From 946dd787f9076ffdbfd2a0d3571559005207dea6 Mon Sep 17 00:00:00 2001
From: ilan-gold <ilanbassgold@gmail.com>
Date: Tue, 22 Jul 2025 15:41:51 +0200
Subject: [PATCH 01/26] fix: keep dtype as `object` for `pd.StringDtype` in
 `safe_cast_to_index`

---
 pyproject.toml         | 1 +
 xarray/core/indexes.py | 2 ++
 2 files changed, 3 insertions(+)

diff --git a/pyproject.toml b/pyproject.toml
index 5e5fd00328b..c7e1e04d85f 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -33,6 +33,7 @@ accel = [
   "numba>=0.59",
   "flox>=0.9",
   "opt_einsum",
+  "numpy<2.3",    # numba has not updated yet: https://github.com/numba/numba/issues/10105
 ]
 complete = ["xarray[accel,etc,io,parallel,viz]"]
 io = [
diff --git a/xarray/core/indexes.py b/xarray/core/indexes.py
index c233c6911e4..1fe36956fc1 100644
--- a/xarray/core/indexes.py
+++ b/xarray/core/indexes.py
@@ -536,6 +536,8 @@ def safe_cast_to_index(array: Any) -> pd.Index:
                 kwargs["dtype"] = "float64"
 
         index = pd.Index(to_numpy(array), **kwargs)
+        if isinstance(index.dtype, pd.StringDtype):
+            index = index.astype("O")
 
     return _maybe_cast_to_cftimeindex(index)
 

From 147e3a7216a632a3520fe9596ece382465e245f4 Mon Sep 17 00:00:00 2001
From: ilan-gold <ilanbassgold@gmail.com>
Date: Tue, 22 Jul 2025 15:42:40 +0200
Subject: [PATCH 02/26] chore: comment

---
 xarray/core/indexes.py | 1 +
 1 file changed, 1 insertion(+)

diff --git a/xarray/core/indexes.py b/xarray/core/indexes.py
index 1fe36956fc1..df7cce3d6e7 100644
--- a/xarray/core/indexes.py
+++ b/xarray/core/indexes.py
@@ -536,6 +536,7 @@ def safe_cast_to_index(array: Any) -> pd.Index:
                 kwargs["dtype"] = "float64"
 
         index = pd.Index(to_numpy(array), **kwargs)
+        # See https://github.com/pydata/xarray/issues/10553
         if isinstance(index.dtype, pd.StringDtype):
             index = index.astype("O")
 

From 3cf92dd966733fdd9eba904c98b01510c68b5bfa Mon Sep 17 00:00:00 2001
From: ilan-gold <ilanbassgold@gmail.com>
Date: Wed, 23 Jul 2025 11:58:48 +0200
Subject: [PATCH 03/26] fix: broader fix

---
 xarray/core/extension_array.py        |  6 +++---
 xarray/core/indexes.py                |  7 ++-----
 xarray/core/indexing.py               | 13 ++++++-------
 xarray/core/utils.py                  | 14 ++++++++++++++
 xarray/core/variable.py               |  8 ++++++--
 xarray/tests/test_pandas_to_xarray.py |  5 ++++-
 6 files changed, 35 insertions(+), 18 deletions(-)

diff --git a/xarray/core/extension_array.py b/xarray/core/extension_array.py
index d85f7e66b55..b752f7141af 100644
--- a/xarray/core/extension_array.py
+++ b/xarray/core/extension_array.py
@@ -11,7 +11,7 @@
 from pandas.api.types import is_extension_array_dtype
 
 from xarray.core.types import DTypeLikeSave, T_ExtensionArray
-from xarray.core.utils import NDArrayMixin
+from xarray.core.utils import NDArrayMixin, is_allowed_extension_array
 
 HANDLED_EXTENSION_ARRAY_FUNCTIONS: dict[Callable, Callable] = {}
 
@@ -101,9 +101,9 @@ def __post_init__(self):
         # This does not use the UNSUPPORTED_EXTENSION_ARRAY_TYPES whitelist because
         # we do support extension arrays from datetime, for example, that need
         # duck array support internally via this class.
-        if isinstance(self.array, pd.arrays.NumpyExtensionArray):
+        if not is_allowed_extension_array(self.array):
             raise TypeError(
-                "`NumpyExtensionArray` should be converted to a numpy array in `xarray` internally."
+                "`NumpyExtensionArray` or string dtype should be converted to a numpy array in `xarray` internally."
             )
 
     def __array_function__(self, func, types, args, kwargs):
diff --git a/xarray/core/indexes.py b/xarray/core/indexes.py
index df7cce3d6e7..d22fc37aa4f 100644
--- a/xarray/core/indexes.py
+++ b/xarray/core/indexes.py
@@ -23,6 +23,7 @@
     Frozen,
     emit_user_level_warning,
     get_valid_numpy_dtype,
+    is_allowed_extension_array_dtype,
     is_dict_like,
     is_scalar,
 )
@@ -536,9 +537,6 @@ def safe_cast_to_index(array: Any) -> pd.Index:
                 kwargs["dtype"] = "float64"
 
         index = pd.Index(to_numpy(array), **kwargs)
-        # See https://github.com/pydata/xarray/issues/10553
-        if isinstance(index.dtype, pd.StringDtype):
-            index = index.astype("O")
 
     return _maybe_cast_to_cftimeindex(index)
 
