Shiro Kawai <xxxxxx@gmail.com> writes:

> Hmm, I see.  Yet, we can require that SRFI-252 generators observe
> current-random-source and still allow the implementations to choose
> distributions.  There are other implementation-dependent factors so we can't
> guarantee reliable reproducible sequences across implementations anyway.
>
> I think keeping the design simple encourages adoption, so I don't insist on
> adding variations of generators.  The user can roll their own using SRFI-194 if
> necessary.   I'm just afraid that the requirement of uniform distribution
> restricts the implementation too much.

The SRFI 252 generator sections say: "The generator should return the
sequence ... then a uniformly random distribution of ...".

Using the RFC 2119 definition of "should", implementations would be free
to change the sequences and distribution.

The use of current-random-source is specified with the word "must" at
the beginning of the section.