hcat/vcat/hvcat bug

[andreasvarga]

The following simple matrix construction fails (joint finding with John Verzani #364):

julia> using LinearAlgebra

julia> [1 2; I]
ERROR: ArgumentError: number of columns of each array must match (got (2, 1))
Stacktrace:
 [1] _typed_vcat(#unused#::Type{Any}, A::Tuple{Matrix{Any}, Matrix{Any}})
   @ Base .\abstractarray.jl:1553
 [2] typed_vcat(::Type{Any}, ::Matrix{Any}, ::Matrix{Any})
   @ Base .\abstractarray.jl:1567
 [3] typed_hvcat(::Type{Any}, ::Tuple{Int64, Int64}, ::Int64, ::Vararg{Any, N} where N)
   @ Base .\abstractarray.jl:1957
 [4] hvcat(::Tuple{Int64, Int64}, ::Int64, ::Vararg{Any, N} where N)
   @ Base .\abstractarray.jl:1932
 [5] top-level scope
   @ REPL[9]:1

but [[1] [2]; I] and [[1 2]; I] work correctly:

julia> [[1] [2]; I]
3×2 Matrix{Int64}:
 1  2
 1  0
 0  1

julia>  [[1 2]; I]
3×2 Matrix{Int64}:
 1  2
 1  0
 0  1

I think, a simple fix would be to replace line 393 of LinearAlgebra.UniformScaling
promote_to_arrays_(n::Int, ::Type, a::Number) = a
with
promote_to_arrays_(n::Int, ::Type, a::Number) = [a]

[ViralBShah]

Can you submit a PR? Seems like it would be good to fix.

[dkarrasch]

I think, a simple fix would be to replace line 393 of LinearAlgebra.UniformScaling
promote_to_arrays_(n::Int, ::Type, a::Number) = a
with
promote_to_arrays_(n::Int, ::Type, a::Number) = [a]

That was my very first attempt in the extension of *cat to numbers, but was considered unfortunate because it creates an (unnecessary) array for each number... which gets eventually included in the concatenated matrix. The issue seems to be that the size of the uniform scaling is deduced incorrectly?

[martinholters]

Fixed on master?

julia> [1 2; I]
3×2 Matrix{Int64}:
 1  2
 1  0
 0  1

[dkarrasch]

Fixed on master?

julia> [1 2; I]
3×2 Matrix{Int64}:
 1  2
 1  0
 0  1

True. I thought JuliaLang/julia@18f2f9f might have been included earlier, but it's going to be released with v1.8.

[andreasvarga]

The following wrong outputs are hopefully also fixed:

julia> [1 I]
1×2 Matrix{Any}:
 1  UniformScaling{Bool}(true)

julia> [1; I]
2-element Vector{Any}:
 1
  UniformScaling{Bool}(true)

julia> [1 1; 1 I]
2×2 Matrix{Any}:
 1  1
 1   UniformScaling{Bool}(true)

[martinholters]

Yes:

julia> [1 I]
1×2 Matrix{Int64}:
 1  1

julia> [1; I]
2×1 Matrix{Int64}:
 1
 1

julia> [1 1; 1 I]
2×2 Matrix{Int64}:
 1  1
 1  1