@@ -669,9 +667,8 @@ def __init__(
 
         self.index = index
         self.dim = dim
-
         if coord_dtype is None:
-            if pd.api.types.is_extension_array_dtype(index.dtype):
+            if is_allowed_extension_array_dtype(index.dtype):
                 cast(pd.api.extensions.ExtensionDtype, index.dtype)
                 coord_dtype = index.dtype
             else:
diff --git a/xarray/core/indexing.py b/xarray/core/indexing.py
index 8e4458fb88f..c98175578f8 100644
--- a/xarray/core/indexing.py
+++ b/xarray/core/indexing.py
@@ -24,6 +24,8 @@
     NDArrayMixin,
     either_dict_or_kwargs,
     get_valid_numpy_dtype,
+    is_allowed_extension_array,
+    is_allowed_extension_array_dtype,
     is_duck_array,
     is_duck_dask_array,
     is_scalar,
@@ -1763,12 +1765,12 @@ def __init__(
         self.array = safe_cast_to_index(array)
 
         if dtype is None:
-            if pd.api.types.is_extension_array_dtype(array.dtype):
+            if is_allowed_extension_array(array):
                 cast(pd.api.extensions.ExtensionDtype, array.dtype)
                 self._dtype = array.dtype
             else:
                 self._dtype = get_valid_numpy_dtype(array)
-        elif pd.api.types.is_extension_array_dtype(dtype):
+        elif is_allowed_extension_array_dtype(dtype):
             self._dtype = cast(pd.api.extensions.ExtensionDtype, dtype)
         else:
             self._dtype = np.dtype(cast(DTypeLike, dtype))
@@ -1816,10 +1818,7 @@ def get_duck_array(self) -> np.ndarray | PandasExtensionArray:
         # We return an PandasExtensionArray wrapper type that satisfies
         # duck array protocols.
         # `NumpyExtensionArray` is excluded
-        if pd.api.types.is_extension_array_dtype(self.array) and not isinstance(
-            self.array.array,
-            pd.arrays.NumpyExtensionArray,  # type: ignore[attr-defined]
-        ):
+        if is_allowed_extension_array(self.array):
             from xarray.core.extension_array import PandasExtensionArray
 
             return PandasExtensionArray(self.array.array)
@@ -1916,7 +1915,7 @@ def copy(self, deep: bool = True) -> Self:
 
     @property
     def nbytes(self) -> int:
-        if pd.api.types.is_extension_array_dtype(self.dtype):
+        if is_allowed_extension_array(self.array):
             return self.array.nbytes
 
         dtype = self._get_numpy_dtype()
diff --git a/xarray/core/utils.py b/xarray/core/utils.py
index 562706a1ac0..6f6cfe18c9d 100644
--- a/xarray/core/utils.py
+++ b/xarray/core/utils.py
@@ -104,6 +104,20 @@
 T = TypeVar("T")
 
 
+def is_allowed_extension_array_dtype(dtype: Any):
+    return pd.api.types.is_extension_array_dtype(dtype) and not isinstance(
+        dtype, pd.StringDtype
+    )
+
+
+def is_allowed_extension_array(array: Any):
+    return (
+        hasattr(array, "dtype")
+        and is_allowed_extension_array_dtype(array.dtype)
+        and not isinstance(array, pd.arrays.NumpyExtensionArray)
+    )
+
+
 def alias_message(old_name: str, new_name: str) -> str:
     return f"{old_name} has been deprecated. Use {new_name} instead."
 
diff --git a/xarray/core/variable.py b/xarray/core/variable.py
index bcc2ca4e460..325dde57e1d 100644
--- a/xarray/core/variable.py
+++ b/xarray/core/variable.py
@@ -40,6 +40,7 @@
     emit_user_level_warning,
     ensure_us_time_resolution,
     infix_dims,
+    is_allowed_extension_array,
     is_dict_like,
     is_duck_array,
     is_duck_dask_array,
@@ -198,7 +199,9 @@ def _maybe_wrap_data(data):
         return PandasIndexingAdapter(data)
     if isinstance(data, UNSUPPORTED_EXTENSION_ARRAY_TYPES):
         return data.to_numpy()
-    if isinstance(data, pd.api.extensions.ExtensionArray):
+    if isinstance(
+        data, pd.api.extensions.ExtensionArray
+    ) and is_allowed_extension_array(data):
         return PandasExtensionArray(data)
     return data
 
@@ -261,7 +264,8 @@ def convert_non_numpy_type(data):
     if isinstance(data, pd.Series | pd.DataFrame):
         if (
             isinstance(data, pd.Series)
-            and pd.api.types.is_extension_array_dtype(data)
+            and is_allowed_extension_array(data.array)
+            # Some datetime types are not allowed as well as backing Variable types
             and not isinstance(data.array, UNSUPPORTED_EXTENSION_ARRAY_TYPES)
         ):
             pandas_data = data.array
diff --git a/xarray/tests/test_pandas_to_xarray.py b/xarray/tests/test_pandas_to_xarray.py
index 111866541eb..8346f5abe21 100644
--- a/xarray/tests/test_pandas_to_xarray.py
+++ b/xarray/tests/test_pandas_to_xarray.py
@@ -37,6 +37,7 @@
 import pandas as pd
 import pandas._testing as tm
 import pytest
+from packaging.version import Version
 from pandas import (
     Categorical,
     CategoricalIndex,
@@ -171,7 +172,9 @@ def test_to_xarray_with_multiindex(self, df):
 
         result = result.to_dataframe()
         expected = df.copy()
-        expected["f"] = expected["f"].astype(object)
+        expected["f"] = expected["f"].astype(
+            object if Version(pd.__version__) < Version("3.0.0dev0") else str
+        )
         expected.columns.name = None
         tm.assert_frame_equal(result, expected)
 

From 287e4bef974f20ce00faaee69d2f12e059b0714b Mon Sep 17 00:00:00 2001
From: ilan-gold <ilanbassgold@gmail.com>
Date: Wed, 23 Jul 2025 12:22:07 +0200
Subject: [PATCH 04/26] feat: ban use of
 `pd.api.types.is_extension_array_dtype`

---
 pyproject.toml                 |  2 ++
 xarray/core/dataset.py         | 10 +++++-----
 xarray/core/dtypes.py          |  4 ++--
 xarray/core/duck_array_ops.py  |  9 +++++----
 xarray/core/extension_array.py |  5 ++---
 xarray/core/utils.py           |  2 +-
 xarray/tests/test_variable.py  |  2 +-
 7 files changed, 18 insertions(+), 16 deletions(-)

diff --git a/pyproject.toml b/pyproject.toml
index c7e1e04d85f..d4901f4d78b 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -325,6 +325,8 @@ known-first-party = ["xarray"]
 [tool.ruff.lint.flake8-tidy-imports]
 # Disallow all relative imports.
 ban-relative-imports = "all"
+[tool.ruff.lint.flake8-tidy-imports.banned-api]
+"pandas.api.types.is_extension_array_dtype".msg = "Use xarray.core.utils.is_allowed_extension_array{_dtype} instead.  Only use the banend API if the incoming data has already been sanitized by xarray"
 
