BandedMatrices Nightly CI failure on Windows with 'could not load library "libblastrampoline"'

[jishnub]

The run: https://github.com/JuliaLinearAlgebra/BandedMatrices.jl/actions/runs/4024201852/jobs/6915956298#step:21:120141
Error message:

Test threw exception
  Expression: Matrix(A * B) ≈ Matrix(A) * Matrix(B)
  could not load library "libblastrampoline"
  The specified module could not be found. 
  Stacktrace:
    [1] gbmv!(trans::Char, m::Int64, n::Int64, kl::Int64, ku::Int64, alpha::Float64, A::Ptr{Float64}, st::Int64, x::Ptr{Float64}, incx::Int64, beta::Float64, y::Ptr{Float64}, incy::Int64)
      @ BandedMatrices D:\a\BandedMatrices.jl\BandedMatrices.jl\src\blas.jl:19
    [2] A11_Btop_Ctop_gbmv!
      @ D:\a\BandedMatrices.jl\BandedMatrices.jl\src\banded\gbmm.jl:39 [inlined]
    [3] _gbmm!(α::Float64, A_data::Matrix{Float64}, B_data::Matrix{Float64}, β::Float64, C_data::SubArray{Float64, 2, Matrix{Float64}, Tuple{UnitRange{Int64}, Base.Slice{Base.OneTo{Int64}}}, false}, ::Tuple{Int64, Int64, Int64}, ::Tuple{Int64, Int64}, ::Tuple{Int64, Int64}, ::Tuple{Int64, Int64})
      @ BandedMatrices D:\a\BandedMatrices.jl\BandedMatrices.jl\src\banded\gbmm.jl:305
    [4] gbmm!(tA::Char, tB::Char, α::Float64, A::BandedMatrix{Float64, Matrix{Float64}, Base.OneTo{Int64}}, B::BandedMatrix{Float64, Matrix{Float64}, Base.OneTo{Int64}}, β::Float64, C::BandedMatrix{Float64, Matrix{Float64}, Base.OneTo{Int64}})
      @ BandedMatrices D:\a\BandedMatrices.jl\BandedMatrices.jl\src\banded\gbmm.jl:288
    [5] _banded_muladd!
      @ D:\a\BandedMatrices.jl\BandedMatrices.jl\src\generic\matmul.jl:196 [inlined]
    [6] materialize!
      @ D:\a\BandedMatrices.jl\BandedMatrices.jl\src\generic\matmul.jl:204 [inlined]
    [7] muladd!
      @ C:\Users\runneradmin\.julia\packages\ArrayLayouts\4IG3b\src\muladd.jl:70 [inlined]
    [8] copyto!
      @ C:\Users\runneradmin\.julia\packages\ArrayLayouts\4IG3b\src\muladd.jl:77 [inlined]
    [9] copy(M::ArrayLayouts.MulAdd{BandedMatrices.BandedColumns{ArrayLayouts.DenseColumnMajor}, BandedMatrices.BandedColumns{ArrayLayouts.DenseColumnMajor}, ArrayLayouts.ZerosLayout, Float64, BandedMatrix{Float64, Matrix{Float64}, Base.OneTo{Int64}}, BandedMatrix{Float64, Matrix{Float64}, Base.OneTo{Int64}}, Zeros{Float64, 2, Tuple{Base.OneTo{Int64}, Base.OneTo{Int64}}}})
      @ ArrayLayouts C:\Users\runneradmin\.julia\packages\ArrayLayouts\4IG3b\src\muladd.jl:72
   [10] copy
      @ C:\Users\runneradmin\.julia\packages\ArrayLayouts\4IG3b\src\mul.jl:119 [inlined]
   [11] materialize
      @ C:\Users\runneradmin\.julia\packages\ArrayLayouts\4IG3b\src\mul.jl:116 [inlined]
   [12] mul
      @ C:\Users\runneradmin\.julia\packages\ArrayLayouts\4IG3b\src\mul.jl:117 [inlined]
   [13] *
      @ C:\Users\runneradmin\.julia\packages\ArrayLayouts\4IG3b\src\mul.jl:204 [inlined]
   [14] macro expansion
      @ C:\hostedtoolcache\windows\julia\nightly\x64\share\julia\stdlib\v1.10\Test\src\Test.jl:477 [inlined]
   [15] macro expansion
      @ D:\a\BandedMatrices.jl\BandedMatrices.jl\test\test_banded.jl:153 [inlined]
   [16] macro expansion
      @ C:\hostedtoolcache\windows\julia\nightly\x64\share\julia\stdlib\v1.10\Test\src\Test.jl:1502 [inlined]
   [17] macro expansion
      @ D:\a\BandedMatrices.jl\BandedMatrices.jl\test\test_banded.jl:149 [inlined]
   [18] macro expansion
      @ C:\hostedtoolcache\windows\julia\nightly\x64\share\julia\stdlib\v1.10\Test\src\Test.jl:1502 [inlined]
   [19] top-level scope
      @ D:\a\BandedMatrices.jl\BandedMatrices.jl\test\test_banded.jl:30

There are many other similar errors.

The package uses Base.libblas_name to get the BLAS library. Tests are passing on other configurations (nightly on linux/max os) and on all platforms for v1.8.5 and v1.6.

Version info:

Julia Version 1.10.0-DEV.443
Commit bd8ffef2b9 (2023-01-27 05:51 UTC)
Platform Info:
  OS: Windows (x86_64-w64-mingw32)
  CPU: 2 × Intel(R) Xeon(R) CPU E5-2673 v3 @ 2.40GHz
  WORD_SIZE: 64
  LIBM: libopenlibm
  LLVM: libLLVM-14.0.6 (ORCJIT, haswell)
  Threads: 1 on 2 virtual cores

[giordano]

https://github.com/JuliaLang/julia/blob/87f8958a218215df0aaffab59736c30eb6f51d2e/base/Base.jl#L190-L191

should be something like

const libblas_name = "libblastrampoline" * (Sys.iswindows() ? "-5" : "")

[giordano]

Ideally we'd get the name from https://github.com/JuliaLang/julia/blob/87f8958a218215df0aaffab59736c30eb6f51d2e/stdlib/libblastrampoline_jll/src/libblastrampoline_jll.jl#L21-L27 but we can't reference JLLs from Base. An alternative would be devising an automated way to generate the name, without having to remember to keep it in sync.

[gbaraldi]

This slipped through the cracks in my PR sorry :(.

[jishnub]

This seems to be happening on v1.9.0 as well, so perhaps this needs to be backported?

Sorry, it seems to have been hardcoded in the package