hyperractangles' compute_bbox seems to break KDTree due to setindex() for non StaticArrays types

[AbdAlazezAhmed]

Last release broke something in PointEvalHandler in Ferrite.jl, I noticed that from CI in termi-official/Thunderbolt.jl#153
MWE:

julia> grid = generate_grid(Quadrilateral, (20,20))
Grid{2, Quadrilateral, Float64} with 400 Quadrilateral cells and 441 nodes

julia> KDTree(reinterpret(Vec{2,Float64}, getnodes(grid)))
ERROR: MethodError: no method matching setindex(::Vector{Float64}, ::Float64, ::Int64)

Closest candidates are:
  setindex(::StaticArrays.SHermitianCompact{N, T, L}, ::Any, ::Int64) where {N, T, L}
   @ StaticArrays ~/.julia/packages/StaticArrays/MSJcA/src/SHermitianCompact.jl:125
  setindex(::StaticArraysCore.StaticArray, ::Any, ::Int64)
   @ StaticArrays ~/.julia/packages/StaticArrays/MSJcA/src/deque.jl:185
  setindex(::StaticArraysCore.StaticArray, ::Any, ::Any...)
   @ StaticArrays ~/.julia/packages/StaticArrays/MSJcA/src/deque.jl:198
  ...

Stacktrace:
 [1] build_KDTree(index::Int64, data::Base.ReinterpretArray{…}, data_reordered::Vector{…}, hyper_rec::NearestNeighbors.HyperRectangle{…}, split_vals::Vector{…}, split_dims::Vector{…}, indices::Vector{…}, indices_reordered::Vector{…}, range::UnitRange{…}, tree_data::NearestNeighbors.TreeData, reorder::Bool)
   @ NearestNeighbors ~/.julia/dev/NearestNeighbors/src/kd_tree.jl:134
 [2] KDTree(data::Base.ReinterpretArray{…}, metric::Euclidean; leafsize::Int64, storedata::Bool, reorder::Bool, reorderbuffer::Vector{…})
   @ NearestNeighbors ~/.julia/dev/NearestNeighbors/src/kd_tree.jl:59
 [3] KDTree
   @ ~/.julia/dev/NearestNeighbors/src/kd_tree.jl:19 [inlined]
 [4] KDTree(data::Base.ReinterpretArray{Vec{2, Float64}, 1, Node{2, Float64}, Vector{Node{2, Float64}}, false})
   @ NearestNeighbors ~/.julia/dev/NearestNeighbors/src/kd_tree.jl:19
 [5] top-level scope
   @ REPL[42]:1
Some type information was truncated. Use `show(err)` to see complete types.

It seems like broadcasting in

NearestNeighbors.jl/src/hyperrectangles.jl

Lines 6 to 10 in a20c107

	function compute_bbox(data::AbstractVector{V}) where {V <: AbstractVector}
	mins = mapreduce(identity, (a, b) -> min.(a, b), data; init=fill(Inf,V))
	maxes = mapreduce(identity, (a, b) -> max.(a, b), data; init=fill(-Inf,V))
	return HyperRectangle(mins, maxes)
	end

results in Vector not V. IIUC it should be sth like

 function compute_bbox(data::AbstractVector{V}) where {V <: AbstractVector} 
     mins = mapreduce(identity, (a, b) -> V(min.(a, b)), data; init=fill(Inf,V)) 
     maxes = mapreduce(identity, (a, b) -> V(max.(a, b)), data; init=fill(-Inf,V)) 
     return HyperRectangle(mins, maxes) 
 end 

*Not sure about performance tho, it's just a quick simple fix *
However, even after preserving the type V here setindex is used in many places which is only defined for <:StaticArray IIRC.
If this is an actual issue not just me misusing the interface I believe we can either solve it by checking for immutability like:

    new_maxes = @inbounds (ismutabletype(V) ? setindex! : setindex)(hyper_rec.maxes, split_val, split_dim)

or maybe have setindex dispatched for Vector?
also, IIUC neither setindex nor setindex! is defined for Tensor. So if the goal was to retain V maybe we need to dispatch setindex for Vec.

For now the simplest workaround I use locally is

function compute_bbox(data::AbstractVector{V}) where {V <: AbstractVector}
    mins = mapreduce(identity, (a, b) -> min.(a, b), data; init=fill(Inf,V))
    maxes = mapreduce(identity, (a, b) -> max.(a, b), data; init=fill(-Inf,V))
    return HyperRectangle(SVector{length(mins)}(mins), SVector{length(maxes)}(maxes))
end

Thanks!

[fredrikekre]

See also #203