static sparse array

[Roger-luo]

I'm wondering if it is a good idea to PR our static sparse array to upstream: https://github.com/QuantumBFS/LuxurySparse.jl/blob/master/src/SSparseMatrixCSC.jl

it is faster than SparseArrays.SparseMatrixCSC sometimes.

[c42f]

Thanks for the suggestion! I think LuxurySparse is a better home so that StaticArrays can remain focused on doing the dense case well.

What benefit do you see in having it here?

[Roger-luo]

no, I'm committing patches in LuxurySparse to upstream. just checking if there is one.

[c42f]

Oh right. The natural place to upstream this is probably stdlib/SparseArrays, by making the colptr rowval and nzval of SparseArrays.SparseMatrixCSC parametric on the array type. This is like what was done with LinearAlgebra.Diagonal a while back to make it capable of holding a static array. (Before that, we needed a special type SDiagonal.)

[Roger-luo]

but will we need a SSparseMatrixCSC after SparseArrays supports this feature for backward compatibility reason (to make it work on v1.0 as well)?

[c42f]

Yes, you'll need something like that if you want backward compatibility.