 [tool.pytest.ini_options]
 addopts = [
diff --git a/xarray/core/dataset.py b/xarray/core/dataset.py
index 26db282c3df..46062c65a62 100644
--- a/xarray/core/dataset.py
+++ b/xarray/core/dataset.py
@@ -26,7 +26,6 @@
 
 import numpy as np
 import pandas as pd
-from pandas.api.types import is_extension_array_dtype
 
 from xarray.coding.calendar_ops import convert_calendar, interp_calendar
 from xarray.coding.cftimeindex import CFTimeIndex, _parse_array_of_cftime_strings
@@ -91,6 +90,7 @@
     either_dict_or_kwargs,
     emit_user_level_warning,
     infix_dims,
+    is_allowed_extension_array,
     is_dict_like,
     is_duck_array,
     is_duck_dask_array,
@@ -6771,7 +6771,7 @@ def reduce(
             elif (
                 # Some reduction functions (e.g. std, var) need to run on variables
                 # that don't have the reduce dims: PR5393
-                not is_extension_array_dtype(var.dtype)
+                not pd.api.types.is_extension_array_dtype(var.dtype)  # noqa: TID251
                 and (
                     not reduce_dims
                     or not numeric_only
@@ -7096,12 +7096,12 @@ def _to_dataframe(self, ordered_dims: Mapping[Any, int]):
         non_extension_array_columns = [
             k
             for k in columns_in_order
-            if not is_extension_array_dtype(self.variables[k].data)
+            if not pd.api.types.is_extension_array_dtype(self.variables[k].data)  # noqa: TID251
         ]
         extension_array_columns = [
             k
             for k in columns_in_order
-            if is_extension_array_dtype(self.variables[k].data)
+            if pd.api.types.is_extension_array_dtype(self.variables[k].data)  # noqa: TID251
         ]
         extension_array_columns_different_index = [
             k
@@ -7293,7 +7293,7 @@ def from_dataframe(cls, dataframe: pd.DataFrame, sparse: bool = False) -> Self:
         arrays = []
         extension_arrays = []
         for k, v in dataframe.items():
-            if not is_extension_array_dtype(v) or isinstance(
+            if not is_allowed_extension_array(v) or isinstance(
                 v.array, UNSUPPORTED_EXTENSION_ARRAY_TYPES
             ):
                 arrays.append((k, np.asarray(v)))
diff --git a/xarray/core/dtypes.py b/xarray/core/dtypes.py
index c959a7f2536..0a7b1722877 100644
--- a/xarray/core/dtypes.py
+++ b/xarray/core/dtypes.py
@@ -4,7 +4,7 @@
 from typing import Any
 
 import numpy as np
-from pandas.api.types import is_extension_array_dtype
+import pandas as pd
 
 from xarray.compat import array_api_compat, npcompat
 from xarray.compat.npcompat import HAS_STRING_DTYPE
@@ -213,7 +213,7 @@ def isdtype(dtype, kind: str | tuple[str, ...], xp=None) -> bool:
 
     if isinstance(dtype, np.dtype):
         return npcompat.isdtype(dtype, kind)
-    elif is_extension_array_dtype(dtype):
+    elif pd.api.types.is_extension_array_dtype(dtype):  # noqa: TID251
         # we never want to match pandas extension array dtypes
         return False
     else:
diff --git a/xarray/core/duck_array_ops.py b/xarray/core/duck_array_ops.py
index 0c7d40113d6..7687e452496 100644
--- a/xarray/core/duck_array_ops.py
+++ b/xarray/core/duck_array_ops.py
@@ -23,7 +23,6 @@
     take,
     unravel_index,  # noqa: F401
 )
-from pandas.api.types import is_extension_array_dtype
 
 from xarray.compat import dask_array_compat, dask_array_ops
 from xarray.compat.array_api_compat import get_array_namespace
@@ -184,7 +183,7 @@ def isnull(data):
         dtype = xp.bool_ if hasattr(xp, "bool_") else xp.bool
         return full_like(data, dtype=dtype, fill_value=False)
     # at this point, array should have dtype=object
-    elif isinstance(data, np.ndarray) or is_extension_array_dtype(data):
+    elif isinstance(data, np.ndarray) or pd.api.types.is_extension_array_dtype(data):  # noqa: TID251
         return pandas_isnull(data)
     else:
         # Not reachable yet, but intended for use with other duck array
@@ -266,9 +265,11 @@ def asarray(data, xp=np, dtype=None):
 
 def as_shared_dtype(scalars_or_arrays, xp=None):
     """Cast arrays to a shared dtype using xarray's type promotion rules."""
-    if any(is_extension_array_dtype(x) for x in scalars_or_arrays):
+    if any(pd.api.types.is_extension_array_dtype(x) for x in scalars_or_arrays):  # noqa: TID251
         extension_array_types = [
-            x.dtype for x in scalars_or_arrays if is_extension_array_dtype(x)
+            x.dtype
+            for x in scalars_or_arrays
+            if pd.api.types.is_extension_array_dtype(x)  # noqa: TID251
         ]
         non_nans = [x for x in scalars_or_arrays if not isna(x)]
         if len(extension_array_types) == len(non_nans) and all(
diff --git a/xarray/core/extension_array.py b/xarray/core/extension_array.py
index b752f7141af..cb7499d91c3 100644
--- a/xarray/core/extension_array.py
+++ b/xarray/core/extension_array.py
@@ -8,7 +8,6 @@
 import numpy as np
 import pandas as pd
 from packaging.version import Version
-from pandas.api.types import is_extension_array_dtype
 
 from xarray.core.types import DTypeLikeSave, T_ExtensionArray
 from xarray.core.utils import NDArrayMixin, is_allowed_extension_array
@@ -126,7 +125,7 @@ def replace_duck_with_extension_array(args) -> list:
         if func not in HANDLED_EXTENSION_ARRAY_FUNCTIONS:
             raise KeyError("Function not registered for pandas extension arrays.")
         res = HANDLED_EXTENSION_ARRAY_FUNCTIONS[func](*args, **kwargs)
-        if is_extension_array_dtype(res):
+        if pd.api.types.is_extension_array_dtype(res):  # noqa: TID251
             return PandasExtensionArray(res)
         return res
 
