针对pulse-transit的工具

dist/client/pandas/tests/extension/array_with_attr/__init__.py

@@ -0,0 +1,6 @@
+from pandas.tests.extension.array_with_attr.array import (
+    FloatAttrArray,
+    FloatAttrDtype,
+)
+
+__all__ = ["FloatAttrArray", "FloatAttrDtype"]

dist/client/pandas/tests/extension/array_with_attr/array.py

@@ -0,0 +1,84 @@
+"""
+Test extension array that has custom attribute information (not stored on the dtype).
+
+"""
+from __future__ import annotations
+
+import numbers
+
+import numpy as np
+
+from pandas._typing import type_t
+
+from pandas.core.dtypes.base import ExtensionDtype
+
+import pandas as pd
+from pandas.core.arrays import ExtensionArray
+
+
+class FloatAttrDtype(ExtensionDtype):
+    type = float
+    name = "float_attr"
+    na_value = np.nan
+
+    @classmethod
+    def construct_array_type(cls) -> type_t[FloatAttrArray]:
+        """
+        Return the array type associated with this dtype.
+
+        Returns
+        -------
+        type
+        """
+        return FloatAttrArray
+
+
+class FloatAttrArray(ExtensionArray):
+    dtype = FloatAttrDtype()
+    __array_priority__ = 1000
+
+    def __init__(self, values, attr=None) -> None:
+        if not isinstance(values, np.ndarray):
+            raise TypeError("Need to pass a numpy array of float64 dtype as values")
+        if not values.dtype == "float64":
+            raise TypeError("Need to pass a numpy array of float64 dtype as values")
+        self.data = values
+        self.attr = attr
+
+    @classmethod
+    def _from_sequence(cls, scalars, dtype=None, copy=False):
+        data = np.array(scalars, dtype="float64", copy=copy)
+        return cls(data)
+
+    def __getitem__(self, item):
+        if isinstance(item, numbers.Integral):
+            return self.data[item]
+        else:
+            # slice, list-like, mask
+            item = pd.api.indexers.check_array_indexer(self, item)
+            return type(self)(self.data[item], self.attr)
+
+    def __len__(self) -> int:
+        return len(self.data)
+
+    def isna(self):
+        return np.isnan(self.data)
+
+    def take(self, indexer, allow_fill=False, fill_value=None):
+        from pandas.api.extensions import take
+
+        data = self.data
+        if allow_fill and fill_value is None:
+            fill_value = self.dtype.na_value
+
+        result = take(data, indexer, fill_value=fill_value, allow_fill=allow_fill)
+        return type(self)(result, self.attr)
+
+    def copy(self):
+        return type(self)(self.data.copy(), self.attr)
+
+    @classmethod
+    def _concat_same_type(cls, to_concat):
+        data = np.concatenate([x.data for x in to_concat])
+        attr = to_concat[0].attr if len(to_concat) else None
+        return cls(data, attr)

dist/client/pandas/tests/extension/array_with_attr/test_array_with_attr.py

@@ -0,0 +1,33 @@
+import numpy as np
+
+import pandas as pd
+import pandas._testing as tm
+from pandas.tests.extension.array_with_attr import FloatAttrArray
+
+
+def test_concat_with_all_na():
+    # https://github.com/pandas-dev/pandas/pull/47762
+    # ensure that attribute of the column array is preserved (when it gets
+    # preserved in reindexing the array) during merge/concat
+    arr = FloatAttrArray(np.array([np.nan, np.nan], dtype="float64"), attr="test")
+
+    df1 = pd.DataFrame({"col": arr, "key": [0, 1]})
+    df2 = pd.DataFrame({"key": [0, 1], "col2": [1, 2]})
+    result = pd.merge(df1, df2, on="key")
+    expected = pd.DataFrame({"col": arr, "key": [0, 1], "col2": [1, 2]})
+    tm.assert_frame_equal(result, expected)
+    assert result["col"].array.attr == "test"
+
+    df1 = pd.DataFrame({"col": arr, "key": [0, 1]})
+    df2 = pd.DataFrame({"key": [0, 2], "col2": [1, 2]})
+    result = pd.merge(df1, df2, on="key")
+    expected = pd.DataFrame({"col": arr.take([0]), "key": [0], "col2": [1]})
+    tm.assert_frame_equal(result, expected)
+    assert result["col"].array.attr == "test"
+
+    result = pd.concat([df1.set_index("key"), df2.set_index("key")], axis=1)
+    expected = pd.DataFrame(
+        {"col": arr.take([0, 1, -1]), "col2": [1, np.nan, 2], "key": [0, 1, 2]}
+    ).set_index("key")
+    tm.assert_frame_equal(result, expected)
+    assert result["col"].array.attr == "test"

dist/client/pandas/tests/extension/arrow/arrays.py

@@ -0,0 +1,225 @@
+"""
+Rudimentary Apache Arrow-backed ExtensionArray.
+
+At the moment, just a boolean array / type is implemented.
+Eventually, we'll want to parametrize the type and support
+multiple dtypes. Not all methods are implemented yet, and the
+current implementation is not efficient.
+"""
+from __future__ import annotations
+
+import copy
+import itertools
+import operator
+
+import numpy as np
+import pyarrow as pa
+
+from pandas._typing import type_t
+
+import pandas as pd
+from pandas.api.extensions import (
+    ExtensionArray,
+    ExtensionDtype,
+    register_extension_dtype,
+    take,
+)
+from pandas.api.types import is_scalar
+from pandas.core.arraylike import OpsMixin
+from pandas.core.construction import extract_array
+
+
+@register_extension_dtype
+class ArrowBoolDtype(ExtensionDtype):
+
+    type = np.bool_
+    kind = "b"
+    name = "arrow_bool"
+    na_value = pa.NULL
+
+    @classmethod
+    def construct_array_type(cls) -> type_t[ArrowBoolArray]:
+        """
+        Return the array type associated with this dtype.
+
+        Returns
+        -------
+        type
+        """
+        return ArrowBoolArray
+
+    @property
+    def _is_boolean(self) -> bool:
+        return True
+
+
+@register_extension_dtype
+class ArrowStringDtype(ExtensionDtype):
+
+    type = str
+    kind = "U"
+    name = "arrow_string"
+    na_value = pa.NULL
+
+    @classmethod
+    def construct_array_type(cls) -> type_t[ArrowStringArray]:
+        """
+        Return the array type associated with this dtype.
+
+        Returns
+        -------
+        type
+        """
+        return ArrowStringArray
+
+
+class ArrowExtensionArray(OpsMixin, ExtensionArray):
+    _data: pa.ChunkedArray
+
+    @classmethod
+    def from_scalars(cls, values):
+        if isinstance(values, cls):
+            # in particular for empty cases the pa.array(np.asarray(...))
+            #  does not round-trip
+            return cls(values._data)
+
+        elif not len(values):
+            if isinstance(values, list):
+                dtype = bool if cls is ArrowBoolArray else str
+                values = np.array([], dtype=dtype)
+
+        arr = pa.chunked_array([pa.array(np.asarray(values))])
+        return cls(arr)
+
+    @classmethod
+    def from_array(cls, arr):
+        assert isinstance(arr, pa.Array)
+        return cls(pa.chunked_array([arr]))
+
+    @classmethod
+    def _from_sequence(cls, scalars, dtype=None, copy=False):
+        return cls.from_scalars(scalars)
+
+    def __repr__(self):
+        return f"{type(self).__name__}({repr(self._data)})"
+
+    def __contains__(self, obj) -> bool:
+        if obj is None or obj is self.dtype.na_value:
+            # None -> EA.__contains__ only checks for self._dtype.na_value, not
+            #  any compatible NA value.
+            # self.dtype.na_value -> <pa.NullScalar:None> isn't recognized by pd.isna
+            return bool(self.isna().any())
+        return bool(super().__contains__(obj))
+
+    def __getitem__(self, item):
+        if is_scalar(item):
+            return self._data.to_pandas()[item]
+        else:
+            vals = self._data.to_pandas()[item]
+            return type(self).from_scalars(vals)
+
+    def __len__(self):
+        return len(self._data)
+
+    def astype(self, dtype, copy=True):
+        # needed to fix this astype for the Series constructor.
+        if isinstance(dtype, type(self.dtype)) and dtype == self.dtype:
+            if copy:
+                return self.copy()
+            return self
+        return super().astype(dtype, copy)
+
+    @property
+    def dtype(self):
+        return self._dtype
+
+    def _logical_method(self, other, op):
+        if not isinstance(other, type(self)):
+            raise NotImplementedError()
+
+        result = op(np.array(self._data), np.array(other._data))
+        return ArrowBoolArray(
+            pa.chunked_array([pa.array(result, mask=pd.isna(self._data.to_pandas()))])
+        )
+
+    def __eq__(self, other):
+        if not isinstance(other, type(self)):
+            # TODO: use some pyarrow function here?
+            return np.asarray(self).__eq__(other)
+
+        return self._logical_method(other, operator.eq)
+
+    @property
+    def nbytes(self) -> int:
+        return sum(
+            x.size
+            for chunk in self._data.chunks
+            for x in chunk.buffers()
+            if x is not None
+        )
+
+    def isna(self):
+        nas = pd.isna(self._data.to_pandas())
+        return type(self).from_scalars(nas)
+
+    def take(self, indices, allow_fill=False, fill_value=None):
+        data = self._data.to_pandas()
+        data = extract_array(data, extract_numpy=True)
+
+        if allow_fill and fill_value is None:
+            fill_value = self.dtype.na_value
+
+        result = take(data, indices, fill_value=fill_value, allow_fill=allow_fill)
+        return self._from_sequence(result, dtype=self.dtype)
+
+    def copy(self):
+        return type(self)(copy.copy(self._data))
+
+    @classmethod
+    def _concat_same_type(cls, to_concat):
+        chunks = list(itertools.chain.from_iterable(x._data.chunks for x in to_concat))
+        arr = pa.chunked_array(chunks)
+        return cls(arr)
+
+    def __invert__(self):
+        return type(self).from_scalars(~self._data.to_pandas())
+
+    def _reduce(self, name: str, *, skipna: bool = True, **kwargs):
+        if skipna:
+            arr = self[~self.isna()]
+        else:
+            arr = self
+
+        try:
+            op = getattr(arr, name)
+        except AttributeError as err:
+            raise TypeError from err
+        return op(**kwargs)
+
+    def any(self, axis=0, out=None):
+        # Explicitly return a plain bool to reproduce GH-34660
+        return bool(self._data.to_pandas().any())
+
+    def all(self, axis=0, out=None):
+        # Explicitly return a plain bool to reproduce GH-34660
+        return bool(self._data.to_pandas().all())
+
+
+class ArrowBoolArray(ArrowExtensionArray):
+    def __init__(self, values):
+        if not isinstance(values, pa.ChunkedArray):
+            raise ValueError
+
+        assert values.type == pa.bool_()
+        self._data = values
+        self._dtype = ArrowBoolDtype()
+
+
+class ArrowStringArray(ArrowExtensionArray):
+    def __init__(self, values):
+        if not isinstance(values, pa.ChunkedArray):
+            raise ValueError
+
+        assert values.type == pa.string()
+        self._data = values
+        self._dtype = ArrowStringDtype()

dist/client/pandas/tests/extension/arrow/test_bool.py

@@ -0,0 +1,113 @@
+import numpy as np
+import pytest
+
+from pandas.compat import (
+    is_ci_environment,
+    is_platform_windows,
+)
+
+import pandas as pd
+import pandas._testing as tm
+from pandas.api.types import is_bool_dtype
+from pandas.tests.extension import base
+
+pytest.importorskip("pyarrow", minversion="1.0.1")
+
+from pandas.tests.extension.arrow.arrays import (  # isort:skip
+    ArrowBoolArray,
+    ArrowBoolDtype,
+)
+
+
+@pytest.fixture
+def dtype():
+    return ArrowBoolDtype()
+
+
+@pytest.fixture
+def data():
+    values = np.random.randint(0, 2, size=100, dtype=bool)
+    values[1] = ~values[0]
+    return ArrowBoolArray.from_scalars(values)
+
+
+@pytest.fixture
+def data_missing():
+    return ArrowBoolArray.from_scalars([None, True])
+
+
+def test_basic_equals(data):
+    # https://github.com/pandas-dev/pandas/issues/34660
+    assert pd.Series(data).equals(pd.Series(data))
+
+
+class BaseArrowTests:
+    pass
+
+
+class TestDtype(BaseArrowTests, base.BaseDtypeTests):
+    pass
+
+
+class TestInterface(BaseArrowTests, base.BaseInterfaceTests):
+    def test_copy(self, data):
+        # __setitem__ does not work, so we only have a smoke-test
+        data.copy()
+
+    def test_view(self, data):
+        # __setitem__ does not work, so we only have a smoke-test
+        data.view()
+
+    @pytest.mark.xfail(
+        raises=AssertionError,
+        reason="Doesn't recognize data._na_value as NA",
+    )
+    def test_contains(self, data, data_missing):
+        super().test_contains(data, data_missing)
+
+
+class TestConstructors(BaseArrowTests, base.BaseConstructorsTests):
+    # seems like some bug in isna on empty BoolArray returning floats.
+    @pytest.mark.xfail(reason="bad is-na for empty data")
+    def test_from_sequence_from_cls(self, data):
+        super().test_from_sequence_from_cls(data)
+
+    @pytest.mark.xfail(reason="pa.NULL is not recognised as scalar, GH-33899")
+    def test_series_constructor_no_data_with_index(self, dtype, na_value):
+        # pyarrow.lib.ArrowInvalid: only handle 1-dimensional arrays
+        super().test_series_constructor_no_data_with_index(dtype, na_value)
+
+    @pytest.mark.xfail(reason="pa.NULL is not recognised as scalar, GH-33899")
+    def test_series_constructor_scalar_na_with_index(self, dtype, na_value):
+        # pyarrow.lib.ArrowInvalid: only handle 1-dimensional arrays
+        super().test_series_constructor_scalar_na_with_index(dtype, na_value)
+
+    @pytest.mark.xfail(reason="ufunc 'invert' not supported for the input types")
+    def test_construct_empty_dataframe(self, dtype):
+        super().test_construct_empty_dataframe(dtype)
+
+    @pytest.mark.xfail(reason="_from_sequence ignores dtype keyword")
+    def test_empty(self, dtype):
+        super().test_empty(dtype)
+
+
+class TestReduce(base.BaseNoReduceTests):
+    def test_reduce_series_boolean(self):
+        pass
+
+
+@pytest.mark.skipif(
+    is_ci_environment() and is_platform_windows(),
+    reason="Causes stack overflow on Windows CI",
+)
+class TestReduceBoolean(base.BaseBooleanReduceTests):
+    pass
+
+
+def test_is_bool_dtype(data):
+    assert is_bool_dtype(data)
+    assert pd.core.common.is_bool_indexer(data)
+    s = pd.Series(range(len(data)))
+    result = s[data]
+    expected = s[np.asarray(data)]
+    tm.assert_series_equal(result, expected)

dist/client/pandas/tests/extension/arrow/test_string.py

@@ -0,0 +1,12 @@
+import pytest
+
+import pandas as pd
+
+pytest.importorskip("pyarrow", minversion="1.0.0")
+
+
+def test_constructor_from_list():
+    # GH 27673
+    result = pd.Series(["E"], dtype=pd.StringDtype(storage="pyarrow"))
+    assert isinstance(result.dtype, pd.StringDtype)
+    assert result.dtype.storage == "pyarrow"

dist/client/pandas/tests/extension/arrow/test_timestamp.py

@@ -0,0 +1,60 @@
+from __future__ import annotations
+
+import datetime
+
+import pytest
+
+from pandas._typing import type_t
+
+import pandas as pd
+from pandas.api.extensions import (
+    ExtensionDtype,
+    register_extension_dtype,
+)
+
+pytest.importorskip("pyarrow", minversion="1.0.1")
+
+import pyarrow as pa  # isort:skip
+
+from pandas.tests.extension.arrow.arrays import ArrowExtensionArray  # isort:skip
+
+
+@register_extension_dtype
+class ArrowTimestampUSDtype(ExtensionDtype):
+
+    type = datetime.datetime
+    kind = "M"
+    name = "arrow_timestamp_us"
+    na_value = pa.NULL
+
+    @classmethod
+    def construct_array_type(cls) -> type_t[ArrowTimestampUSArray]:
+        """
+        Return the array type associated with this dtype.
+
+        Returns
+        -------
+        type
+        """
+        return ArrowTimestampUSArray
+
+
+class ArrowTimestampUSArray(ArrowExtensionArray):
+    def __init__(self, values):
+        if not isinstance(values, pa.ChunkedArray):
+            raise ValueError
+
+        assert values.type == pa.timestamp("us")
+        self._data = values
+        self._dtype = ArrowTimestampUSDtype()
+
+
+def test_constructor_extensionblock():
+    # GH 34986
+    pd.DataFrame(
+        {
+            "timestamp": ArrowTimestampUSArray.from_scalars(
+                [None, datetime.datetime(2010, 9, 8, 7, 6, 5, 4)]
+            )
+        }
+    )

dist/client/pandas/tests/extension/base/__init__.py

@@ -0,0 +1,71 @@
+"""
+Base test suite for extension arrays.
+
+These tests are intended for third-party libraries to subclass to validate
+that their extension arrays and dtypes satisfy the interface. Moving or
+renaming the tests should not be done lightly.
+
+Libraries are expected to implement a few pytest fixtures to provide data
+for the tests. The fixtures may be located in either
+
+* The same module as your test class.
+* A ``conftest.py`` in the same directory as your test class.
+
+The full list of fixtures may be found in the ``conftest.py`` next to this
+file.
+
+.. code-block:: python
+
+   import pytest
+   from pandas.tests.extension.base import BaseDtypeTests
+
+
+   @pytest.fixture
+   def dtype():
+       return MyDtype()
+
+
+   class TestMyDtype(BaseDtypeTests):
+       pass
+
+
+Your class ``TestDtype`` will inherit all the tests defined on
+``BaseDtypeTests``. pytest's fixture discover will supply your ``dtype``
+wherever the test requires it. You're free to implement additional tests.
+
+All the tests in these modules use ``self.assert_frame_equal`` or
+``self.assert_series_equal`` for dataframe or series comparisons. By default,
+they use the usual ``pandas.testing.assert_frame_equal`` and
+``pandas.testing.assert_series_equal``. You can override the checks used
+by defining the staticmethods ``assert_frame_equal`` and
+``assert_series_equal`` on your base test class.
+
+"""
+from pandas.tests.extension.base.casting import BaseCastingTests  # noqa
+from pandas.tests.extension.base.constructors import BaseConstructorsTests  # noqa
+from pandas.tests.extension.base.dim2 import (  # noqa
+    Dim2CompatTests,
+    NDArrayBacked2DTests,
+)
+from pandas.tests.extension.base.dtype import BaseDtypeTests  # noqa
+from pandas.tests.extension.base.getitem import BaseGetitemTests  # noqa
+from pandas.tests.extension.base.groupby import BaseGroupbyTests  # noqa
+from pandas.tests.extension.base.index import BaseIndexTests  # noqa
+from pandas.tests.extension.base.interface import BaseInterfaceTests  # noqa
+from pandas.tests.extension.base.io import BaseParsingTests  # noqa
+from pandas.tests.extension.base.methods import BaseMethodsTests  # noqa
+from pandas.tests.extension.base.missing import BaseMissingTests  # noqa
+from pandas.tests.extension.base.ops import (  # noqa
+    BaseArithmeticOpsTests,
+    BaseComparisonOpsTests,
+    BaseOpsUtil,
+    BaseUnaryOpsTests,
+)
+from pandas.tests.extension.base.printing import BasePrintingTests  # noqa
+from pandas.tests.extension.base.reduce import (  # noqa
+    BaseBooleanReduceTests,
+    BaseNoReduceTests,
+    BaseNumericReduceTests,
+)
+from pandas.tests.extension.base.reshaping import BaseReshapingTests  # noqa
+from pandas.tests.extension.base.setitem import BaseSetitemTests  # noqa

