Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Posting-Version: version B 2.10.2 9/18/84 SMI; site sun.uucp
Path: utzoo!watmath!clyde!burl!ulysses!mhuxr!mhuxt!houxm!mtuxo!mtunh!mtung!mtunf!ariel!vax135!petsd!pesnta!amd!amdcad!decwrl!sun!guy
From: guy@sun.uucp (Guy Harris)
Newsgroups: net.unix
Subject: Re: Precision of rand()
Message-ID: <2470@sun.uucp>
Date: Tue, 23-Jul-85 06:19:47 EDT
Article-I.D.: sun.2470
Posted: Tue Jul 23 06:19:47 1985
Date-Received: Thu, 25-Jul-85 08:04:27 EDT
References: <1244@hound.UUCP> <11392@brl-tgr.ARPA> <2399@sun.uucp> <156@ukecc.UUCP> <157@ukecc.UUCP> <211@twitch.UUCP>
Organization: Sun Microsystems, Inc.
Lines: 14

> > 	It seems that I spoke too soon. Ints on our 3B20S are indeed 4 bytes
> > long. rand() returns a 4 byte value BUT (this is the catch) the value is
> > masked to its lower 15 bits yeilding 0 <= rand() <= 32767.
> 
> The length of the calculation has been 32 bits for a long time, even on
> the PDP-11.  The static variable, randx, is a long.  See the source
> routine, which is /usr/src/lib/libc/port/gen/rand.c on our machine.

Yes, but the same source code reveals that the result of the calculation is
ANDed with 0x7fff before being returned.  All results that "rand" returns,
except on UNIX/32V and its descendants (i.e., 4.xBSD), and maybe on V7 ports
with 32-bit "int"s (are there any out there?), will fit in 16 bits.

	Guy Harris