On 2026-05-31 08:12 -1000, Shiro Kawai wrote:
> I think this covers the use cases I supposed.   Trivial port with "too
> trivial" pattern (e.g. just all 1's) may result degraded quality when
> passed to some-other-random-algorithm-port, but it is easily mitigated by
> the users using longer and better pattern.
>
> Shall we mention certain patterns (e.g. all zeros) that definitely
> fail with certain algorithms?

I'm glad you think this approach handles the cases we've been
discussing.

There is a new draft of SRFI 271 waiting to be merged, so I'll try to
add a note about unacceptable patterns.  "All zeros" is definitely a
problem and must not be used.  Otherwise, I think algorithms SHOULD
accept any initialization sequence but MUST signal a random port
initialization error if they can't use a given sequence.

--
Wolfgang Corcoran-Mathe  <xxxxxx@sigwinch.xyz>