fix and test mixed CartesianIndex

[rafaqz]

Fixes #259

@lupemba hopefully this solves your issue

[lupemba]

I just did a small test and discovered that it doesn't work for the views.

using DiskArrays: ReshapedDiskArray, PermutedDiskArray
using DiskArrays.TestTypes

a = AccessCountDiskArray(reshape(1:24, 2, 3, 4), chunksize=(2, 2, 2))
a[CartesianIndex(1, 2), 1:end] # works now
@view a[CartesianIndex(1, 2), 1:end] # fails

I don't think I need the functionality but I just wanted to share it right away before I forget.

[rafaqz]

In the latest push? I tried to fix view too, it works in the tests now. Have to check why this argument combo doesn't work.

[lupemba]

Yes in the latest commit on this branch. The error comes from "DiskArrays.jl/src/subarray.jl:65"

i2 = _replace_colon.(size(a), i) size(a) has length 3 and i only has length 2 so broadcasting fails.

ERROR: DimensionMismatch: arrays could not be broadcast to a common size: a has axes Base.OneTo(3) and b has axes Base.OneTo(2)
Stacktrace:
  [1] _bcs1
    @ ./broadcast.jl:528 [inlined]
  [2] _bcs
    @ ./broadcast.jl:522 [inlined]
  [3] broadcast_shape(::Tuple{Base.OneTo{Int64}}, ::Tuple{Base.OneTo{Int64}})
    @ Base.Broadcast ./broadcast.jl:516
  [4] combine_axes
    @ ./broadcast.jl:497 [inlined]
  [5] _axes
    @ ./broadcast.jl:236 [inlined]
  [6] axes
    @ ./broadcast.jl:234 [inlined]
  [7] copy
    @ ./broadcast.jl:1099 [inlined]
  [8] materialize
    @ ./broadcast.jl:872 [inlined]
  [9] view(::AccessCountDiskArray{…}, ::CartesianIndex{…}, ::UnitRange{…})
    @ DiskArrays ~/Documents/git repos/DiskArrays.jl/src/subarray.jl:65
 [10] top-level scope
    @ ~/Documents/git repos/DiskArrays.jl/small_test.jl:6
Some type information was truncated. Use `show(err)` to see complete types.

[rafaqz]

Ok thanks, I'll fix that tomorrow or monday

[rafaqz]

Ok so it turns out this fails with any array at all! Its just julias end doesn't work with CartesianIndex in the mix, it will lower to lastindex(a, 2) rather than lastindex(a, 3).

@view a[CartesianIndex(1, 2), 1:end] # fails

[rafaqz]

@meggart could you have a look at this? it would be good to merge for @lupemba s dependent PR

[lupemba]

Ok so it turns out this fails with any array at all! Its just julias end doesn't work with CartesianIndex in the mix, it will lower to lastindex(a, 2) rather than lastindex(a, 3).

Interesting, I will add an issue to the Julia repository to document this.

a = reshape(1:(2*3*4),(4,3,2))

a[CartesianIndex(1, 2), :] # works
a[1,2, 1:end] # works
a[CartesianIndex(1, 2), 1:end] # fails 

[meggart]

Thanks a lot for digging into this. Looks good in general, except the implementation of process_index

[meggart]

Ok, I finally understood the bug. It was not view-related but could be induced by a getindex on a normal DiskArray as well when indexing with a longer CartesianIndex than there were actual dimensions. I re-implemented the fix with a (probably unnecessary) additional argument check. Thanks for taking care.