dist/client/pandas/tests/extension/base/base.py

@@ -0,0 +1,21 @@
+import pandas._testing as tm
+
+
+class BaseExtensionTests:
+    # classmethod and different signature is needed
+    # to make inheritance compliant with mypy
+    @classmethod
+    def assert_equal(cls, left, right, **kwargs):
+        return tm.assert_equal(left, right, **kwargs)
+
+    @classmethod
+    def assert_series_equal(cls, left, right, *args, **kwargs):
+        return tm.assert_series_equal(left, right, *args, **kwargs)
+
+    @classmethod
+    def assert_frame_equal(cls, left, right, *args, **kwargs):
+        return tm.assert_frame_equal(left, right, *args, **kwargs)
+
+    @classmethod
+    def assert_extension_array_equal(cls, left, right, *args, **kwargs):
+        return tm.assert_extension_array_equal(left, right, *args, **kwargs)

dist/client/pandas/tests/extension/base/casting.py

@@ -0,0 +1,87 @@
+import numpy as np
+import pytest
+
+from pandas.compat import np_version_under1p20
+import pandas.util._test_decorators as td
+
+import pandas as pd
+from pandas.core.internals import ObjectBlock
+from pandas.tests.extension.base.base import BaseExtensionTests
+
+
+class BaseCastingTests(BaseExtensionTests):
+    """Casting to and from ExtensionDtypes"""
+
+    def test_astype_object_series(self, all_data):
+        ser = pd.Series(all_data, name="A")
+        result = ser.astype(object)
+        assert result.dtype == np.dtype(object)
+        if hasattr(result._mgr, "blocks"):
+            assert isinstance(result._mgr.blocks[0], ObjectBlock)
+        assert isinstance(result._mgr.array, np.ndarray)
+        assert result._mgr.array.dtype == np.dtype(object)
+
+    def test_astype_object_frame(self, all_data):
+        df = pd.DataFrame({"A": all_data})
+
+        result = df.astype(object)
+        if hasattr(result._mgr, "blocks"):
+            blk = result._data.blocks[0]
+            assert isinstance(blk, ObjectBlock), type(blk)
+        assert isinstance(result._mgr.arrays[0], np.ndarray)
+        assert result._mgr.arrays[0].dtype == np.dtype(object)
+
+        # earlier numpy raises TypeError on e.g. np.dtype(np.int64) == "Int64"
+        #  instead of returning False
+        if not np_version_under1p20:
+            # check that we can compare the dtypes
+            comp = result.dtypes == df.dtypes
+            assert not comp.any()
+
+    def test_tolist(self, data):
+        result = pd.Series(data).tolist()
+        expected = list(data)
+        assert result == expected
+
+    def test_astype_str(self, data):
+        result = pd.Series(data[:5]).astype(str)
+        expected = pd.Series([str(x) for x in data[:5]], dtype=str)
+        self.assert_series_equal(result, expected)
+
+    @pytest.mark.parametrize(
+        "nullable_string_dtype",
+        [
+            "string[python]",
+            pytest.param(
+                "string[pyarrow]", marks=td.skip_if_no("pyarrow", min_version="1.0.0")
+            ),
+        ],
+    )
+    def test_astype_string(self, data, nullable_string_dtype):
+        # GH-33465
+        result = pd.Series(data[:5]).astype(nullable_string_dtype)
+        expected = pd.Series([str(x) for x in data[:5]], dtype=nullable_string_dtype)
+        self.assert_series_equal(result, expected)
+
+    def test_to_numpy(self, data):
+        expected = np.asarray(data)
+
+        result = data.to_numpy()
+        self.assert_equal(result, expected)
+
+        result = pd.Series(data).to_numpy()
+        self.assert_equal(result, expected)
+
+    def test_astype_empty_dataframe(self, dtype):
+        # https://github.com/pandas-dev/pandas/issues/33113
+        df = pd.DataFrame()
+        result = df.astype(dtype)
+        self.assert_frame_equal(result, df)
+
+    @pytest.mark.parametrize("copy", [True, False])
+    def test_astype_own_type(self, data, copy):
+        # ensure that astype returns the original object for equal dtype and copy=False
+        # https://github.com/pandas-dev/pandas/issues/28488
+        result = data.astype(data.dtype, copy=copy)
+        assert (result is data) is (not copy)
+        self.assert_extension_array_equal(result, data)

dist/client/pandas/tests/extension/base/constructors.py

@@ -0,0 +1,142 @@
+import numpy as np
+import pytest
+
+import pandas as pd
+from pandas.api.extensions import ExtensionArray
+from pandas.core.internals.blocks import EABackedBlock
+from pandas.tests.extension.base.base import BaseExtensionTests
+
+
+class BaseConstructorsTests(BaseExtensionTests):
+    def test_from_sequence_from_cls(self, data):
+        result = type(data)._from_sequence(data, dtype=data.dtype)
+        self.assert_extension_array_equal(result, data)
+
+        data = data[:0]
+        result = type(data)._from_sequence(data, dtype=data.dtype)
+        self.assert_extension_array_equal(result, data)
+
+    def test_array_from_scalars(self, data):
+        scalars = [data[0], data[1], data[2]]
+        result = data._from_sequence(scalars)
+        assert isinstance(result, type(data))
+
+    def test_series_constructor(self, data):
+        result = pd.Series(data)
+        assert result.dtype == data.dtype
+        assert len(result) == len(data)
+        if hasattr(result._mgr, "blocks"):
+            assert isinstance(result._mgr.blocks[0], EABackedBlock)
+        assert result._mgr.array is data
+
+        # Series[EA] is unboxed / boxed correctly
+        result2 = pd.Series(result)
+        assert result2.dtype == data.dtype
+        if hasattr(result._mgr, "blocks"):
+            assert isinstance(result2._mgr.blocks[0], EABackedBlock)
+
+    def test_series_constructor_no_data_with_index(self, dtype, na_value):
+        result = pd.Series(index=[1, 2, 3], dtype=dtype)
+        expected = pd.Series([na_value] * 3, index=[1, 2, 3], dtype=dtype)
+        self.assert_series_equal(result, expected)
+
+        # GH 33559 - empty index
+        result = pd.Series(index=[], dtype=dtype)
+        expected = pd.Series([], index=pd.Index([], dtype="object"), dtype=dtype)
+        self.assert_series_equal(result, expected)
+
+    def test_series_constructor_scalar_na_with_index(self, dtype, na_value):
+        result = pd.Series(na_value, index=[1, 2, 3], dtype=dtype)
+        expected = pd.Series([na_value] * 3, index=[1, 2, 3], dtype=dtype)
+        self.assert_series_equal(result, expected)
+
+    def test_series_constructor_scalar_with_index(self, data, dtype):
+        scalar = data[0]
+        result = pd.Series(scalar, index=[1, 2, 3], dtype=dtype)
+        expected = pd.Series([scalar] * 3, index=[1, 2, 3], dtype=dtype)
+        self.assert_series_equal(result, expected)
+
+        result = pd.Series(scalar, index=["foo"], dtype=dtype)
+        expected = pd.Series([scalar], index=["foo"], dtype=dtype)
+        self.assert_series_equal(result, expected)
+
+    @pytest.mark.parametrize("from_series", [True, False])
+    def test_dataframe_constructor_from_dict(self, data, from_series):
+        if from_series:
+            data = pd.Series(data)
+        result = pd.DataFrame({"A": data})
+        assert result.dtypes["A"] == data.dtype
+        assert result.shape == (len(data), 1)
+        if hasattr(result._mgr, "blocks"):
+            assert isinstance(result._mgr.blocks[0], EABackedBlock)
+        assert isinstance(result._mgr.arrays[0], ExtensionArray)
+
+    def test_dataframe_from_series(self, data):
+        result = pd.DataFrame(pd.Series(data))
+        assert result.dtypes[0] == data.dtype
+        assert result.shape == (len(data), 1)
+        if hasattr(result._mgr, "blocks"):
+            assert isinstance(result._mgr.blocks[0], EABackedBlock)
+        assert isinstance(result._mgr.arrays[0], ExtensionArray)
+
+    def test_series_given_mismatched_index_raises(self, data):
+        msg = r"Length of values \(3\) does not match length of index \(5\)"
+        with pytest.raises(ValueError, match=msg):
+            pd.Series(data[:3], index=[0, 1, 2, 3, 4])
+
+    def test_from_dtype(self, data):
+        # construct from our dtype & string dtype
+        dtype = data.dtype
+
+        expected = pd.Series(data)
+        result = pd.Series(list(data), dtype=dtype)
+        self.assert_series_equal(result, expected)
+
+        result = pd.Series(list(data), dtype=str(dtype))
+        self.assert_series_equal(result, expected)
+
+        # gh-30280
+
+        expected = pd.DataFrame(data).astype(dtype)
+        result = pd.DataFrame(list(data), dtype=dtype)
+        self.assert_frame_equal(result, expected)
+
+        result = pd.DataFrame(list(data), dtype=str(dtype))
+        self.assert_frame_equal(result, expected)
+
+    def test_pandas_array(self, data):
+        # pd.array(extension_array) should be idempotent...
+        result = pd.array(data)
+        self.assert_extension_array_equal(result, data)
+
+    def test_pandas_array_dtype(self, data):
+        # ... but specifying dtype will override idempotency
+        result = pd.array(data, dtype=np.dtype(object))
+        expected = pd.arrays.PandasArray(np.asarray(data, dtype=object))
+        self.assert_equal(result, expected)
+
+    def test_construct_empty_dataframe(self, dtype):
+        # GH 33623
+        result = pd.DataFrame(columns=["a"], dtype=dtype)
+        expected = pd.DataFrame(
+            {"a": pd.array([], dtype=dtype)}, index=pd.Index([], dtype="object")
+        )
+        self.assert_frame_equal(result, expected)
+
+    def test_empty(self, dtype):
+        cls = dtype.construct_array_type()
+        result = cls._empty((4,), dtype=dtype)
+        assert isinstance(result, cls)
+        assert result.dtype == dtype
+        assert result.shape == (4,)
+
+        # GH#19600 method on ExtensionDtype
+        result2 = dtype.empty((4,))
+        assert isinstance(result2, cls)
+        assert result2.dtype == dtype
+        assert result2.shape == (4,)
+
+        result2 = dtype.empty(4)
+        assert isinstance(result2, cls)
+        assert result2.dtype == dtype
+        assert result2.shape == (4,)

dist/client/pandas/tests/extension/base/dim2.py

@@ -0,0 +1,301 @@
+"""
+Tests for 2D compatibility.
+"""
+import numpy as np
+import pytest
+
+from pandas._libs.missing import is_matching_na
+
+import pandas as pd
+from pandas.core.arrays.integer import INT_STR_TO_DTYPE
+from pandas.tests.extension.base.base import BaseExtensionTests
+
+
+class Dim2CompatTests(BaseExtensionTests):
+    def test_transpose(self, data):
+        arr2d = data.repeat(2).reshape(-1, 2)
+        shape = arr2d.shape
+        assert shape[0] != shape[-1]  # otherwise the rest of the test is useless
+
+        assert arr2d.T.shape == shape[::-1]
+
+    def test_frame_from_2d_array(self, data):
+        arr2d = data.repeat(2).reshape(-1, 2)
+
+        df = pd.DataFrame(arr2d)
+        expected = pd.DataFrame({0: arr2d[:, 0], 1: arr2d[:, 1]})
+        self.assert_frame_equal(df, expected)
+
+    def test_swapaxes(self, data):
+        arr2d = data.repeat(2).reshape(-1, 2)
+
+        result = arr2d.swapaxes(0, 1)
+        expected = arr2d.T
+        self.assert_extension_array_equal(result, expected)
+
+    def test_delete_2d(self, data):
+        arr2d = data.repeat(3).reshape(-1, 3)
+
+        # axis = 0
+        result = arr2d.delete(1, axis=0)
+        expected = data.delete(1).repeat(3).reshape(-1, 3)
+        self.assert_extension_array_equal(result, expected)
+
+        # axis = 1
+        result = arr2d.delete(1, axis=1)
+        expected = data.repeat(2).reshape(-1, 2)
+        self.assert_extension_array_equal(result, expected)
+
+    def test_take_2d(self, data):
+        arr2d = data.reshape(-1, 1)
+
+        result = arr2d.take([0, 0, -1], axis=0)
+
+        expected = data.take([0, 0, -1]).reshape(-1, 1)
+        self.assert_extension_array_equal(result, expected)
+
+    def test_repr_2d(self, data):
+        # this could fail in a corner case where an element contained the name
+        res = repr(data.reshape(1, -1))
+        assert res.count(f"<{type(data).__name__}") == 1
+
+        res = repr(data.reshape(-1, 1))
+        assert res.count(f"<{type(data).__name__}") == 1
+
+    def test_reshape(self, data):
+        arr2d = data.reshape(-1, 1)
+        assert arr2d.shape == (data.size, 1)
+        assert len(arr2d) == len(data)
+
+        arr2d = data.reshape((-1, 1))
+        assert arr2d.shape == (data.size, 1)
+        assert len(arr2d) == len(data)
+
+        with pytest.raises(ValueError):
+            data.reshape((data.size, 2))
+        with pytest.raises(ValueError):
+            data.reshape(data.size, 2)
+
+    def test_getitem_2d(self, data):
+        arr2d = data.reshape(1, -1)
+
+        result = arr2d[0]
+        self.assert_extension_array_equal(result, data)
+
+        with pytest.raises(IndexError):
+            arr2d[1]
+
+        with pytest.raises(IndexError):
+            arr2d[-2]
+
+        result = arr2d[:]
+        self.assert_extension_array_equal(result, arr2d)
+
+        result = arr2d[:, :]
+        self.assert_extension_array_equal(result, arr2d)
+
+        result = arr2d[:, 0]
+        expected = data[[0]]
+        self.assert_extension_array_equal(result, expected)
+
+        # dimension-expanding getitem on 1D
+        result = data[:, np.newaxis]
+        self.assert_extension_array_equal(result, arr2d.T)
+
+    def test_iter_2d(self, data):
+        arr2d = data.reshape(1, -1)
+
+        objs = list(iter(arr2d))
+        assert len(objs) == arr2d.shape[0]
+
+        for obj in objs:
+            assert isinstance(obj, type(data))
+            assert obj.dtype == data.dtype
+            assert obj.ndim == 1
+            assert len(obj) == arr2d.shape[1]
+
+    def test_tolist_2d(self, data):
+        arr2d = data.reshape(1, -1)
+
+        result = arr2d.tolist()
+        expected = [data.tolist()]
+
+        assert isinstance(result, list)
+        assert all(isinstance(x, list) for x in result)
+
+        assert result == expected
+
+    def test_concat_2d(self, data):
+        left = type(data)._concat_same_type([data, data]).reshape(-1, 2)
+        right = left.copy()
+
+        # axis=0
+        result = left._concat_same_type([left, right], axis=0)
+        expected = data._concat_same_type([data] * 4).reshape(-1, 2)
+        self.assert_extension_array_equal(result, expected)
+
+        # axis=1
+        result = left._concat_same_type([left, right], axis=1)
+        assert result.shape == (len(data), 4)
+        self.assert_extension_array_equal(result[:, :2], left)
+        self.assert_extension_array_equal(result[:, 2:], right)
+
+        # axis > 1 -> invalid
+        msg = "axis 2 is out of bounds for array of dimension 2"
+        with pytest.raises(ValueError, match=msg):
+            left._concat_same_type([left, right], axis=2)
+
+    @pytest.mark.parametrize("method", ["backfill", "pad"])
+    def test_fillna_2d_method(self, data_missing, method):
+        arr = data_missing.repeat(2).reshape(2, 2)
+        assert arr[0].isna().all()
+        assert not arr[1].isna().any()
+
+        result = arr.fillna(method=method)
+
+        expected = data_missing.fillna(method=method).repeat(2).reshape(2, 2)
+        self.assert_extension_array_equal(result, expected)
+
+    @pytest.mark.parametrize("method", ["mean", "median", "var", "std", "sum", "prod"])
+    def test_reductions_2d_axis_none(self, data, method):
+        arr2d = data.reshape(1, -1)
+
+        err_expected = None
+        err_result = None
+        try:
+            expected = getattr(data, method)()
+        except Exception as err:
+            # if the 1D reduction is invalid, the 2D reduction should be as well
+            err_expected = err
+            try:
+                result = getattr(arr2d, method)(axis=None)
+            except Exception as err2:
+                err_result = err2
+
+        else:
+            result = getattr(arr2d, method)(axis=None)
+
+        if err_result is not None or err_expected is not None:
+            assert type(err_result) == type(err_expected)
+            return
+
+        assert is_matching_na(result, expected) or result == expected
+
+    @pytest.mark.parametrize("method", ["mean", "median", "var", "std", "sum", "prod"])
+    def test_reductions_2d_axis0(self, data, method):
+        arr2d = data.reshape(1, -1)
+
+        kwargs = {}
+        if method == "std":
+            # pass ddof=0 so we get all-zero std instead of all-NA std
+            kwargs["ddof"] = 0
+
+        try:
+            result = getattr(arr2d, method)(axis=0, **kwargs)
+        except Exception as err:
+            try:
+                getattr(data, method)()
+            except Exception as err2:
+                assert type(err) == type(err2)
+                return
+            else:
+                raise AssertionError("Both reductions should raise or neither")
+
+        def get_reduction_result_dtype(dtype):
+            # windows and 32bit builds will in some cases have int32/uint32
+            #  where other builds will have int64/uint64.
+            if dtype.itemsize == 8:
+                return dtype
+            elif dtype.kind in "ib":
+                return INT_STR_TO_DTYPE[np.dtype(int).name]
+            else:
+                # i.e. dtype.kind == "u"
+                return INT_STR_TO_DTYPE[np.dtype(np.uint).name]
+
+        if method in ["mean", "median", "sum", "prod"]:
+            # std and var are not dtype-preserving
+            expected = data
+            if method in ["sum", "prod"] and data.dtype.kind in "iub":
+                dtype = get_reduction_result_dtype(data.dtype)
+
+                expected = data.astype(dtype)
+                if data.dtype.kind == "b" and method in ["sum", "prod"]:
+                    # We get IntegerArray instead of BooleanArray
+                    pass
+                else:
+                    assert type(expected) == type(data), type(expected)
+                assert dtype == expected.dtype
+
+            self.assert_extension_array_equal(result, expected)
+        elif method == "std":
+            self.assert_extension_array_equal(result, data - data)
+        # punt on method == "var"
+
+    @pytest.mark.parametrize("method", ["mean", "median", "var", "std", "sum", "prod"])
+    def test_reductions_2d_axis1(self, data, method):
+        arr2d = data.reshape(1, -1)
+
+        try:
+            result = getattr(arr2d, method)(axis=1)
+        except Exception as err:
+            try:
+                getattr(data, method)()
+            except Exception as err2:
+                assert type(err) == type(err2)
+                return
+            else:
+                raise AssertionError("Both reductions should raise or neither")
+
+        # not necessarily type/dtype-preserving, so weaker assertions
+        assert result.shape == (1,)
+        expected_scalar = getattr(data, method)()
+        res = result[0]
+        assert is_matching_na(res, expected_scalar) or res == expected_scalar
+
+
+class NDArrayBacked2DTests(Dim2CompatTests):
+    # More specific tests for NDArrayBackedExtensionArray subclasses
+
+    def test_copy_order(self, data):
+        # We should be matching numpy semantics for the "order" keyword in 'copy'
+        arr2d = data.repeat(2).reshape(-1, 2)
+        assert arr2d._ndarray.flags["C_CONTIGUOUS"]
+
+        res = arr2d.copy()
+        assert res._ndarray.flags["C_CONTIGUOUS"]
+
+        res = arr2d[::2, ::2].copy()
+        assert res._ndarray.flags["C_CONTIGUOUS"]
+
+        res = arr2d.copy("F")
+        assert not res._ndarray.flags["C_CONTIGUOUS"]
+        assert res._ndarray.flags["F_CONTIGUOUS"]
+
+        res = arr2d.copy("K")
+        assert res._ndarray.flags["C_CONTIGUOUS"]
+
+        res = arr2d.T.copy("K")
+        assert not res._ndarray.flags["C_CONTIGUOUS"]
+        assert res._ndarray.flags["F_CONTIGUOUS"]
+
+        # order not accepted by numpy
+        msg = r"order must be one of 'C', 'F', 'A', or 'K' \(got 'Q'\)"
+        with pytest.raises(ValueError, match=msg):
+            arr2d.copy("Q")
+
+        # neither contiguity
+        arr_nc = arr2d[::2]
+        assert not arr_nc._ndarray.flags["C_CONTIGUOUS"]
+        assert not arr_nc._ndarray.flags["F_CONTIGUOUS"]
+
+        assert arr_nc.copy()._ndarray.flags["C_CONTIGUOUS"]
+        assert not arr_nc.copy()._ndarray.flags["F_CONTIGUOUS"]
+
+        assert arr_nc.copy("C")._ndarray.flags["C_CONTIGUOUS"]
+        assert not arr_nc.copy("C")._ndarray.flags["F_CONTIGUOUS"]
+
+        assert not arr_nc.copy("F")._ndarray.flags["C_CONTIGUOUS"]
+        assert arr_nc.copy("F")._ndarray.flags["F_CONTIGUOUS"]
+
+        assert arr_nc.copy("K")._ndarray.flags["C_CONTIGUOUS"]
+        assert not arr_nc.copy("K")._ndarray.flags["F_CONTIGUOUS"]