@@ -135,7 +134,7 @@ def __array_ufunc__(self, ufunc, method, *inputs, **kwargs):
 
     def __getitem__(self, key) -> PandasExtensionArray[T_ExtensionArray]:
         item = self.array[key]
-        if is_extension_array_dtype(item):
+        if pd.api.types.is_extension_array_dtype(item):  # noqa: TID251
             return PandasExtensionArray(item)
         if np.isscalar(item) or isinstance(key, int):
             return PandasExtensionArray(type(self.array)._from_sequence([item]))  # type: ignore[call-arg,attr-defined,unused-ignore]
diff --git a/xarray/core/utils.py b/xarray/core/utils.py
index 6f6cfe18c9d..f51a73fb32b 100644
--- a/xarray/core/utils.py
+++ b/xarray/core/utils.py
@@ -105,7 +105,7 @@
 
 
 def is_allowed_extension_array_dtype(dtype: Any):
-    return pd.api.types.is_extension_array_dtype(dtype) and not isinstance(
+    return pd.api.types.is_extension_array_dtype(dtype) and not isinstance(  # noqa: TID251
         dtype, pd.StringDtype
     )
 
diff --git a/xarray/tests/test_variable.py b/xarray/tests/test_variable.py
index 2f67e97522c..e2f4a3154f3 100644
--- a/xarray/tests/test_variable.py
+++ b/xarray/tests/test_variable.py
@@ -1594,7 +1594,7 @@ def test_pandas_categorical_dtype(self):
         data = pd.Categorical(np.arange(10, dtype="int64"))
         v = self.cls("x", data)
         print(v)  # should not error
-        assert pd.api.types.is_extension_array_dtype(v.dtype)
+        assert isinstance(v.dtype, pd.CategoricalDtype)
 
     def test_squeeze(self):
         v = Variable(["x", "y"], [[1]])

From ffc905d1a10f19c2cc3206950e867e8f09dcb90d Mon Sep 17 00:00:00 2001
From: ilan-gold <ilanbassgold@gmail.com>
Date: Wed, 23 Jul 2025 13:33:26 +0200
Subject: [PATCH 05/26] fix: type ignore

---
 xarray/core/utils.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/xarray/core/utils.py b/xarray/core/utils.py
index f51a73fb32b..2de57b6d213 100644
--- a/xarray/core/utils.py
+++ b/xarray/core/utils.py
@@ -114,7 +114,7 @@ def is_allowed_extension_array(array: Any):
     return (
         hasattr(array, "dtype")
         and is_allowed_extension_array_dtype(array.dtype)
-        and not isinstance(array, pd.arrays.NumpyExtensionArray)
+        and not isinstance(array, pd.arrays.NumpyExtensionArray)  # type: ignore[attr-defined]
     )
 
 

From aa68745ace8227bede2a036eb61930fc26e39dd5 Mon Sep 17 00:00:00 2001
From: ilan-gold <ilanbassgold@gmail.com>
Date: Thu, 24 Jul 2025 11:31:56 +0200
Subject: [PATCH 06/26] fix: `pd.Series` in `pandas>=3` does not preserve
 object dtype metadata [test-upstream]

---
 xarray/core/variable.py | 9 +++++++++
 1 file changed, 9 insertions(+)

diff --git a/xarray/core/variable.py b/xarray/core/variable.py
index 325dde57e1d..eff332539ab 100644
--- a/xarray/core/variable.py
+++ b/xarray/core/variable.py
@@ -13,6 +13,7 @@
 import numpy as np
 import pandas as pd
 from numpy.typing import ArrayLike
+from packaging.version import Version
 
 import xarray as xr  # only for Dataset and DataArray
 from xarray.compat.array_api_compat import to_like_array
@@ -217,6 +218,14 @@ def _possibly_convert_objects(values):
     """
     as_series = pd.Series(values.ravel(), copy=False)
     result = np.asarray(as_series).reshape(values.shape)
+    # FIXME: Why does pd.Series no longer preserve data type metadata for object dtype?
+    if (
+        result.dtype.kind == "O"
+        and values.dtype.kind == "O"
+        and Version(pd.__version__) >= Version("3.0.0dev0")
+    ):
+        result = np.asarray(as_series, copy=True).reshape(values.shape)
+        result.dtype = values.dtype
     if not result.flags.writeable:
         # GH8843, pandas copy-on-write mode creates read-only arrays by default
         try:

From 29c5224383a9d50a1d87a5737d8228ac8f3092d7 Mon Sep 17 00:00:00 2001
From: ilan-gold <ilanbassgold@gmail.com>
Date: Thu, 24 Jul 2025 11:38:25 +0200
Subject: [PATCH 07/26] fix: bytes for catgeorical repr [test-upstream]
 [test-upstream]

---
 xarray/tests/test_dataset.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/xarray/tests/test_dataset.py b/xarray/tests/test_dataset.py
index 3e0734c8a1a..ed1a16b3dfb 100644
--- a/xarray/tests/test_dataset.py
+++ b/xarray/tests/test_dataset.py
@@ -299,7 +299,7 @@ def test_repr(self) -> None:
                 var1     (dim1, dim2) float64 576B -0.9891 -0.3678 1.288 ... -0.2116 0.364
                 var2     (dim1, dim2) float64 576B 0.953 1.52 1.704 ... 0.1347 -0.6423
                 var3     (dim3, dim1) float64 640B 0.4107 0.9941 0.1665 ... 0.716 1.555
-                var4     (dim1) category 32B b c b a c a c a{var5}
+                var4     (dim1) category 36B b c b a c a c a{var5}
             Attributes:
                 foo:      bar"""
         )

