Newsgroups: comp.lang.c
Path: utzoo!sq!msb
From: msb@sq.sq.com (Mark Brader)
Subject: Re: Portability vs. Endianness
Message-ID: <1991Mar24.005821.8399@sq.sq.com>
Organization: SoftQuad Inc., Toronto, Canada
References: <1991Mar12.105451.19488@dit.upm.es> <2628@ksr.com> <829@saxony.pa.reuter.COM>
Distribution: na
Date: Sun, 24 Mar 91 00:58:21 GMT
Lines: 44

> >	Bytes[0] = (var >> 24) & 0xFF;
> >	Bytes[1] = (var >> 16) & 0xFF;
> >	Bytes[2] = (var >> 8)  & 0xFF;
> >	Bytes[3] =  var        & 0xFF;
> >This code is guaranteed.
> 
> I don't think that the standard guarantees that chars are eight bits.
> I will agree that that's probably even more common than 4-byte longs, but
> even this assumption about word sizes cannot, IMHO, be guaranteed to be
> portable. ...

The original question asked that the output be in "68000 format", or some
such words.  Therefore it *is* correct to pull off 8 bits at a time,
even if the code is running on a machine where chars are larger than
8 bits.  (They aren't allowed to be smaller.)

However, the >> operation is implementation-defined when applied to a
signed integer variable whose value is negative.  To avoid problems, the
original value should be copied into a variable of type unsigned long.
This will also in effect convert it to 2's complement format, as desired
for the "68000 format" result, no matter what format the machine uses.
(Guaranteed in ANSI C.)

However, it isn't correct to assume that the original long fits in 32 bits.
That is, the original value may not be convertible to the desired format,
and the code should check for this.  There are several ways to do so, none
of them particularly pretty.  For instance, supposing that var is the
original input and uvar is the result of converting it to unsigned long:

	if (((uvar >> 31) >> 1) != (((var < 0)? -1UL : 0UL) >> 31) >> 1)
		complain_about_overflow();

There may be easier ways that I haven't thought of.  Note that it is not
portable to shift right by 32 bits, as shifts by the entire number of
bits in a type are undefined.  Also, I have not tested the above expression,
since I don't have access to a machine with longs wider than 32 bits.
Old compilers will require a cast rather than the UL suffix, but the
original poster asked for ANSI C.

-- 
Mark Brader, SoftQuad Inc., Toronto, utzoo!sq!msb, msb@sq.com
#define	MSB(type)	(~(((unsigned type)-1)>>1))

This article is in the public domain.