Help with Array dispatch (what methods do I need (or don't I need) ?

[Patrick Kofod Mogensen]

I'm trying to fix https://github.com/JuliaLang/julia/issues/16661 but I have a dispatch problem.

Basically, I am trying to add functionality such that sparse vector and sparse matrix concatenation returns a sparse matrix, and I thought I would make sparse matrix and vector + sparse vector and vector return sparse as well while I was at it.

The sparse / dense, sparse / sparse combinations all work, but! For some reason concatenation on an Array of SparseVector s return a SparseMatrixCSC not a SparseVector. I don't really understand why. It is the following test that doesn't work

let m = 80, n = 100
    A = Array{SparseVector{Float64,Int}}(n)
    tnnz = 0
    for i = 1:length(A)
        A[i] = sprand(m, 0.3)
        tnnz += nnz(A[i])
    end

    H = hcat(A...)
    @test isa(H, SparseMatrixCSC{Float64,Int})
    @test size(H) == (m, n)
    @test nnz(H) == tnnz
    Hr = zeros(m, n)
    for j = 1:n
        Hr[:,j] = full(A[j])
    end
    @test full(H) == Hr

    V = vcat(A...)
    @test isa(V, SparseVector{Float64,Int})
    @test length(V) == m * n
    Vr = vec(Hr)
    @test full(V) == Vr
end

I am happy to explain anything that is not clear. Thanks

[Patrick Kofod Mogensen]

Oops, forgot the gist link. Will post when I'm at a computer.

[Patrick Kofod Mogensen]

[Patrick Kofod Mogensen]

Here are the changes I made in copy+pastable form. https://gist.github.com/pkofod/5536ae52c23db5f34ba35cedbf0b74e9

[Patrick Kofod Mogensen]

I think I figured it out. The hcat for sparse vector/matrix was more specific than the sparse vector only *cats, so it would convert them to SparseMatrixCSC in the method I defined. Very sorry for the noise. (btw, I know there are many superfluous allocations, I was just figured out which method definitions were needed).