Dense and sparse cholfact handles deviations from symmetry differently

[andreasnoack]

Right now sparse Cholesky factorization checks for symmetry in contrast to dense Cholesky.

julia> A = speye(4);A[1,2] = eps();

julia> cholfact(A)
ERROR: ArgumentError: sparse matrix is not symmetric/Hermitian
 in cholfact at sparse/cholmod.jl:980
 in cholfact at sparse/cholmod.jl:1055

julia> cholfact(full(A))
Base.LinAlg.Cholesky{Float64,Array{Float64,2},:U} with factor:
4x4 Base.LinAlg.UpperTriangular{Float64,Array{Float64,2}}:
 1.0  2.22045e-16  0.0  0.0
 0.0  1.0          0.0  0.0
 0.0  0.0          1.0  0.0
 0.0  0.0          0.0  1.0

It is an easy fix to make this consistent, but which solution is preferred. Right now I lean towards the sparse solution where the matrix is checked for symmetry. If the user wants to bypass the check, there should be cholfact(Symmetric) method defined which is already the case for cholfact(Sparse).

[tkelman]

Did this get changed with any of the recent cholmod refactorings, or is there still work to do here?

[andreasnoack]

There is still work to do here.

[tkelman]

Will any of it happen before 0.5 or should this be pushed to a later milestone?

[andreasnoack]

It's a relatively small fix but can break some code so let me fix it in this cycle.