Eigen / SV decomposition fails with Measurements

[BambOoxX]

Hello, I have attempted to use GenericLinearAlgebra along with Measurements but that seems to fail in a fresh environment.
See JuliaPhysics/Measurements.jl#189 for first analysis by @giordano.

The environment

Status `D:\bamboo\AppData\Local\Temp\jl_hvNm0T\Project.toml`
  [14197337] GenericLinearAlgebra v0.3.18
  [eff96d63] Measurements v2.14.0

The example

] add Measurements GenericLinearAlgebra
using LinearAlgebra, GenericLinearAlgebra, Measurements
Ar = rand(10,10) .± rand(10,10)
Ai = rand(10,10) .± rand(10,10)
A = Ar + im * Ai
eigvals(Ar) # fails
eigvals(A) # fails

The reproduced stacktrace

julia> eigvals(Ar)
ERROR: MethodError: no method matching Float64(::Measurement{Float64})
The type `Float64` exists, but no method is defined for this combination of argument types when trying to construct it.

Closest candidates are:
  (::Type{T})(::Real, ::RoundingMode) where T<:AbstractFloat
   @ Base rounding.jl:265
  (::Type{T})(::T) where T<:Number
   @ Core boot.jl:900
  Float64(::Irrational{:SQRT_HALF})
   @ Random irrationals.jl:251
  ...

Stacktrace:
  [1] givensAlgorithm(f::Measurement{Float64}, g::Measurement{Float64})
    @ LinearAlgebra D:\bamboo\AppData\Local\Programs\Julia-1.11.5\share\julia\stdlib\v1.11\LinearAlgebra\src\givens.jl:80
  [2] givens(f::Measurement{Float64}, g::Measurement{Float64}, i1::Int64, i2::Int64)
    @ LinearAlgebra D:\bamboo\AppData\Local\Programs\Julia-1.11.5\share\julia\stdlib\v1.11\LinearAlgebra\src\givens.jl:292
  [3] doubleShiftQR!(HH::Matrix{…}, τ::LinearAlgebra.Rotation{…}, shiftTrace::Measurement{…}, shiftDeterminant::Measurement{…}, istart::Int64, iend::Int64)
    @ GenericLinearAlgebra D:\bamboo\.julia\packages\GenericLinearAlgebra\jlOvW\src\eigenGeneral.jl:241
  [4] _schur!(H::GenericLinearAlgebra.HessenbergFactorization{…}; tol::Float64, shiftmethod::Symbol, maxiter::Int64)
    @ GenericLinearAlgebra D:\bamboo\.julia\packages\GenericLinearAlgebra\jlOvW\src\eigenGeneral.jl:161
  [5] _schur!
    @ D:\bamboo\.julia\packages\GenericLinearAlgebra\jlOvW\src\eigenGeneral.jl:89 [inlined]
  [6] _schur!
    @ D:\bamboo\.julia\packages\GenericLinearAlgebra\jlOvW\src\eigenGeneral.jl:182 [inlined]
  [7] _eigvals!
    @ D:\bamboo\.julia\packages\GenericLinearAlgebra\jlOvW\src\eigenGeneral.jl:281 [inlined]
  [8] eigvals!(A::Matrix{Measurement{Float64}}; sortby::typeof(LinearAlgebra.eigsortby), kwargs::@Kwargs{})
    @ GenericLinearAlgebra D:\bamboo\.julia\packages\GenericLinearAlgebra\jlOvW\src\eigenGeneral.jl:294
  [9] eigvals!
    @ D:\bamboo\.julia\packages\GenericLinearAlgebra\jlOvW\src\eigenGeneral.jl:285 [inlined]
 [10] eigvals(A::Matrix{Measurement{Float64}}; kws::@Kwargs{})
    @ LinearAlgebra D:\bamboo\AppData\Local\Programs\Julia-1.11.5\share\julia\stdlib\v1.11\LinearAlgebra\src\eigen.jl:343
 [11] eigvals(A::Matrix{Measurement{Float64}})
    @ LinearAlgebra D:\bamboo\AppData\Local\Programs\Julia-1.11.5\share\julia\stdlib\v1.11\LinearAlgebra\src\eigen.jl:343
 [12] top-level scope
    @ REPL[12]:1
Some type information was truncated. Use `show(err)` to see complete types.

The problem seems to be that GenericLinearAlgebra forwards Measurements to LinearAlgebra instead of standard float types, but that is not really helpful to solve the problem :)

[andreasnoack]

The problem seems to be that GenericLinearAlgebra forwards Measurements to LinearAlgebra instead of standard float types, but that is not really helpful to solve the problem :)

I'm not following. What is the alternative to computing the Givens rotations between the Measurements? I believe the root cause is JuliaLang/julia@7fb758a#r167343271

[stevengj]

Probably https://github.com/JuliaLang/LinearAlgebra.jl/blob/7b21cabb1e81b719e93ab612937c0ded809fbe6e/src/givens.jl#L80 should be zeropar = zero(onepar)

[andreasnoack]

@BambOoxX would you be able to try out that change and open a PR to LinearAlgebra.jl if it works?

