`nodes()` returns the full grid when given a smaller view of a `Field`

[tomchor]

I'm not sure this is an issue per se, but I noticed that when a view of a Field is given to nodes() it always returns the nodes for the full grid, regardless of what the view is. For example:

julia> grid = RectilinearGrid(size=(8, 2, 2), extent=(1, 1, 1));

julia> c = XFaceField(grid);

julia> nodes(c)
(0.0:0.125:0.875, 0.25:0.5:0.75, -0.75:0.5:-0.25)

julia> nodes(view(c, 2:4, :, :))
(0.0:0.125:0.875, 0.25:0.5:0.75, -0.75:0.5:-0.25)

Looking at the code I see maybe this was always the intended the behavior, but I think it might be more useful to return the nodes corresponding to the indices of the view in this case. Curious about what everyone thinks.

[simone-silvestri]

I guess we never really implemented nodes for windowed fields, but I agree that it would be more useful to return the nodes of the actual windowed field.

[glwagner]

I think that would be a great addition. Windowing is design to be fully general, but is currently mostly used for "slices" (ie taking a single vertical level, plus the entirety of the other dimensions). As a result, this limitation of node hasn't come up. But if we want to properly support windowing in more than one direction, we need a corresponding nodes that extracts nodal values on the Field's indices.

[tomchor]

Okay, I'll probably tackle that soon then. I'll wait for some of PRs get merged since they're starting to pile up.

For context, this came up while coding #4730, but I suspect it can also make the code for our Makie extension simpler.

[glwagner]

I actually think that fields which are windowed in two dimensions cannot be plotted with the Makie extension. We already use:

Oceananigans.jl/ext/OceananigansMakieExt.jl

Line 150 in 5284729

fnodes = nodes(f)

and we only accommodate windowing in 1D.

However, an additional simplification would be to also adapt the tuple of coordinates; eg returning just two coordinates for 2D fields, etc. This is not an issue of windowing but rather an issue of inferring dimensionality via either Flat topology or Nothing location. I think the windowing will be simple to implement, but improvements to how we conceive of field dimensionality may run deeper and require more effort to fully clean up.

[navidcy]

This is what causes that I believe:

Oceananigans.jl/src/Fields/abstract_field.jl

Line 115 in 5284729

nodes(ψ::AbstractField; kwargs...) = nodes(ψ.grid, instantiated_location(ψ); kwargs...)

[glwagner]

Hmm yes. We need a more specific version that dispatches on Field (note, we should keep the version that uses AbstractField)