Fix string reductions and isin across string dtype flavors

python/cudf/cudf/core/column/numerical_base.py

@@ -90,13 +90,13 @@ def kurtosis(self, skipna: bool = True) -> float:
             )
 
         if len(self) == 0 or self._can_return_nan(skipna=skipna):
-            return _get_nan_for_dtype(self.dtype)  # type: ignore[return-value]
+            return _get_nan_for_dtype(self.dtype)
 
         original_dtype = self.dtype
         self = self.nans_to_nulls().dropna()
 
         if len(self) < 4:
-            return _get_nan_for_dtype(original_dtype)  # type: ignore[return-value]
+            return _get_nan_for_dtype(original_dtype)
 
         if original_dtype.kind == "f" and original_dtype.itemsize < 8:
             # Compute in float64 to match pandas precision for narrower
@@ -206,7 +206,7 @@ def quantile(
                 except (TypeError, ValueError):
                     pass
             return (
-                _get_nan_for_dtype(self.dtype)  # type: ignore[return-value]
+                _get_nan_for_dtype(self.dtype)
                 if scalar_result is NA
                 else scalar_result
             )
@@ -221,7 +221,7 @@ def median(
             )
 
         if self._can_return_nan(skipna=skipna):
-            return _get_nan_for_dtype(self.dtype)  # type: ignore[return-value]
+            return _get_nan_for_dtype(self.dtype)
 
         # enforce linear in case the default ever changes
         result = self.quantile(
@@ -240,21 +240,21 @@ def cov(self, other: NumericalBaseColumn) -> float:
             or len(other) == 0
             or (len(self) == 1 and len(other) == 1)
         ):
-            return _get_nan_for_dtype(self.dtype)  # type: ignore[return-value]
+            return _get_nan_for_dtype(self.dtype)
 
         result = (self - self.mean()) * (other - other.mean())
         cov_sample = result.sum() / (len(self) - 1)
         return cov_sample
 
     def corr(self, other: NumericalBaseColumn) -> float:
         if len(self) == 0 or len(other) == 0:
-            return _get_nan_for_dtype(self.dtype)  # type: ignore[return-value]
+            return _get_nan_for_dtype(self.dtype)
 
         cov = self.cov(other)
         lhs_std, rhs_std = self.std(), other.std()
 
         if not cov or lhs_std == 0 or rhs_std == 0:
-            return _get_nan_for_dtype(self.dtype)  # type: ignore[return-value]
+            return _get_nan_for_dtype(self.dtype)
         return cov / lhs_std / rhs_std
 
     def round(

python/cudf/cudf/core/column/string.py

@@ -29,6 +29,7 @@
     pylibcudf_result_dtype_policy,
     same_dtype_policy,
 )
+from cudf.core.dtype.validators import is_dtype_obj_string
 from cudf.core.mixins import Scannable
 from cudf.errors import MixedTypeError
 from cudf.utils.dtypes import (
@@ -43,7 +44,7 @@
 from cudf.utils.utils import is_na_like
 
 if TYPE_CHECKING:
-    from collections.abc import Callable, Iterable, Mapping
+    from collections.abc import Callable, Iterable, Mapping, Sequence
 
     import cupy as cp
 
@@ -204,11 +205,11 @@ def sum(
         if min_count > 0 and col.valid_count < min_count:
             return pd.NA
 
-        return (
-            0
-            if len(col) == 0
-            else col.join_strings("", None).element_indexing(0)
-        )
+        if len(col) == 0:
+            # pandas sums an empty/all-null object series to the numeric
+            # identity 0; genuine string dtypes concatenate to "".
+            return 0 if self.dtype == np.dtype("object") else ""
+        return col.join_strings("", None).element_indexing(0)
 
     def any(
         self, skipna: bool = True, min_count: int = 0, **kwargs: Any
@@ -397,6 +398,18 @@ def as_string_column(self, dtype: DtypeObj) -> Self:
             return cast("Self", ColumnBase.create(self.plc_column, dtype))
         return self
 
