`pd.concat` with `GeoDataFrame`s

[zmoon]

import geopandas as gpd
import pandas as pd
from shapely import Point

gdf1 = gpd.GeoDataFrame(geometry=[Point(0, 0)])
gdf2 = gpd.GeoDataFrame(geometry=[Point(1, 1)])

gdf: gpd.GeoDataFrame = pd.concat([gdf1, gdf2], ignore_index=True)
assert isinstance(gdf, gpd.GeoDataFrame)
print(gdf)

      geometry
0  POINT (0 0)
1  POINT (1 1)

t.py:8: error: Incompatible types in assignment (expression has type "DataFrame", variable has type "GeoDataFrame")  [assignment]
Found 1 error in 1 file (checked 1 source file)

• pandas-stubs 2.3.2.250926
• mypy 1.18.2
• geopandas 1.1.1

[Dr-Irv]

I'm not sure that we can support this usage. You are passing subclasses of DataFrame and expecting that concat return that same subclass for typing purposes. But maybe there is a way to do this within our stubs via some clever use of TypeVar, but I'm concerned it will gunk up the stubs.

[cmp0xff]

It might be a good idea to be tolerant and give covariant typing, such that pd.conat([identical subclasses of DataFrame]) -> that subclass of DataFrame.

If we are too strict, we are at the risk of enforcing something like np.ndarray + pd.Series -> np.ndarray, which the downstream modules cannot override from their side.

[loicdiridollou]

I think the trick here with a TypeVar is how we defined the bound of that variable, we could add an overload for:

DataFrameCompatType = TypeVar("DataFrameCompatType", bound=...)

@overload
def concat(
    objs: Sequence[DataFrameCompatType],
    ...
) -> DataFrameCompatType

I think that should work, let me know @Dr-Irv and @cmp0xff if that is something we would want to do in the stubs.

[Dr-Irv]

I'm OK with adding something like the TypeVar, but am worried that we fix it for concat, and then we have to start fixing this issue in many places.

And testing becomes a challenge, because we have to create some "fake" subclass of DataFrame that tests concat (and probably other methods and functions)