Solver fail when jac_prototype is BandedMatrix{BigFloat}

[ErikQQY]

Describe the bug 🐞

When the jac_prototype is a BandedMatrix{BigFloat}, solvers would throw some UndefVarError, I think this is related to JuliaLinearAlgebra/BandedMatrices.jl#121

Expected behavior

Should successfully solve the problem

Minimal Reproducible Example 👇

Without MRE, we would only be able to help you to a limited extent, and attention to the issue would be limited. to know more about MRE refer to wikipedia and stackoverflow.

function test!(du, u, p)
    du[1] = u[1] - 1
    du[2] = u[2] - 1
    du[3] = u[3] - 1
end
nlf = NonlinearFunction(test!, jac_prototype = BandedMatrix{BigFloat}(ones(3,3), (0,0)))
prob = NonlinearProblem(nlf, BigFloat[0.0, 0.0, 0.0], (0.0, 1.0))
sol = solve(prob, NewtonRaphson())

Error & Stacktrace ⚠️

ERROR: UndefVarError: `banded_lufact!` not defined in `BandedMatrices`
Suggestion: check for spelling errors or missing imports.
Stacktrace:
  [1] lu!_layout(::BandedMatrices.BandedColumns{…}, axes::Tuple{…}, A::BandedMatrix{…}, pivot::Val{…}; check::Bool)
    @ BandedMatrices ~/.julia/packages/BandedMatrices/i60tK/src/banded/BandedLU.jl:108
  [2] lu!_layout
    @ ~/.julia/packages/BandedMatrices/i60tK/src/banded/BandedLU.jl:108 [inlined]
  [3] lu!
    @ ~/.julia/packages/ArrayLayouts/B2wRU/src/factorizations.jl:429 [inlined]
  [4] #_lu#22
    @ ~/.julia/packages/BandedMatrices/i60tK/src/banded/BandedLU.jl:113 [inlined]
  [5] lu_layout (repeats 2 times)
    @ ~/.julia/packages/BandedMatrices/i60tK/src/banded/BandedLU.jl:111 [inlined]
  [6] lu
    @ ~/.julia/packages/ArrayLayouts/B2wRU/src/factorizations.jl:428 [inlined]
  [7] init_cacheval
    @ ~/.julia/packages/LinearSolve/KN05h/ext/LinearSolveBandedMatricesExt.jl:47 [inlined]
  [8] macro expansion
    @ ~/.julia/packages/LinearSolve/KN05h/src/default.jl:308 [inlined]
  [9] init_cacheval(alg::LinearSolve.DefaultLinearSolver, A::BandedMatrix{…}, b::Vector{…}, u::Vector{…}, Pl::IdentityOperator, Pr::IdentityOperator, maxiters::Int64, abstol::BigFloat, reltol::BigFloat, verbose::Bool, assump::OperatorAssumptions{…})
    @ LinearSolve ~/.julia/packages/LinearSolve/KN05h/src/default.jl:291
 [10] init(::LinearProblem{…}, ::LinearSolve.DefaultLinearSolver; alias::LinearAliasSpecifier, abstol::BigFloat, reltol::BigFloat, maxiters::Int64, verbose::Bool, Pl::Nothing, Pr::Nothing, assumptions::OperatorAssumptions{…}, sensealg::LinearSolveAdjoint{…}, kwargs::@Kwargs{})
    @ LinearSolve ~/.julia/packages/LinearSolve/KN05h/src/common.jl:239
 [11] init
    @ ~/.julia/packages/LinearSolve/KN05h/src/common.jl:140 [inlined]
 [12] #init#95
    @ ~/.julia/packages/LinearSolve/KN05h/src/default.jl:270 [inlined]
 [13] init
    @ ~/.julia/packages/LinearSolve/KN05h/src/default.jl:266 [inlined]
 [14] construct_linear_solver(alg::NewtonDescent{…}, linsolve::Nothing, A::BandedMatrix{…}, b::Vector{…}, u::Vector{…}; stats::SciMLBase.NLStats, kwargs::@Kwargs{…})
    @ NonlinearSolveBase ~/.julia/packages/NonlinearSolveBase/5ckRl/src/linear_solve.jl:77
 [15] construct_linear_solver
    @ ~/.julia/packages/NonlinearSolveBase/5ckRl/src/linear_solve.jl:61 [inlined]
 [16] init(prob::NonlinearProblem{…}, alg::NewtonDescent{…}, J::BandedMatrix{…}, fu::Vector{…}, u::Vector{…}; stats::SciMLBase.NLStats, shared::Val{…}, pre_inverted::Val{…}, linsolve_kwargs::@NamedTuple{…}, abstol::BigFloat, reltol::BigFloat, timer::Nothing, kwargs::@Kwargs{…})
    @ NonlinearSolveBase ~/.julia/packages/NonlinearSolveBase/5ckRl/src/descent/newton.jl:41
 [17] __init(::NonlinearProblem{…}, ::GeneralizedFirstOrderAlgorithm{…}; stats::SciMLBase.NLStats, alias_u0::Bool, maxiters::Int64, abstol::Nothing, reltol::Nothing, maxtime::Nothing, termination_condition::Nothing, internalnorm::Function, linsolve_kwargs::@NamedTuple{}, initializealg::NonlinearSolveBase.NonlinearSolveDefaultInit, kwargs::@Kwargs{})
    @ NonlinearSolveFirstOrder ~/.julia/packages/NonlinearSolveFirstOrder/yeYrV/src/solve.jl:167
 [18] __init
    @ ~/.julia/packages/NonlinearSolveFirstOrder/yeYrV/src/solve.jl:121 [inlined]
 [19] __solve(::NonlinearProblem{…}, ::GeneralizedFirstOrderAlgorithm{…}; kwargs::@Kwargs{})
    @ NonlinearSolveBase ~/.julia/packages/NonlinearSolveBase/5ckRl/src/solve.jl:5
 [20] __solve
    @ ~/.julia/packages/NonlinearSolveBase/5ckRl/src/solve.jl:1 [inlined]
 [21] #solve_call#35
    @ ~/.julia/packages/DiffEqBase/zYZst/src/solve.jl:635 [inlined]
 [22] solve_call
    @ ~/.julia/packages/DiffEqBase/zYZst/src/solve.jl:592 [inlined]
 [23] #solve_up#44
    @ ~/.julia/packages/DiffEqBase/zYZst/src/solve.jl:1126 [inlined]
 [24] solve_up
    @ ~/.julia/packages/DiffEqBase/zYZst/src/solve.jl:1120 [inlined]
 [25] #solve#43
    @ ~/.julia/packages/DiffEqBase/zYZst/src/solve.jl:1114 [inlined]
 [26] solve(prob::NonlinearProblem{…}, args::GeneralizedFirstOrderAlgorithm{…})
    @ DiffEqBase ~/.julia/packages/DiffEqBase/zYZst/src/solve.jl:1104
 [27] top-level scope
    @ REPL[252]:1
Some type information was truncated. Use `show(err)` to see complete types.

[oscardssmith]

I think JuliaLinearAlgebra/BandedMatrices.jl#457 is probably a better place to track this.

[ErikQQY]

Oh, actually it’s related to issue JuliaLinearAlgebra/BandedMatrices.jl#121, i think it’s the linear solve inside triggered this issue, and when the jac_prototype is a BandedMatrix{BigFloat}, we should use qr instead of lu.