1.12-rc3: Overload Array(::BunchKaufman) to fix Array(factorization(A))

[dlfivefifty]

I ran into a missing overload in MatrixFactorizations.jl. I believe Array should be overloaded for BunchKaufman just like it is for Cholesky:

julia> A = Symmetric(randn(5,5)) + 100I;

julia> Array(cholesky(A))
5×5 Matrix{Float64}:
 100.064      -0.513683   0.0526112   -0.734536    0.402664
  -0.513683   99.7659    -0.520904    -0.025515   -0.299517
   0.0526112  -0.520904  98.3513       0.157766    0.239067
  -0.734536   -0.025515   0.157766   100.985      -1.34571
   0.402664   -0.299517   0.239067    -1.34571   100.884

julia> Array(cholesky(A)) ≈ A
true

julia> Array(bunchkaufman(A)) ≈ A
ERROR: MethodError: no method matching Array(::BunchKaufman{Float64, Matrix{Float64}, Vector{Int64}})
The type `Array` exists, but no method is defined for this combination of argument types when trying to construct it.

Closest candidates are:
  Array(::Tridiagonal)
   @ LinearAlgebra ~/Projects/julia-1.12/usr/share/julia/stdlib/v1.12/LinearAlgebra/src/tridiag.jl:624
  Array(::Schur)
   @ LinearAlgebra ~/Projects/julia-1.12/usr/share/julia/stdlib/v1.12/LinearAlgebra/src/schur.jl:446
  Array(::LDLt)
   @ LinearAlgebra ~/Projects/julia-1.12/usr/share/julia/stdlib/v1.12/LinearAlgebra/src/ldlt.jl:224
  ...

Stacktrace:
 [1] top-level scope
   @ REPL[10]:1

[dlfivefifty]

Actually this didn't exist in Julia v1.11. I'm not sure why the tests for MatrixFactorizations.jl are now failing because of this on v1.12, but in any case it doesn't needed to be fixed urgently.

[dlfivefifty]

I realised the issue. In Julia v1.11 we had:

julia> a = randn(5,5); A = a'a; F = factorize(A);

julia> typeof(F)
Cholesky{Float64, Matrix{Float64}}

julia> Array(F)
5×5 Matrix{Float64}:
  0.979196  -0.482452   0.157173  1.47145   0.853471
 -0.482452   6.22114   -2.27973   1.03153   0.811198
  0.157173  -2.27973    1.49942   0.547494  0.148516
  1.47145    1.03153    0.547494  9.5565    7.51031
  0.853471   0.811198   0.148516  7.51031   6.43662

This now behaves like this:

julia> a = randn(5,5); A = a'a; F = factorize(A);

julia> typeof(F)
BunchKaufman{Float64, Matrix{Float64}, Vector{Int64}}

julia> Array(F)
ERROR: MethodError: no method matching Array(::BunchKaufman{Float64, Matrix{Float64}, Vector{Int64}})
The type `Array` exists, but no method is defined for this combination of argument types when trying to construct it.

Closest candidates are:
  Array(::QL)
   @ MatrixFactorizations ~/Projects/MatrixFactorizations.jl/src/ql.jl:204
  Array(::Eigen)
   @ LinearAlgebra ~/Projects/julia-1.12/usr/share/julia/stdlib/v1.12/LinearAlgebra/src/eigen.jl:682
  Array(::UniformScaling, ::Tuple{Int64, Int64})
   @ LinearAlgebra ~/Projects/julia-1.12/usr/share/julia/stdlib/v1.12/LinearAlgebra/src/uniformscaling.jl:437
  ...

Stacktrace:
 [1] top-level scope
   @ REPL[52]:1

I would therefore consider this a regression and something that should be fixed before v1.12 is finalised.

[araujoms]

Shouldn't this have been closed by #1466?