Path: utzoo!attcan!uunet!lll-winken!ames!xanth!nic.MR.NET!thor!mackenzi
From: mackenzi@thor.stolaf.edu (David MacKenzie)
Newsgroups: comp.lang.c
Subject: preventing sign extension
Keywords: bitmask sign extension
Message-ID: <905@thor.stolaf.edu>
Date: 10 Jan 89 05:40:30 GMT
Reply-To: edf@rocky2.rockefeller.edu
Distribution: na
Organization: Environmental Defense Fund
Lines: 47


Now that my confusion about varargs has been taken care of, I have run
into another problem with "my" port of Holub's integer doprnt to Unix
(it's getting to be a group project . . .).

Here are some code fragments:

/*
 * INTMASK is a portable way to mask off the bottom N bits
 * of a long, where N is the width of an int.
 */
#define INTMASK	(long) ((unsigned) (0))
	:
	:
long lnum;			/* used to hold numeric arguments */
	:
	:
	/* Fetch a long or int sized argument off the
	 * stack as appropriate.  If the fetched number
	 * is a base 10 int then mask off the top
	 * bits to prevent sign extension.
	 */
	:
	:
	[skip to where it gets an int sized argument]
	lnum = (long) va_arg (args, int);
	:
	:
	[if (print in unsigned format)]
		lnum &= INTMASK;

I'm not quite sure what this was supposed to do, but on my 68000 box
with Green Hills C, it does what it looks like it's doing: a bitwise
AND with zero.  The result is that
	printf ("%d\n", 4);
produces the correct output, but
	printf ("%u\n", 4);
displays 0.

I'm a bit fuzzy on the conditions under which sign extension takes
place; I had hoped I'd never have to deal with it.  Would any of
you C wizards like to explain what Allen was trying to do, and
why it presumably works under MS-DOS but doesn't under Unix?  And
how to make it work under Unix?

David MacKenzie
edf@rocky2.rockefeller.edu