SubArray getindex doesn't work on immutable array types

[jakebolewski]

Should this method be renamed to sortrows! / sortcols? It currently fails for AbstractMatrix subtypes which do not define setindex!.

Ex.

julia> d = Diagonal(rand(10,10))
10x10 Base.LinAlg.Diagonal{Float64}:
 0.633063  0.0       0.0       0.0        0.0       0.0       0.0       0.0       0.0       0.0
 0.0       0.719781  0.0       0.0        0.0       0.0       0.0       0.0       0.0       0.0
 0.0       0.0       0.779313  0.0        0.0       0.0       0.0       0.0       0.0       0.0
 0.0       0.0       0.0       0.0480023  0.0       0.0       0.0       0.0       0.0       0.0
 0.0       0.0       0.0       0.0        0.747617  0.0       0.0       0.0       0.0       0.0
 0.0       0.0       0.0       0.0        0.0       0.188713  0.0       0.0       0.0       0.0
 0.0       0.0       0.0       0.0        0.0       0.0       0.572225  0.0       0.0       0.0
 0.0       0.0       0.0       0.0        0.0       0.0       0.0       0.840128  0.0       0.0
 0.0       0.0       0.0       0.0        0.0       0.0       0.0       0.0       0.317382  0.0
 0.0       0.0       0.0       0.0        0.0       0.0       0.0       0.0       0.0       0.688265

julia> sortrows(d)
ERROR: MethodError: `getindex` has no method matching getindex(::Base.LinAlg.Diagonal{Float64}, ::Array{Int64,1}, ::UnitRange{Int64})
Closest candidates are:
  getindex(::AbstractArray{T,2}, ::Any, ::Any, ::Any, ::Any...)
  getindex{T,N,P,IV}(::SubArray{T,N,P,IV,LD}, ::Union(Array{Int64,1},Range{Int64},Colon,Int64)...)
  getindex{T,N,P,IV}(::SubArray{T,N,P,IV,LD}, ::Union(AbstractArray{T,1},Real), ::Union(AbstractArray{T,1},Real))
  ...

 in sortrows at sort.jl:370

julia> s = sub(d, 1:5, 1:5)
5x5 SubArray{Float64,2,Base.LinAlg.Diagonal{Float64},(UnitRange{Int64},UnitRange{Int64}),0}:
 0.633063  0.0       0.0       0.0        0.0
 0.0       0.719781  0.0       0.0        0.0
 0.0       0.0       0.779313  0.0        0.0
 0.0       0.0       0.0       0.0480023  0.0
 0.0       0.0       0.0       0.0        0.747617

julia> sortrows(s)
ERROR: MethodError: `setindex!` has no method matching setindex!(::Base.LinAlg.Diagonal{Float64}, ::Float64, ::Int64, ::Int64)
Closest candidates are:
  setindex!(::AbstractArray{T,2}, ::Any, ::Any, ::Any, ::Any, ::Any...)
  setindex!{T}(::Array{T,N}, ::Any, ::Real, ::Real)
  setindex!{T}(::Array{T,N}, ::Any, ::Real, ::Real, ::Real)
  ...

[simonster]

The line of sortrows that's throwing is just A[p,:], where p::Vector{Int}. I think it's reasonable to expect this to work, but it doesn't for Diagonal:

julia> d = Diagonal(rand(10,10));

julia> d[[1, 2, 3, 4, 5], :]
ERROR: MethodError: `getindex` has no method matching getindex(::Base.LinAlg.Diagonal{Float64}, ::Array{Int64,1}, ::UnitRange{Int64})
Closest candidates are:
  getindex(::AbstractArray{T,2}, ::Any, ::Any, ::Any, ::Any...)
  getindex{T,N,P,IV}(::SubArray{T,N,P,IV,LD}, ::Union(Array{Int64,1},Range{Int64},Int64,Colon)...)
  getindex{T,N,P,IV}(::SubArray{T,N,P,IV,LD}, ::Union(AbstractArray{T,1},Real), ::Union(AbstractArray{T,1},Real))
  ...


julia> s = sub(d, 1:5, 1:5);

julia> s[[1, 2, 3, 4, 5], :]
ERROR: MethodError: `setindex!` has no method matching setindex!(::Base.LinAlg.Diagonal{Float64}, ::Float64, ::Int64, ::Int64)
Closest candidates are:
  setindex!(::AbstractArray{T,2}, ::Any, ::Any, ::Any, ::Any, ::Any...)
  setindex!{T}(::Array{T,N}, ::Any, ::Real, ::Real)
  setindex!{T}(::Array{T,N}, ::Any, ::Real, ::Real, ::Real)
  ...

 in copy! at multidimensional.jl:436
 in getindex at subarray2.jl:82

[JeffBezanson]

Yes, the "mutation" is an implementation of copy that works by calling copy! from the existing array to a new similar one.

[mbauman]

This seems to be evidence that the solution to JuliaLang/LinearAlgebra.jl#168 should be to provide a full array from similar.

[tkelman]

provide a full array from similar.

From similar of what, a Diagonal or a subarray of a Diagonal? For the former I don't like the idea of promoting to dense, for the latter I'm not sure what would be best to do.

[mbauman]

(response in #11574)

[mbauman]

Fixed by #11582. All the examples above look much better now:

julia> d = Diagonal(rand(10,10));

julia> s = sub(d, 1:5, 1:5);

julia> sortrows(d)
ERROR: ArgumentError: cannot set an off-diagonal index (10, 1) to a nonzero value (0.05324786837334572)

julia> sortrows(s)
ERROR: ArgumentError: cannot set an off-diagonal index (5, 1) to a nonzero value (0.05324786837334572)

julia> d[[1, 2, 3, 4, 5], :]
ERROR: ArgumentError: diagonal matrix must be square

julia> s[[1, 2, 3, 4, 5], :]
5x5 Diagonal{Float64}:
 0.160014   ⋅         ⋅          ⋅         ⋅
  ⋅        0.574236   ⋅          ⋅         ⋅
  ⋅         ⋅        0.0339231   ⋅         ⋅
  ⋅         ⋅         ⋅         0.151776   ⋅
  ⋅         ⋅         ⋅          ⋅        0.509368