answers. The first is that this version has side effects.
A macro foo(a,b,c) should reference each of a, b, and c exactly once.
If the stdio macro writers understood this, putc(c,*f++) would work.

The second is a more general religious argument: Macros should feel like
functions. As above, they should reference each argument exactly once;
if an argument is modified, the macro should use a pointer to it instead;
and so on. The only exception is all-caps macros, which can do anything
they want. I wouldn't mind seeing this convention more widely observed.

---Dan