add convenience constructors for grids

[simonbyrne]

cc @Sbozzolo

• Code follows the style guidelines OR N/A.
• Unit tests are included OR N/A.
• Code is exercised in an integration test OR N/A.
• Documentation has been added/updated OR N/A.

[simonbyrne]

The first question is: what should we name the aliases? Currently I've gone with the simplest possible names I could come up with:

• ColumnGrid for a single column in the z direction
  • requires renaming an existing object, but that wasn't exported
• PlaneGrid for a XZ plane
• BoxGrid for a XYZ box
• ExtrudedCubedSphereGrid for a cubed sphere
  • any thoughts on a simpler name?

One question is whether we should be a bit more specific on the names:

• should we include the coordinate in the names (say XZPlaneGrid instead of PlaneGrid?)
• should we include the discretization type (FDColumnGrid, FDSEPlaneGrid?)

[simonbyrne]

• PlaneGrid for a XZ plane

@OsKnoth suggested calling a SliceGrid

[simonbyrne]

The other interface questions:

• how to handle stretching: currently I have a z_stretch argument (not dz_X args like in ClimaAtmos). Default is Uniform()
• topography is not yet handled, but will probably be a single argument (topography=Flat() default)

[simonbyrne]

I am a bit reluctant to simply call it a CubedSphereGrid, as a sphere is defined as a 2D manifold, not a 3D object. The domain we're operating in is usually called a spherical shell.

Some options:

• CubedSphere3DGrid
• CubedSphereShellGrid

[Sbozzolo]

I am a bit reluctant to simply call it a CubedSphereGrid, as a sphere is defined as a 2D manifold, not a 3D object. The domain we're operating in is usually called a spherical shell.

Some options:

• CubedSphere3DGrid
• CubedSphereShellGrid

CubedSphere3DGrid would be different from all the other ones and invites the question of "what about CubedSphere2DGrid. CubedSphereShellGrid is good enough (better than ExtrudedCubedSphereGrid in my opinion).
I'd dare to say that maybe it could be just SphereShellGrid. While technically incorrect, the word Cubed conveys no extra meaning once it is established that all the spherical shells in ClimaCore/ClimaAtmos are always cubed.

[OsKnoth]

Cubed Sphere Grid is used in the community, it is a discrete manifold, which approximates the sphere, also extruded grid is used normally, in all cases we start with two-dimensional manifolds and than go to the third dimension, furthermore this construction is connected to the function spaces (spectral versus piecewise constant in the third dimension, or piecewise linear for w), it makes no sense to look for new namings outside the community.

[simonbyrne]

@Sbozzolo any more thoughts on this?

[Sbozzolo]

@Sbozzolo any more thoughts on this?

Can you please provide a short description on how I would go about identifying the Grid that is being used (e.g., for dispatch in ClimaAtmos)?

[simonbyrne]

f(grid::Grids.PlaneGrid) = ...
f(grid::Grids.BoxGrid) = ...
f(grid::Grids.ExtrudedCubedSphereGrid) = ...

[Sbozzolo]

f(grid::Grids.PlaneGrid) = ...
f(grid::Grids.BoxGrid) = ...
f(grid::Grids.ExtrudedCubedSphereGrid) = ...

Are we going with Plane or Slice?

I still think that ExtrudedCubedSphereGrid, while descriptive, is unnecessarily complex for an important user-facing type. (But I am fine keeping it).

As a next step, we should implement accessors: it should be trivial to get the radius from a sphere, the heigth of a column, and so on.

[simonbyrne]

I still think that ExtrudedCubedSphereGrid, while descriptive, is unnecessarily complex for an important user-facing type. (But I am fine keeping it).

How about CubedSphereShellGrid?

[simonbyrne]

Okay, I've renamed them.

[charleskawczynski]

@Sbozzolo what do you think about adding doc strings to all of these convenience constructors, so that users know what the kwargs are? I think just copying the code itself into the doc string should be pretty self explanatory, and it'll be easy to keep up-to-date.

[szy21]

@charleskawczynski What is the status of this PR?

[charleskawczynski]

I need to rebase this, and make a few changes

[charleskawczynski]

Superseded by #1848.