dist/client/pandas/tests/extension/base/dtype.py

@@ -0,0 +1,137 @@
+import warnings
+
+import numpy as np
+import pytest
+
+import pandas as pd
+from pandas.api.types import (
+    infer_dtype,
+    is_object_dtype,
+    is_string_dtype,
+)
+from pandas.tests.extension.base.base import BaseExtensionTests
+
+
+class BaseDtypeTests(BaseExtensionTests):
+    """Base class for ExtensionDtype classes"""
+
+    def test_name(self, dtype):
+        assert isinstance(dtype.name, str)
+
+    def test_kind(self, dtype):
+        valid = set("biufcmMOSUV")
+        assert dtype.kind in valid
+
+    def test_construct_from_string_own_name(self, dtype):
+        result = dtype.construct_from_string(dtype.name)
+        assert type(result) is type(dtype)
+
+        # check OK as classmethod
+        result = type(dtype).construct_from_string(dtype.name)
+        assert type(result) is type(dtype)
+
+    def test_is_dtype_from_name(self, dtype):
+        result = type(dtype).is_dtype(dtype.name)
+        assert result is True
+
+    def test_is_dtype_unboxes_dtype(self, data, dtype):
+        assert dtype.is_dtype(data) is True
+
+    def test_is_dtype_from_self(self, dtype):
+        result = type(dtype).is_dtype(dtype)
+        assert result is True
+
+    def test_is_dtype_other_input(self, dtype):
+        assert dtype.is_dtype([1, 2, 3]) is False
+
+    def test_is_not_string_type(self, dtype):
+        return not is_string_dtype(dtype)
+
+    def test_is_not_object_type(self, dtype):
+        return not is_object_dtype(dtype)
+
+    def test_eq_with_str(self, dtype):
+        assert dtype == dtype.name
+        assert dtype != dtype.name + "-suffix"
+
+    def test_eq_with_numpy_object(self, dtype):
+        assert dtype != np.dtype("object")
+
+    def test_eq_with_self(self, dtype):
+        assert dtype == dtype
+        assert dtype != object()
+
+    def test_array_type(self, data, dtype):
+        assert dtype.construct_array_type() is type(data)
+
+    def test_check_dtype(self, data):
+        dtype = data.dtype
+
+        # check equivalency for using .dtypes
+        df = pd.DataFrame(
+            {"A": pd.Series(data, dtype=dtype), "B": data, "C": "foo", "D": 1}
+        )
+
+        # TODO(numpy-1.20): This warnings filter and if block can be removed
+        # once we require numpy>=1.20
+        with warnings.catch_warnings():
+            warnings.simplefilter("ignore", DeprecationWarning)
+            result = df.dtypes == str(dtype)
+            # NumPy>=1.20.0, but not pandas.compat.numpy till there
+            # is a wheel available with this change.
+            try:
+                new_numpy_behavior = np.dtype("int64") != "Int64"
+            except TypeError:
+                new_numpy_behavior = True
+
+        if dtype.name == "Int64" and not new_numpy_behavior:
+            expected = pd.Series([True, True, False, True], index=list("ABCD"))
+        else:
+            expected = pd.Series([True, True, False, False], index=list("ABCD"))
+
+        self.assert_series_equal(result, expected)
+
+        expected = pd.Series([True, True, False, False], index=list("ABCD"))
+        result = df.dtypes.apply(str) == str(dtype)
+        self.assert_series_equal(result, expected)
+
+    def test_hashable(self, dtype):
+        hash(dtype)  # no error
+
+    def test_str(self, dtype):
+        assert str(dtype) == dtype.name
+
+    def test_eq(self, dtype):
+        assert dtype == dtype.name
+        assert dtype != "anonther_type"
+
+    def test_construct_from_string(self, dtype):
+        dtype_instance = type(dtype).construct_from_string(dtype.name)
+        assert isinstance(dtype_instance, type(dtype))
+
+    def test_construct_from_string_another_type_raises(self, dtype):
+        msg = f"Cannot construct a '{type(dtype).__name__}' from 'another_type'"
+        with pytest.raises(TypeError, match=msg):
+            type(dtype).construct_from_string("another_type")
+
+    def test_construct_from_string_wrong_type_raises(self, dtype):
+        with pytest.raises(
+            TypeError,
+            match="'construct_from_string' expects a string, got <class 'int'>",
+        ):
+            type(dtype).construct_from_string(0)
+
+    def test_get_common_dtype(self, dtype):
+        # in practice we will not typically call this with a 1-length list
+        # (we shortcut to just use that dtype as the common dtype), but
+        # still testing as good practice to have this working (and it is the
+        # only case we can test in general)
+        assert dtype._get_common_dtype([dtype]) == dtype
+
+    @pytest.mark.parametrize("skipna", [True, False])
+    def test_infer_dtype(self, data, data_missing, skipna):
+        # only testing that this works without raising an error
+        res = infer_dtype(data, skipna=skipna)
+        assert isinstance(res, str)
+        res = infer_dtype(data_missing, skipna=skipna)
+        assert isinstance(res, str)

dist/client/pandas/tests/extension/base/getitem.py

@@ -0,0 +1,486 @@
+import numpy as np
+import pytest
+
+import pandas as pd
+import pandas._testing as tm
+from pandas.tests.extension.base.base import BaseExtensionTests
+
+
+class BaseGetitemTests(BaseExtensionTests):
+    """Tests for ExtensionArray.__getitem__."""
+
+    def test_iloc_series(self, data):
+        ser = pd.Series(data)
+        result = ser.iloc[:4]
+        expected = pd.Series(data[:4])
+        self.assert_series_equal(result, expected)
+
+        result = ser.iloc[[0, 1, 2, 3]]
+        self.assert_series_equal(result, expected)
+
+    def test_iloc_frame(self, data):
+        df = pd.DataFrame({"A": data, "B": np.arange(len(data), dtype="int64")})
+        expected = pd.DataFrame({"A": data[:4]})
+
+        # slice -> frame
+        result = df.iloc[:4, [0]]
+        self.assert_frame_equal(result, expected)
+
+        # sequence -> frame
+        result = df.iloc[[0, 1, 2, 3], [0]]
+        self.assert_frame_equal(result, expected)
+
+        expected = pd.Series(data[:4], name="A")
+
+        # slice -> series
+        result = df.iloc[:4, 0]
+        self.assert_series_equal(result, expected)
+
+        # sequence -> series
+        result = df.iloc[:4, 0]
+        self.assert_series_equal(result, expected)
+
+        # GH#32959 slice columns with step
+        result = df.iloc[:, ::2]
+        self.assert_frame_equal(result, df[["A"]])
+        result = df[["B", "A"]].iloc[:, ::2]
+        self.assert_frame_equal(result, df[["B"]])
+
+    def test_iloc_frame_single_block(self, data):
+        # GH#32959 null slice along index, slice along columns with single-block
+        df = pd.DataFrame({"A": data})
+
+        result = df.iloc[:, :]
+        self.assert_frame_equal(result, df)
+
+        result = df.iloc[:, :1]
+        self.assert_frame_equal(result, df)
+
+        result = df.iloc[:, :2]
+        self.assert_frame_equal(result, df)
+
+        result = df.iloc[:, ::2]
+        self.assert_frame_equal(result, df)
+
+        result = df.iloc[:, 1:2]
+        self.assert_frame_equal(result, df.iloc[:, :0])
+
+        result = df.iloc[:, -1:]
+        self.assert_frame_equal(result, df)
+
+    def test_loc_series(self, data):
+        ser = pd.Series(data)
+        result = ser.loc[:3]
+        expected = pd.Series(data[:4])
+        self.assert_series_equal(result, expected)
+
+        result = ser.loc[[0, 1, 2, 3]]
+        self.assert_series_equal(result, expected)
+
+    def test_loc_frame(self, data):
+        df = pd.DataFrame({"A": data, "B": np.arange(len(data), dtype="int64")})
+        expected = pd.DataFrame({"A": data[:4]})
+
+        # slice -> frame
+        result = df.loc[:3, ["A"]]
+        self.assert_frame_equal(result, expected)
+
+        # sequence -> frame
+        result = df.loc[[0, 1, 2, 3], ["A"]]
+        self.assert_frame_equal(result, expected)
+
+        expected = pd.Series(data[:4], name="A")
+
+        # slice -> series
+        result = df.loc[:3, "A"]
+        self.assert_series_equal(result, expected)
+
+        # sequence -> series
+        result = df.loc[:3, "A"]
+        self.assert_series_equal(result, expected)
+
+    def test_loc_iloc_frame_single_dtype(self, data):
+        # GH#27110 bug in ExtensionBlock.iget caused df.iloc[n] to incorrectly
+        #  return a scalar
+        df = pd.DataFrame({"A": data})
+        expected = pd.Series([data[2]], index=["A"], name=2, dtype=data.dtype)
+
+        result = df.loc[2]
+        self.assert_series_equal(result, expected)
+
+        expected = pd.Series(
+            [data[-1]], index=["A"], name=len(data) - 1, dtype=data.dtype
+        )
+        result = df.iloc[-1]
+        self.assert_series_equal(result, expected)
+
+    def test_getitem_scalar(self, data):
+        result = data[0]
+        assert isinstance(result, data.dtype.type)
+
+        result = pd.Series(data)[0]
+        assert isinstance(result, data.dtype.type)
+
+    def test_getitem_invalid(self, data):
+        # TODO: box over scalar, [scalar], (scalar,)?
+
+        msg = (
+            r"only integers, slices \(`:`\), ellipsis \(`...`\), numpy.newaxis "
+            r"\(`None`\) and integer or boolean arrays are valid indices"
+        )
+        with pytest.raises(IndexError, match=msg):
+            data["foo"]
+        with pytest.raises(IndexError, match=msg):
+            data[2.5]
+
+        ub = len(data)
+        msg = "|".join(
+            [
+                "list index out of range",  # json
+                "index out of bounds",  # pyarrow
+                "Out of bounds access",  # Sparse
+                f"loc must be an integer between -{ub} and {ub}",  # Sparse
+                f"index {ub+1} is out of bounds for axis 0 with size {ub}",
+                f"index -{ub+1} is out of bounds for axis 0 with size {ub}",
+            ]
+        )
+        with pytest.raises(IndexError, match=msg):
+            data[ub + 1]
+        with pytest.raises(IndexError, match=msg):
+            data[-ub - 1]
+
+    def test_getitem_scalar_na(self, data_missing, na_cmp, na_value):
+        result = data_missing[0]
+        assert na_cmp(result, na_value)
+
+    def test_getitem_empty(self, data):
+        # Indexing with empty list
+        result = data[[]]
+        assert len(result) == 0
+        assert isinstance(result, type(data))
+
+        expected = data[np.array([], dtype="int64")]
+        self.assert_extension_array_equal(result, expected)
+
+    def test_getitem_mask(self, data):
+        # Empty mask, raw array
+        mask = np.zeros(len(data), dtype=bool)
+        result = data[mask]
+        assert len(result) == 0
+        assert isinstance(result, type(data))
+
+        # Empty mask, in series
+        mask = np.zeros(len(data), dtype=bool)
+        result = pd.Series(data)[mask]
+        assert len(result) == 0
+        assert result.dtype == data.dtype
+
+        # non-empty mask, raw array
+        mask[0] = True
+        result = data[mask]
+        assert len(result) == 1
+        assert isinstance(result, type(data))
+
+        # non-empty mask, in series
+        result = pd.Series(data)[mask]
+        assert len(result) == 1
+        assert result.dtype == data.dtype
+
+    def test_getitem_mask_raises(self, data):
+        mask = np.array([True, False])
+        msg = f"Boolean index has wrong length: 2 instead of {len(data)}"
+        with pytest.raises(IndexError, match=msg):
+            data[mask]
+
+        mask = pd.array(mask, dtype="boolean")
+        with pytest.raises(IndexError, match=msg):
+            data[mask]
+
+    def test_getitem_boolean_array_mask(self, data):
+        mask = pd.array(np.zeros(data.shape, dtype="bool"), dtype="boolean")
+        result = data[mask]
+        assert len(result) == 0
+        assert isinstance(result, type(data))
+
+        result = pd.Series(data)[mask]
+        assert len(result) == 0
+        assert result.dtype == data.dtype
+
+        mask[:5] = True
+        expected = data.take([0, 1, 2, 3, 4])
+        result = data[mask]
+        self.assert_extension_array_equal(result, expected)
+
+        expected = pd.Series(expected)
+        result = pd.Series(data)[mask]
+        self.assert_series_equal(result, expected)
+
+    def test_getitem_boolean_na_treated_as_false(self, data):
+        # https://github.com/pandas-dev/pandas/issues/31503
+        mask = pd.array(np.zeros(data.shape, dtype="bool"), dtype="boolean")
+        mask[:2] = pd.NA
+        mask[2:4] = True
+
+        result = data[mask]
+        expected = data[mask.fillna(False)]
+
+        self.assert_extension_array_equal(result, expected)
+
+        s = pd.Series(data)
+
+        result = s[mask]
+        expected = s[mask.fillna(False)]
+
+        self.assert_series_equal(result, expected)
+
+    @pytest.mark.parametrize(
+        "idx",
+        [[0, 1, 2], pd.array([0, 1, 2], dtype="Int64"), np.array([0, 1, 2])],
+        ids=["list", "integer-array", "numpy-array"],
+    )
+    def test_getitem_integer_array(self, data, idx):
+        result = data[idx]
+        assert len(result) == 3
+        assert isinstance(result, type(data))
+        expected = data.take([0, 1, 2])
+        self.assert_extension_array_equal(result, expected)
+
+        expected = pd.Series(expected)
+        result = pd.Series(data)[idx]
+        self.assert_series_equal(result, expected)
+
+    @pytest.mark.parametrize(
+        "idx",
+        [[0, 1, 2, pd.NA], pd.array([0, 1, 2, pd.NA], dtype="Int64")],
+        ids=["list", "integer-array"],
+    )
+    def test_getitem_integer_with_missing_raises(self, data, idx):
+        msg = "Cannot index with an integer indexer containing NA values"
+        with pytest.raises(ValueError, match=msg):
+            data[idx]
+
+    @pytest.mark.xfail(
+        reason="Tries label-based and raises KeyError; "
+        "in some cases raises when calling np.asarray"
+    )
+    @pytest.mark.parametrize(
+        "idx",
+        [[0, 1, 2, pd.NA], pd.array([0, 1, 2, pd.NA], dtype="Int64")],
+        ids=["list", "integer-array"],
+    )
+    def test_getitem_series_integer_with_missing_raises(self, data, idx):
+        msg = "Cannot index with an integer indexer containing NA values"
+        # TODO: this raises KeyError about labels not found (it tries label-based)
+
+        ser = pd.Series(data, index=[tm.rands(4) for _ in range(len(data))])
+        with pytest.raises(ValueError, match=msg):
+            ser[idx]
+
+    def test_getitem_slice(self, data):
+        # getitem[slice] should return an array
+        result = data[slice(0)]  # empty
+        assert isinstance(result, type(data))
+
+        result = data[slice(1)]  # scalar
+        assert isinstance(result, type(data))
+
+    def test_getitem_ellipsis_and_slice(self, data):
+        # GH#40353 this is called from getitem_block_index
+        result = data[..., :]
+        self.assert_extension_array_equal(result, data)
+
+        result = data[:, ...]
+        self.assert_extension_array_equal(result, data)
+
+        result = data[..., :3]
+        self.assert_extension_array_equal(result, data[:3])
+
+        result = data[:3, ...]
+        self.assert_extension_array_equal(result, data[:3])
+
+        result = data[..., ::2]
+        self.assert_extension_array_equal(result, data[::2])
+
+        result = data[::2, ...]
+        self.assert_extension_array_equal(result, data[::2])
+
+    def test_get(self, data):
+        # GH 20882
+        s = pd.Series(data, index=[2 * i for i in range(len(data))])
+        assert s.get(4) == s.iloc[2]
+
+        result = s.get([4, 6])
+        expected = s.iloc[[2, 3]]
+        self.assert_series_equal(result, expected)
+
+        result = s.get(slice(2))
+        expected = s.iloc[[0, 1]]
+        self.assert_series_equal(result, expected)
+
+        assert s.get(-1) is None
+        assert s.get(s.index.max() + 1) is None
+
+        s = pd.Series(data[:6], index=list("abcdef"))
+        assert s.get("c") == s.iloc[2]
+
+        result = s.get(slice("b", "d"))
+        expected = s.iloc[[1, 2, 3]]
+        self.assert_series_equal(result, expected)
+
+        result = s.get("Z")
+        assert result is None
+
+        assert s.get(4) == s.iloc[4]
+        assert s.get(-1) == s.iloc[-1]
+        assert s.get(len(s)) is None
+
+        # GH 21257
+        s = pd.Series(data)
+        s2 = s[::2]
+        assert s2.get(1) is None
+
+    def test_take_sequence(self, data):
+        result = pd.Series(data)[[0, 1, 3]]
+        assert result.iloc[0] == data[0]
+        assert result.iloc[1] == data[1]
+        assert result.iloc[2] == data[3]
+
+    def test_take(self, data, na_value, na_cmp):
+        result = data.take([0, -1])
+        assert result.dtype == data.dtype
+        assert result[0] == data[0]
+        assert result[1] == data[-1]
+
+        result = data.take([0, -1], allow_fill=True, fill_value=na_value)
+        assert result[0] == data[0]
+        assert na_cmp(result[1], na_value)
+
+        with pytest.raises(IndexError, match="out of bounds"):
+            data.take([len(data) + 1])
+
+    def test_take_empty(self, data, na_value, na_cmp):
+        empty = data[:0]
+
+        result = empty.take([-1], allow_fill=True)
+        assert na_cmp(result[0], na_value)
+
+        msg = "cannot do a non-empty take from an empty axes|out of bounds"
+
+        with pytest.raises(IndexError, match=msg):
+            empty.take([-1])
+
+        with pytest.raises(IndexError, match="cannot do a non-empty take"):
+            empty.take([0, 1])
+
+    def test_take_negative(self, data):
+        # https://github.com/pandas-dev/pandas/issues/20640
+        n = len(data)
+        result = data.take([0, -n, n - 1, -1])
+        expected = data.take([0, 0, n - 1, n - 1])
+        self.assert_extension_array_equal(result, expected)
+
+    def test_take_non_na_fill_value(self, data_missing):
+        fill_value = data_missing[1]  # valid
+        na = data_missing[0]
+
+        arr = data_missing._from_sequence(
+            [na, fill_value, na], dtype=data_missing.dtype
+        )
+        result = arr.take([-1, 1], fill_value=fill_value, allow_fill=True)
+        expected = arr.take([1, 1])
+        self.assert_extension_array_equal(result, expected)
+
+    def test_take_pandas_style_negative_raises(self, data, na_value):
+        with pytest.raises(ValueError, match=""):
+            data.take([0, -2], fill_value=na_value, allow_fill=True)
+
+    @pytest.mark.parametrize("allow_fill", [True, False])
+    def test_take_out_of_bounds_raises(self, data, allow_fill):
+        arr = data[:3]
+
+        with pytest.raises(IndexError, match="out of bounds|out-of-bounds"):
+            arr.take(np.asarray([0, 3]), allow_fill=allow_fill)
+
+    def test_take_series(self, data):
+        s = pd.Series(data)
+        result = s.take([0, -1])
+        expected = pd.Series(
+            data._from_sequence([data[0], data[len(data) - 1]], dtype=s.dtype),
+            index=[0, len(data) - 1],
+        )
+        self.assert_series_equal(result, expected)
+
+    def test_reindex(self, data, na_value):
+        s = pd.Series(data)
+        result = s.reindex([0, 1, 3])
+        expected = pd.Series(data.take([0, 1, 3]), index=[0, 1, 3])
+        self.assert_series_equal(result, expected)
+
+        n = len(data)
+        result = s.reindex([-1, 0, n])
+        expected = pd.Series(
+            data._from_sequence([na_value, data[0], na_value], dtype=s.dtype),
+            index=[-1, 0, n],
+        )
+        self.assert_series_equal(result, expected)
+
+        result = s.reindex([n, n + 1])
+        expected = pd.Series(
+            data._from_sequence([na_value, na_value], dtype=s.dtype), index=[n, n + 1]
+        )
+        self.assert_series_equal(result, expected)
+
+    def test_reindex_non_na_fill_value(self, data_missing):
+        valid = data_missing[1]
+        na = data_missing[0]
+
+        arr = data_missing._from_sequence([na, valid], dtype=data_missing.dtype)
+        ser = pd.Series(arr)
+        result = ser.reindex([0, 1, 2], fill_value=valid)
+        expected = pd.Series(
+            data_missing._from_sequence([na, valid, valid], dtype=data_missing.dtype)
+        )
+
+        self.assert_series_equal(result, expected)
+
+    def test_loc_len1(self, data):
+        # see GH-27785 take_nd with indexer of len 1 resulting in wrong ndim
+        df = pd.DataFrame({"A": data})
+        res = df.loc[[0], "A"]
+        assert res.ndim == 1
+        assert res._mgr.arrays[0].ndim == 1
+        if hasattr(res._mgr, "blocks"):
+            assert res._mgr._block.ndim == 1
+
+    def test_item(self, data):
+        # https://github.com/pandas-dev/pandas/pull/30175
+        s = pd.Series(data)
+        result = s[:1].item()
+        assert result == data[0]
+
+        msg = "can only convert an array of size 1 to a Python scalar"
+        with pytest.raises(ValueError, match=msg):
+            s[:0].item()
+
+        with pytest.raises(ValueError, match=msg):
+            s.item()
+
+    def test_ellipsis_index(self):
+        # GH42430 1D slices over extension types turn into N-dimensional slices over
+        #  ExtensionArrays
+        class CapturingStringArray(pd.arrays.StringArray):
+            """Extend StringArray to capture arguments to __getitem__"""
+
+            def __getitem__(self, item):
+                self.last_item_arg = item
+                return super().__getitem__(item)
+
+        df = pd.DataFrame(
+            {"col1": CapturingStringArray(np.array(["hello", "world"], dtype=object))}
+        )
+        _ = df.iloc[:1]
+
+        # String comparison because there's no native way to compare slices.
+        # Before the fix for GH42430, last_item_arg would get set to the 2D slice
+        # (Ellipsis, slice(None, 1, None))
+        self.assert_equal(str(df["col1"].array.last_item_arg), "slice(None, 1, None)")

