Interval creation with .. returns inconsistent types

[remi-garcia]

I find it strange that:

julia> typeof(Float16(0)..Float32(0.5))
Interval{Float32}

julia> typeof(Int(0)..Float32(0.5))
Interval{Float64}

I would think that typeof(Int(0)..Float32(0.5)) == Interval{Float32}

[dpsanders]

You're right that that seems wrong according to Julia's standard promotion mechanism:

julia> promote(1, 3.5f0)
(1.0f0, 3.5f0)

[dpsanders]

That is due to using ... If you use interval it behaves as expected:

julia> interval(Int(0), Float32(0.5)) |> typeof
Interval{Float32}

[dpsanders]

But yes this should be fixed.

[remi-garcia]

I don't find what should be the differences between Interval(a, b), interval(a, b) and a..b. To solve this issue that information would help me.

[Kolaru]

Interval is the direct construsctor, interval checks the bounds and is slightly safer and slower but otherwise equivalent. .. in the other hand tries to be clever, and thus its behavior is not very well defined, but as mentionned in #396 we should probably not do it and rather rely on macro for the cleverness.

[lbenet]

Just to clarify with an example what @Kolaru wrote above, interval(2, 1) throws an error because the lower bound is larger than the upper one, while Interval(2, 1) does not throw any error.

Regarding .., the next example is a nice illustration:

julia> using IntervalArithmetic

julia> setformat(:full);

julia>  a = Interval(0.1,0.25)
Interval(0.1, 0.25)

julia> b = 0.1 .. 0.25
Interval(0.09999999999999999, 0.25)

julia> 1//10 ∈ a
false

julia> 1//10 ∈ b
true

So the "cleverness" to which @Kolaru referred is related to "correct" (outer) rounding.

[dpsanders]

This is explained in the docs: https://juliaintervals.github.io/IntervalArithmetic.jl/stable/construction/

[remi-garcia]

I'll put a RTFM on the side of my screen. Sorry.