+    def _process_values_for_isin(
+        self, values: Sequence | ColumnBase
+    ) -> tuple[ColumnBase, ColumnBase]:
+        lhs, rhs = super()._process_values_for_isin(values)
+        if lhs.dtype != rhs.dtype and is_dtype_obj_string(rhs.dtype):
+            # Strings of any dtype flavor (object, pd.StringDtype with
+            # python/pyarrow storage and NaN/NA na_value, pd.ArrowDtype
+            # string) hold comparable values; align rhs with lhs's dtype
+            # so ColumnBase.isin does not treat them as disjoint types.
+            rhs = rhs.astype(lhs.dtype)
+        return lhs, rhs
+
     @property
     def values(self) -> cp.ndarray:
         """

python/cudf/cudf/pandas/fast_slow_proxy.py

@@ -1477,6 +1477,19 @@ def _transform_arg(
         transformed: list[Any] = [
             _transform_arg(a, attribute_name, seen) for a in arg.flat
         ]
+        if all(
+            new is old for new, old in zip(transformed, arg.flat, strict=True)
+        ):
+            # No element needed transforming: return the original array
+            # (as we already do for non-object ndarrays below) to
+            # preserve buffer identity. Rebuilding an equivalent copy
+            # breaks aliasing checks such as
+            # ``np.may_share_memory(np.asarray(x), x)`` that numpy uses
+            # (e.g. in ``Generator.permutation``) to decide whether to
+            # defensively copy before an in-place shuffle; the views
+            # pandas returns under copy-on-write are read-only, so
+            # skipping that copy raises "ValueError: array is read-only".
+            return arg
         # Keep the same memory layout as arg (the default is C_CONTIGUOUS)
         if arg.flags["F_CONTIGUOUS"] and not arg.flags["C_CONTIGUOUS"]:
             order = "F"