dist/client/pandas/tests/extension/base/groupby.py

@@ -0,0 +1,106 @@
+import pytest
+
+import pandas as pd
+import pandas._testing as tm
+from pandas.tests.extension.base.base import BaseExtensionTests
+
+
+class BaseGroupbyTests(BaseExtensionTests):
+    """Groupby-specific tests."""
+
+    def test_grouping_grouper(self, data_for_grouping):
+        df = pd.DataFrame(
+            {"A": ["B", "B", None, None, "A", "A", "B", "C"], "B": data_for_grouping}
+        )
+        gr1 = df.groupby("A").grouper.groupings[0]
+        gr2 = df.groupby("B").grouper.groupings[0]
+
+        tm.assert_numpy_array_equal(gr1.grouping_vector, df.A.values)
+        tm.assert_extension_array_equal(gr2.grouping_vector, data_for_grouping)
+
+    @pytest.mark.parametrize("as_index", [True, False])
+    def test_groupby_extension_agg(self, as_index, data_for_grouping):
+        df = pd.DataFrame({"A": [1, 1, 2, 2, 3, 3, 1, 4], "B": data_for_grouping})
+        result = df.groupby("B", as_index=as_index).A.mean()
+        _, uniques = pd.factorize(data_for_grouping, sort=True)
+
+        if as_index:
+            index = pd.Index._with_infer(uniques, name="B")
+            expected = pd.Series([3.0, 1.0, 4.0], index=index, name="A")
+            self.assert_series_equal(result, expected)
+        else:
+            expected = pd.DataFrame({"B": uniques, "A": [3.0, 1.0, 4.0]})
+            self.assert_frame_equal(result, expected)
+
+    def test_groupby_agg_extension(self, data_for_grouping):
+        # GH#38980 groupby agg on extension type fails for non-numeric types
+        df = pd.DataFrame({"A": [1, 1, 2, 2, 3, 3, 1, 4], "B": data_for_grouping})
+
+        expected = df.iloc[[0, 2, 4, 7]]
+        expected = expected.set_index("A")
+
+        result = df.groupby("A").agg({"B": "first"})
+        self.assert_frame_equal(result, expected)
+
+        result = df.groupby("A").agg("first")
+        self.assert_frame_equal(result, expected)
+
+        result = df.groupby("A").first()
+        self.assert_frame_equal(result, expected)
+
+    def test_groupby_extension_no_sort(self, data_for_grouping):
+        df = pd.DataFrame({"A": [1, 1, 2, 2, 3, 3, 1, 4], "B": data_for_grouping})
+        result = df.groupby("B", sort=False).A.mean()
+        _, index = pd.factorize(data_for_grouping, sort=False)
+
+        index = pd.Index._with_infer(index, name="B")
+        expected = pd.Series([1.0, 3.0, 4.0], index=index, name="A")
+        self.assert_series_equal(result, expected)
+
+    def test_groupby_extension_transform(self, data_for_grouping):
+        valid = data_for_grouping[~data_for_grouping.isna()]
+        df = pd.DataFrame({"A": [1, 1, 3, 3, 1, 4], "B": valid})
+
+        result = df.groupby("B").A.transform(len)
+        expected = pd.Series([3, 3, 2, 2, 3, 1], name="A")
+
+        self.assert_series_equal(result, expected)
+
+    def test_groupby_extension_apply(self, data_for_grouping, groupby_apply_op):
+        df = pd.DataFrame({"A": [1, 1, 2, 2, 3, 3, 1, 4], "B": data_for_grouping})
+        df.groupby("B").apply(groupby_apply_op)
+        df.groupby("B").A.apply(groupby_apply_op)
+        df.groupby("A").apply(groupby_apply_op)
+        df.groupby("A").B.apply(groupby_apply_op)
+
+    def test_groupby_apply_identity(self, data_for_grouping):
+        df = pd.DataFrame({"A": [1, 1, 2, 2, 3, 3, 1, 4], "B": data_for_grouping})
+        result = df.groupby("A").B.apply(lambda x: x.array)
+        expected = pd.Series(
+            [
+                df.B.iloc[[0, 1, 6]].array,
+                df.B.iloc[[2, 3]].array,
+                df.B.iloc[[4, 5]].array,
+                df.B.iloc[[7]].array,
+            ],
+            index=pd.Index([1, 2, 3, 4], name="A"),
+            name="B",
+        )
+        self.assert_series_equal(result, expected)
+
+    def test_in_numeric_groupby(self, data_for_grouping):
+        df = pd.DataFrame(
+            {
+                "A": [1, 1, 2, 2, 3, 3, 1, 4],
+                "B": data_for_grouping,
+                "C": [1, 1, 1, 1, 1, 1, 1, 1],
+            }
+        )
+        result = df.groupby("A").sum().columns
+
+        if data_for_grouping.dtype._is_numeric:
+            expected = pd.Index(["B", "C"])
+        else:
+            expected = pd.Index(["C"])
+
+        tm.assert_index_equal(result, expected)

dist/client/pandas/tests/extension/base/index.py

@@ -0,0 +1,20 @@
+"""
+Tests for Indexes backed by arbitrary ExtensionArrays.
+"""
+import pandas as pd
+from pandas.tests.extension.base.base import BaseExtensionTests
+
+
+class BaseIndexTests(BaseExtensionTests):
+    """Tests for Index object backed by an ExtensionArray"""
+
+    def test_index_from_array(self, data):
+        idx = pd.Index(data)
+        assert data.dtype == idx.dtype
+
+    def test_index_from_listlike_with_dtype(self, data):
+        idx = pd.Index(data, dtype=data.dtype)
+        assert idx.dtype == data.dtype
+
+        idx = pd.Index(list(data), dtype=data.dtype)
+        assert idx.dtype == data.dtype

dist/client/pandas/tests/extension/base/interface.py

@@ -0,0 +1,127 @@
+import numpy as np
+
+from pandas.core.dtypes.common import is_extension_array_dtype
+from pandas.core.dtypes.dtypes import ExtensionDtype
+
+import pandas as pd
+import pandas._testing as tm
+from pandas.tests.extension.base.base import BaseExtensionTests
+
+
+class BaseInterfaceTests(BaseExtensionTests):
+    """Tests that the basic interface is satisfied."""
+
+    # ------------------------------------------------------------------------
+    # Interface
+    # ------------------------------------------------------------------------
+
+    def test_len(self, data):
+        assert len(data) == 100
+
+    def test_size(self, data):
+        assert data.size == 100
+
+    def test_ndim(self, data):
+        assert data.ndim == 1
+
+    def test_can_hold_na_valid(self, data):
+        # GH-20761
+        assert data._can_hold_na is True
+
+    def test_contains(self, data, data_missing):
+        # GH-37867
+        # Tests for membership checks. Membership checks for nan-likes is tricky and
+        # the settled on rule is: `nan_like in arr` is True if nan_like is
+        # arr.dtype.na_value and arr.isna().any() is True. Else the check returns False.
+
+        na_value = data.dtype.na_value
+        # ensure data without missing values
+        data = data[~data.isna()]
+
+        # first elements are non-missing
+        assert data[0] in data
+        assert data_missing[0] in data_missing
+
+        # check the presence of na_value
+        assert na_value in data_missing
+        assert na_value not in data
+
+        # the data can never contain other nan-likes than na_value
+        for na_value_obj in tm.NULL_OBJECTS:
+            if na_value_obj is na_value or type(na_value_obj) == type(na_value):
+                # type check for e.g. two instances of Decimal("NAN")
+                continue
+            assert na_value_obj not in data
+            assert na_value_obj not in data_missing
+
+    def test_memory_usage(self, data):
+        s = pd.Series(data)
+        result = s.memory_usage(index=False)
+        assert result == s.nbytes
+
+    def test_array_interface(self, data):
+        result = np.array(data)
+        assert result[0] == data[0]
+
+        result = np.array(data, dtype=object)
+        expected = np.array(list(data), dtype=object)
+        tm.assert_numpy_array_equal(result, expected)
+
+    def test_is_extension_array_dtype(self, data):
+        assert is_extension_array_dtype(data)
+        assert is_extension_array_dtype(data.dtype)
+        assert is_extension_array_dtype(pd.Series(data))
+        assert isinstance(data.dtype, ExtensionDtype)
+
+    def test_no_values_attribute(self, data):
+        # GH-20735: EA's with .values attribute give problems with internal
+        # code, disallowing this for now until solved
+        assert not hasattr(data, "values")
+        assert not hasattr(data, "_values")
+
+    def test_is_numeric_honored(self, data):
+        result = pd.Series(data)
+        if hasattr(result._mgr, "blocks"):
+            assert result._mgr.blocks[0].is_numeric is data.dtype._is_numeric
+
+    def test_isna_extension_array(self, data_missing):
+        # If your `isna` returns an ExtensionArray, you must also implement
+        # _reduce. At the *very* least, you must implement any and all
+        na = data_missing.isna()
+        if is_extension_array_dtype(na):
+            assert na._reduce("any")
+            assert na.any()
+
+            assert not na._reduce("all")
+            assert not na.all()
+
+            assert na.dtype._is_boolean
+
+    def test_copy(self, data):
+        # GH#27083 removing deep keyword from EA.copy
+        assert data[0] != data[1]
+        result = data.copy()
+
+        data[1] = data[0]
+        assert result[1] != result[0]
+
+    def test_view(self, data):
+        # view with no dtype should return a shallow copy, *not* the same
+        #  object
+        assert data[1] != data[0]
+
+        result = data.view()
+        assert result is not data
+        assert type(result) == type(data)
+
+        result[1] = result[0]
+        assert data[1] == data[0]
+
+        # check specifically that the `dtype` kwarg is accepted
+        data.view(dtype=None)
+
+    def test_tolist(self, data):
+        result = data.tolist()
+        expected = list(data)
+        assert isinstance(result, list)
+        assert result == expected

dist/client/pandas/tests/extension/base/io.py

@@ -0,0 +1,19 @@
+from io import StringIO
+
+import numpy as np
+import pytest
+
+import pandas as pd
+from pandas.tests.extension.base.base import BaseExtensionTests
+
+
+class BaseParsingTests(BaseExtensionTests):
+    @pytest.mark.parametrize("engine", ["c", "python"])
+    def test_EA_types(self, engine, data):
+        df = pd.DataFrame({"with_dtype": pd.Series(data, dtype=str(data.dtype))})
+        csv_output = df.to_csv(index=False, na_rep=np.nan)
+        result = pd.read_csv(
+            StringIO(csv_output), dtype={"with_dtype": str(data.dtype)}, engine=engine
+        )
+        expected = df
+        self.assert_frame_equal(result, expected)