From ed86c095dde859b848f61a6b0fae7bc7edf98ad8 Mon Sep 17 00:00:00 2001
From: Ilan Gold <ilanbassgold@gmail.com>
Date: Thu, 24 Jul 2025 11:48:51 +0200
Subject: [PATCH 08/26] Update xarray/core/extension_array.py

Co-authored-by: Deepak Cherian <dcherian@users.noreply.github.com>
---
 xarray/core/extension_array.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/xarray/core/extension_array.py b/xarray/core/extension_array.py
index cb7499d91c3..7a51ba09084 100644
--- a/xarray/core/extension_array.py
+++ b/xarray/core/extension_array.py
@@ -102,7 +102,7 @@ def __post_init__(self):
         # duck array support internally via this class.
         if not is_allowed_extension_array(self.array):
             raise TypeError(
-                "`NumpyExtensionArray` or string dtype should be converted to a numpy array in `xarray` internally."
+                f"{self.array.dtype!r} should be converted to a numpy array in `xarray` internally."
             )
 
     def __array_function__(self, func, types, args, kwargs):

From f603aa09ff5757927fcea2bb6fc342f43c04f87e Mon Sep 17 00:00:00 2001
From: ilan-gold <ilanbassgold@gmail.com>
Date: Thu, 24 Jul 2025 11:50:22 +0200
Subject: [PATCH 09/26] fix: repr

---
 xarray/tests/test_dataset.py | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/xarray/tests/test_dataset.py b/xarray/tests/test_dataset.py
index ed1a16b3dfb..20c5d653187 100644
--- a/xarray/tests/test_dataset.py
+++ b/xarray/tests/test_dataset.py
@@ -13,6 +13,7 @@
 import numpy as np
 import pandas as pd
 import pytest
+from packaging.version import Version
 from pandas.core.indexes.datetimes import DatetimeIndex
 
 # remove once numpy 2.0 is the oldest supported version
@@ -299,7 +300,7 @@ def test_repr(self) -> None:
                 var1     (dim1, dim2) float64 576B -0.9891 -0.3678 1.288 ... -0.2116 0.364
                 var2     (dim1, dim2) float64 576B 0.953 1.52 1.704 ... 0.1347 -0.6423
                 var3     (dim3, dim1) float64 640B 0.4107 0.9941 0.1665 ... 0.716 1.555
-                var4     (dim1) category 36B b c b a c a c a{var5}
+                var4     (dim1) category 3{6 if Version(pd.__version__) >= Version("3.0.0dev0") else 2}B b c b a c a c a{var5}
             Attributes:
                 foo:      bar"""
         )

From 02f2496f037b28731f29d0dd3f49d429cbf930cb Mon Sep 17 00:00:00 2001
From: Ilan Gold <ilanbassgold@gmail.com>
Date: Fri, 25 Jul 2025 15:25:31 +0200
Subject: [PATCH 10/26] Update variable.py

---
 xarray/core/variable.py | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)

diff --git a/xarray/core/variable.py b/xarray/core/variable.py
index eff332539ab..9f2d00f879f 100644
--- a/xarray/core/variable.py
+++ b/xarray/core/variable.py
@@ -217,8 +217,7 @@ def _possibly_convert_objects(values):
     * pd.Timedelta
     """
     as_series = pd.Series(values.ravel(), copy=False)
-    result = np.asarray(as_series).reshape(values.shape)
-    # FIXME: Why does pd.Series no longer preserve data type metadata for object dtype?
+    # For why we need this behavior: https://github.com/pandas-dev/pandas/issues/61938
     if (
         result.dtype.kind == "O"
         and values.dtype.kind == "O"
@@ -226,6 +225,8 @@ def _possibly_convert_objects(values):
     ):
         result = np.asarray(as_series, copy=True).reshape(values.shape)
         result.dtype = values.dtype
+    else:
+        result = np.asarray(as_series).reshape(values.shape)
     if not result.flags.writeable:
         # GH8843, pandas copy-on-write mode creates read-only arrays by default
         try:

From 9ea9d8d7b576ba9867236cbbef5c8fe1a097ea08 Mon Sep 17 00:00:00 2001
From: ilan-gold <ilanbassgold@gmail.com>
Date: Fri, 25 Jul 2025 17:55:25 +0200
Subject: [PATCH 11/26] ? mypy

---
 xarray/computation/ops.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/xarray/computation/ops.py b/xarray/computation/ops.py
index 61834a85acf..70a7f916c8b 100644
--- a/xarray/computation/ops.py
+++ b/xarray/computation/ops.py
@@ -159,7 +159,7 @@ def fillna(data, other, join="left", dataset_join="left"):
 
 
 # Unsure why we get a mypy error here
-def where_method(self, cond, other=dtypes.NA):  # type: ignore[has-type]
+def where_method(self, cond, other=dtypes.NA):
     """Return elements from `self` or `other` depending on `cond`.
 
     Parameters

From 7dc9662d91e4add6167dc6002cdbf78be7474c2d Mon Sep 17 00:00:00 2001
From: ilan-gold <ilanbassgold@gmail.com>
Date: Fri, 25 Jul 2025 17:57:04 +0200
Subject: [PATCH 12/26] fix: remove comment

---
 xarray/computation/ops.py | 1 -
 1 file changed, 1 deletion(-)

diff --git a/xarray/computation/ops.py b/xarray/computation/ops.py
index 70a7f916c8b..8067ebb3a5f 100644
--- a/xarray/computation/ops.py
+++ b/xarray/computation/ops.py
@@ -158,7 +158,6 @@ def fillna(data, other, join="left", dataset_join="left"):
     )
 
 
-# Unsure why we get a mypy error here
 def where_method(self, cond, other=dtypes.NA):
     """Return elements from `self` or `other` depending on `cond`.
 

