LU factorization on matrix with Inf/NaN errors in 1.9

[liushang0322]

I want to port my code base from 1.7.3 to 1.9.0beta4 and test it, found some problems

Below is my version information

julia> versioninfo()
Julia Version 1.9.0-beta4
Commit b75ddb787f (2023-02-07 21:53 UTC)
Platform Info:
  OS: Windows (x86_64-w64-mingw32)
  CPU: 12 × Intel(R) Core(TM) i7-10850H CPU @ 2.70GHz
  WORD_SIZE: 64
  LIBM: libopenlibm
  LLVM: libLLVM-14.0.6 (ORCJIT, skylake)
  Threads: 1 on 12 virtual cores
Environment:
  JULIA_DEPOT_PATH = C:\Users\Public\TongYuan\.julia
  JULIA_DIR = D:\MWorks.Syslab 2022\Tools\julia-1.7.0
  JULIA_IMAGE_THREADS = 1
  JULIA_NUM_THREADS-- = 0x0000000000000001
  JULIA_PKG_USE_CLI_GIT = true
  JULIA_PYTHONCALL_EXE = @PyCall

I locate the problem in the following code

using LinearAlgebra
JwJ = [NaN NaN; NaN NaN] 
lambda = 786.0229394073194 
Jwy = [NaN; NaN;;]
h = (JwJ + lambda * diagm(diag(JwJ))) \ Jwy

Julia1.9.0beta4 gave me the following error：

ERROR: ArgumentError: matrix contains Infs or NaNs
Stacktrace:
 [1] chkfinite
   @ E:\Julia-1.9.0-beta4\share\julia\stdlib\v1.9\LinearAlgebra\src\lapack.jl:86 [inlined]
 [2] getrf!(A::Matrix{Float64})
   @ LinearAlgebra.LAPACK E:\Julia-1.9.0-beta4\share\julia\stdlib\v1.9\LinearAlgebra\src\lapack.jl:559
 [3] #lu!#170
   @ E:\Julia-1.9.0-beta4\share\julia\stdlib\v1.9\LinearAlgebra\src\lu.jl:81 [inlined]
 [4] #lu#176
   @ E:\Julia-1.9.0-beta4\share\julia\stdlib\v1.9\LinearAlgebra\src\lu.jl:299 [inlined]
 [5] lu (repeats 2 times)
   @ E:\Julia-1.9.0-beta4\share\julia\stdlib\v1.9\LinearAlgebra\src\lu.jl:298 [inlined]
 [6] \(A::Matrix{Float64}, B::Matrix{Float64})
   @ LinearAlgebra E:\Julia-1.9.0-beta4\share\julia\stdlib\v1.9\LinearAlgebra\src\generic.jl:1115
 [7] top-level scope
   @ REPL[216]:1

I locate this problem in\function.

Realized that the \ function is performed using lu decomposition, so I tried lu decomposition again, and also reported the following error to me

julia> luh = lu(JwJ + lambda * diagm(diag(JwJ)))
ERROR: ArgumentError: matrix contains Infs or NaNs
Stacktrace:
 [1] chkfinite
   @ E:\Julia-1.9.0-beta4\share\julia\stdlib\v1.9\LinearAlgebra\src\lapack.jl:86 [inlined]
 [2] getrf!(A::Matrix{Float64})
   @ LinearAlgebra.LAPACK E:\Julia-1.9.0-beta4\share\julia\stdlib\v1.9\LinearAlgebra\src\lapack.jl:559
 [3] #lu!#170
   @ E:\Julia-1.9.0-beta4\share\julia\stdlib\v1.9\LinearAlgebra\src\lu.jl:81 [inlined]
 [4] #lu#176
   @ E:\Julia-1.9.0-beta4\share\julia\stdlib\v1.9\LinearAlgebra\src\lu.jl:299 [inlined]
 [5] lu
   @ E:\Julia-1.9.0-beta4\share\julia\stdlib\v1.9\LinearAlgebra\src\lu.jl:298 [inlined]
 [6] lu(A::Matrix{Float64})
   @ LinearAlgebra E:\Julia-1.9.0-beta4\share\julia\stdlib\v1.9\LinearAlgebra\src\lu.jl:298
 [7] top-level scope
   @ REPL[217]:1

This made me realize that it might be Julia1.9.0beta4 calling lapack getrf! Caused by not returning and processing Inf or NaN in advance before the function

Below is what I get in 1.7.3

julia> (JwJ + lambda * diagm(diag(JwJ)))
2×2 Matrix{Float64}:
 NaN  NaN
 NaN  NaN

julia> Jwy
2×1 Matrix{Float64}:
 NaN
 NaN

julia> (JwJ + lambda * diagm(diag(JwJ)))
2×2 Matrix{Float64}:
 NaN  NaN
 NaN  NaN

julia> h = lu(JwJ + lambda * diagm(diag(JwJ)))
LU{Float64, Matrix{Float64}}
L factor:
2×2 Matrix{Float64}:
   1.0  0.0
 NaN    1.0
U factor:
2×2 Matrix{Float64}:
 NaN    NaN
   0.0  NaN
julia> JwJ + lambda * diagm(diag(JwJ))2×2 Matrix{Float64}:
 NaN  NaN
 NaN  NaN

[dkarrasch]

That was changed on purpose, see JuliaLang/julia#47098.

[dkarrasch]

What's the expected behavior? Is there anything actionable here or can we close this?

[oscardssmith]

At the very least, this check should respect the check keyword. If it does this, it can be closed IMO.

[dkarrasch]

At the very least, this check should respect the check keyword. If it does this, it can be closed IMO.

This was done in JuliaLang/julia#50134.