dist/client/pandas/tests/extension/base/methods.py

@@ -0,0 +1,583 @@
+import inspect
+import operator
+
+import numpy as np
+import pytest
+
+from pandas.core.dtypes.common import is_bool_dtype
+
+import pandas as pd
+import pandas._testing as tm
+from pandas.core.sorting import nargsort
+from pandas.tests.extension.base.base import BaseExtensionTests
+
+
+class BaseMethodsTests(BaseExtensionTests):
+    """Various Series and DataFrame methods."""
+
+    def test_value_counts_default_dropna(self, data):
+        # make sure we have consistent default dropna kwarg
+        if not hasattr(data, "value_counts"):
+            pytest.skip("value_counts is not implemented")
+        sig = inspect.signature(data.value_counts)
+        kwarg = sig.parameters["dropna"]
+        assert kwarg.default is True
+
+    @pytest.mark.parametrize("dropna", [True, False])
+    def test_value_counts(self, all_data, dropna):
+        all_data = all_data[:10]
+        if dropna:
+            other = np.array(all_data[~all_data.isna()])
+        else:
+            other = all_data
+
+        result = pd.Series(all_data).value_counts(dropna=dropna).sort_index()
+        expected = pd.Series(other).value_counts(dropna=dropna).sort_index()
+
+        self.assert_series_equal(result, expected)
+
+    def test_value_counts_with_normalize(self, data):
+        # GH 33172
+        data = data[:10].unique()
+        values = np.array(data[~data.isna()])
+        ser = pd.Series(data, dtype=data.dtype)
+
+        result = ser.value_counts(normalize=True).sort_index()
+
+        if not isinstance(data, pd.Categorical):
+            expected = pd.Series([1 / len(values)] * len(values), index=result.index)
+        else:
+            expected = pd.Series(0.0, index=result.index)
+            expected[result > 0] = 1 / len(values)
+
+        self.assert_series_equal(result, expected)
+
+    def test_count(self, data_missing):
+        df = pd.DataFrame({"A": data_missing})
+        result = df.count(axis="columns")
+        expected = pd.Series([0, 1])
+        self.assert_series_equal(result, expected)
+
+    def test_series_count(self, data_missing):
+        # GH#26835
+        ser = pd.Series(data_missing)
+        result = ser.count()
+        expected = 1
+        assert result == expected
+
+    def test_apply_simple_series(self, data):
+        result = pd.Series(data).apply(id)
+        assert isinstance(result, pd.Series)
+
+    def test_argsort(self, data_for_sorting):
+        result = pd.Series(data_for_sorting).argsort()
+        # argsort result gets passed to take, so should be np.intp
+        expected = pd.Series(np.array([2, 0, 1], dtype=np.intp))
+        self.assert_series_equal(result, expected)
+
+    def test_argsort_missing_array(self, data_missing_for_sorting):
+        result = data_missing_for_sorting.argsort()
+        # argsort result gets passed to take, so should be np.intp
+        expected = np.array([2, 0, 1], dtype=np.intp)
+        tm.assert_numpy_array_equal(result, expected)
+
+    def test_argsort_missing(self, data_missing_for_sorting):
+        result = pd.Series(data_missing_for_sorting).argsort()
+        expected = pd.Series(np.array([1, -1, 0], dtype=np.intp))
+        self.assert_series_equal(result, expected)
+
+    def test_argmin_argmax(self, data_for_sorting, data_missing_for_sorting, na_value):
+        # GH 24382
+
+        # data_for_sorting -> [B, C, A] with A < B < C
+        assert data_for_sorting.argmax() == 1
+        assert data_for_sorting.argmin() == 2
+
+        # with repeated values -> first occurrence
+        data = data_for_sorting.take([2, 0, 0, 1, 1, 2])
+        assert data.argmax() == 3
+        assert data.argmin() == 0
+
+        # with missing values
+        # data_missing_for_sorting -> [B, NA, A] with A < B and NA missing.
+        assert data_missing_for_sorting.argmax() == 0
+        assert data_missing_for_sorting.argmin() == 2
+
+    @pytest.mark.parametrize("method", ["argmax", "argmin"])
+    def test_argmin_argmax_empty_array(self, method, data):
+        # GH 24382
+        err_msg = "attempt to get"
+        with pytest.raises(ValueError, match=err_msg):
+            getattr(data[:0], method)()
+
+    @pytest.mark.parametrize("method", ["argmax", "argmin"])
+    def test_argmin_argmax_all_na(self, method, data, na_value):
+        # all missing with skipna=True is the same as empty
+        err_msg = "attempt to get"
+        data_na = type(data)._from_sequence([na_value, na_value], dtype=data.dtype)
+        with pytest.raises(ValueError, match=err_msg):
+            getattr(data_na, method)()
+
+    @pytest.mark.parametrize(
+        "op_name, skipna, expected",
+        [
+            ("idxmax", True, 0),
+            ("idxmin", True, 2),
+            ("argmax", True, 0),
+            ("argmin", True, 2),
+            ("idxmax", False, np.nan),
+            ("idxmin", False, np.nan),
+            ("argmax", False, -1),
+            ("argmin", False, -1),
+        ],
+    )
+    def test_argreduce_series(
+        self, data_missing_for_sorting, op_name, skipna, expected
+    ):
+        # data_missing_for_sorting -> [B, NA, A] with A < B and NA missing.
+        ser = pd.Series(data_missing_for_sorting)
+        result = getattr(ser, op_name)(skipna=skipna)
+        tm.assert_almost_equal(result, expected)
+
+    def test_argmax_argmin_no_skipna_notimplemented(self, data_missing_for_sorting):
+        # GH#38733
+        data = data_missing_for_sorting
+
+        with pytest.raises(NotImplementedError, match=""):
+            data.argmin(skipna=False)
+
+        with pytest.raises(NotImplementedError, match=""):
+            data.argmax(skipna=False)
+
+    @pytest.mark.parametrize(
+        "na_position, expected",
+        [
+            ("last", np.array([2, 0, 1], dtype=np.dtype("intp"))),
+            ("first", np.array([1, 2, 0], dtype=np.dtype("intp"))),
+        ],
+    )
+    def test_nargsort(self, data_missing_for_sorting, na_position, expected):
+        # GH 25439
+        result = nargsort(data_missing_for_sorting, na_position=na_position)
+        tm.assert_numpy_array_equal(result, expected)
+
+    @pytest.mark.parametrize("ascending", [True, False])
+    def test_sort_values(self, data_for_sorting, ascending, sort_by_key):
+        ser = pd.Series(data_for_sorting)
+        result = ser.sort_values(ascending=ascending, key=sort_by_key)
+        expected = ser.iloc[[2, 0, 1]]
+        if not ascending:
+            # GH 35922. Expect stable sort
+            if ser.nunique() == 2:
+                expected = ser.iloc[[0, 1, 2]]
+            else:
+                expected = ser.iloc[[1, 0, 2]]
+
+        self.assert_series_equal(result, expected)
+
+    @pytest.mark.parametrize("ascending", [True, False])
+    def test_sort_values_missing(
+        self, data_missing_for_sorting, ascending, sort_by_key
+    ):
+        ser = pd.Series(data_missing_for_sorting)
+        result = ser.sort_values(ascending=ascending, key=sort_by_key)
+        if ascending:
+            expected = ser.iloc[[2, 0, 1]]
+        else:
+            expected = ser.iloc[[0, 2, 1]]
+        self.assert_series_equal(result, expected)
+
+    @pytest.mark.parametrize("ascending", [True, False])
+    def test_sort_values_frame(self, data_for_sorting, ascending):
+        df = pd.DataFrame({"A": [1, 2, 1], "B": data_for_sorting})
+        result = df.sort_values(["A", "B"])
+        expected = pd.DataFrame(
+            {"A": [1, 1, 2], "B": data_for_sorting.take([2, 0, 1])}, index=[2, 0, 1]
+        )
+        self.assert_frame_equal(result, expected)
+
+    @pytest.mark.parametrize("box", [pd.Series, lambda x: x])
+    @pytest.mark.parametrize("method", [lambda x: x.unique(), pd.unique])
+    def test_unique(self, data, box, method):
+        duplicated = box(data._from_sequence([data[0], data[0]]))
+
+        result = method(duplicated)
+
+        assert len(result) == 1
+        assert isinstance(result, type(data))
+        assert result[0] == duplicated[0]
+
+    @pytest.mark.parametrize("na_sentinel", [-1, -2])
+    def test_factorize(self, data_for_grouping, na_sentinel):
+        codes, uniques = pd.factorize(data_for_grouping, na_sentinel=na_sentinel)
+        expected_codes = np.array(
+            [0, 0, na_sentinel, na_sentinel, 1, 1, 0, 2], dtype=np.intp
+        )
+        expected_uniques = data_for_grouping.take([0, 4, 7])
+
+        tm.assert_numpy_array_equal(codes, expected_codes)
+        self.assert_extension_array_equal(uniques, expected_uniques)
+
+    @pytest.mark.parametrize("na_sentinel", [-1, -2])
+    def test_factorize_equivalence(self, data_for_grouping, na_sentinel):
+        codes_1, uniques_1 = pd.factorize(data_for_grouping, na_sentinel=na_sentinel)
+        codes_2, uniques_2 = data_for_grouping.factorize(na_sentinel=na_sentinel)
+
+        tm.assert_numpy_array_equal(codes_1, codes_2)
+        self.assert_extension_array_equal(uniques_1, uniques_2)
+        assert len(uniques_1) == len(pd.unique(uniques_1))
+        assert uniques_1.dtype == data_for_grouping.dtype
+
+    def test_factorize_empty(self, data):
+        codes, uniques = pd.factorize(data[:0])
+        expected_codes = np.array([], dtype=np.intp)
+        expected_uniques = type(data)._from_sequence([], dtype=data[:0].dtype)
+
+        tm.assert_numpy_array_equal(codes, expected_codes)
+        self.assert_extension_array_equal(uniques, expected_uniques)
+
+    def test_fillna_copy_frame(self, data_missing):
+        arr = data_missing.take([1, 1])
+        df = pd.DataFrame({"A": arr})
+
+        filled_val = df.iloc[0, 0]
+        result = df.fillna(filled_val)
+
+        assert df.A.values is not result.A.values
+
+    def test_fillna_copy_series(self, data_missing):
+        arr = data_missing.take([1, 1])
+        ser = pd.Series(arr)
+
+        filled_val = ser[0]
+        result = ser.fillna(filled_val)
+
+        assert ser._values is not result._values
+        assert ser._values is arr
+
+    def test_fillna_length_mismatch(self, data_missing):
+        msg = "Length of 'value' does not match."
+        with pytest.raises(ValueError, match=msg):
+            data_missing.fillna(data_missing.take([1]))
+
+    def test_combine_le(self, data_repeated):
+        # GH 20825
+        # Test that combine works when doing a <= (le) comparison
+        orig_data1, orig_data2 = data_repeated(2)
+        s1 = pd.Series(orig_data1)
+        s2 = pd.Series(orig_data2)
+        result = s1.combine(s2, lambda x1, x2: x1 <= x2)
+        expected = pd.Series(
+            [a <= b for (a, b) in zip(list(orig_data1), list(orig_data2))]
+        )
+        self.assert_series_equal(result, expected)
+
+        val = s1.iloc[0]
+        result = s1.combine(val, lambda x1, x2: x1 <= x2)
+        expected = pd.Series([a <= val for a in list(orig_data1)])
+        self.assert_series_equal(result, expected)
+
+    def test_combine_add(self, data_repeated):
+        # GH 20825
+        orig_data1, orig_data2 = data_repeated(2)
+        s1 = pd.Series(orig_data1)
+        s2 = pd.Series(orig_data2)
+        result = s1.combine(s2, lambda x1, x2: x1 + x2)
+        with np.errstate(over="ignore"):
+            expected = pd.Series(
+                orig_data1._from_sequence(
+                    [a + b for (a, b) in zip(list(orig_data1), list(orig_data2))]
+                )
+            )
+        self.assert_series_equal(result, expected)
+
+        val = s1.iloc[0]
+        result = s1.combine(val, lambda x1, x2: x1 + x2)
+        expected = pd.Series(
+            orig_data1._from_sequence([a + val for a in list(orig_data1)])
+        )
+        self.assert_series_equal(result, expected)
+
+    def test_combine_first(self, data):
+        # https://github.com/pandas-dev/pandas/issues/24147
+        a = pd.Series(data[:3])
+        b = pd.Series(data[2:5], index=[2, 3, 4])
+        result = a.combine_first(b)
+        expected = pd.Series(data[:5])
+        self.assert_series_equal(result, expected)
+
+    @pytest.mark.parametrize("frame", [True, False])
+    @pytest.mark.parametrize(
+        "periods, indices",
+        [(-2, [2, 3, 4, -1, -1]), (0, [0, 1, 2, 3, 4]), (2, [-1, -1, 0, 1, 2])],
+    )
+    def test_container_shift(self, data, frame, periods, indices):
+        # https://github.com/pandas-dev/pandas/issues/22386
+        subset = data[:5]
+        data = pd.Series(subset, name="A")
+        expected = pd.Series(subset.take(indices, allow_fill=True), name="A")
+
+        if frame:
+            result = data.to_frame(name="A").assign(B=1).shift(periods)
+            expected = pd.concat(
+                [expected, pd.Series([1] * 5, name="B").shift(periods)], axis=1
+            )
+            compare = self.assert_frame_equal
+        else:
+            result = data.shift(periods)
+            compare = self.assert_series_equal
+
+        compare(result, expected)
+
+    def test_shift_0_periods(self, data):
+        # GH#33856 shifting with periods=0 should return a copy, not same obj
+        result = data.shift(0)
+        assert data[0] != data[1]  # otherwise below is invalid
+        data[0] = data[1]
+        assert result[0] != result[1]  # i.e. not the same object/view
+
+    @pytest.mark.parametrize("periods", [1, -2])
+    def test_diff(self, data, periods):
+        data = data[:5]
+        if is_bool_dtype(data.dtype):
+            op = operator.xor
+        else:
+            op = operator.sub
+        try:
+            # does this array implement ops?
+            op(data, data)
+        except Exception:
+            pytest.skip(f"{type(data)} does not support diff")
+        s = pd.Series(data)
+        result = s.diff(periods)
+        expected = pd.Series(op(data, data.shift(periods)))
+        self.assert_series_equal(result, expected)
+
+        df = pd.DataFrame({"A": data, "B": [1.0] * 5})
+        result = df.diff(periods)
+        if periods == 1:
+            b = [np.nan, 0, 0, 0, 0]
+        else:
+            b = [0, 0, 0, np.nan, np.nan]
+        expected = pd.DataFrame({"A": expected, "B": b})
+        self.assert_frame_equal(result, expected)
+
+    @pytest.mark.parametrize(
+        "periods, indices",
+        [[-4, [-1, -1]], [-1, [1, -1]], [0, [0, 1]], [1, [-1, 0]], [4, [-1, -1]]],
+    )
+    def test_shift_non_empty_array(self, data, periods, indices):
+        # https://github.com/pandas-dev/pandas/issues/23911
+        subset = data[:2]
+        result = subset.shift(periods)
+        expected = subset.take(indices, allow_fill=True)
+        self.assert_extension_array_equal(result, expected)
+
+    @pytest.mark.parametrize("periods", [-4, -1, 0, 1, 4])
+    def test_shift_empty_array(self, data, periods):
+        # https://github.com/pandas-dev/pandas/issues/23911
+        empty = data[:0]
+        result = empty.shift(periods)
+        expected = empty
+        self.assert_extension_array_equal(result, expected)
+
+    def test_shift_zero_copies(self, data):
+        result = data.shift(0)
+        assert result is not data
+
+        result = data[:0].shift(2)
+        assert result is not data
+
+    def test_shift_fill_value(self, data):
+        arr = data[:4]
+        fill_value = data[0]
+        result = arr.shift(1, fill_value=fill_value)
+        expected = data.take([0, 0, 1, 2])
+        self.assert_extension_array_equal(result, expected)
+
+        result = arr.shift(-2, fill_value=fill_value)
+        expected = data.take([2, 3, 0, 0])
+        self.assert_extension_array_equal(result, expected)
+
+    def test_not_hashable(self, data):
+        # We are in general mutable, so not hashable
+        with pytest.raises(TypeError, match="unhashable type"):
+            hash(data)
+
+    def test_hash_pandas_object_works(self, data, as_frame):
+        # https://github.com/pandas-dev/pandas/issues/23066
+        data = pd.Series(data)
+        if as_frame:
+            data = data.to_frame()
+        a = pd.util.hash_pandas_object(data)
+        b = pd.util.hash_pandas_object(data)
+        self.assert_equal(a, b)
+
+    def test_searchsorted(self, data_for_sorting, as_series):
+        b, c, a = data_for_sorting
+        arr = data_for_sorting.take([2, 0, 1])  # to get [a, b, c]
+
+        if as_series:
+            arr = pd.Series(arr)
+        assert arr.searchsorted(a) == 0
+        assert arr.searchsorted(a, side="right") == 1
+
+        assert arr.searchsorted(b) == 1
+        assert arr.searchsorted(b, side="right") == 2
+
+        assert arr.searchsorted(c) == 2
+        assert arr.searchsorted(c, side="right") == 3
+
+        result = arr.searchsorted(arr.take([0, 2]))
+        expected = np.array([0, 2], dtype=np.intp)
+
+        tm.assert_numpy_array_equal(result, expected)
+
+        # sorter
+        sorter = np.array([1, 2, 0])
+        assert data_for_sorting.searchsorted(a, sorter=sorter) == 0
+
+    def test_where_series(self, data, na_value, as_frame):
+        assert data[0] != data[1]
+        cls = type(data)
+        a, b = data[:2]
+
+        ser = pd.Series(cls._from_sequence([a, a, b, b], dtype=data.dtype))
+        cond = np.array([True, True, False, False])
+
+        if as_frame:
+            ser = ser.to_frame(name="a")
+            cond = cond.reshape(-1, 1)
+
+        result = ser.where(cond)
+        expected = pd.Series(
+            cls._from_sequence([a, a, na_value, na_value], dtype=data.dtype)
+        )
+
+        if as_frame:
+            expected = expected.to_frame(name="a")
+        self.assert_equal(result, expected)
+
+        # array other
+        cond = np.array([True, False, True, True])
+        other = cls._from_sequence([a, b, a, b], dtype=data.dtype)
+        if as_frame:
+            other = pd.DataFrame({"a": other})
+            cond = pd.DataFrame({"a": cond})
+        result = ser.where(cond, other)
+        expected = pd.Series(cls._from_sequence([a, b, b, b], dtype=data.dtype))
+        if as_frame:
+            expected = expected.to_frame(name="a")
+        self.assert_equal(result, expected)
+
+    @pytest.mark.parametrize("repeats", [0, 1, 2, [1, 2, 3]])
+    def test_repeat(self, data, repeats, as_series, use_numpy):
+        arr = type(data)._from_sequence(data[:3], dtype=data.dtype)
+        if as_series:
+            arr = pd.Series(arr)
+
+        result = np.repeat(arr, repeats) if use_numpy else arr.repeat(repeats)
+
+        repeats = [repeats] * 3 if isinstance(repeats, int) else repeats
+        expected = [x for x, n in zip(arr, repeats) for _ in range(n)]
+        expected = type(data)._from_sequence(expected, dtype=data.dtype)
+        if as_series:
+            expected = pd.Series(expected, index=arr.index.repeat(repeats))
+
+        self.assert_equal(result, expected)
+
+    @pytest.mark.parametrize(
+        "repeats, kwargs, error, msg",
+        [
+            (2, {"axis": 1}, ValueError, "axis"),
+            (-1, {}, ValueError, "negative"),
+            ([1, 2], {}, ValueError, "shape"),
+            (2, {"foo": "bar"}, TypeError, "'foo'"),
+        ],
+    )
+    def test_repeat_raises(self, data, repeats, kwargs, error, msg, use_numpy):
+        with pytest.raises(error, match=msg):
+            if use_numpy:
+                np.repeat(data, repeats, **kwargs)
+            else:
+                data.repeat(repeats, **kwargs)
+
+    def test_delete(self, data):
+        result = data.delete(0)
+        expected = data[1:]
+        self.assert_extension_array_equal(result, expected)
+
+        result = data.delete([1, 3])
+        expected = data._concat_same_type([data[[0]], data[[2]], data[4:]])
+        self.assert_extension_array_equal(result, expected)
+
+    def test_insert(self, data):
+        # insert at the beginning
+        result = data[1:].insert(0, data[0])
+        self.assert_extension_array_equal(result, data)
+
+        result = data[1:].insert(-len(data[1:]), data[0])
+        self.assert_extension_array_equal(result, data)
+
+        # insert at the middle
+        result = data[:-1].insert(4, data[-1])
+
+        taker = np.arange(len(data))
+        taker[5:] = taker[4:-1]
+        taker[4] = len(data) - 1
+        expected = data.take(taker)
+        self.assert_extension_array_equal(result, expected)
+
+    def test_insert_invalid(self, data, invalid_scalar):
+        item = invalid_scalar
+
+        with pytest.raises((TypeError, ValueError)):
+            data.insert(0, item)
+
+        with pytest.raises((TypeError, ValueError)):
+            data.insert(4, item)
+
+        with pytest.raises((TypeError, ValueError)):
+            data.insert(len(data) - 1, item)
+
+    def test_insert_invalid_loc(self, data):
+        ub = len(data)
+
+        with pytest.raises(IndexError):
+            data.insert(ub + 1, data[0])
+
+        with pytest.raises(IndexError):
+            data.insert(-ub - 1, data[0])
+
+        with pytest.raises(TypeError):
+            # we expect TypeError here instead of IndexError to match np.insert
+            data.insert(1.5, data[0])
+
+    @pytest.mark.parametrize("box", [pd.array, pd.Series, pd.DataFrame])
+    def test_equals(self, data, na_value, as_series, box):
+        data2 = type(data)._from_sequence([data[0]] * len(data), dtype=data.dtype)
+        data_na = type(data)._from_sequence([na_value] * len(data), dtype=data.dtype)
+
+        data = tm.box_expected(data, box, transpose=False)
+        data2 = tm.box_expected(data2, box, transpose=False)
+        data_na = tm.box_expected(data_na, box, transpose=False)
+
+        # we are asserting with `is True/False` explicitly, to test that the
+        # result is an actual Python bool, and not something "truthy"
+
+        assert data.equals(data) is True
+        assert data.equals(data.copy()) is True
+
+        # unequal other data
+        assert data.equals(data2) is False
+        assert data.equals(data_na) is False
+
+        # different length
+        assert data[:2].equals(data[:3]) is False
+
+        # empty are equal
+        assert data[:0].equals(data[:0]) is True
+
+        # other types
+        assert data.equals(None) is False
+        assert data[[0]].equals(data[0]) is False

dist/client/pandas/tests/extension/base/missing.py

@@ -0,0 +1,160 @@
+import numpy as np
+import pytest
+
+import pandas as pd
+import pandas._testing as tm
+from pandas.api.types import is_sparse
+from pandas.tests.extension.base.base import BaseExtensionTests
+
+
+class BaseMissingTests(BaseExtensionTests):
+    def test_isna(self, data_missing):
+        expected = np.array([True, False])
+
+        result = pd.isna(data_missing)
+        tm.assert_numpy_array_equal(result, expected)
+
+        result = pd.Series(data_missing).isna()
+        expected = pd.Series(expected)
+        self.assert_series_equal(result, expected)
+
+        # GH 21189
+        result = pd.Series(data_missing).drop([0, 1]).isna()
+        expected = pd.Series([], dtype=bool)
+        self.assert_series_equal(result, expected)
+
+    @pytest.mark.parametrize("na_func", ["isna", "notna"])
+    def test_isna_returns_copy(self, data_missing, na_func):
+        result = pd.Series(data_missing)
+        expected = result.copy()
+        mask = getattr(result, na_func)()
+        if is_sparse(mask):
+            mask = np.array(mask)
+
+        mask[:] = True
+        self.assert_series_equal(result, expected)
+
+    def test_dropna_array(self, data_missing):
+        result = data_missing.dropna()
+        expected = data_missing[[1]]
+        self.assert_extension_array_equal(result, expected)
+
+    def test_dropna_series(self, data_missing):
+        ser = pd.Series(data_missing)
+        result = ser.dropna()
+        expected = ser.iloc[[1]]
+        self.assert_series_equal(result, expected)
+
+    def test_dropna_frame(self, data_missing):
+        df = pd.DataFrame({"A": data_missing})
+
+        # defaults
+        result = df.dropna()
+        expected = df.iloc[[1]]
+        self.assert_frame_equal(result, expected)
+
+        # axis = 1
+        result = df.dropna(axis="columns")
+        expected = pd.DataFrame(index=[0, 1])
+        self.assert_frame_equal(result, expected)
+
+        # multiple
+        df = pd.DataFrame({"A": data_missing, "B": [1, np.nan]})
+        result = df.dropna()
+        expected = df.iloc[:0]
+        self.assert_frame_equal(result, expected)
+
+    def test_fillna_scalar(self, data_missing):
+        valid = data_missing[1]
+        result = data_missing.fillna(valid)
+        expected = data_missing.fillna(valid)
+        self.assert_extension_array_equal(result, expected)
+
+    def test_fillna_limit_pad(self, data_missing):
+        arr = data_missing.take([1, 0, 0, 0, 1])
+        result = pd.Series(arr).fillna(method="ffill", limit=2)
+        expected = pd.Series(data_missing.take([1, 1, 1, 0, 1]))
+        self.assert_series_equal(result, expected)
+
+    def test_fillna_limit_backfill(self, data_missing):
+        arr = data_missing.take([1, 0, 0, 0, 1])
+        result = pd.Series(arr).fillna(method="backfill", limit=2)
+        expected = pd.Series(data_missing.take([1, 0, 1, 1, 1]))
+        self.assert_series_equal(result, expected)
+
+    def test_fillna_no_op_returns_copy(self, data):
+        data = data[~data.isna()]
+
+        valid = data[0]
+        result = data.fillna(valid)
+        assert result is not data
+        self.assert_extension_array_equal(result, data)
+
+        result = data.fillna(method="backfill")
+        assert result is not data
+        self.assert_extension_array_equal(result, data)
+
+    def test_fillna_series(self, data_missing):
+        fill_value = data_missing[1]
+        ser = pd.Series(data_missing)
+
+        result = ser.fillna(fill_value)
+        expected = pd.Series(
+            data_missing._from_sequence(
+                [fill_value, fill_value], dtype=data_missing.dtype
+            )
+        )
+        self.assert_series_equal(result, expected)
+
+        # Fill with a series
+        result = ser.fillna(expected)
+        self.assert_series_equal(result, expected)
+
+        # Fill with a series not affecting the missing values
+        result = ser.fillna(ser)
+        self.assert_series_equal(result, ser)
+
+    def test_fillna_series_method(self, data_missing, fillna_method):
+        fill_value = data_missing[1]
+
+        if fillna_method == "ffill":
+            data_missing = data_missing[::-1]
+
+        result = pd.Series(data_missing).fillna(method=fillna_method)
+        expected = pd.Series(
+            data_missing._from_sequence(
+                [fill_value, fill_value], dtype=data_missing.dtype
+            )
+        )
+
+        self.assert_series_equal(result, expected)
+
+    def test_fillna_frame(self, data_missing):
+        fill_value = data_missing[1]
+
+        result = pd.DataFrame({"A": data_missing, "B": [1, 2]}).fillna(fill_value)
+
+        expected = pd.DataFrame(
+            {
+                "A": data_missing._from_sequence(
+                    [fill_value, fill_value], dtype=data_missing.dtype
+                ),
+                "B": [1, 2],
+            }
+        )
+
+        self.assert_frame_equal(result, expected)
+
+    def test_fillna_fill_other(self, data):
+        result = pd.DataFrame({"A": data, "B": [np.nan] * len(data)}).fillna({"B": 0.0})
+
+        expected = pd.DataFrame({"A": data, "B": [0.0] * len(result)})
+
+        self.assert_frame_equal(result, expected)
+
+    def test_use_inf_as_na_no_effect(self, data_missing):
+        ser = pd.Series(data_missing)
+        expected = ser.isna()
+        with pd.option_context("mode.use_inf_as_na", True):
+            result = ser.isna()
+        self.assert_series_equal(result, expected)

