Newsgroups: comp.lang.pascal
Path: utzoo!utgpu!watserv1!maytag!watstat.waterloo.edu!dmurdoch
From: dmurdoch@watstat.waterloo.edu (Duncan Murdoch)
Subject: Re: TP 6.0 Random Number Generator
Message-ID: <1991Jan15.163811.15107@maytag.waterloo.edu>
Sender: daemon@maytag.waterloo.edu (Admin)
Organization: University of Waterloo
References: <1991Jan14.213834.26624@maytag.waterloo.edu>
Date: Tue, 15 Jan 91 16:38:11 GMT
Lines: 24

In article <1991Jan14.213834.26624@maytag.waterloo.edu> I wrote:
>
>3.  The extended (floating point) version of random is calculated by dividing 
>    the absolute value of the seed by 2^31, after updating it.
>
>This has a really nasty consequence.  If you choose to calculate your own
>integer valued random numbers by the formula
>
> myrandom(n) := trunc(n*random);
>
>you'll have random coming out to 1 once in 2^32 tries (it's the value 
>you'll get first if you start with randseed = -1498392781), and so
>myrandom(n) will give n.  If it's indexing into an array[0..n-1], this
>will give an out of bounds error.

I've now taken a look at the real version of random (which you get if you
compile with the $N- option), and find that it doesn't suffer from this
flaw.  It appears to calculate the value by dividing an unsigned long
version of randseed by 2^32; this means it can never return 1.  It does
return 0 if the seed is 0 after updating; you can get this by setting randseed
to 649090867.

Duncan Murdoch
dmurdoch@watstat.waterloo.edu