`iterate` returns tuple for non-existent state about julia

[Tortar]

mmh, ok but maybe returning nothing could a bit better anyway? It would be easy in this case

function iterate(r::OrdinalRange{T}, i) where {T}
    @inline
    i >= last(r) && return nothing # >= instead of ==
    next = convert(T, i + step(r))
    (next, next)
end

[KristofferC]

What iterate does after nothing has been returned is more or less arbitrary.

[Tortar]

I think that anyway the docstring should say that in this case the returned value is arbitrary

[Tortar]

Actually the previous implementation wasn't totally correct because of decreasing ranges, this would return nothing whenever the state is outside the range:

function iterate(r::OrdinalRange{T}, i) where {T}
    @inline
    rf, rl = first(r), last(r)
    rf < rl && (i < rf || i >= rl) && return nothing
    rf >= rl && (i <= rl || i > rf) && return nothing
    next = convert(T, i + step(r))
    (next, next)
end

I would have assumed that this is more costly and so probably not worth it but I see on 1.10 and 1.11 this:

julia> function f(x)
           s = 0
           for i in 1:x
               s += rand() < 0.5 ? 1 : 2
           end
       end
f (generic function with 1 method)

julia> @btime f(1000) # before
  581.742 ns (0 allocations: 0 bytes)

julia> @btime f(1000) # after
  489.426 ns (0 allocations: 0 bytes)

which indicates the opposite, does someone know why?

[Tortar]

ok sorry I wasn't returning s, now I see same performance

[Tortar]

but it is fighting a losing battle trying to make this safer without compromising performance, e.g. still a step greater than 1 should be taken care of