rdiv! issue?

[MikaelSlevinsky]

julia> using BandedMatrices, LinearAlgebra

julia> A = BandedMatrix(rand(2, 1), (1, 0))
2×1 BandedMatrix{Float64} with bandwidths (1, 0):
 0.9360463133092783
 0.973726372662548

julia> B = Diagonal([1.0])
1×1 Diagonal{Float64, Vector{Float64}}:
 1.0

julia> A/B
ERROR: BandError: attempt to access Matrix{Float64} with bandwidths (0, 0) at band -1
Stacktrace:
  [1] banded_setindex!
    @ ~/.julia/dev/BandedMatrices/src/banded/BandedMatrix.jl:614 [inlined]
  [2] setindex!
    @ ~/.julia/dev/BandedMatrices/src/banded/BandedMatrix.jl:623 [inlined]
  [3] _setindex!
    @ ./abstractarray.jl:1426 [inlined]
  [4] setindex!
    @ ./abstractarray.jl:1396 [inlined]
  [5] macro expansion
    @ ./broadcast.jl:1004 [inlined]
  [6] macro expansion
    @ ./simdloop.jl:77 [inlined]
  [7] copyto!
    @ ./broadcast.jl:1003 [inlined]
  [8] copyto!
    @ ./broadcast.jl:956 [inlined]
  [9] copyto!(dest::BandedMatrix{Float64, Matrix{…}, Base.OneTo{…}}, bc::Base.Broadcast.Broadcasted{BandedMatrices.BandedStyle, Tuple{…}, typeof(*), Tuple{…}})
    @ BandedMatrices ~/.julia/dev/BandedMatrices/src/generic/broadcast.jl:55
 [10] copy
    @ ./broadcast.jl:928 [inlined]
 [11] materialize
    @ ./broadcast.jl:903 [inlined]
 [12] copy(M::ArrayLayouts.Rdiv{BandedMatrices.BandedColumns{…}, ArrayLayouts.DiagonalLayout{…}, BandedMatrix{…}, Diagonal{…}})
    @ ArrayLayouts ~/.julia/packages/ArrayLayouts/B2wRU/src/diagonal.jl:53
 [13] materialize
    @ ~/.julia/packages/ArrayLayouts/B2wRU/src/ldiv.jl:22 [inlined]
 [14] rdiv
    @ ~/.julia/packages/ArrayLayouts/B2wRU/src/ldiv.jl:99 [inlined]
 [15] /(A::BandedMatrix{Float64, Matrix{Float64}, Base.OneTo{Int64}}, D::Diagonal{Float64, Vector{Float64}})
    @ ArrayLayouts ~/.julia/packages/ArrayLayouts/B2wRU/src/ldiv.jl:190
 [16] top-level scope
    @ REPL[15]:1
Some type information was truncated. Use `show(err)` to see complete types.

julia> Matrix(A)/Matrix(B)
2×1 Matrix{Float64}:
 0.9360463133092783
 0.973726372662548

julia> Matrix(A)/B
2×1 Matrix{Float64}:
 0.9360463133092783
 0.973726372662548

julia> A*inv(B) # this seems wrong
2×1 BandedMatrix{Float64} with bandwidths (0, 0):
 0.9360463133092783
  ⋅ 

julia> 

[MikaelSlevinsky]

Seems to also be an issue with rmul!

[MikaelSlevinsky]

Going through the stack, it seems like one issue is that mul for Banded * Diagonal gets converted to a broadcasting over the BandedMatrix with a matrix of permuted data from the diagonal matrix. For 1x1, that equivalent to the incorrect behaviour below:

julia> A*B
2×1 BandedMatrix{Float64} with bandwidths (0, 0):
 0.9360463133092783
  ⋅ 

julia> A .* Matrix(B)
2×1 BandedMatrix{Float64} with bandwidths (0, 0):
 0.9360463133092783
  ⋅ 

julia> Matrix(A) .* Matrix(B)
2×1 Matrix{Float64}:
 0.9360463133092783
 0.973726372662548

julia> Matrix(A) * Matrix(B)
2×1 Matrix{Float64}:
 0.9360463133092783
 0.973726372662548

julia> 

[dlfivefifty]

Looks like it’s just a typo for the bandwidth

[MikaelSlevinsky]

Yes, it would seem like it's in this file https://github.com/JuliaLinearAlgebra/BandedMatrices.jl/blob/master/src/generic/broadcast.jl but not sure exactly where yet