Broadcasted setindex (surprisingly) silently doesn't do anything

[kshyatt]

MWE:

julia> using BlockArrays, LinearAlgebra

julia> A = mortar(reshape([randn(2, 2), randn(2, 2), randn(2, 2), randn(2, 2)], (2, 2)))
2×2-blocked 4×4 BlockMatrix{Float64}:
 -0.915837  -0.333415   │  -0.458008   0.521461
 -0.566032   1.29096    │   0.515333  -0.969364
 ───────────────────────┼──────────────────────
  0.449189  -0.0617162  │  -0.96811   -1.22779 
 -0.155907   1.19382    │  -2.00874   -0.362436

julia> for ii in axes(A, 1), jj in axes(A, 2)
          A[ii, jj] = ifelse( ii == jj, 2.0, 1.0)
       end

julia> A
2×2-blocked 4×4 BlockMatrix{Float64}:
 2.0  1.0  │  1.0  1.0
 1.0  2.0  │  1.0  1.0
 ──────────┼──────────
 1.0  1.0  │  2.0  1.0
 1.0  1.0  │  1.0  2.0

julia> A[diagind(A)] .= -1.0
4-element view(reshape(::BlockedArray{…,::Matrix{Float64}}, 16), 1:5:16) with eltype Float64:
 -1.0
 -1.0
 -1.0
 -1.0

julia> A
2×2-blocked 4×4 BlockMatrix{Float64}:
 2.0  1.0  │  1.0  1.0
 1.0  2.0  │  1.0  1.0
 ──────────┼──────────
 1.0  1.0  │  2.0  1.0
 1.0  1.0  │  1.0  2.0

It seems like A should've been modified in these last two steps, or warned me that I was doing something unsupported...

[dlfivefifty]

I think here's a simpler version of the bug:

julia> view(A,diagind(A))[1] = -1
-1

julia> A
2×2-blocked 4×4 BlockMatrix{Float64}:
 2.0  1.0  │  1.0  1.0
 1.0  2.0  │  1.0  1.0
 ──────────┼──────────
 1.0  1.0  │  2.0  1.0
 1.0  1.0  │  1.0  2.0

[dlfivefifty]

The problem is reshape currently makes a copy:

BlockArrays.jl/src/blockarray.jl

Line 586 in cd73466

Base.reshape(block_array::BlockArray, axes::Tuple{Union{Integer,Base.OneTo}, Vararg{Union{Integer,Base.OneTo}}}) =

I'm not entirely sure why it doesn't just use ReshapedArray. But perhaps the solution is to just change it to ReshapedArray, which make require adding support for block arrays