@interval should always return an interval

[dpsanders]

julia> x = 1..2.1
[1, 2.10001]

julia> @interval mid(x)
1.55

julia> macroexpand(:(@interval mid(x)))
:(mid((ValidatedNumerics.convert)(Interval{parameters.precision_type},x)))

Maybe just wrap everything in Interval(...) in the @interval macro?

[lbenet]

I see your point. Yet, I think the way it works is actually the way we wanted it to behave.

If I recall correctly, the idea was that @interval sin(0.1) computed the sin of the correctly rounded interval 0.1. This should be contrasted with Interval(sin(0.1)), which yields a thin interval.

So, @interval has to "act" inside the function, and then the function is computed. In the case of mid, it (re)computes the interval x and then computed the mid of x. So, the solution is either Interval(mid(x)), or

julia> xmid = mid(x);

julia> @interval(xx)
[1.54999, 1.55001]

[dpsanders]

My point is that @interval mid(x) should definitely return an interval, since that was always the desired behaviour of @interval.

[lbenet]

My point is that @interval f(x) yields f(@interval(x)).

[dpsanders]

Yes, I see what is happening, but it seems to be wrong that something called interval doesn't return one!

[lbenet]

I agree with you: it's strange. However, I can't figure out how to solve this. The same will happen with diam, or whatever function that returns something different from an interval...

[dpsanders]

One way (I believe, without having tried) is just to modify the @interval macro to always wrap the result in Interval(...) at the end.

[dpsanders]

Fixed in #188.