[WIP] Clean up global grid parameters and grid shape

[Copilot]

• Clean up GlobalGridParams dataclass (model/common/src/icon4py/model/common/grid/icon.py):
  • Make grid_shape required (non-Optional)
  • Remove global_num_cells field (theoretical full-sphere count, not used in production)
  • Remove num_edges and num_vertices fields (only used in test definitions, grid size info belongs in GridConfig)
  • Remove characteristic_length field (dead field, never populated - computed in geometry.py)
  • Remove limited_area field (redundant with GridConfig.limited_area)
  • Remove compute_mean_cell_area_for_sphere function (dead code, never called)
  • Clean up __post_init__ to reflect removed fields
  • Update geometry_type and subdivision properties (no Optional return)
• Move fields to GridDescription (model/testing/src/icon4py/model/testing/definitions.py):
  • Add num_edges, num_vertices, limited_area fields to GridDescription
  • Update all grid description instances
• Update grid_manager.py: Remove global_num_cells from params construction
• Update initialization_utils.py (driver): Update GlobalGridParams construction
• Update py2fgen common.py: Update GlobalGridParams() construction
• Update test files:
  • test_icon.py: Remove global_num_cells, mean_cell_area from test parameters
  • test_grid_manager.py: Update test that checks global_num_cells, update params.num_edges/params.num_vertices references
  • test_gridfile.py: Update to use grid_descriptor fields
  • test_grid_refinement.py: Update params.limited_area references
  • MPI test files: Update params.limited_area references
• Run tests to verify changes
• Code review and security scan

[jenkins-apn]

Hi there, this is jenkins continuous integration...
Do you want me to verify this patch?

[jenkins-apn]

Hi there, this is jenkins continuous integration...
Do you want me to verify this patch?

[msimberg]

What I had more in mind for this is something like:

• torus_or_icosahedron_params: IcosahedronParams | TorusParams (with a better variable name)
• num_cells: int
• num_edges: int
• num_vertices: int
• limited_area: bool

More concretely:

• no None parameters, we should almost always be able to figure out what the parameters are
• following from the previous, torus/icosahedron parameters should be variant-like, not sometimes set, sometimes not set
• either all of num cells/edges/vertices are stored here, or none are
• limited area: not sure if this belongs here or not
• charecteristic length: probably belongs next to the mean edge lengths etc., it's directly derived from them

@jcanton @nfarabullini @OngChia do you have opinions on what this should look like?

[nfarabullini]

Here are my 2 cents:

• why did you add None to radius if it anyway has a default value? The Nonetype is only specified because the default value is None
• how about meshParams as an alternative to torus_or_icosahedron_params?
• where would you place limited_area instead?
• I'm ok if you move charecteristic_length

[msimberg]

• Because it doesn't make sense for torus to have a radius. It was a poor solution.
• Mesh params sounds like a more promising name, I'm also considering keeping something related to "grid" and "shape"... I think we've tried to avoid mixing "grid" and "mesh" since they are mostly the same as far as I know, and so far we prefer "grid"
• It's also in GridConfig. I don't know yet if we can keep only one of these.
• 👍 for characteristic length

[msimberg]

One more: num_cells, num_edges, num_vertices also have a helper class: HorizontalGridSize. We should probably reuse that here, or remove it from here if HorizontalGridSize anyway lives on its own.

[nfarabullini]

Ok then how about CustomGridParams because simply GridParams might be too generic. Thank you for answer the other questions. As far as I'm concerned, you can go ahead with the changes

[msimberg]

Yeah, agree on GridParams being quite generic. Not sure Custom makes it less generic, but I'll have a think about it. Thanks for the suggestions!