dist/client/pandas/tests/extension/base/ops.py

@@ -0,0 +1,217 @@
+from __future__ import annotations
+
+import numpy as np
+import pytest
+
+import pandas as pd
+import pandas._testing as tm
+from pandas.core import ops
+from pandas.tests.extension.base.base import BaseExtensionTests
+
+
+class BaseOpsUtil(BaseExtensionTests):
+    def get_op_from_name(self, op_name: str):
+        return tm.get_op_from_name(op_name)
+
+    def check_opname(self, ser: pd.Series, op_name: str, other, exc=Exception):
+        op = self.get_op_from_name(op_name)
+
+        self._check_op(ser, op, other, op_name, exc)
+
+    def _combine(self, obj, other, op):
+        if isinstance(obj, pd.DataFrame):
+            if len(obj.columns) != 1:
+                raise NotImplementedError
+            expected = obj.iloc[:, 0].combine(other, op).to_frame()
+        else:
+            expected = obj.combine(other, op)
+        return expected
+
+    def _check_op(
+        self, ser: pd.Series, op, other, op_name: str, exc=NotImplementedError
+    ):
+        if exc is None:
+            result = op(ser, other)
+            expected = self._combine(ser, other, op)
+            assert isinstance(result, type(ser))
+            self.assert_equal(result, expected)
+        else:
+            with pytest.raises(exc):
+                op(ser, other)
+
+    def _check_divmod_op(self, ser: pd.Series, op, other, exc=Exception):
+        # divmod has multiple return values, so check separately
+        if exc is None:
+            result_div, result_mod = op(ser, other)
+            if op is divmod:
+                expected_div, expected_mod = ser // other, ser % other
+            else:
+                expected_div, expected_mod = other // ser, other % ser
+            self.assert_series_equal(result_div, expected_div)
+            self.assert_series_equal(result_mod, expected_mod)
+        else:
+            with pytest.raises(exc):
+                divmod(ser, other)
+
+
+class BaseArithmeticOpsTests(BaseOpsUtil):
+    """
+    Various Series and DataFrame arithmetic ops methods.
+
+    Subclasses supporting various ops should set the class variables
+    to indicate that they support ops of that kind
+
+    * series_scalar_exc = TypeError
+    * frame_scalar_exc = TypeError
+    * series_array_exc = TypeError
+    * divmod_exc = TypeError
+    """
+
+    series_scalar_exc: type[TypeError] | None = TypeError
+    frame_scalar_exc: type[TypeError] | None = TypeError
+    series_array_exc: type[TypeError] | None = TypeError
+    divmod_exc: type[TypeError] | None = TypeError
+
+    def test_arith_series_with_scalar(self, data, all_arithmetic_operators):
+        # series & scalar
+        op_name = all_arithmetic_operators
+        ser = pd.Series(data)
+        self.check_opname(ser, op_name, ser.iloc[0], exc=self.series_scalar_exc)
+
+    def test_arith_frame_with_scalar(self, data, all_arithmetic_operators):
+        # frame & scalar
+        op_name = all_arithmetic_operators
+        df = pd.DataFrame({"A": data})
+        self.check_opname(df, op_name, data[0], exc=self.frame_scalar_exc)
+
+    def test_arith_series_with_array(self, data, all_arithmetic_operators):
+        # ndarray & other series
+        op_name = all_arithmetic_operators
+        ser = pd.Series(data)
+        self.check_opname(
+            ser, op_name, pd.Series([ser.iloc[0]] * len(ser)), exc=self.series_array_exc
+        )
+
+    def test_divmod(self, data):
+        ser = pd.Series(data)
+        self._check_divmod_op(ser, divmod, 1, exc=self.divmod_exc)
+        self._check_divmod_op(1, ops.rdivmod, ser, exc=self.divmod_exc)
+
+    def test_divmod_series_array(self, data, data_for_twos):
+        ser = pd.Series(data)
+        self._check_divmod_op(ser, divmod, data)
+
+        other = data_for_twos
+        self._check_divmod_op(other, ops.rdivmod, ser)
+
+        other = pd.Series(other)
+        self._check_divmod_op(other, ops.rdivmod, ser)
+
+    def test_add_series_with_extension_array(self, data):
+        ser = pd.Series(data)
+        result = ser + data
+        expected = pd.Series(data + data)
+        self.assert_series_equal(result, expected)
+
+    @pytest.mark.parametrize("box", [pd.Series, pd.DataFrame])
+    def test_direct_arith_with_ndframe_returns_not_implemented(
+        self, request, data, box
+    ):
+        # EAs should return NotImplemented for ops with Series/DataFrame
+        # Pandas takes care of unboxing the series and calling the EA's op.
+        other = pd.Series(data)
+        if box is pd.DataFrame:
+            other = other.to_frame()
+        if not hasattr(data, "__add__"):
+            request.node.add_marker(
+                pytest.mark.xfail(
+                    reason=f"{type(data).__name__} does not implement add"
+                )
+            )
+        result = data.__add__(other)
+        assert result is NotImplemented
+
+
+class BaseComparisonOpsTests(BaseOpsUtil):
+    """Various Series and DataFrame comparison ops methods."""
+
+    def _compare_other(self, ser: pd.Series, data, op, other):
+
+        if op.__name__ in ["eq", "ne"]:
+            # comparison should match point-wise comparisons
+            result = op(ser, other)
+            expected = ser.combine(other, op)
+            self.assert_series_equal(result, expected)
+
+        else:
+            exc = None
+            try:
+                result = op(ser, other)
+            except Exception as err:
+                exc = err
+
+            if exc is None:
+                # Didn't error, then should match pointwise behavior
+                expected = ser.combine(other, op)
+                self.assert_series_equal(result, expected)
+            else:
+                with pytest.raises(type(exc)):
+                    ser.combine(other, op)
+
+    def test_compare_scalar(self, data, comparison_op):
+        ser = pd.Series(data)
+        self._compare_other(ser, data, comparison_op, 0)
+
+    def test_compare_array(self, data, comparison_op):
+        ser = pd.Series(data)
+        other = pd.Series([data[0]] * len(data))
+        self._compare_other(ser, data, comparison_op, other)
+
+    @pytest.mark.parametrize("box", [pd.Series, pd.DataFrame])
+    def test_direct_arith_with_ndframe_returns_not_implemented(self, data, box):
+        # EAs should return NotImplemented for ops with Series/DataFrame
+        # Pandas takes care of unboxing the series and calling the EA's op.
+        other = pd.Series(data)
+        if box is pd.DataFrame:
+            other = other.to_frame()
+
+        if hasattr(data, "__eq__"):
+            result = data.__eq__(other)
+            assert result is NotImplemented
+        else:
+            raise pytest.skip(f"{type(data).__name__} does not implement __eq__")
+
+        if hasattr(data, "__ne__"):
+            result = data.__ne__(other)
+            assert result is NotImplemented
+        else:
+            raise pytest.skip(f"{type(data).__name__} does not implement __ne__")
+
+
+class BaseUnaryOpsTests(BaseOpsUtil):
+    def test_invert(self, data):
+        ser = pd.Series(data, name="name")
+        result = ~ser
+        expected = pd.Series(~data, name="name")
+        self.assert_series_equal(result, expected)
+
+    @pytest.mark.parametrize("ufunc", [np.positive, np.negative, np.abs])
+    def test_unary_ufunc_dunder_equivalence(self, data, ufunc):
+        # the dunder __pos__ works if and only if np.positive works,
+        #  same for __neg__/np.negative and __abs__/np.abs
+        attr = {np.positive: "__pos__", np.negative: "__neg__", np.abs: "__abs__"}[
+            ufunc
+        ]
+
+        exc = None
+        try:
+            result = getattr(data, attr)()
+        except Exception as err:
+            exc = err
+
+            # if __pos__ raised, then so should the ufunc
+            with pytest.raises((type(exc), TypeError)):
+                ufunc(data)
+        else:
+            alt = ufunc(data)
+            self.assert_extension_array_equal(result, alt)

dist/client/pandas/tests/extension/base/printing.py

@@ -0,0 +1,42 @@
+import io
+
+import pytest
+
+import pandas as pd
+from pandas.tests.extension.base.base import BaseExtensionTests
+
+
+class BasePrintingTests(BaseExtensionTests):
+    """Tests checking the formatting of your EA when printed."""
+
+    @pytest.mark.parametrize("size", ["big", "small"])
+    def test_array_repr(self, data, size):
+        if size == "small":
+            data = data[:5]
+        else:
+            data = type(data)._concat_same_type([data] * 5)
+
+        result = repr(data)
+        assert type(data).__name__ in result
+        assert f"Length: {len(data)}" in result
+        assert str(data.dtype) in result
+        if size == "big":
+            assert "..." in result
+
+    def test_array_repr_unicode(self, data):
+        result = str(data)
+        assert isinstance(result, str)
+
+    def test_series_repr(self, data):
+        ser = pd.Series(data)
+        assert data.dtype.name in repr(ser)
+
+    def test_dataframe_repr(self, data):
+        df = pd.DataFrame({"A": data})
+        repr(df)
+
+    def test_dtype_name_in_info(self, data):
+        buf = io.StringIO()
+        pd.DataFrame({"A": data}).info(buf=buf)
+        result = buf.getvalue()
+        assert data.dtype.name in result

dist/client/pandas/tests/extension/base/reduce.py

@@ -0,0 +1,69 @@
+import warnings
+
+import pytest
+
+import pandas as pd
+import pandas._testing as tm
+from pandas.tests.extension.base.base import BaseExtensionTests
+
+
+class BaseReduceTests(BaseExtensionTests):
+    """
+    Reduction specific tests. Generally these only
+    make sense for numeric/boolean operations.
+    """
+
+    def check_reduce(self, s, op_name, skipna):
+        result = getattr(s, op_name)(skipna=skipna)
+        expected = getattr(s.astype("float64"), op_name)(skipna=skipna)
+        tm.assert_almost_equal(result, expected)
+
+
+class BaseNoReduceTests(BaseReduceTests):
+    """we don't define any reductions"""
+
+    @pytest.mark.parametrize("skipna", [True, False])
+    def test_reduce_series_numeric(self, data, all_numeric_reductions, skipna):
+        op_name = all_numeric_reductions
+        s = pd.Series(data)
+
+        msg = (
+            "[Cc]annot perform|Categorical is not ordered for operation|"
+            "does not support reduction|"
+        )
+
+        with pytest.raises(TypeError, match=msg):
+            getattr(s, op_name)(skipna=skipna)
+
+    @pytest.mark.parametrize("skipna", [True, False])
+    def test_reduce_series_boolean(self, data, all_boolean_reductions, skipna):
+        op_name = all_boolean_reductions
+        s = pd.Series(data)
+
+        msg = (
+            "[Cc]annot perform|Categorical is not ordered for operation|"
+            "does not support reduction|"
+        )
+
+        with pytest.raises(TypeError, match=msg):
+            getattr(s, op_name)(skipna=skipna)
+
+
+class BaseNumericReduceTests(BaseReduceTests):
+    @pytest.mark.parametrize("skipna", [True, False])
+    def test_reduce_series(self, data, all_numeric_reductions, skipna):
+        op_name = all_numeric_reductions
+        s = pd.Series(data)
+
+        # min/max with empty produce numpy warnings
+        with warnings.catch_warnings():
+            warnings.simplefilter("ignore", RuntimeWarning)
+            self.check_reduce(s, op_name, skipna)
+
+
+class BaseBooleanReduceTests(BaseReduceTests):
+    @pytest.mark.parametrize("skipna", [True, False])
+    def test_reduce_series(self, data, all_boolean_reductions, skipna):
+        op_name = all_boolean_reductions
+        s = pd.Series(data)
+        self.check_reduce(s, op_name, skipna)

