On 5/5/20 4:55 PM, Shiro Kawai wrote:
> I guess we could say it is safer to exclude boundaries to avoid
> singularities but the user must be careful for the
> range nevertheless.
Because of rounding, one cannot ever guarantee that
a+(b-a)*(random-real) will never return a or b.
Oh, what I was assuming is that if the spec says boundary is exclusive, then the implementation must check the result and reject and retry if the produced value falls on the boundary.
If we allow a loose implementation that doesn't reject it, then the spec should say the boundary may inclusive or exclusive, it's implementation dependent.