🚀 go-pugleaf

I wrote:

>> No, if you read Knuth, he says using the same generator is better worst-
>> case and at least as good on average.

And Mike Meyer <mwm@eris.berkeley.edu> took me to task.

My memory was slightly faulty.  I was thinking about algorithm B (algorithm M
with X_n = Y_{n+/-1}, I can't be bothered to work out which) and algorithm M
with X_n and Y_n *sucessive* values from the same random-number stream, as
described at the end of section 3.2.2, which references F. Gebhardt [Math. Comp.
21 (1967), 708-709].

Algorithm B Knuth say, "...can be recommended for use in combination with any
other random-number generator."  Start with an array V of size k and a variable
Y set to random values.  Then,

int good_rand()
{
	int j;

	j = Y*k/MAX_RAND;  /* Or Y%k, but the high bits are usually better */
	Y = V[j];
	V[j] = bad_rand();
	return Y;
}

j = bad_rand()%k; also works well, I believe, but Mike is right that I don't
have Knuth backing me up on that.  I wish Knuth wasn't written in such a
ridiculous assembly language.  It would make the algorithms much more
accessible.
-- 
	-Colin