[ruby-talk:4551] Re: RFC: Ruby extension for Random numbers

Brian Fundakowski Feldman writes:

>I'd like to ask that you provide a class that uses /dev/random, too!
>FreeBSD's /dev/random in 5.0 uses Yarrow, so it can be used as just
>about the highest-quality RNG you could think of :)
>
Sorry but /dev/random is OS-specific. We plan on implementing Yarrow
though...

>Also, how about one that uses the arc4random.c-type generator?  That
>is quite fast, and is pretty darn good if you add some real entropy
>every once in a while ;)
>
We don't plan to add many different RNG's in the first version but we've
been contemplating R250 since it's known to be fast and good. Have you got
any info/data/papers/links relating arc4random.c to R250 and/or MT19937?

>I'd like it if there was a RandBest (or something) class that would
>use whatever the highest quality RNG you have on your system is.
>
Difficult to say what is highest quality (speed, "randomness",
"randomness"/speeed etc). However we will supply one Rng or Prng alias for
the default or recommended RNG. rand() and srand() might delegate to this
object?!

Anyone strongly against using the mersenne twister as the default one? If
it's to slow for some application we might have an alias RngFast for a
fast (but still reasonably good) one.

>Count me as supportive of a good RNG for the base Ruby (or even just
>RAA, too :) The most important features for me are that it's of higher
>quality than rand() or random() and can read an arbitrary number of
>bytes into a buffer.  Being of cryptographically strong quality is a
>major plus!
>
We will add at least one crypto-strength RNG (probably Yarrow or have you
got any other ones we should consider?).

How would you like the byte filling method? Should it return an Array (of 
what)? Array of Fixnums => only 31 bit? etc... Please be concrete and you
might get exactly what you want. 8-)

/Robert