[BambOoxX]

Sorry for the late reply. I'll try.

[BambOoxX]

@andreasnoack I could not ] dev LinearAlgebra. I'm guessing it is because it is a standard library. I don't think I would know how to develop that.

[araujoms]

Indeed. You need to follow the instructions in the README of the github repository: https://github.com/julialang/linearalgebra.jl

[BambOoxX]

@araujoms I tried to use the techniques in the readme, but due to my company's IT policy I can't switch julia versions easily and cannot use juliaup... Finally the PackageCompiler technique fails as well wit the following error...

PS D:\bamboo.julia\dev\LinearAlgebra.jl> julia -Jnew_sysimage.dll
fatal: error thrown and no exception handler available.
Core.InitError(mod=:LinearAlgebra, error=Core.UndefVarError(var=:add_dependency!, world=0x000000000000760e, scope=libblastrampoline_jll))
ijl_undefined_var_error at C:/workdir/src\rtutils.c:166
getproperty at D:\bamboo\AppData\Local\Programs\Julia-1.12.4\share\julia\base\Base_compiler.jl:47 [inlined]
init at d:\bamboo.julia\dev\LinearAlgebra.jl\src\LinearAlgebra.jl:862
jfptr___init___59040 at D:\bamboo.julia\dev\LinearAlgebra.jl\new_sysimage.DLL (unknown line)
jl_apply at C:/workdir/src\julia.h:2391 [inlined]
jl_module_run_initializer at C:/workdir/src\toplevel.c:68
finish_jl_init at C:/workdir/src\init.c:625
ijl_init_ at C:/workdir/src\init.c:776
jl_repl_entrypoint at C:/workdir/src\jlapi.c:1132
mainCRTStartup at C:/workdir/cli\loader_exe.c:58
BaseThreadInitThunk at C:\windows\System32\KERNEL32.DLL (unknown line)
RtlUserThreadStart at C:\windows\SYSTEM32\ntdll.dll (unknown line)

I don't think I can do much about this as it is.

[BambOoxX]

Sorry to bump, but I still could not find a way to make this work.

[BambOoxX]

Another observation : svd on a Measurements Array fails with the following

julia> svd(AA)
ERROR: MethodError: no method matching svd!(::Matrix{Measurement{Float64}}; full::Bool, alg::LinearAlgebra.DivideAndConquer)
The function `svd!` exists, but no method is defined for this combination of argument types.

Closest candidates are:
  svd!(::Bidiagonal{var"#s4809", V} where {var"#s4809"<:Union{Float32, Float64}, V<:AbstractVector{var"#s4809"}}; full) got unsupported keyword argument "alg"
   @ LinearAlgebra D:\bamboo\.julia\juliaup\julia-1.12.5+0.x64.w64.mingw32\share\julia\stdlib\v1.12\LinearAlgebra\src\bidiag.jl:240
  svd!(::StridedMatrix{T}, ::StridedMatrix{T}) where T<:Union{Float32, Float64, ComplexF64, ComplexF32} got unsupported keyword arguments "full", "alg"
   @ LinearAlgebra D:\bamboo\.julia\juliaup\julia-1.12.5+0.x64.w64.mingw32\share\julia\stdlib\v1.12\LinearAlgebra\src\svd.jl:415
  svd!(::StridedVector{T}; full, alg) where T<:Union{Float32, Float64, ComplexF64, ComplexF32}
   @ LinearAlgebra D:\bamboo\.julia\juliaup\julia-1.12.5+0.x64.w64.mingw32\share\julia\stdlib\v1.12\LinearAlgebra\src\svd.jl:109
  ...

Stacktrace:
 [1] svd(A::Matrix{Measurement{Float64}}; full::Bool, alg::LinearAlgebra.DivideAndConquer)
   @ LinearAlgebra D:\bamboo\.julia\juliaup\julia-1.12.5+0.x64.w64.mingw32\share\julia\stdlib\v1.12\LinearAlgebra\src\svd.jl:180
 [2] svd(A::Matrix{Measurement{Float64}})
   @ LinearAlgebra D:\bamboo\.julia\juliaup\julia-1.12.5+0.x64.w64.mingw32\share\julia\stdlib\v1.12\LinearAlgebra\src\svd.jl:179
 [3] top-level scope
   @ REPL[372]:1

I does not look related directly, more like there is a missing dispatch somewhere...

[araujoms]

Method 1 is the easiest one to develop LinearAlgebra. You don't need to install juliaup, you can just manually download the julia nightly, and replace the command "julia +nightly" with whatever command you use to start the nightly.

Usually this kind of issue is left open for beginners to get involved with contributing. If you want I can fix it for you though.

The SVD issue is unrelated, please open another issue for it.

[BambOoxX]

@araujoms Well, if by beginner you mean I've never contributed to a julia standard library package, you are right on point. Frankly it took me forever to make it work, but I managed to install julia nightly through juliaup this week-end after checking with the developpers. I managed to dev LinearAlgebra and test @stevengj 's proposal.
It seems to solve the issue indeed, for eigvals, but also the svd, so I'll leave the two topics here.
I'll push a PR right now.