BlockArray to Sparse quite slow

Hi,

I noticed the following is quite slow

using LinearAlgebra, SparseArrays, BlockArrays

M = 30
N = 500
J = BlockArray(spzeros(M * N, M * N), N * ones(Int64,M),  N * ones(Int64,M))

for ii=2:M
    setblock!(J, sprand(N,N,1/N), ii, ii)
    setblock!(J, sprand(N,N,1/N), ii,ii-1)
end

J1 = @time sparse(J) #5seconds

whereas

function createSP(J::BlockArray)
    nl, nc = size(J.blocks)
    N = size(J[Block(1,1)])[1]
    res = spzeros(N,N*nc)
    for i=1:nl
        line = J[Block(i,1)]
        for j=2:nc
            line = hcat(line,J[Block(i,j)])
        end
        res = vcat(res,line)
    end
    return res[N+1:end,:]
end

Jb = @time createSP(J)

takes 0.23s.

1. Am I missing a point?
2. Any chance to put createSP in your package?

Thank you,

Best regards

Start with the Julia reproduction in the issue and profile or inspect the sparse(J) conversion for the BlockArray example. Compare its behavior and output with createSP(J), then determine whether the conversion should be optimized or whether a helper belongs in the package. Done means a documented performance improvement or a clear resolution of the proposed helper, backed by the reproduction.