Bug with Makie extension

[navidcy]

The code snippet below is not working as expected. I explain.

This works fine:

using Oceananigans, CairoMakie

underlying_grid = RectilinearGrid(size = (250, 125),
                                  x = (-1000e3, 1000e3),
                                  z = (-2e3, 0),
                                  halo = (4, 4),
                                  topology = (Periodic, Flat, Bounded))

bottom(x) = - underlying_grid.Lz + 500 * exp(-x^2 / (2*(20e3)^2))

grid = ImmersedBoundaryGrid(underlying_grid, GridFittedBottom(bottom))

lines(grid.immersed_boundary.bottom_height)

and gives:

which is all good! But:

x, y, z = nodes(grid.immersed_boundary.bottom_height)

lines(x, grid.immersed_boundary.bottom_height)

Perhaps this is a bit informative

lines(x/1e3, grid.immersed_boundary.bottom_height)

Originally posted by @navidcy in #3725 (comment)

[glwagner]

Not sure but I'm guessing to fix this we have to overload lines. It seems like it computes that it's going to be making a 3D plot at some point before convert_arguments is called.. not sure

[glwagner]

What if you make the axis first?

[navidcy]

Actually that worked!

using Oceananigans, CairoMakie

underlying_grid = RectilinearGrid(size = (250, 125),
                                  x = (-1000e3, 1000e3),
                                  z = (-2e3, 0),
                                  halo = (4, 4),
                                  topology = (Periodic, Flat, Bounded))

bottom(x) = - underlying_grid.Lz + 500 * exp(-x^2 / (2*(20e3)^2))

grid = ImmersedBoundaryGrid(underlying_grid, GridFittedBottom(bottom))

x, y, z = nodes(grid.immersed_boundary.bottom_height)

fig = Figure()
ax = Axis(fig[1, 1])
lines!(ax, x/1e3, grid.immersed_boundary.bottom_height)
fig

OK. I suppose if you don't create the axis Makie thinks this is a 3D field and then creates Axis3 object... So it's good to point out somewhere that to visualize it's better to create axis first?

[glwagner]

Or we can dig into Makie and maybe try to fix the wrong axis assignment?

[glwagner]

I think I figured it out, basically there is a function args_preferred_axis that determines which Axis type to use (when it's not specified), and because Field isa AbstractArray{3} it ends up choosing LScene. So the simple fix here is just to say that Field does not have a preferred axis type. As a result, when the axis is not provided, it is determined later on (by the converted arguments, rather than the un-converted Field).

We could in the future attempt to figure out the "effective dimensionality" of a Field and then use that to determine whether it has a preferred axis type.