From f05e703ca10f664daa17dcf359dc25ab53e274a2 Mon Sep 17 00:00:00 2001
From: ilan-gold <ilanbassgold@gmail.com>
Date: Mon, 28 Jul 2025 15:20:56 +0200
Subject: [PATCH 13/26] try blanket ignore

---
 xarray/computation/ops.py | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/xarray/computation/ops.py b/xarray/computation/ops.py
index 8067ebb3a5f..e91b5ef2164 100644
--- a/xarray/computation/ops.py
+++ b/xarray/computation/ops.py
@@ -158,7 +158,8 @@ def fillna(data, other, join="left", dataset_join="left"):
     )
 
 
-def where_method(self, cond, other=dtypes.NA):
+# Unsure why we get a mypy error here, or why
+def where_method(self, cond, other=dtypes.NA):  # type: ignore  # noqa: PGH003
     """Return elements from `self` or `other` depending on `cond`.
 
     Parameters

From 7343f6dd950cdd1acc25ae4b9e0308e77c6960e2 Mon Sep 17 00:00:00 2001
From: ilan-gold <ilanbassgold@gmail.com>
Date: Mon, 28 Jul 2025 15:26:59 +0200
Subject: [PATCH 14/26] try blanket ignore

---
 xarray/computation/ops.py | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/xarray/computation/ops.py b/xarray/computation/ops.py
index 8067ebb3a5f..e91b5ef2164 100644
--- a/xarray/computation/ops.py
+++ b/xarray/computation/ops.py
@@ -158,7 +158,8 @@ def fillna(data, other, join="left", dataset_join="left"):
     )
 
 
-def where_method(self, cond, other=dtypes.NA):
+# Unsure why we get a mypy error here, or why
+def where_method(self, cond, other=dtypes.NA):  # type: ignore  # noqa: PGH003
     """Return elements from `self` or `other` depending on `cond`.
 
     Parameters

From cc1776fb3382ac9a87fabcd1519030ee21185ac6 Mon Sep 17 00:00:00 2001
From: ilan-gold <ilanbassgold@gmail.com>
Date: Mon, 28 Jul 2025 15:33:45 +0200
Subject: [PATCH 15/26] try blanket ignore again

---
 xarray/computation/ops.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/xarray/computation/ops.py b/xarray/computation/ops.py
index e91b5ef2164..cae7cef841a 100644
--- a/xarray/computation/ops.py
+++ b/xarray/computation/ops.py
@@ -159,7 +159,7 @@ def fillna(data, other, join="left", dataset_join="left"):
 
 
 # Unsure why we get a mypy error here, or why
-def where_method(self, cond, other=dtypes.NA):  # type: ignore  # noqa: PGH003
+def where_method(self, cond, other=dtypes.NA):  # type: ignore[unused-ignore]
     """Return elements from `self` or `other` depending on `cond`.
 
     Parameters

From 2e8ed6781a3e4d31d91e54b9fa151d4e4fae6d8c Mon Sep 17 00:00:00 2001
From: ilan-gold <ilanbassgold@gmail.com>
Date: Mon, 28 Jul 2025 16:17:00 +0200
Subject: [PATCH 16/26] fix: mypy

---
 xarray/computation/ops.py | 11 ++++++++---
 1 file changed, 8 insertions(+), 3 deletions(-)

diff --git a/xarray/computation/ops.py b/xarray/computation/ops.py
index cae7cef841a..fbc14040671 100644
--- a/xarray/computation/ops.py
+++ b/xarray/computation/ops.py
@@ -8,12 +8,18 @@
 from __future__ import annotations
 
 import operator
-from typing import Literal
+from typing import TYPE_CHECKING, Literal
 
 import numpy as np
 
 from xarray.core import dtypes, duck_array_ops
 
+if TYPE_CHECKING:
+    from typing import Any
+
+    from xarray.core.dataarray import DataArray
+    from xarray.core.dataset import Dataset
+
 try:
     import bottleneck as bn
 
@@ -158,8 +164,7 @@ def fillna(data, other, join="left", dataset_join="left"):
     )
 
 
-# Unsure why we get a mypy error here, or why
-def where_method(self, cond, other=dtypes.NA):  # type: ignore[unused-ignore]
+def where_method(self, cond: DataArray | Dataset, other: Any = dtypes.NA):
     """Return elements from `self` or `other` depending on `cond`.
 
     Parameters

From ac77713f3b84349e2a382e58feb280cb71b7e89e Mon Sep 17 00:00:00 2001
From: ilan-gold <ilanbassgold@gmail.com>
Date: Mon, 28 Jul 2025 16:21:06 +0200
Subject: [PATCH 17/26] use Any

---
 xarray/computation/ops.py | 6 ++----
 1 file changed, 2 insertions(+), 4 deletions(-)

diff --git a/xarray/computation/ops.py b/xarray/computation/ops.py
index fbc14040671..9c7fd96c12f 100644
--- a/xarray/computation/ops.py
+++ b/xarray/computation/ops.py
@@ -17,9 +17,6 @@
 if TYPE_CHECKING:
     from typing import Any
 
-    from xarray.core.dataarray import DataArray
-    from xarray.core.dataset import Dataset
-
 try:
     import bottleneck as bn
 
@@ -164,7 +161,8 @@ def fillna(data, other, join="left", dataset_join="left"):
     )
 
 
-def where_method(self, cond: DataArray | Dataset, other: Any = dtypes.NA):
+# TODO: type this properly
+def where_method(self: Any, cond: Any, other: Any = dtypes.NA):
     """Return elements from `self` or `other` depending on `cond`.
 
