Attempt to add support for getindex(::DiskArray, Int[])

[charleskawczynski]

Closes #186.

I'm running into some code paths that I'm not sure how to fix. The current error (in the added test suite) is:

julia> a[Int[]]
ERROR: ArgumentError: Indices can only be omitted for trailing singleton dimensions
Stacktrace:
 [1] _resolve_indices
   @ ~/Dropbox/Caltech/work/dev/clones/forks/DiskArrays.jl/src/diskarray.jl:111 [inlined]
 [2] _resolve_indices
   @ ~/Dropbox/Caltech/work/dev/clones/forks/DiskArrays.jl/src/diskarray.jl:97 [inlined]
 [3] resolve_indices
   @ ~/Dropbox/Caltech/work/dev/clones/forks/DiskArrays.jl/src/diskarray.jl:44 [inlined]
 [4] resolve_indices(a::AccessCountDiskArray{…}, i::Tuple{…}, batchstrategy::DiskArrays.NoBatch{…})
   @ DiskArrays ~/Dropbox/Caltech/work/dev/clones/forks/DiskArrays.jl/src/diskarray.jl:51
 [5] getindex_disk_nobatch!(out::Nothing, a::AccessCountDiskArray{…}, i::Tuple{…})
   @ DiskArrays ~/Dropbox/Caltech/work/dev/clones/forks/DiskArrays.jl/src/diskarray.jl:233
 [6] getindex_disk!(out::Nothing, a::AccessCountDiskArray{…}, i::Vector{…})
   @ DiskArrays ~/Dropbox/Caltech/work/dev/clones/forks/DiskArrays.jl/src/diskarray.jl:254
 [7] getindex_disk
   @ ~/Dropbox/Caltech/work/dev/clones/forks/DiskArrays.jl/src/diskarray.jl:214 [inlined]
 [8] getindex(a::AccessCountDiskArray{Int64, 3, Base.ReshapedArray{…}, DiskArrays.ChunkRead{…}}, i::Vector{Int64})
   @ DiskArrays ~/Dropbox/Caltech/work/dev/clones/forks/DiskArrays.jl/src/diskarray.jl:352
 [9] top-level scope
   @ REPL[23]:1
Some type information was truncated. Use `show(err)` to see complete types.

Can someone offer some tips? @rafaqz / @meggart ?

[rafaqz]

Probably just needs to allow 0 as well as 1 on line 111 of diskarray.jl. may hit other errors after that tho

[charleskawczynski]

Probably just needs to allow 0 as well as 1 on line 111 of diskarray.jl. may hit other errors after that tho

With the current set of changes, @show arraysize_from_chunksize(csnow) prints out 3 in _resolve_indices(cs, ::Tuple{}, indices_pre::DiskIndex, nb::ChunkStrategy). If I just comment this check out, the next failure is Number of indices is not correct.

[rafaqz]

Ok. Probably will have to wait for @meggart to get back unless you can work through the logic

[meggart]

Thanks a lot for starting the PR I have applied some fixes so that it works and tests pass. I am not fully happy yet, because in the case of these empty reads there is still a readblock! call executed where backends need to catch the empty input slices. I will make a fix for this as well.

[charleskawczynski]

Thanks a lot for starting the PR I have applied some fixes so that it works and tests pass. I am not fully happy yet, because in the case of these empty reads there is still a readblock! call executed where backends need to catch the empty input slices. I will make a fix for this as well.

Thanks for following up with this @meggart! Please let mw know if there's any thing I can do to help