Allow for constructing arrays with CartesianRange

[AzamatB]

Something like

OffsetArray{Float64}(undef, CartesianIndex(3,4,-1):CartesianIndex(7,9,5))

[timholy]

Good idea. PR would be welcome!

[johnnychen94]

Generally, we don't know if a given CartesianIndex array has dense indices, and we don't want to check that for such a basic construction operation. Hence what can be provided without loss of performance is

OffsetArray{Float64}(undef, i_first::CartesianIndex, i_last::CartesianIndex)

Note that this usage doesn't make sense for Array type because Array is 1-based.

If we support this, then we might also want to give supports to fill!, fill, zeros, ones, which are type piracies IIUC.

[timholy]

if a given CartesianIndex array has dense indices

It's not actually an array:

julia> R = CartesianIndex(3,4,-1):CartesianIndex(7,9,5);

julia> typeof(R)
CartesianIndices{3,Tuple{UnitRange{Int64},UnitRange{Int64},UnitRange{Int64}}}

julia> R.indices
(3:7, 4:9, -1:5)

So if we dispatch on CartesianIndices all is well.

[jishnub]

Is R.indices safe to use (ie. won't break)? Or should one use the roundabout map(UnitRange, Tuple(first(R)), Tuple(last(R)) )?

Edit: Actually this isn't a great idea as it converts the ranges to UnitRanges

[timholy]

R.indices is probably OK but could also use convert(Tuple{Vararg{AbstractUnitRange{Int}}}, iter).