Fix bug for categorical pandas index with categories with EA dtype

doc/whats-new.rst

@@ -42,6 +42,8 @@ Deprecations
 Bug fixes
 ~~~~~~~~~
 
+- Fix dtype inference for ``pd.CategoricalIndex`` when categories are backed by a ``pd.ExtensionDtype`` (:pull:`8481`)
+
 
 Documentation
 ~~~~~~~~~~~~~

xarray/core/utils.py

@@ -114,6 +114,8 @@ def get_valid_numpy_dtype(array: np.ndarray | pd.Index):
     elif hasattr(array, "categories"):
         # category isn't a real numpy dtype
         dtype = array.categories.dtype
+        if not is_valid_numpy_dtype(dtype):
+            dtype = np.dtype("O")
     elif not is_valid_numpy_dtype(array.dtype):
         dtype = np.dtype("O")
     else:

xarray/tests/test_dataset.py

@@ -4697,6 +4697,17 @@ def test_from_dataframe_categorical(self) -> None:
         assert len(ds["i1"]) == 2
         assert len(ds["i2"]) == 2
 
+    def test_from_dataframe_categorical_string_categories(self):
+        cat = pd.CategoricalIndex(
+            pd.Categorical.from_codes(
+                np.array([1, 1, 0, 2]),
+                categories=pd.Index(["foo", "bar", "baz"], dtype="string"),
+            )
+        )
+        ser = pd.Series(1, index=cat)
+        ds = ser.to_xarray()
+        assert ds.coords.dtypes["index"] == np.dtype("O")
+
     @requires_sparse
     def test_from_dataframe_sparse(self) -> None:
         import sparse