     Parameters

From bfbe24485fe5d59e2571d170614ff014f545a7f1 Mon Sep 17 00:00:00 2001
From: ilan-gold <ilanbassgold@gmail.com>
Date: Tue, 29 Jul 2025 16:15:20 +0200
Subject: [PATCH 18/26] chore: add comment

---
 xarray/core/extension_array.py | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/xarray/core/extension_array.py b/xarray/core/extension_array.py
index 7a51ba09084..134d3dd8cee 100644
--- a/xarray/core/extension_array.py
+++ b/xarray/core/extension_array.py
@@ -99,7 +99,8 @@ def __post_init__(self):
             raise TypeError(f"{self.array} is not an pandas ExtensionArray.")
         # This does not use the UNSUPPORTED_EXTENSION_ARRAY_TYPES whitelist because
         # we do support extension arrays from datetime, for example, that need
-        # duck array support internally via this class.
+        # duck array support internally via this class.  These can appear from `DatetimeIndex`
+        # wrapped by `PandasIndex` internally, for example.
         if not is_allowed_extension_array(self.array):
             raise TypeError(
                 f"{self.array.dtype!r} should be converted to a numpy array in `xarray` internally."

From 594c164ef61378eb758ee8e5648bec8a9da0c120 Mon Sep 17 00:00:00 2001
From: Ilan Gold <ilanbassgold@gmail.com>
Date: Tue, 29 Jul 2025 16:17:10 +0200
Subject: [PATCH 19/26] Update xarray/core/utils.py

Co-authored-by: Deepak Cherian <dcherian@users.noreply.github.com>
---
 xarray/core/utils.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/xarray/core/utils.py b/xarray/core/utils.py
index 2de57b6d213..386f1e346de 100644
--- a/xarray/core/utils.py
+++ b/xarray/core/utils.py
@@ -110,7 +110,7 @@ def is_allowed_extension_array_dtype(dtype: Any):
     )
 
 
-def is_allowed_extension_array(array: Any):
+def is_allowed_extension_array(array: Any) -> bool:
     return (
         hasattr(array, "dtype")
         and is_allowed_extension_array_dtype(array.dtype)

From a436d787d226fa817eb22d8bdeaa5908052bf88b Mon Sep 17 00:00:00 2001
From: ilan-gold <ilanbassgold@gmail.com>
Date: Tue, 29 Jul 2025 16:21:02 +0200
Subject: [PATCH 20/26] refactor: use is_allowed_extension_array more

---
 xarray/core/extension_array.py | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/xarray/core/extension_array.py b/xarray/core/extension_array.py
index 134d3dd8cee..9262982d4cb 100644
--- a/xarray/core/extension_array.py
+++ b/xarray/core/extension_array.py
@@ -126,7 +126,7 @@ def replace_duck_with_extension_array(args) -> list:
         if func not in HANDLED_EXTENSION_ARRAY_FUNCTIONS:
             raise KeyError("Function not registered for pandas extension arrays.")
         res = HANDLED_EXTENSION_ARRAY_FUNCTIONS[func](*args, **kwargs)
-        if pd.api.types.is_extension_array_dtype(res):  # noqa: TID251
+        if is_allowed_extension_array(res):
             return PandasExtensionArray(res)
         return res
 
@@ -135,7 +135,7 @@ def __array_ufunc__(self, ufunc, method, *inputs, **kwargs):
 
     def __getitem__(self, key) -> PandasExtensionArray[T_ExtensionArray]:
         item = self.array[key]
-        if pd.api.types.is_extension_array_dtype(item):  # noqa: TID251
+        if is_allowed_extension_array(item):
             return PandasExtensionArray(item)
         if np.isscalar(item) or isinstance(key, int):
             return PandasExtensionArray(type(self.array)._from_sequence([item]))  # type: ignore[call-arg,attr-defined,unused-ignore]

From 5e603a793580ea086466a9cbd282732484c0a552 Mon Sep 17 00:00:00 2001
From: ilan-gold <ilanbassgold@gmail.com>
Date: Tue, 29 Jul 2025 16:23:31 +0200
Subject: [PATCH 21/26] fix: remove one of the loops

---
 xarray/core/duck_array_ops.py | 12 ++++++------
 1 file changed, 6 insertions(+), 6 deletions(-)

diff --git a/xarray/core/duck_array_ops.py b/xarray/core/duck_array_ops.py
index 7687e452496..b8a4011a72e 100644
--- a/xarray/core/duck_array_ops.py
+++ b/xarray/core/duck_array_ops.py
@@ -265,12 +265,12 @@ def asarray(data, xp=np, dtype=None):
 
 def as_shared_dtype(scalars_or_arrays, xp=None):
     """Cast arrays to a shared dtype using xarray's type promotion rules."""
-    if any(pd.api.types.is_extension_array_dtype(x) for x in scalars_or_arrays):  # noqa: TID251
-        extension_array_types = [
-            x.dtype
-            for x in scalars_or_arrays
-            if pd.api.types.is_extension_array_dtype(x)  # noqa: TID251
-        ]
+    extension_array_types = [
+        x.dtype
+        for x in scalars_or_arrays
+        if pd.api.types.is_extension_array_dtype(x)  # noqa: TID251
+    ]
+    if len(extension_array_types) >= 1:
         non_nans = [x for x in scalars_or_arrays if not isna(x)]
         if len(extension_array_types) == len(non_nans) and all(
             isinstance(x, type(extension_array_types[0])) for x in extension_array_types

From a8d1aa17189aa0193469dbe17fbd83419a638ac6 Mon Sep 17 00:00:00 2001
From: Ilan Gold <ilanbassgold@gmail.com>
Date: Tue, 29 Jul 2025 17:51:39 +0200
Subject: [PATCH 22/26] Update xarray/computation/ops.py

Co-authored-by: Deepak Cherian <dcherian@users.noreply.github.com>
---
 xarray/computation/ops.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/xarray/computation/ops.py b/xarray/computation/ops.py
index 9c7fd96c12f..49e7406f409 100644
--- a/xarray/computation/ops.py
+++ b/xarray/computation/ops.py
@@ -162,7 +162,7 @@ def fillna(data, other, join="left", dataset_join="left"):
 
 
 # TODO: type this properly
-def where_method(self: Any, cond: Any, other: Any = dtypes.NA):
+def where_method(self, cond, other=dtypes.NA):  # type: ignore[has-type]
     """Return elements from `self` or `other` depending on `cond`.
 
