| Date: Tue, 31 Aug 2010 09:49:33 -0300
 | From: Jeronimo Pellegrini <j_p@aleph0.info>
 |
 | I was wondering if it isn't worth to include the timing part of
 | SRFI-18 in the small version of Scheme. The procedures that I
 | suppose would be interesting are:
 |
 | - (current-time)
 | - (time->seconds time)
 | - (seconds->time x)
 | - (time? obj)
 |
 | Just that. No threads, mutexes or signals.
 |
 | The reason for this is that those procedures would be useful
 | to implement two important didactic examples:
 |
 | - A PRNG:
 |   I agree with those that think that including simple randomness
 |   in "small" Scheme is not encessary, since it's easy to write a
 |   PRNG that would work fine for didatic purposes: a linear
 |   congruential PRNG is trivial to write, easy to understand and
 |   a nice example of procedure. It's also a nice example of a
 |   referentially transparent procedure if you don't hide its
 |   state -- and later in the course it's possible to talk about
 |   encapsulation of state using it again as an example.
 |   However, it would be good to have some way to seed the PRNG
 |   with a minimally unpredictable number, which could be obtained
 |   from (time->seconds (current-time))  (for simple randomness
 |   for students this is usually good enough).

Linear-congruential PRNGs do not pass Marsaglia's (Diehard) PRNG
tests.  RC4, which is as easy to write as LCPRNGs and is included in
SLIB, does pass Diehard
<http://people.csail.mit.edu/jaffer/slib_5.html#SEC119>.  RC4 does 5
byte reads, a byte addition, and 5 byte writes per byte of output.
RC4 effectively computes a cryptographic hash of its key; you can't
improve on that.

We do students no favor by getting them accustomed to the flawed PRNGs
from the twentieth-century.

_______________________________________________
Scheme-reports mailing list
Scheme-reports@scheme-reports.org
http://lists.scheme-reports.org/cgi-bin/mailman/listinfo/scheme-reports