python/cudf/cudf/pandas/scripts/pandas-testing-plugin.py

@@ -193,7 +193,6 @@ def pytest_unconfigure(config):
     "tests/arrays/categorical/test_indexing.py::TestCategoricalIndexing::test_where_new_category_raises": "TODO: Add a reason for failure",
     "tests/arrays/categorical/test_missing.py::TestCategoricalMissing::test_compare_categorical_with_missing[a10-a20-categories0]": "AssertionError: Attributes of Series are different",
     "tests/arrays/categorical/test_replace.py::test_replace_categorical_ea_dtype": "TODO: Add a reason for failure",
-    "tests/arrays/categorical/test_replace.py::test_replace_categorical_ea_dtype_different_cats_raises": "Failed: DID NOT RAISE <class 'TypeError'>",
     "tests/arrays/categorical/test_sorting.py::TestCategoricalSort::test_sort_values": "TODO: Add a reason for failure",
     "tests/arrays/datetimes/test_constructors.py::TestDatetimeArrayConstructor::test_copy": "TODO: Add a reason for failure",
     "tests/arrays/floating/test_astype.py::test_astype_copy": "TODO: Add a reason for failure",
@@ -266,27 +265,6 @@ def pytest_unconfigure(config):
     "tests/arrays/sparse/test_array.py::TestSparseArrayAnalytics::test_ufunc_args": "TODO: Add a reason for failure",
     "tests/arrays/sparse/test_array.py::test_array_interface": "TODO: Add a reason for failure",
     "tests/arrays/sparse/test_constructors.py::TestConstructors::test_constructor_copy": "TODO: Add a reason for failure",
-    "tests/arrays/string_/test_string.py::test_isin[string=str[python]]": "TODO: Add a reason for failure",
-    "tests/arrays/string_/test_string.py::test_isin[string=string[pyarrow]]": "TODO: Add a reason for failure",
-    "tests/arrays/string_/test_string.py::test_isin[string=string[python]]": "TODO: Add a reason for failure",
-    "tests/arrays/string_/test_string.py::test_min_max[False-string=str[pyarrow]-max]": "assert np.float64(nan) is nan",
-    "tests/arrays/string_/test_string.py::test_min_max[False-string=str[pyarrow]-min]": "assert np.float64(nan) is nan",
-    "tests/arrays/string_/test_string.py::test_min_max[False-string=str[python]-max]": "AssertionError: assert np.float64(nan) is nan",
-    "tests/arrays/string_/test_string.py::test_min_max[False-string=str[python]-min]": "AssertionError: assert np.float64(nan) is nan",
-    "tests/arrays/string_/test_string.py::test_min_max[False-string=string[pyarrow]-max]": "assert np.float64(nan) is <NA>",
-    "tests/arrays/string_/test_string.py::test_min_max[False-string=string[pyarrow]-min]": "assert np.float64(nan) is <NA>",
-    "tests/arrays/string_/test_string.py::test_min_max[False-string=string[python]-max]": "AssertionError: assert np.float64(nan) is <NA>",
-    "tests/arrays/string_/test_string.py::test_min_max[False-string=string[python]-min]": "AssertionError: assert np.float64(nan) is <NA>",
-    "tests/arrays/string_/test_string.py::test_numpy_random_permute[string=str[python]-Series]": "ValueError: array is read-only",
-    "tests/arrays/string_/test_string.py::test_numpy_random_permute[string=string[python]-Series]": "ValueError: array is read-only",
-    "tests/arrays/string_/test_string.py::test_reduce_empty[False-string=str[pyarrow]-0]": "AssertionError: assert 0 == ''",
-    "tests/arrays/string_/test_string.py::test_reduce_empty[False-string=str[python]-0]": "AssertionError: assert 0 == ''",
-    "tests/arrays/string_/test_string.py::test_reduce_empty[False-string=string[pyarrow]-0]": "AssertionError: assert 0 == ''",
-    "tests/arrays/string_/test_string.py::test_reduce_empty[False-string=string[python]-0]": "AssertionError: assert 0 == ''",
-    "tests/arrays/string_/test_string.py::test_reduce_empty[True-string=str[pyarrow]-0]": "AssertionError: assert 0 == ''",
-    "tests/arrays/string_/test_string.py::test_reduce_empty[True-string=str[python]-0]": "AssertionError: assert 0 == ''",
-    "tests/arrays/string_/test_string.py::test_reduce_empty[True-string=string[pyarrow]-0]": "AssertionError: assert 0 == ''",
-    "tests/arrays/string_/test_string.py::test_reduce_empty[True-string=string[python]-0]": "AssertionError: assert 0 == ''",
     "tests/arrays/string_/test_string_arrow.py::test_pickle_roundtrip[na_value0]": "https://github.com/rapidsai/cudf/issues/18659#issuecomment-3710985854",
     "tests/arrays/string_/test_string_arrow.py::test_pickle_roundtrip[nan]": "https://github.com/rapidsai/cudf/issues/18659#issuecomment-3710985854",
     "tests/arrays/test_datetimelike.py::TestDatetimeArray::test_array_interface[B]": "TODO: Add a reason for failure",
@@ -366,9 +344,6 @@ def pytest_unconfigure(config):
     "tests/copy_view/test_array.py::test_dataframe_values[array]": "TODO: Add a reason for failure",
     "tests/copy_view/test_array.py::test_dataframe_values[asarray]": "TODO: Add a reason for failure",
     "tests/copy_view/test_array.py::test_dataframe_values[values]": "TODO: Add a reason for failure",
-    "tests/copy_view/test_array.py::test_series_array_string_dtype[string=object]": "AssertionError: assert False",
-    "tests/copy_view/test_array.py::test_series_array_string_dtype[string=str[python]]": "AssertionError: assert False",
-    "tests/copy_view/test_array.py::test_series_array_string_dtype[string=string[python]]": "AssertionError: assert False",
     "tests/copy_view/test_array.py::test_series_to_numpy": "TODO: Add a reason for failure",
     "tests/copy_view/test_array.py::test_series_values[array]": "TODO: Add a reason for failure",
     "tests/copy_view/test_array.py::test_series_values[np.array]": "AssertionError: assert False",

python/cudf/cudf/tests/series/methods/test_isin.py

@@ -1,8 +1,9 @@
-# SPDX-FileCopyrightText: Copyright (c) 2025-2026, NVIDIA CORPORATION.
+# SPDX-FileCopyrightText: Copyright (c) 2025-2026, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 # SPDX-License-Identifier: Apache-2.0
 
