bisect fails for infinite interval

[dpsanders]

julia> x = y = -∞..∞
[-∞, ∞]

julia> X = x × y
[-∞, ∞] × [-∞, ∞]

julia> bisect(X)
([NaN, NaN] × [-∞, ∞],[NaN, NaN] × [-∞, ∞])

Although it's not clear that this is actually a useful thing to do.

[dpsanders]

Nor is it clear what the correct bisection of a semi-infinite interval would mean. Maybe just throw an error?

An alternative is to have a way to just shave off infinity from the interval to leave a finite one.

[dpsanders]

Nor is it clear what the correct bisection of a semi-infinite interval would mean. Maybe just throw an error?

An alternative is to have a way to just shave off infinity from the interval to leave a finite one.

[lbenet]

I think the reason bisect fails is because it does not check if the interval is unbounded or empty; see mid for reference.

As a side remark, I just noticed that in the standard table 10.2 (page 39) states that mid is defined as "no value" for an empty or unbounded interval. Yet, in page 64 it states that for the interval [-∞, ∞] it should return 0 (as we do), but for intervals like [-∞,x] with finite x it should return the "finite largest T number of largest maginitude", i.e., nextfloat(-Inf), and something alike for [x,∞]; we don't follow this, and i think there is no test in the ITF1788 suite that checks this.

[dpsanders]

Yes, you're right. I'll change it.