Add helper function bounding_box

[lijas]

I think it can be nice to add some grid utils. In this PR I am adding compute_bouding_box() , which I have recently needed for some pre-processing stuff.

Note. The current code is type unstable for some reason. Both max_vertex and min_vertex are both Core.Box. Anyone understands why?

[KnutAM]

Interesting problem - don't know why it is type unstable but could solve it if Tensors would support map

Base.map(f, iters::T...) where {T<:AbstractTensor} = T(map(f, Tensors.get_data.(iters)...))
function compute_bounding_box(grid::AbstractGrid{dim}) where {dim}
    T = get_coordinate_eltype(grid)
    min_vertex = Vec{dim,T}(i->T(+Inf))
    max_vertex = Vec{dim,T}(i->T(-Inf))
    for node in getnodes(grid)
        x = get_node_coordinate(node)
        max_vertex = map(max, max_vertex, x)
        min_vertex = map(min, min_vertex, x)
    end
    return min_vertex, max_vertex
end

Somehow defining an output tmp = max(max_vertex[1], x[1]) and only considering the first node, tmp was set as Any if max_vertex and min_vertex were updated and outputted, but type stable otherwise - so perhaps it is just inference giving up for some reason?

[lijas]

I asked on slack helpdesk, and got this reply:

This will be slow, because Julia will need to pass the value boxed into the closure to make sure it gets the value of the right type.
This is a bug in Julia, because type inference could be used to determine that the box is unnecessary, but currently in Julia, the compiler step that creates the box happens before type inference, so it can't be optimised away (or so I'm told - I don't understand it).
The solution is:
Either, typeassert your variables so that the box have less impact on performance (but it'll still allocate)
Or, use a let statement