dist/client/pandas/tests/extension/base/reshaping.py

@@ -0,0 +1,370 @@
+import itertools
+
+import numpy as np
+import pytest
+
+import pandas as pd
+from pandas.api.extensions import ExtensionArray
+from pandas.core.internals.blocks import EABackedBlock
+from pandas.tests.extension.base.base import BaseExtensionTests
+
+
+class BaseReshapingTests(BaseExtensionTests):
+    """Tests for reshaping and concatenation."""
+
+    @pytest.mark.parametrize("in_frame", [True, False])
+    def test_concat(self, data, in_frame):
+        wrapped = pd.Series(data)
+        if in_frame:
+            wrapped = pd.DataFrame(wrapped)
+        result = pd.concat([wrapped, wrapped], ignore_index=True)
+
+        assert len(result) == len(data) * 2
+
+        if in_frame:
+            dtype = result.dtypes[0]
+        else:
+            dtype = result.dtype
+
+        assert dtype == data.dtype
+        if hasattr(result._mgr, "blocks"):
+            assert isinstance(result._mgr.blocks[0], EABackedBlock)
+        assert isinstance(result._mgr.arrays[0], ExtensionArray)
+
+    @pytest.mark.parametrize("in_frame", [True, False])
+    def test_concat_all_na_block(self, data_missing, in_frame):
+        valid_block = pd.Series(data_missing.take([1, 1]), index=[0, 1])
+        na_block = pd.Series(data_missing.take([0, 0]), index=[2, 3])
+        if in_frame:
+            valid_block = pd.DataFrame({"a": valid_block})
+            na_block = pd.DataFrame({"a": na_block})
+        result = pd.concat([valid_block, na_block])
+        if in_frame:
+            expected = pd.DataFrame({"a": data_missing.take([1, 1, 0, 0])})
+            self.assert_frame_equal(result, expected)
+        else:
+            expected = pd.Series(data_missing.take([1, 1, 0, 0]))
+            self.assert_series_equal(result, expected)
+
+    def test_concat_mixed_dtypes(self, data):
+        # https://github.com/pandas-dev/pandas/issues/20762
+        df1 = pd.DataFrame({"A": data[:3]})
+        df2 = pd.DataFrame({"A": [1, 2, 3]})
+        df3 = pd.DataFrame({"A": ["a", "b", "c"]}).astype("category")
+        dfs = [df1, df2, df3]
+
+        # dataframes
+        result = pd.concat(dfs)
+        expected = pd.concat([x.astype(object) for x in dfs])
+        self.assert_frame_equal(result, expected)
+
+        # series
+        result = pd.concat([x["A"] for x in dfs])
+        expected = pd.concat([x["A"].astype(object) for x in dfs])
+        self.assert_series_equal(result, expected)
+
+        # simple test for just EA and one other
+        result = pd.concat([df1, df2.astype(object)])
+        expected = pd.concat([df1.astype("object"), df2.astype("object")])
+        self.assert_frame_equal(result, expected)
+
+        result = pd.concat([df1["A"], df2["A"].astype(object)])
+        expected = pd.concat([df1["A"].astype("object"), df2["A"].astype("object")])
+        self.assert_series_equal(result, expected)
+
+    def test_concat_columns(self, data, na_value):
+        df1 = pd.DataFrame({"A": data[:3]})
+        df2 = pd.DataFrame({"B": [1, 2, 3]})
+
+        expected = pd.DataFrame({"A": data[:3], "B": [1, 2, 3]})
+        result = pd.concat([df1, df2], axis=1)
+        self.assert_frame_equal(result, expected)
+        result = pd.concat([df1["A"], df2["B"]], axis=1)
+        self.assert_frame_equal(result, expected)
+
+        # non-aligned
+        df2 = pd.DataFrame({"B": [1, 2, 3]}, index=[1, 2, 3])
+        expected = pd.DataFrame(
+            {
+                "A": data._from_sequence(list(data[:3]) + [na_value], dtype=data.dtype),
+                "B": [np.nan, 1, 2, 3],
+            }
+        )
+
+        result = pd.concat([df1, df2], axis=1)
+        self.assert_frame_equal(result, expected)
+        result = pd.concat([df1["A"], df2["B"]], axis=1)
+        self.assert_frame_equal(result, expected)
+
+    def test_concat_extension_arrays_copy_false(self, data, na_value):
+        # GH 20756
+        df1 = pd.DataFrame({"A": data[:3]})
+        df2 = pd.DataFrame({"B": data[3:7]})
+        expected = pd.DataFrame(
+            {
+                "A": data._from_sequence(list(data[:3]) + [na_value], dtype=data.dtype),
+                "B": data[3:7],
+            }
+        )
+        result = pd.concat([df1, df2], axis=1, copy=False)
+        self.assert_frame_equal(result, expected)
+
+    def test_concat_with_reindex(self, data):
+        # GH-33027
+        a = pd.DataFrame({"a": data[:5]})
+        b = pd.DataFrame({"b": data[:5]})
+        result = pd.concat([a, b], ignore_index=True)
+        expected = pd.DataFrame(
+            {
+                "a": data.take(list(range(5)) + ([-1] * 5), allow_fill=True),
+                "b": data.take(([-1] * 5) + list(range(5)), allow_fill=True),
+            }
+        )
+        self.assert_frame_equal(result, expected)
+
+    def test_align(self, data, na_value):
+        a = data[:3]
+        b = data[2:5]
+        r1, r2 = pd.Series(a).align(pd.Series(b, index=[1, 2, 3]))
+
+        # Assumes that the ctor can take a list of scalars of the type
+        e1 = pd.Series(data._from_sequence(list(a) + [na_value], dtype=data.dtype))
+        e2 = pd.Series(data._from_sequence([na_value] + list(b), dtype=data.dtype))
+        self.assert_series_equal(r1, e1)
+        self.assert_series_equal(r2, e2)
+
+    def test_align_frame(self, data, na_value):
+        a = data[:3]
+        b = data[2:5]
+        r1, r2 = pd.DataFrame({"A": a}).align(pd.DataFrame({"A": b}, index=[1, 2, 3]))
+
+        # Assumes that the ctor can take a list of scalars of the type
+        e1 = pd.DataFrame(
+            {"A": data._from_sequence(list(a) + [na_value], dtype=data.dtype)}
+        )
+        e2 = pd.DataFrame(
+            {"A": data._from_sequence([na_value] + list(b), dtype=data.dtype)}
+        )
+        self.assert_frame_equal(r1, e1)
+        self.assert_frame_equal(r2, e2)
+
+    def test_align_series_frame(self, data, na_value):
+        # https://github.com/pandas-dev/pandas/issues/20576
+        ser = pd.Series(data, name="a")
+        df = pd.DataFrame({"col": np.arange(len(ser) + 1)})
+        r1, r2 = ser.align(df)
+
+        e1 = pd.Series(
+            data._from_sequence(list(data) + [na_value], dtype=data.dtype),
+            name=ser.name,
+        )
+
+        self.assert_series_equal(r1, e1)
+        self.assert_frame_equal(r2, df)
+
+    def test_set_frame_expand_regular_with_extension(self, data):
+        df = pd.DataFrame({"A": [1] * len(data)})
+        df["B"] = data
+        expected = pd.DataFrame({"A": [1] * len(data), "B": data})
+        self.assert_frame_equal(df, expected)
+
+    def test_set_frame_expand_extension_with_regular(self, data):
+        df = pd.DataFrame({"A": data})
+        df["B"] = [1] * len(data)
+        expected = pd.DataFrame({"A": data, "B": [1] * len(data)})
+        self.assert_frame_equal(df, expected)
+
+    def test_set_frame_overwrite_object(self, data):
+        # https://github.com/pandas-dev/pandas/issues/20555
+        df = pd.DataFrame({"A": [1] * len(data)}, dtype=object)
+        df["A"] = data
+        assert df.dtypes["A"] == data.dtype
+
+    def test_merge(self, data, na_value):
+        # GH-20743
+        df1 = pd.DataFrame({"ext": data[:3], "int1": [1, 2, 3], "key": [0, 1, 2]})
+        df2 = pd.DataFrame({"int2": [1, 2, 3, 4], "key": [0, 0, 1, 3]})
+
+        res = pd.merge(df1, df2)
+        exp = pd.DataFrame(
+            {
+                "int1": [1, 1, 2],
+                "int2": [1, 2, 3],
+                "key": [0, 0, 1],
+                "ext": data._from_sequence(
+                    [data[0], data[0], data[1]], dtype=data.dtype
+                ),
+            }
+        )
+        self.assert_frame_equal(res, exp[["ext", "int1", "key", "int2"]])
+
+        res = pd.merge(df1, df2, how="outer")
+        exp = pd.DataFrame(
+            {
+                "int1": [1, 1, 2, 3, np.nan],
+                "int2": [1, 2, 3, np.nan, 4],
+                "key": [0, 0, 1, 2, 3],
+                "ext": data._from_sequence(
+                    [data[0], data[0], data[1], data[2], na_value], dtype=data.dtype
+                ),
+            }
+        )
+        self.assert_frame_equal(res, exp[["ext", "int1", "key", "int2"]])
+
+    def test_merge_on_extension_array(self, data):
+        # GH 23020
+        a, b = data[:2]
+        key = type(data)._from_sequence([a, b], dtype=data.dtype)
+
+        df = pd.DataFrame({"key": key, "val": [1, 2]})
+        result = pd.merge(df, df, on="key")
+        expected = pd.DataFrame({"key": key, "val_x": [1, 2], "val_y": [1, 2]})
+        self.assert_frame_equal(result, expected)
+
+        # order
+        result = pd.merge(df.iloc[[1, 0]], df, on="key")
+        expected = expected.iloc[[1, 0]].reset_index(drop=True)
+        self.assert_frame_equal(result, expected)
+
+    def test_merge_on_extension_array_duplicates(self, data):
+        # GH 23020
+        a, b = data[:2]
+        key = type(data)._from_sequence([a, b, a], dtype=data.dtype)
+        df1 = pd.DataFrame({"key": key, "val": [1, 2, 3]})
+        df2 = pd.DataFrame({"key": key, "val": [1, 2, 3]})
+
+        result = pd.merge(df1, df2, on="key")
+        expected = pd.DataFrame(
+            {
+                "key": key.take([0, 0, 0, 0, 1]),
+                "val_x": [1, 1, 3, 3, 2],
+                "val_y": [1, 3, 1, 3, 2],
+            }
+        )
+        self.assert_frame_equal(result, expected)
+
+    @pytest.mark.parametrize(
+        "columns",
+        [
+            ["A", "B"],
+            pd.MultiIndex.from_tuples(
+                [("A", "a"), ("A", "b")], names=["outer", "inner"]
+            ),
+        ],
+    )
+    def test_stack(self, data, columns):
+        df = pd.DataFrame({"A": data[:5], "B": data[:5]})
+        df.columns = columns
+        result = df.stack()
+        expected = df.astype(object).stack()
+        # we need a second astype(object), in case the constructor inferred
+        # object -> specialized, as is done for period.
+        expected = expected.astype(object)
+
+        if isinstance(expected, pd.Series):
+            assert result.dtype == df.iloc[:, 0].dtype
+        else:
+            assert all(result.dtypes == df.iloc[:, 0].dtype)
+
+        result = result.astype(object)
+        self.assert_equal(result, expected)
+
+    @pytest.mark.parametrize(
+        "index",
+        [
+            # Two levels, uniform.
+            pd.MultiIndex.from_product(([["A", "B"], ["a", "b"]]), names=["a", "b"]),
+            # non-uniform
+            pd.MultiIndex.from_tuples([("A", "a"), ("A", "b"), ("B", "b")]),
+            # three levels, non-uniform
+            pd.MultiIndex.from_product([("A", "B"), ("a", "b", "c"), (0, 1, 2)]),
+            pd.MultiIndex.from_tuples(
+                [
+                    ("A", "a", 1),
+                    ("A", "b", 0),
+                    ("A", "a", 0),
+                    ("B", "a", 0),
+                    ("B", "c", 1),
+                ]
+            ),
+        ],
+    )
+    @pytest.mark.parametrize("obj", ["series", "frame"])
+    def test_unstack(self, data, index, obj):
+        data = data[: len(index)]
+        if obj == "series":
+            ser = pd.Series(data, index=index)
+        else:
+            ser = pd.DataFrame({"A": data, "B": data}, index=index)
+
+        n = index.nlevels
+        levels = list(range(n))
+        # [0, 1, 2]
+        # [(0,), (1,), (2,), (0, 1), (0, 2), (1, 0), (1, 2), (2, 0), (2, 1)]
+        combinations = itertools.chain.from_iterable(
+            itertools.permutations(levels, i) for i in range(1, n)
+        )
+
+        for level in combinations:
+            result = ser.unstack(level=level)
+            assert all(
+                isinstance(result[col].array, type(data)) for col in result.columns
+            )
+
+            if obj == "series":
+                # We should get the same result with to_frame+unstack+droplevel
+                df = ser.to_frame()
+
+                alt = df.unstack(level=level).droplevel(0, axis=1)
+                self.assert_frame_equal(result, alt)
+
+            expected = ser.astype(object).unstack(
+                level=level, fill_value=data.dtype.na_value
+            )
+            if obj == "series" and not isinstance(ser.dtype, pd.SparseDtype):
+                # GH#34457 SparseArray.astype(object) gives Sparse[object]
+                #  instead of np.dtype(object)
+                assert (expected.dtypes == object).all()
+
+            result = result.astype(object)
+
+            self.assert_frame_equal(result, expected)
+
+    def test_ravel(self, data):
+        # as long as EA is 1D-only, ravel is a no-op
+        result = data.ravel()
+        assert type(result) == type(data)
+
+        # Check that we have a view, not a copy
+        result[0] = result[1]
+        assert data[0] == data[1]
+
+    def test_transpose(self, data):
+        result = data.transpose()
+        assert type(result) == type(data)
+
+        # check we get a new object
+        assert result is not data
+
+        # If we ever _did_ support 2D, shape should be reversed
+        assert result.shape == data.shape[::-1]
+
+        # Check that we have a view, not a copy
+        result[0] = result[1]
+        assert data[0] == data[1]
+
+    def test_transpose_frame(self, data):
+        df = pd.DataFrame({"A": data[:4], "B": data[:4]}, index=["a", "b", "c", "d"])
+        result = df.T
+        expected = pd.DataFrame(
+            {
+                "a": type(data)._from_sequence([data[0]] * 2, dtype=data.dtype),
+                "b": type(data)._from_sequence([data[1]] * 2, dtype=data.dtype),
+                "c": type(data)._from_sequence([data[2]] * 2, dtype=data.dtype),
+                "d": type(data)._from_sequence([data[3]] * 2, dtype=data.dtype),
+            },
+            index=["A", "B"],
+        )
+        self.assert_frame_equal(result, expected)
+        self.assert_frame_equal(np.transpose(np.transpose(df)), df)
+        self.assert_frame_equal(np.transpose(np.transpose(df[["A"]])), df[["A"]])

dist/client/pandas/tests/extension/base/setitem.py

@@ -0,0 +1,428 @@
+import numpy as np
+import pytest
+
+from pandas.core.dtypes.dtypes import (
+    DatetimeTZDtype,
+    IntervalDtype,
+    PandasDtype,
+    PeriodDtype,
+)
+
+import pandas as pd
+import pandas._testing as tm
+from pandas.tests.extension.base.base import BaseExtensionTests
+
+
+class BaseSetitemTests(BaseExtensionTests):
+    @pytest.fixture(
+        params=[
+            lambda x: x.index,
+            lambda x: list(x.index),
+            lambda x: slice(None),
+            lambda x: slice(0, len(x)),
+            lambda x: range(len(x)),
+            lambda x: list(range(len(x))),
+            lambda x: np.ones(len(x), dtype=bool),
+        ],
+        ids=[
+            "index",
+            "list[index]",
+            "null_slice",
+            "full_slice",
+            "range",
+            "list(range)",
+            "mask",
+        ],
+    )
+    def full_indexer(self, request):
+        """
+        Fixture for an indexer to pass to obj.loc to get/set the full length of the
+        object.
+
+        In some cases, assumes that obj.index is the default RangeIndex.
+        """
+        return request.param
+
+    def test_setitem_scalar_series(self, data, box_in_series):
+        if box_in_series:
+            data = pd.Series(data)
+        data[0] = data[1]
+        assert data[0] == data[1]
+
+    def test_setitem_sequence(self, data, box_in_series):
+        if box_in_series:
+            data = pd.Series(data)
+        original = data.copy()
+
+        data[[0, 1]] = [data[1], data[0]]
+        assert data[0] == original[1]
+        assert data[1] == original[0]
+
+    def test_setitem_sequence_mismatched_length_raises(self, data, as_array):
+        ser = pd.Series(data)
+        original = ser.copy()
+        value = [data[0]]
+        if as_array:
+            value = data._from_sequence(value)
+
+        xpr = "cannot set using a {} indexer with a different length"
+        with pytest.raises(ValueError, match=xpr.format("list-like")):
+            ser[[0, 1]] = value
+        # Ensure no modifications made before the exception
+        self.assert_series_equal(ser, original)
+
+        with pytest.raises(ValueError, match=xpr.format("slice")):
+            ser[slice(3)] = value
+        self.assert_series_equal(ser, original)
+
+    def test_setitem_empty_indexer(self, data, box_in_series):
+        if box_in_series:
+            data = pd.Series(data)
+        original = data.copy()
+        data[np.array([], dtype=int)] = []
+        self.assert_equal(data, original)
+
+    def test_setitem_sequence_broadcasts(self, data, box_in_series):
+        if box_in_series:
+            data = pd.Series(data)
+        data[[0, 1]] = data[2]
+        assert data[0] == data[2]
+        assert data[1] == data[2]
+
+    @pytest.mark.parametrize("setter", ["loc", "iloc"])
+    def test_setitem_scalar(self, data, setter):
+        arr = pd.Series(data)
+        setter = getattr(arr, setter)
+        setter[0] = data[1]
+        assert arr[0] == data[1]
+
+    def test_setitem_loc_scalar_mixed(self, data):
+        df = pd.DataFrame({"A": np.arange(len(data)), "B": data})
+        df.loc[0, "B"] = data[1]
+        assert df.loc[0, "B"] == data[1]
+
+    def test_setitem_loc_scalar_single(self, data):
+        df = pd.DataFrame({"B": data})
+        df.loc[10, "B"] = data[1]
+        assert df.loc[10, "B"] == data[1]
+
+    def test_setitem_loc_scalar_multiple_homogoneous(self, data):
+        df = pd.DataFrame({"A": data, "B": data})
+        df.loc[10, "B"] = data[1]
+        assert df.loc[10, "B"] == data[1]
+
+    def test_setitem_iloc_scalar_mixed(self, data):
+        df = pd.DataFrame({"A": np.arange(len(data)), "B": data})
+        df.iloc[0, 1] = data[1]
+        assert df.loc[0, "B"] == data[1]
+
+    def test_setitem_iloc_scalar_single(self, data):
+        df = pd.DataFrame({"B": data})
+        df.iloc[10, 0] = data[1]
+        assert df.loc[10, "B"] == data[1]
+
+    def test_setitem_iloc_scalar_multiple_homogoneous(self, data):
+        df = pd.DataFrame({"A": data, "B": data})
+        df.iloc[10, 1] = data[1]
+        assert df.loc[10, "B"] == data[1]
+
+    @pytest.mark.parametrize(
+        "mask",
+        [
+            np.array([True, True, True, False, False]),
+            pd.array([True, True, True, False, False], dtype="boolean"),
+            pd.array([True, True, True, pd.NA, pd.NA], dtype="boolean"),
+        ],
+        ids=["numpy-array", "boolean-array", "boolean-array-na"],
+    )
+    def test_setitem_mask(self, data, mask, box_in_series):
+        arr = data[:5].copy()
+        expected = arr.take([0, 0, 0, 3, 4])
+        if box_in_series:
+            arr = pd.Series(arr)
+            expected = pd.Series(expected)
+        arr[mask] = data[0]
+        self.assert_equal(expected, arr)
+
+    def test_setitem_mask_raises(self, data, box_in_series):
+        # wrong length
+        mask = np.array([True, False])
+
+        if box_in_series:
+            data = pd.Series(data)
+
+        with pytest.raises(IndexError, match="wrong length"):
+            data[mask] = data[0]
+
+        mask = pd.array(mask, dtype="boolean")
+        with pytest.raises(IndexError, match="wrong length"):
+            data[mask] = data[0]
+
+    def test_setitem_mask_boolean_array_with_na(self, data, box_in_series):
+        mask = pd.array(np.zeros(data.shape, dtype="bool"), dtype="boolean")
+        mask[:3] = True
+        mask[3:5] = pd.NA
+
+        if box_in_series:
+            data = pd.Series(data)
+
+        data[mask] = data[0]
+
+        assert (data[:3] == data[0]).all()
+
+    @pytest.mark.parametrize(
+        "idx",
+        [[0, 1, 2], pd.array([0, 1, 2], dtype="Int64"), np.array([0, 1, 2])],
+        ids=["list", "integer-array", "numpy-array"],
+    )
+    def test_setitem_integer_array(self, data, idx, box_in_series):
+        arr = data[:5].copy()
+        expected = data.take([0, 0, 0, 3, 4])
+
+        if box_in_series:
+            arr = pd.Series(arr)
+            expected = pd.Series(expected)
+
+        arr[idx] = arr[0]
+        self.assert_equal(arr, expected)
+
+    @pytest.mark.parametrize(
+        "idx, box_in_series",
+        [
+            ([0, 1, 2, pd.NA], False),
+            pytest.param(
+                [0, 1, 2, pd.NA], True, marks=pytest.mark.xfail(reason="GH-31948")
+            ),
+            (pd.array([0, 1, 2, pd.NA], dtype="Int64"), False),
+            (pd.array([0, 1, 2, pd.NA], dtype="Int64"), False),
+        ],
+        ids=["list-False", "list-True", "integer-array-False", "integer-array-True"],
+    )
+    def test_setitem_integer_with_missing_raises(self, data, idx, box_in_series):
+        arr = data.copy()
+
+        # TODO(xfail) this raises KeyError about labels not found (it tries label-based)
+        # for list of labels with Series
+        if box_in_series:
+            arr = pd.Series(data, index=[tm.rands(4) for _ in range(len(data))])
+
+        msg = "Cannot index with an integer indexer containing NA values"
+        with pytest.raises(ValueError, match=msg):
+            arr[idx] = arr[0]
+
+    @pytest.mark.parametrize("as_callable", [True, False])
+    @pytest.mark.parametrize("setter", ["loc", None])
+    def test_setitem_mask_aligned(self, data, as_callable, setter):
+        ser = pd.Series(data)
+        mask = np.zeros(len(data), dtype=bool)
+        mask[:2] = True
+
+        if as_callable:
+            mask2 = lambda x: mask
+        else:
+            mask2 = mask
+
+        if setter:
+            # loc
+            target = getattr(ser, setter)
+        else:
+            # Series.__setitem__
+            target = ser
+
+        target[mask2] = data[5:7]
+
+        ser[mask2] = data[5:7]
+        assert ser[0] == data[5]
+        assert ser[1] == data[6]
+
+    @pytest.mark.parametrize("setter", ["loc", None])
+    def test_setitem_mask_broadcast(self, data, setter):
+        ser = pd.Series(data)
+        mask = np.zeros(len(data), dtype=bool)
+        mask[:2] = True
+
+        if setter:  # loc
+            target = getattr(ser, setter)
+        else:  # __setitem__
+            target = ser
+
+        target[mask] = data[10]
+        assert ser[0] == data[10]
+        assert ser[1] == data[10]
+
+    def test_setitem_expand_columns(self, data):
+        df = pd.DataFrame({"A": data})
+        result = df.copy()
+        result["B"] = 1
+        expected = pd.DataFrame({"A": data, "B": [1] * len(data)})
+        self.assert_frame_equal(result, expected)
+
+        result = df.copy()
+        result.loc[:, "B"] = 1
+        self.assert_frame_equal(result, expected)
+
+        # overwrite with new type
+        result["B"] = data
+        expected = pd.DataFrame({"A": data, "B": data})
+        self.assert_frame_equal(result, expected)
+
+    def test_setitem_expand_with_extension(self, data):
+        df = pd.DataFrame({"A": [1] * len(data)})
+        result = df.copy()
+        result["B"] = data
+        expected = pd.DataFrame({"A": [1] * len(data), "B": data})
+        self.assert_frame_equal(result, expected)
+
+        result = df.copy()
+        result.loc[:, "B"] = data
+        self.assert_frame_equal(result, expected)
+
+    def test_setitem_frame_invalid_length(self, data):
+        df = pd.DataFrame({"A": [1] * len(data)})
+        xpr = (
+            rf"Length of values \({len(data[:5])}\) "
+            rf"does not match length of index \({len(df)}\)"
+        )
+        with pytest.raises(ValueError, match=xpr):
+            df["B"] = data[:5]
+
+    def test_setitem_tuple_index(self, data):
+        ser = pd.Series(data[:2], index=[(0, 0), (0, 1)])
+        expected = pd.Series(data.take([1, 1]), index=ser.index)
+        ser[(0, 0)] = data[1]
+        self.assert_series_equal(ser, expected)
+
+    def test_setitem_slice(self, data, box_in_series):
+        arr = data[:5].copy()
+        expected = data.take([0, 0, 0, 3, 4])
+        if box_in_series:
+            arr = pd.Series(arr)
+            expected = pd.Series(expected)
+
+        arr[:3] = data[0]
+        self.assert_equal(arr, expected)
+
+    def test_setitem_loc_iloc_slice(self, data):
+        arr = data[:5].copy()
+        s = pd.Series(arr, index=["a", "b", "c", "d", "e"])
+        expected = pd.Series(data.take([0, 0, 0, 3, 4]), index=s.index)
+
+        result = s.copy()
+        result.iloc[:3] = data[0]
+        self.assert_equal(result, expected)
+
+        result = s.copy()
+        result.loc[:"c"] = data[0]
+        self.assert_equal(result, expected)
+
+    def test_setitem_slice_mismatch_length_raises(self, data):
+        arr = data[:5]
+        with pytest.raises(ValueError):
+            arr[:1] = arr[:2]
+
+    def test_setitem_slice_array(self, data):
+        arr = data[:5].copy()
+        arr[:5] = data[-5:]
+        self.assert_extension_array_equal(arr, data[-5:])
+
+    def test_setitem_scalar_key_sequence_raise(self, data):
+        arr = data[:5].copy()
+        with pytest.raises(ValueError):
+            arr[0] = arr[[0, 1]]
+
+    def test_setitem_preserves_views(self, data):
+        # GH#28150 setitem shouldn't swap the underlying data
+        view1 = data.view()
+        view2 = data[:]
+
+        data[0] = data[1]
+        assert view1[0] == data[1]
+        assert view2[0] == data[1]
+
+    def test_setitem_with_expansion_dataframe_column(self, data, full_indexer):
+        # https://github.com/pandas-dev/pandas/issues/32395
+        df = expected = pd.DataFrame({"data": pd.Series(data)})
+        result = pd.DataFrame(index=df.index)
+
+        key = full_indexer(df)
+        result.loc[key, "data"] = df["data"]
+
+        self.assert_frame_equal(result, expected)
+
+    def test_setitem_with_expansion_row(self, data, na_value):
+        df = pd.DataFrame({"data": data[:1]})
+
+        df.loc[1, "data"] = data[1]
+        expected = pd.DataFrame({"data": data[:2]})
+        self.assert_frame_equal(df, expected)
+
+        # https://github.com/pandas-dev/pandas/issues/47284
+        df.loc[2, "data"] = na_value
+        expected = pd.DataFrame(
+            {"data": pd.Series([data[0], data[1], na_value], dtype=data.dtype)}
+        )
+        self.assert_frame_equal(df, expected)
+
+    def test_setitem_series(self, data, full_indexer):
+        # https://github.com/pandas-dev/pandas/issues/32395
+        ser = pd.Series(data, name="data")
+        result = pd.Series(index=ser.index, dtype=object, name="data")
+
+        # because result has object dtype, the attempt to do setting inplace
+        #  is successful, and object dtype is retained
+        key = full_indexer(ser)
+        result.loc[key] = ser
+
+        expected = pd.Series(
+            data.astype(object), index=ser.index, name="data", dtype=object
+        )
+        self.assert_series_equal(result, expected)
+
+    def test_setitem_frame_2d_values(self, data, request):
+        # GH#44514
+        df = pd.DataFrame({"A": data})
+
+        # Avoiding using_array_manager fixture
+        #  https://github.com/pandas-dev/pandas/pull/44514#discussion_r754002410
+        using_array_manager = isinstance(df._mgr, pd.core.internals.ArrayManager)
+        if using_array_manager:
+            if not isinstance(
+                data.dtype, (PandasDtype, PeriodDtype, IntervalDtype, DatetimeTZDtype)
+            ):
+                # These dtypes have non-broken implementations of _can_hold_element
+                mark = pytest.mark.xfail(reason="Goes through split path, loses dtype")
+                request.node.add_marker(mark)
+
+        df = pd.DataFrame({"A": data})
+        orig = df.copy()
+
+        df.iloc[:] = df
+        self.assert_frame_equal(df, orig)
+
+        df.iloc[:-1] = df.iloc[:-1]
+        self.assert_frame_equal(df, orig)
+
+        df.iloc[:] = df.values
+        self.assert_frame_equal(df, orig)
+
+        df.iloc[:-1] = df.values[:-1]
+        self.assert_frame_equal(df, orig)
+
+    def test_delitem_series(self, data):
+        # GH#40763
+        ser = pd.Series(data, name="data")
+
+        taker = np.arange(len(ser))
+        taker = np.delete(taker, 1)
+
+        expected = ser[taker]
+        del ser[1]
+        self.assert_series_equal(ser, expected)
+
+    def test_setitem_invalid(self, data, invalid_scalar):
+        msg = ""  # messages vary by subclass, so we do not test it
+        with pytest.raises((ValueError, TypeError), match=msg):
+            data[0] = invalid_scalar
+
+        with pytest.raises((ValueError, TypeError), match=msg):
+            data[:] = invalid_scalar

