refactor gis cell init to accept kwargs

[wang-boyu]

This PR updates constructors of raster cell subclasses, from for example

class LakeCell(mesa_geo.Cell):
    def __init__(self, model, pos=None, indices=None):
        super().__init__(model, pos, indices)
        ...

to

class LakeCell(mesa_geo.Cell):
    def __init__(self, model=None, **kwargs):
        super().__init__(model, **kwargs)
        ...

This is mainly due to changes to raster cell constructor from mesa/mesa-geo#299, where we deprecated cell.indices to cell.rowcol, and added cell.xy. In addition, when we initialize cells when creating a raster layer, this is how cell constructor is called now (link to code):

cell = self.cell_cls(
    model,
    pos=(grid_x, grid_y),
    rowcol=(row_idx, col_idx),
    xy=xy,
)

where cell_cls is the user-defined subclass of raster cell, such as the LakeCell above.

By putting all parameters but model into **kwargs, this PR demonstrates how to subclass mesa-geo's raster cell class in case there are further changes to constructor API in the future.

[EwoutH]

Thanks for the PR! I fixed another CI issue in 9567bdf, so CI should display the state of the example models again.

It seems to fix the main issue, however, this does expose a new issue:

DeprecationWarning: Cell.pos setter is deprecated and will be read-only in a future release.

Is that also related to this PR?

Further, can we do anything about this change being breaking to user models (see mesa/mesa-geo#299 (comment))?

[wang-boyu]

Yes this warning was introduced in mesa/mesa-geo#299. Since cells don't move, I intend to make their pos attribute to be read-only. Previously both setter and getter are available.

Could you point me to the failed github action / tests?

[EwoutH]

https://github.com/mesa/mesa-examples/actions/runs/21634731514/job/63382599658?pr=320

[wang-boyu]

This is interesting. It's probably because there's a deprecation warning from mesa-geo on pos setter, while mesa's base agent class still tries to set agent.pos = None by default.

I can remove the deprecation warning, since it conflicts with base agent class in Mesa. In the future for Mesa-Geo it might be better not to treat cells as agents (that is, not part of AgentSet) but have a dedicated cell container such as CellCollection that behaves like AgentSet, but with no automatically assigned agent id and fixed position.

[EwoutH]

mesa's base agent class still tries to set agent.pos = None by default.

@quaquel can/should we remove this?

[quaquel]

@quaquel can/should we remove this?

Not for 3.5 because of semver, but yes in 4 were it will be replaced with the Locatable protocol. No idea how that will work downstream for either agent or cell. We might just have a generic LocatableAgent class.