NetNews Usenet Archive 1992 #27/NN_1992_27.iso/spool/comp/lang/pascal/6698

Text File | 1992-11-18 | 2.6 KB | 61 lines

Newsgroups: comp.lang.pascal
Path: sparky!uunet!utcsri!skule.ecf!torn!news.ccs.queensu.ca!mast.queensu.ca!dmurdoch
From: [e-mail] (Duncan Murdoch)
Subject: Re: Encryption and Random Numbers
Message-ID: <[e-mail]>
Lines: 49
Sender: [e-mail] (Netnews control)
Organization: Queen's University
References: <[e-mail]> <[e-mail]>
Date: Wed, 18 Nov 1992 20:52:41 GMT

>In article <[e-mail]> [e-mail] (Christopher Schmidt) writes:
>>generator.  It's a nice technique, except that if Borland
>>modifies the TP random number generator, then you and your
>>customers will be locked out of your encrypted data.

I've just checked in BP 7.  It appears that the Randseed update algorithm is
the same, but the translation from Randseed to Random(N) has changed.

Here's what happens now in a call to Random(N):

First, Randseed is updated.  The update formula is

  randseed := randseed*134775813 + 1;

Then, treating randseed as an unsigned 32 bit quantity, Random(N) returns

  Random := trunc(N*randseed/2^32);

The calculations inside the parentheses above must *not* be allowed to
overflow; this makes implementation sort of tricky.  One way to do it is

  if Randseed < 0 then
    Random := trunc(N*(randseed + 4294967296.0))/4294967296.0)
  else
    Random := trunc(N*double(randseed)/4294967296.0);

I don't have the code handy to simulate the TP 6.0 Random(N) call, but it
returns the high word of RandSeed mod N. (This may have changed between
releases of 6.0; there were lots of complaints about the lack of uniformity
of the distibution.)

In the special case of N=256 that I used in my encryption filter, the
difference is that TP 6 gives you the 2nd most significant byte of Randseed
as Random(256), while BP 7 gives you the most significant byte.

Finally, there's a difference between TP 6 and BP 7 in the way the
floating point Random values are generated.  BP 7 treats Randseed as a
Longint, and returns (RandSeed+2^31)/2^32.  TP 6 also treated Randseed as a
Longint, but returned Abs(RandSeed)/2^31.  (This was in $N+ mode; in $N-
mode, TP 6 treated Randseed as unsigned, and returned RandSeed/2^32.)

It appears to me both of these changes are good.  There were two bugs
listed in my TP 6 bug list about random number generation:  first, the non-
uniformity of Random(N) which has been remedied to a great extent, and
second, the possibility that Random = 1.0, which has now been removed.


Duncan Murdoch
[e-mail]