`resize!` failure after using ldiv! on v1.0

[ChrisRackauckas]

MWE:

using LinearAlgebra
function resizer()
  v = rand(7)
  w = similar(v)
  W = rand(7,7)
  luw = lu(W)
  ldiv!(vec(w),luw,v)
  resize!(v,3)
end
resizer()

The resize! fails because of the view on v which is built internally. Patching with

function LinearAlgebra.ldiv!(Y::AbstractVecOrMat, A::Factorization, B::AbstractVecOrMat)
    @assert !LinearAlgebra.has_offset_axes(Y, B)
    m, n = size(A, 1), size(A, 2)
    if m > n
        ldiv!(A, B)
        return copyto!(Y, B)
    else
        return ldiv!(A, copyto!(Y, B))
    end
end

fixes the issue. Note that this is a regression from v0.6 where

function resizer()
  v = rand(7)
  w = similar(v)
  W = rand(7,7)
  luw = lufact(W)
  A_ldiv_B!(w,luw,v)
  resize!(v,3)
end
resizer()

works.

Is that viewless version fine? Should I PR it? I don't really understand what supporting using only part of B is for. For reference the dispatch is

https://github.com/JuliaLang/julia/blob/f104ea4ec352519de76e52bf65ea7b3ed2dc6155/stdlib/LinearAlgebra/src/factorization.jl#L98-L108

Why not generally use B?

[andreasnoack]

Why not generally use B?

Because they are not the same size when the system is underdetermined.

[fredrikekre]

dup of #577?

[andreasnoack]

dup of #577?

No that one is the overdetermined branch

It's actually surprising to me that ldiv!(A, copyto!(Y, view(B, 1:m, :))) works. I hadn't realized copyto! behaved like this

julia> Ones = ones(2,2);

julia> OnesView = view(Ones, 1:2, :);

julia> Ones == OnesView && size(Ones) == size(OnesView)
true

julia> copyto!(zeros(3,2), Ones)
3×2 Array{Float64,2}:
 1.0  1.0
 1.0  0.0
 1.0  0.0

julia> copyto!(zeros(3,2), OnesView)
3×2 Array{Float64,2}:
 1.0  1.0
 1.0  1.0
 0.0  0.0

[JeffBezanson]

Uhoh.