Replace isfinite check in ranges with lo ≤ x ≤ hi

[LilithHafner]

This isfinite check is a problem in at least two cases:

• Working with infinite ranges. If I'm working in nonstandard arithmetic with nonstandard element a, a ∈ 1:a^2 should hold
• Working with ranges over non-numbers The docstring of AbstractRange states "Supertype for ranges with elements of type T. UnitRange and other types are subtypes of this." (T need not be a number) whereas isfinite's docstring states "Test whether a number is finite." This check breaks range membership checks for non-numeric elements.

The check does not make n = round(Integer, (x - first(r)) / step(r)) + 1 much safer because it is possible to have a number that is too large to round to an Integer and still finite.

The new check makes n = round(Integer, (x - first(r)) / step(r)) + 1 much safer, but it could still throw for ranges with length greater than typemax(Int). Fixes #45747.

[giordano]

Needs a test probably?

[ViralBShah]

@oscardssmith Review?

[oscardssmith]

This looks good, but ranges are very finicky, so this probably should get a pkgeval.

[vyu]

Suggested change

	if iszero(step(r))
	isnan(x) && return false
	if iszero(step(r))

The isfinite check implicitly covered NaNs. I think an explicit check for NaN is necessary if it's removed? Otherwise NaN in 1.0:2.0 would throw instead of returning false.

[LilithHafner]

Good catch! Unfortunately, that fix will not work because we don't have isnan for non-numeric ranges.

[LilithHafner]

I used first(r) <= x <= last(r) instead.

[LilithHafner]

Here's a related issue inspired by @vyu's comment: 1e40 ∈ 0:1.0 throws on conversion to Integer.

[LilithHafner]

This should now also support empty ranges with undefined first.

[LilithHafner]

Is the set of assumptions we can make about an AbstractRange documented somewhere?
For example, can we assume first(r) + (length(r)-1) * step(r) ≈ last(r)?

[LilithHafner]

This looks good, but ranges are very finicky, so this probably should get a pkgeval.

@nanosoldier runtests(ALL, vs = ":master")

[nanosoldier]

Your package evaluation job has completed - possible new issues were detected. A full report can be found here.

[LilithHafner]

@nanosoldier runtests(["BasisFunctions", "HorseML", "OutlierDetectionData", "SimpleChains", "StableTrees", "TensorBoardLogger"], vs = ":master")

[nanosoldier]

Your package evaluation job has completed - no new issues were detected. A full report can be found here.

[LilithHafner]

PkgEval game back positive and I think this is ready to go if someone is willing to give it a final look over.

[oscardssmith]

looks good to me.