Fix pandas 3 compatibility

openeo/extra/job_management/_manager.py

@@ -94,6 +94,25 @@ def normalize_df(self, df: pd.DataFrame) -> pd.DataFrame:
         """
         new_columns = {col: req.default for (col, req) in self._requirements.items() if col not in df.columns}
         df = df.assign(**new_columns)
+        # Apply dtype conversions to ensure compatibility with pandas 3's stricter type checking.
+        # This is especially important for columns that may contain NaN values but need to be strings.
+        # Only convert columns where the current dtype doesn't match the required dtype.
+        # For string columns (dtype="str"), we convert to "object" to maintain backward compatibility
+        # (pandas 3.x "str" creates StringDtype, pandas 2.x "str" was an alias for object).
+        dtype_conversions = {}
+        for col, req in self._requirements.items():
+            if col in df.columns:
+                current_dtype = str(df[col].dtype)
+                required_dtype = req.dtype
+                # Only convert if the current dtype is different and not already compatible
+                if required_dtype == "str" and current_dtype not in {"object", "string", "str"}:
+                    # Convert to object for backward compatibility
+                    dtype_conversions[col] = "object"
+                elif required_dtype not in {"object", "str"} and current_dtype != required_dtype:
+                    # For other dtypes (e.g., float64), convert as specified
+                    dtype_conversions[col] = required_dtype
+        if dtype_conversions:
+            df = df.astype(dtype_conversions)
         return df
 
     def dtype_mapping(self) -> Dict[str, str]:

openeo/udf/run_code.py

@@ -77,7 +77,16 @@ def _get_annotation_str(annotation: Union[str, type]) -> str:
 
 
 def _annotation_is_pandas_series(annotation) -> bool:
-    return annotation in {pandas.Series, _get_annotation_str(pandas.Series)}
+    # Support both pandas 2.x ("pandas.core.series.Series") and pandas 3.x ("pandas.Series")
+    # We explicitly list both paths to ensure cross-version compatibility:
+    # - UDFs written with pandas 2.x annotations should work in pandas 3.x environments
+    # - UDFs written with pandas 3.x annotations should work in pandas 2.x environments
+    return annotation in {
+        pandas.Series,
+        _get_annotation_str(pandas.Series),  # Current pandas version's path
+        "pandas.core.series.Series",  # Explicit pandas 2.x path for cross-version support
+        "pandas.Series",  # Explicit pandas 3.x path for cross-version support
+    }
 
 
 def _annotation_is_udf_datacube(annotation) -> bool:

setup.py

@@ -80,7 +80,7 @@
         "shapely>=1.6.4",
         "numpy>=1.17.0",
         "xarray>=0.12.3,<2025.01.2",  # TODO #721 xarray non-nanosecond support
-        "pandas>0.20.0,<3.0.0",  # TODO pandas 3 compatibility https://github.com/Open-EO/openeo-python-client/issues/856
+        "pandas>0.20.0",  # pandas 3 compatibility added
         # TODO #578: pystac 1.5.0 is highest version available for lowest Python version we still support (3.7).
         "pystac>=1.5.0",
         "deprecated>=1.2.12",

tests/udf/test_run_code.py

@@ -28,7 +28,10 @@
 
 @pytest.mark.parametrize(["annotation", "expected"], [
     ("str", "str"),
-    (pandas.Series, "pandas.core.series.Series"),
+    # pandas.Series module path changed in pandas 3.0:
+    # pandas 2.x: "pandas.core.series.Series"
+    # pandas 3.x: "pandas.Series"
+    (pandas.Series, _get_annotation_str(pandas.Series)),
     (XarrayDataCube, "openeo.udf.xarraydatacube.XarrayDataCube"),
     (UdfData, "openeo.udf.udf_data.UdfData"),
     (str, "str"),
@@ -40,7 +43,9 @@ def test_get_annotation_str(annotation, expected):
 
 def test_annotation_is_pandas_series():
     assert _annotation_is_pandas_series(pandas.Series) is True
-    assert _annotation_is_pandas_series("pandas.core.series.Series") is True
+    # Support both pandas 2.x and 3.x module paths
+    assert _annotation_is_pandas_series("pandas.core.series.Series") is True  # pandas 2.x
+    assert _annotation_is_pandas_series("pandas.Series") is True  # pandas 3.x
 
 
 def test_annotation_is_udf_datacube():