Cannot convert an object of type Matrix{Float32} to an object of type UpperTriangular{Float32, Matrix{Float32}} when testing PDMats

[KristofferC]

PkgEval log: https://s3.amazonaws.com/julialang-reports/nanosoldier/pkgeval/by_hash/7825364_vs_d63aded/PDMats.primary.log

* kron 
Kronecker product: Error During Test at /home/pkgeval/.julia/packages/PDMats/iX2rk/test/kron.jl:14
  Got exception outside of a @test
  MethodError: Cannot `convert` an object of type Matrix{Float32} to an object of type UpperTriangular{Float32, Matrix{Float32}}
  The function `convert` exists, but no method is defined for this combination of argument types.
  
  Closest candidates are:
    UpperTriangular{T, S}(::Any) where {T, S<:AbstractMatrix{T}}
     @ LinearAlgebra /opt/julia/share/julia/stdlib/v1.12/LinearAlgebra/src/triangular.jl:20
    convert(::Type{<:UpperTriangular}, !Matched::SparseArrays.AbstractSparseMatrixCSC)
     @ SparseArrays /opt/julia/share/julia/stdlib/v1.12/SparseArrays/src/sparsematrix.jl:990
    convert(::Type{T}, !Matched::Union{LinearAlgebra.AbstractTriangular, Bidiagonal, Diagonal}) where T<:UpperTriangular
     @ LinearAlgebra /opt/julia/share/julia/stdlib/v1.12/LinearAlgebra/src/special.jl:75
    ...
  
  Stacktrace:
    [1] PDMat{Float32, UpperTriangular{Float32, Matrix{Float32}}}(m::Matrix{Float32}, c::Cholesky{Float32, UpperTriangular{Float32, Matrix{Float32}}})
      @ PDMats ~/.julia/packages/PDMats/iX2rk/src/pdmat.jl:15
    [2] PDMat(mat::Matrix{Float32}, chol::Cholesky{Float32, UpperTriangular{Float32, Matrix{Float32}}})
      @ PDMats ~/.julia/packages/PDMats/iX2rk/src/pdmat.jl:22
    [3] kron(A::PDMat{Float32, Matrix{Float32}}, B::PDMat{Float32, Matrix{Float32}})
      @ PDMats ~/.julia/packages/PDMats/iX2rk/src/pdmat.jl:116
    [4] _pd_kron_compare(A::PDMat{Float32, Matrix{Float32}}, B::PDMat{Float32, Matrix{Float32}})
      @ Main ~/.julia/packages/PDMats/iX2rk/test/kron.jl:6
    [5] macro expansion

[dkarrasch]

This is caused by the fact that kron of two UpperTriangular matrices returns an UpperTriangular matrix now. So, in

Base.kron(A::PDMat, B::PDMat) = PDMat(kron(A.mat, B.mat), Cholesky(kron(A.chol.U, B.chol.U), 'U', A.chol.info))

we get from kron(A.chol.U, B.chol.U) an UpperTriangular, which then in

struct PDMat{T<:Real,S<:AbstractMatrix{T}} <: AbstractPDMat{T}
    mat::S
    chol::Cholesky{T,S}
...

forces S<:UpperTriangular, so it also tries to convert kron(A.mat, B.mat) to UpperTriangular. I see three options: (a) unwrap the triangular wrapper inside Cholesky, or (b) relax the type equality between mat and chol, or (c) introduce a method _full! which does nothing on v1.12-, and calls full! on v1.12+, and call it on the result of kron(A.chol.U, B.chol.U).

@jishnub @andreasnoack

[jishnub]

(b) makes sense to me. I doubt there will always be a correspondence between the matrix types, and the other options seem like patches to fix this specific case.

[andreasnoack]

I prefer something like (c). Unwrapping the triangular wrapper after kron seems right to me. Then you'll benefit from the the structure during kron but when wrapping the result in Cholesky, the wrapper should be gone as it it is implied by the Cholesky. I think the type equality makes sense and should be enforced.

[dkarrasch]

Just to be clear: both options (b) and (c) would need to done within PDMats.jl. I'll open an issue and hope somebody picks it up.

[devmotion]

This issue can be closed, the problem in PDMats is fixed on the master branch.