     Parameters

From bc27b15731265d000ca36fcc2a66a5b5aac55d0a Mon Sep 17 00:00:00 2001
From: "pre-commit-ci[bot]"
 <66853113+pre-commit-ci[bot]@users.noreply.github.com>
Date: Tue, 29 Jul 2025 15:52:00 +0000
Subject: [PATCH 23/26] [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci
---
 xarray/computation/ops.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/xarray/computation/ops.py b/xarray/computation/ops.py
index 49e7406f409..7d1d933af79 100644
--- a/xarray/computation/ops.py
+++ b/xarray/computation/ops.py
@@ -15,7 +15,7 @@
 from xarray.core import dtypes, duck_array_ops
 
 if TYPE_CHECKING:
-    from typing import Any
+    pass
 
 try:
     import bottleneck as bn

From f914dc701c9c7dea2055095e18f0d74373d5792b Mon Sep 17 00:00:00 2001
From: ilan-gold <ilanbassgold@gmail.com>
Date: Tue, 29 Jul 2025 18:25:20 +0200
Subject: [PATCH 24/26] fix: result handling

---
 xarray/core/variable.py | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/xarray/core/variable.py b/xarray/core/variable.py
index 9f2d00f879f..8ff155e8c87 100644
--- a/xarray/core/variable.py
+++ b/xarray/core/variable.py
@@ -218,15 +218,15 @@ def _possibly_convert_objects(values):
     """
     as_series = pd.Series(values.ravel(), copy=False)
     # For why we need this behavior: https://github.com/pandas-dev/pandas/issues/61938
+    result = np.asarray(as_series).reshape(values.shape)
     if (
         result.dtype.kind == "O"
         and values.dtype.kind == "O"
         and Version(pd.__version__) >= Version("3.0.0dev0")
     ):
+        # need to copy to be able to override `dtype`
         result = np.asarray(as_series, copy=True).reshape(values.shape)
         result.dtype = values.dtype
-    else:
-        result = np.asarray(as_series).reshape(values.shape)
     if not result.flags.writeable:
         # GH8843, pandas copy-on-write mode creates read-only arrays by default
         try:

From 8c2f73ec830aca9dda5db502bf3cd0d6f642a3dd Mon Sep 17 00:00:00 2001
From: ilan-gold <ilanbassgold@gmail.com>
Date: Wed, 30 Jul 2025 16:47:39 +0200
Subject: [PATCH 25/26] fix: avoid extra copy + comment

---
 xarray/core/variable.py | 24 +++++++++++++++---------
 1 file changed, 15 insertions(+), 9 deletions(-)

diff --git a/xarray/core/variable.py b/xarray/core/variable.py
index 8ff155e8c87..031ca2c3fde 100644
--- a/xarray/core/variable.py
+++ b/xarray/core/variable.py
@@ -209,7 +209,10 @@ def _maybe_wrap_data(data):
 
 def _possibly_convert_objects(values):
     """Convert object arrays into datetime64 and timedelta64 according
-    to the pandas convention.
+    to the pandas convention.  For backwards compat, as of 3.0.0 pandas,
+    object dtype inputs are cast to strings by `pandas.Series`
+    but we output them as object dtype with the input metadata preserved as well.
+
 
     * datetime.datetime
     * datetime.timedelta
@@ -217,22 +220,25 @@ def _possibly_convert_objects(values):
     * pd.Timedelta
     """
     as_series = pd.Series(values.ravel(), copy=False)
-    # For why we need this behavior: https://github.com/pandas-dev/pandas/issues/61938
     result = np.asarray(as_series).reshape(values.shape)
+    if not result.flags.writeable:
+        # GH8843, pandas copy-on-write mode creates read-only arrays by default
+        try:
+            result.flags.writeable = True
+        except ValueError:
+            result = result.copy()
+    # For why we need this behavior: https://github.com/pandas-dev/pandas/issues/61938
+    # Object datatype inputs that are strings
+    # will be converted to strings by `pandas.Series`, and as of 3.0.0, lose
+    # metadata.  If the roundtrip back to numpy in this function yields an
+    # object array again, the dtype.metadata will be preserved.
     if (
         result.dtype.kind == "O"
         and values.dtype.kind == "O"
         and Version(pd.__version__) >= Version("3.0.0dev0")
     ):
         # need to copy to be able to override `dtype`
-        result = np.asarray(as_series, copy=True).reshape(values.shape)
         result.dtype = values.dtype
-    if not result.flags.writeable:
-        # GH8843, pandas copy-on-write mode creates read-only arrays by default
-        try:
-            result.flags.writeable = True
-        except ValueError:
-            result = result.copy()
     return result
 
 

From 3cf3b884f15ec3c3ef191e49c48c4169135c87b2 Mon Sep 17 00:00:00 2001
From: Deepak Cherian <dcherian@users.noreply.github.com>
Date: Wed, 30 Jul 2025 08:58:23 -0600
Subject: [PATCH 26/26] Apply suggestions from code review

---
 xarray/core/variable.py | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)

diff --git a/xarray/core/variable.py b/xarray/core/variable.py
index 031ca2c3fde..06d7218fe7c 100644
--- a/xarray/core/variable.py
+++ b/xarray/core/variable.py
@@ -230,14 +230,13 @@ def _possibly_convert_objects(values):
     # For why we need this behavior: https://github.com/pandas-dev/pandas/issues/61938
     # Object datatype inputs that are strings
     # will be converted to strings by `pandas.Series`, and as of 3.0.0, lose
-    # metadata.  If the roundtrip back to numpy in this function yields an
+    # `dtype.metadata`.  If the roundtrip back to numpy in this function yields an
     # object array again, the dtype.metadata will be preserved.
     if (
         result.dtype.kind == "O"
         and values.dtype.kind == "O"
         and Version(pd.__version__) >= Version("3.0.0dev0")
     ):
-        # need to copy to be able to override `dtype`
         result.dtype = values.dtype
     return result