 import numpy as np
 import pandas as pd
+import pyarrow as pa
 import pytest
 
 import cudf
@@ -159,3 +160,98 @@ def test_isin_categorical(data, values):
     got = gsr.isin(values)
     expected = psr.isin(values)
     assert_eq(got, expected)
+
+
+@pytest.mark.parametrize(
+    "dtype",
+    [
+        np.dtype("object"),
+        pd.StringDtype(storage="python", na_value=pd.NA),
+        pd.StringDtype(storage="python", na_value=np.nan),
+        pd.StringDtype(storage="pyarrow", na_value=pd.NA),
+        pd.StringDtype(storage="pyarrow", na_value=np.nan),
+        pd.ArrowDtype(pa.string()),
+        pd.ArrowDtype(pa.large_string()),
+    ],
+)
+@pytest.mark.parametrize(
+    "values",
+    [
+        ["b", "x"],
+        np.array(["b", "x"], dtype=object),
+        pd.array(
+            ["b", "x"], dtype=pd.StringDtype(storage="python", na_value=pd.NA)
+        ),
+        pd.array(
+            ["b", "x"],
+            dtype=pd.StringDtype(storage="python", na_value=np.nan),
+        ),
+        pd.array(
+            ["b", "x"],
+            dtype=pd.StringDtype(storage="pyarrow", na_value=pd.NA),
+        ),
+        pd.array(
+            ["b", "x"],
+            dtype=pd.StringDtype(storage="pyarrow", na_value=np.nan),
+        ),
+        pd.array(["b", "x"], dtype=pd.ArrowDtype(pa.string())),
+        pa.array(["b", "x"]),
+    ],
+)
+def test_isin_string_dtype_flavors(dtype, values):
+    # isin must match on element values regardless of which string dtype
+    # flavor the series and the values use (object, pd.StringDtype with
+    # python/pyarrow storage and NaN/NA na_value, pd.ArrowDtype string).
+    data = ["a", "b", "c"]
+    psr = pd.Series(data, dtype=dtype)
+    gsr = cudf.Series(data, dtype=dtype)
+
+    got = gsr.isin(values)
+    if isinstance(values, pa.Array) and (
+        dtype == np.dtype("object")
+        or (isinstance(dtype, pd.StringDtype) and dtype.storage == "python")
+    ):
+        # pandas does not match pyarrow.Array values against
+        # object/python-storage series (returns all-False); cudf matches
+        # on element values.
+        assert got.to_pandas().tolist() == [False, True, False]
+    else:
+        expected = psr.isin(values)
+        assert_eq(got, expected)
+
+
+@pytest.mark.parametrize(
+    "dtype",
+    [
+        np.dtype("object"),
+        pd.StringDtype(storage="python", na_value=pd.NA),
+        pd.StringDtype(storage="python", na_value=np.nan),
+        pd.StringDtype(storage="pyarrow", na_value=pd.NA),
+        pd.StringDtype(storage="pyarrow", na_value=np.nan),
+        pd.ArrowDtype(pa.string()),
+        pd.ArrowDtype(pa.large_string()),
+    ],
+)
+def test_isin_string_null_values(dtype):
+    # cudf treats every NA-like value in ``values`` (None, pd.NA) as a
+    # match for nulls in the series. Results are compared with pandas
+    # except where pandas diverges: object dtype distinguishes the
+    # stored None from pd.NA, and ArrowDtype raises ArrowTypeError when
+    # the value set contains nulls.
+    data = ["a", "b", None]
+    psr = pd.Series(data, dtype=dtype)
+    gsr = cudf.Series(data, dtype=dtype)
+
+    got = gsr.isin([None])
+    if isinstance(dtype, pd.ArrowDtype):
+        assert got.to_pandas().tolist() == [False, False, True]
+    else:
+        assert_eq(got, psr.isin([None]))
+
+    got = gsr.isin(["a", pd.NA])
+    if isinstance(dtype, pd.ArrowDtype) or dtype == np.dtype("object"):
+        assert got.to_pandas().tolist() == [True, False, True]
+    else:
+        assert_eq(got, psr.isin(["a", pd.NA]))
+
+    assert_eq(gsr.isin([]), psr.isin([]))