Conversion from Symmetric/Hermitian to sparse uses abstract fallback

[KristofferC]

The way to hit this in real code is to do Symmetric(A) \ B, have the Cholesky factorization fail, and then get hit by the conversion just before the lu fallback at:

julia/stdlib/SuiteSparse/src/cholmod.jl

Line 1737 in fa7d1a9

return \(lu(SparseMatrixCSC{eltype(A), SuiteSparse_long}(A)), B)

which will never finish for real problems, leaving you wonder why the factorization is so slow.

julia> using SparseArrays

julia> using LinearAlgebra

julia> s = Symmetric(sprand(10^4, 10^4, 1/10^4));

julia> @time sparse(s);
  1.156863 seconds (10.03 k allocations: 1.188 MiB)

julia> nnz(s.data)
10089

[KristofferC]

Seems fixed