Iteration failure with CartesianIndex

[timholy]

julia> A = rand(5,5,5);

julia> S = Slices(A, True(), False(), False());

julia> size(S)
(5, 5)

julia> for s in S
           @show size(s)
       end
ERROR: ArgumentError: tuple must be non-empty
Stacktrace:
 [1] first(::Tuple{}) at ./tuple.jl:77
 [2] setindex(::Tuple{Base.OneTo{Int64}}, ::Tuple{}, ::Tuple{True}) at /home/tim/.julia/packages/JuliennedArrays/t66eT/src/JuliennedArrays.jl:39
 [3] setindex(::Tuple{Base.OneTo{Int64},Base.OneTo{Int64}}, ::Tuple{CartesianIndex{2}}, ::Tuple{True,True}) at /home/tim/.julia/packages/JuliennedArrays/t66eT/src/JuliennedArrays.jl:43 (repeats 2 times)
 [4] getindex at /home/tim/.julia/packages/JuliennedArrays/t66eT/src/JuliennedArrays.jl:58 [inlined]
 [5] iterate at ./abstractarray.jl:860 [inlined]
 [6] iterate(::Slices{SubArray{Float64,1,Array{Float64,3},Tuple{Base.OneTo{Int64},Int64,Int64},true},2,Array{Float64,3},Tuple{True,False,False}}) at ./abstractarray.jl:858
 [7] top-level scope at ./none:0

These kinds of things could be caught in real tests 😉.

[bramtayl]

Hmm, ok, is the solution just to define:

@propagate_inbounds getindex(it::Slices, index::CartesianIndex) = 
    getindex(it, index.I...)

?

If so, should there be a note in here about supporting ::CartesianIndex?

[bramtayl]

Never mind, index::Int... seems to do the trick. Fixed on master.

[johnnychen94]

Never mind, index::Int... seems to do the trick. Fixed on master.

This would incorrectly capture Linear Indexing getindex(S, 1) and some weird indexinggetindex(S, 1, 2, 3, 4, 5). Because Align/Slices internally uses the cartesian index to calculate where to insert the dimensions, we need to use the fallback version of getindex for linear indexing cases.

Patch submitted in #28.