Adding interior of a field with indices in OutputWriter

[navidcy]

Attempt to close #3260.

using Oceananigans

model = HydrostaticFreeSurfaceModel(grid = RectilinearGrid(size = (5, 5, 4), x = (-1e3, 1e3), y = (-1e3, 1e3), z = (-1e3, 0)))

ηᵢ(x, y, z) = 1 * exp(-(x^2 + y^2) / 2 * (2e2)^2)

set!(model, η = ηᵢ)

simulation = Simulation(model, Δt=100, stop_time = 1000)

simulation.output_writers[:surface] = JLD2OutputWriter(model, (η=model.free_surface.η,),
                                                       schedule = TimeInterval(200),
                                                       filename = "surface",
                                                       overwrite_existing = true)

JLD2OutputWriter scheduled on TimeInterval(3.333 minutes):
├── filepath: ./surface.jld2
├── 1 outputs: η
├── array type: Array{Float64}
├── including: [:grid, :coriolis, :buoyancy, :closure]
└── max filesize: Inf YiB

(with @glwagner)

[glwagner]

Still some test to fix

[simone-silvestri]

here we can probably just return Colon?

[glwagner]

No, basically this returns either 1:N (for center fields) or 1:N+1 (for face fields) --- the range of interior points.

This convenience feature is a little annoying. Basically the user can control index ranges through two arguments. One specifies indices. The other specifies whether to chop halos or not. If the user askes for index::Colon but no halos, we have to return something like 1:N when the field covers all indices. The complication addressed in this PR is when the available indices are restricted (ie the output field is windowed)

[simone-silvestri]

Ok, (maybe not in this PR) but I wonder if we can make with_halo=true the default. An example of where this "restricing to the interior" might have unintended (or silent) consequences is when we try to compute derivatives on post-processed fields and find out that we cannot express derivatives across the whole field but we are restricted to one point less because the field has been effectively chopped off to only the interior silently (I am guilty of something like this happening to me). This seems more like a behavior we want to explicitly decide to have rather than it being implicitly enforced as a default.

[navidcy]

@glwagner I did this here as a quick workaround.

But why the NetCDFOutputWriter calls restrict_to_interior? Isn't this the job of the construct_output method to do that?

[glwagner]

My memory is hazy and I don't work with it much so I could make a mistake but I think NetCDFOutputWriter needs to have indices to generate dimensions (unlike JLD2OutputWriter).

[glwagner]

Doesn't follow YASGuide but we can make an exception...

[navidcy]

Damn.. Sorry... I was trying to make things cleaner by having the positional argument names

[glwagner]

ahaha no its ok, lets keep it.

[glwagner]

Does this supercede #2771 ?