dist/client/pandas/tests/extension/conftest.py

@@ -0,0 +1,195 @@
+import operator
+
+import pytest
+
+from pandas import Series
+
+
+@pytest.fixture
+def dtype():
+    """A fixture providing the ExtensionDtype to validate."""
+    raise NotImplementedError
+
+
+@pytest.fixture
+def data():
+    """
+    Length-100 array for this type.
+
+    * data[0] and data[1] should both be non missing
+    * data[0] and data[1] should not be equal
+    """
+    raise NotImplementedError
+
+
+@pytest.fixture
+def data_for_twos():
+    """Length-100 array in which all the elements are two."""
+    raise NotImplementedError
+
+
+@pytest.fixture
+def data_missing():
+    """Length-2 array with [NA, Valid]"""
+    raise NotImplementedError
+
+
+@pytest.fixture(params=["data", "data_missing"])
+def all_data(request, data, data_missing):
+    """Parametrized fixture giving 'data' and 'data_missing'"""
+    if request.param == "data":
+        return data
+    elif request.param == "data_missing":
+        return data_missing
+
+
+@pytest.fixture
+def data_repeated(data):
+    """
+    Generate many datasets.
+
+    Parameters
+    ----------
+    data : fixture implementing `data`
+
+    Returns
+    -------
+    Callable[[int], Generator]:
+        A callable that takes a `count` argument and
+        returns a generator yielding `count` datasets.
+    """
+
+    def gen(count):
+        for _ in range(count):
+            yield data
+
+    return gen
+
+
+@pytest.fixture
+def data_for_sorting():
+    """
+    Length-3 array with a known sort order.
+
+    This should be three items [B, C, A] with
+    A < B < C
+    """
+    raise NotImplementedError
+
+
+@pytest.fixture
+def data_missing_for_sorting():
+    """
+    Length-3 array with a known sort order.
+
+    This should be three items [B, NA, A] with
+    A < B and NA missing.
+    """
+    raise NotImplementedError
+
+
+@pytest.fixture
+def na_cmp():
+    """
+    Binary operator for comparing NA values.
+
+    Should return a function of two arguments that returns
+    True if both arguments are (scalar) NA for your type.
+
+    By default, uses ``operator.is_``
+    """
+    return operator.is_
+
+
+@pytest.fixture
+def na_value():
+    """The scalar missing value for this type. Default 'None'"""
+    return None
+
+
+@pytest.fixture
+def data_for_grouping():
+    """
+    Data for factorization, grouping, and unique tests.
+
+    Expected to be like [B, B, NA, NA, A, A, B, C]
+
+    Where A < B < C and NA is missing
+    """
+    raise NotImplementedError
+
+
+@pytest.fixture(params=[True, False])
+def box_in_series(request):
+    """Whether to box the data in a Series"""
+    return request.param
+
+
+@pytest.fixture(
+    params=[
+        lambda x: 1,
+        lambda x: [1] * len(x),
+        lambda x: Series([1] * len(x)),
+        lambda x: x,
+    ],
+    ids=["scalar", "list", "series", "object"],
+)
+def groupby_apply_op(request):
+    """
+    Functions to test groupby.apply().
+    """
+    return request.param
+
+
+@pytest.fixture(params=[True, False])
+def as_frame(request):
+    """
+    Boolean fixture to support Series and Series.to_frame() comparison testing.
+    """
+    return request.param
+
+
+@pytest.fixture(params=[True, False])
+def as_series(request):
+    """
+    Boolean fixture to support arr and Series(arr) comparison testing.
+    """
+    return request.param
+
+
+@pytest.fixture(params=[True, False])
+def use_numpy(request):
+    """
+    Boolean fixture to support comparison testing of ExtensionDtype array
+    and numpy array.
+    """
+    return request.param
+
+
+@pytest.fixture(params=["ffill", "bfill"])
+def fillna_method(request):
+    """
+    Parametrized fixture giving method parameters 'ffill' and 'bfill' for
+    Series.fillna(method=<method>) testing.
+    """
+    return request.param
+
+
+@pytest.fixture(params=[True, False])
+def as_array(request):
+    """
+    Boolean fixture to support ExtensionDtype _from_sequence method testing.
+    """
+    return request.param
+
+
+@pytest.fixture
+def invalid_scalar(data):
+    """
+    A scalar that *cannot* be held by this ExtensionArray.
+
+    The default should work for most subclasses, but is not guaranteed.
+
+    If the array can hold any item (i.e. object dtype), then use pytest.skip.
+    """
+    return object.__new__(object)

dist/client/pandas/tests/extension/date/__init__.py

@@ -0,0 +1,6 @@
+from pandas.tests.extension.date.array import (
+    DateArray,
+    DateDtype,
+)
+
+__all__ = ["DateArray", "DateDtype"]

dist/client/pandas/tests/extension/date/array.py

@@ -0,0 +1,183 @@
+import datetime as dt
+from typing import (
+    Any,
+    Optional,
+    Sequence,
+    Tuple,
+    Union,
+    cast,
+)
+
+import numpy as np
+
+from pandas._typing import (
+    Dtype,
+    PositionalIndexer,
+)
+
+from pandas.core.dtypes.dtypes import register_extension_dtype
+
+from pandas.api.extensions import (
+    ExtensionArray,
+    ExtensionDtype,
+)
+from pandas.api.types import pandas_dtype
+
+
+@register_extension_dtype
+class DateDtype(ExtensionDtype):
+    @property
+    def type(self):
+        return dt.date
+
+    @property
+    def name(self):
+        return "DateDtype"
+
+    @classmethod
+    def construct_from_string(cls, string: str):
+        if not isinstance(string, str):
+            raise TypeError(
+                f"'construct_from_string' expects a string, got {type(string)}"
+            )
+
+        if string == cls.__name__:
+            return cls()
+        else:
+            raise TypeError(f"Cannot construct a '{cls.__name__}' from '{string}'")
+
+    @classmethod
+    def construct_array_type(cls):
+        return DateArray
+
+    @property
+    def na_value(self):
+        return dt.date.min
+
+    def __repr__(self) -> str:
+        return self.name
+
+
+class DateArray(ExtensionArray):
+    def __init__(
+        self,
+        dates: Union[
+            dt.date,
+            Sequence[dt.date],
+            Tuple[np.ndarray, np.ndarray, np.ndarray],
+            np.ndarray,
+        ],
+    ) -> None:
+        if isinstance(dates, dt.date):
+            self._year = np.array([dates.year])
+            self._month = np.array([dates.month])
+            self._day = np.array([dates.year])
+            return
+
+        ldates = len(dates)
+        if isinstance(dates, list):
+            # pre-allocate the arrays since we know the size before hand
+            self._year = np.zeros(ldates, dtype=np.uint16)  # 65535 (0, 9999)
+            self._month = np.zeros(ldates, dtype=np.uint8)  # 255 (1, 31)
+            self._day = np.zeros(ldates, dtype=np.uint8)  # 255 (1, 12)
+            # populate them
+            for i, (y, m, d) in enumerate(
+                map(lambda date: (date.year, date.month, date.day), dates)
+            ):
+                self._year[i] = y
+                self._month[i] = m
+                self._day[i] = d
+
+        elif isinstance(dates, tuple):
+            # only support triples
+            if ldates != 3:
+                raise ValueError("only triples are valid")
+            # check if all elements have the same type
+            if any(map(lambda x: not isinstance(x, np.ndarray), dates)):
+                raise TypeError("invalid type")
+            ly, lm, ld = (len(cast(np.ndarray, d)) for d in dates)
+            if not ly == lm == ld:
+                raise ValueError(
+                    f"tuple members must have the same length: {(ly, lm, ld)}"
+                )
+            self._year = dates[0].astype(np.uint16)
+            self._month = dates[1].astype(np.uint8)
+            self._day = dates[2].astype(np.uint8)
+
+        elif isinstance(dates, np.ndarray) and dates.dtype == "U10":
+            self._year = np.zeros(ldates, dtype=np.uint16)  # 65535 (0, 9999)
+            self._month = np.zeros(ldates, dtype=np.uint8)  # 255 (1, 31)
+            self._day = np.zeros(ldates, dtype=np.uint8)  # 255 (1, 12)
+
+            # "object_" object is not iterable  [misc]
+            for (i,), (y, m, d) in np.ndenumerate(  # type: ignore[misc]
+                np.char.split(dates, sep="-")
+            ):
+                self._year[i] = int(y)
+                self._month[i] = int(m)
+                self._day[i] = int(d)
+
+        else:
+            raise TypeError(f"{type(dates)} is not supported")
+
+    @property
+    def dtype(self) -> ExtensionDtype:
+        return DateDtype()
+
+    def astype(self, dtype, copy=True):
+        dtype = pandas_dtype(dtype)
+
+        if isinstance(dtype, DateDtype):
+            data = self.copy() if copy else self
+        else:
+            data = self.to_numpy(dtype=dtype, copy=copy, na_value=dt.date.min)
+
+        return data
+
+    @property
+    def nbytes(self) -> int:
+        return self._year.nbytes + self._month.nbytes + self._day.nbytes
+
+    def __len__(self) -> int:
+        return len(self._year)  # all 3 arrays are enforced to have the same length
+
+    def __getitem__(self, item: PositionalIndexer):
+        if isinstance(item, int):
+            return dt.date(self._year[item], self._month[item], self._day[item])
+        else:
+            raise NotImplementedError("only ints are supported as indexes")
+
+    def __setitem__(self, key: Union[int, slice, np.ndarray], value: Any):
+        if not isinstance(key, int):
+            raise NotImplementedError("only ints are supported as indexes")
+
+        if not isinstance(value, dt.date):
+            raise TypeError("you can only set datetime.date types")
+
+        self._year[key] = value.year
+        self._month[key] = value.month
+        self._day[key] = value.day
+
+    def __repr__(self) -> str:
+        return f"DateArray{list(zip(self._year, self._month, self._day))}"
+
+    def copy(self) -> "DateArray":
+        return DateArray((self._year.copy(), self._month.copy(), self._day.copy()))
+
+    def isna(self) -> np.ndarray:
+        return np.logical_and(
+            np.logical_and(
+                self._year == dt.date.min.year, self._month == dt.date.min.month
+            ),
+            self._day == dt.date.min.day,
+        )
+
+    @classmethod
+    def _from_sequence(cls, scalars, *, dtype: Optional[Dtype] = None, copy=False):
+        if isinstance(scalars, dt.date):
+            pass
+        elif isinstance(scalars, DateArray):
+            pass
+        elif isinstance(scalars, np.ndarray):
+            scalars = scalars.astype("U10")  # 10 chars for yyyy-mm-dd
+            return DateArray(scalars)

dist/client/pandas/tests/extension/decimal/__init__.py

@@ -0,0 +1,8 @@
+from pandas.tests.extension.decimal.array import (
+    DecimalArray,
+    DecimalDtype,
+    make_data,
+    to_decimal,
+)
+
+__all__ = ["DecimalArray", "DecimalDtype", "to_decimal", "make_data"]

dist/client/pandas/tests/extension/decimal/array.py

@@ -0,0 +1,272 @@
+from __future__ import annotations
+
+import decimal
+import numbers
+import random
+import sys
+
+import numpy as np
+
+from pandas._typing import type_t
+
+from pandas.core.dtypes.base import ExtensionDtype
+from pandas.core.dtypes.common import (
+    is_dtype_equal,
+    is_float,
+    pandas_dtype,
+)
+
+import pandas as pd
+from pandas.api.extensions import (
+    no_default,
+    register_extension_dtype,
+)
+from pandas.api.types import (
+    is_list_like,
+    is_scalar,
+)
+from pandas.core import arraylike
+from pandas.core.arraylike import OpsMixin
+from pandas.core.arrays import (
+    ExtensionArray,
+    ExtensionScalarOpsMixin,
+)
+from pandas.core.indexers import check_array_indexer
+
+
+@register_extension_dtype
+class DecimalDtype(ExtensionDtype):
+    type = decimal.Decimal
+    name = "decimal"
+    na_value = decimal.Decimal("NaN")
+    _metadata = ("context",)
+
+    def __init__(self, context=None):
+        self.context = context or decimal.getcontext()
+
+    def __repr__(self) -> str:
+        return f"DecimalDtype(context={self.context})"
+
+    @classmethod
+    def construct_array_type(cls) -> type_t[DecimalArray]:
+        """
+        Return the array type associated with this dtype.
+
+        Returns
+        -------
+        type
+        """
+        return DecimalArray
+
+    @property
+    def _is_numeric(self) -> bool:
+        return True
+
+
+class DecimalArray(OpsMixin, ExtensionScalarOpsMixin, ExtensionArray):
+    __array_priority__ = 1000
+
+    def __init__(self, values, dtype=None, copy=False, context=None):
+        for i, val in enumerate(values):
+            if is_float(val):
+                if np.isnan(val):
+                    values[i] = DecimalDtype.na_value
+                else:
+                    values[i] = DecimalDtype.type(val)
+            elif not isinstance(val, decimal.Decimal):
+                raise TypeError("All values must be of type " + str(decimal.Decimal))
+        values = np.asarray(values, dtype=object)
+
+        self._data = values
+        # Some aliases for common attribute names to ensure pandas supports
+        # these
+        self._items = self.data = self._data
+        # those aliases are currently not working due to assumptions
+        # in internal code (GH-20735)
+        # self._values = self.values = self.data
+        self._dtype = DecimalDtype(context)
+
+    @property
+    def dtype(self):
+        return self._dtype
+
+    @classmethod
+    def _from_sequence(cls, scalars, dtype=None, copy=False):
+        return cls(scalars)
+
+    @classmethod
+    def _from_sequence_of_strings(cls, strings, dtype=None, copy=False):
+        return cls._from_sequence([decimal.Decimal(x) for x in strings], dtype, copy)
+
+    @classmethod
+    def _from_factorized(cls, values, original):
+        return cls(values)
+
+    _HANDLED_TYPES = (decimal.Decimal, numbers.Number, np.ndarray)
+
+    def to_numpy(
+        self, dtype=None, copy: bool = False, na_value=no_default, decimals=None
+    ) -> np.ndarray:
+        result = np.asarray(self, dtype=dtype)
+        if decimals is not None:
+            result = np.asarray([round(x, decimals) for x in result])
+        return result
+
+    def __array_ufunc__(self, ufunc: np.ufunc, method: str, *inputs, **kwargs):
+        #
+        if not all(
+            isinstance(t, self._HANDLED_TYPES + (DecimalArray,)) for t in inputs
+        ):
+            return NotImplemented
+
+        inputs = tuple(x._data if isinstance(x, DecimalArray) else x for x in inputs)
+        result = getattr(ufunc, method)(*inputs, **kwargs)
+
+        if method == "reduce":
+            result = arraylike.dispatch_reduction_ufunc(
+                self, ufunc, method, *inputs, **kwargs
+            )
+            if result is not NotImplemented:
+                return result
+
+        def reconstruct(x):
+            if isinstance(x, (decimal.Decimal, numbers.Number)):
+                return x
+            else:
+                return DecimalArray._from_sequence(x)
+
+        if ufunc.nout > 1:
+            return tuple(reconstruct(x) for x in result)
+        else:
+            return reconstruct(result)
+
+    def __getitem__(self, item):
+        if isinstance(item, numbers.Integral):
+            return self._data[item]
+        else:
+            # array, slice.
+            item = pd.api.indexers.check_array_indexer(self, item)
+            return type(self)(self._data[item])
+
+    def take(self, indexer, allow_fill=False, fill_value=None):
+        from pandas.api.extensions import take
+
+        data = self._data
+        if allow_fill and fill_value is None:
+            fill_value = self.dtype.na_value
+
+        result = take(data, indexer, fill_value=fill_value, allow_fill=allow_fill)
+        return self._from_sequence(result)
+
+    def copy(self):
+        return type(self)(self._data.copy(), dtype=self.dtype)
+
+    def astype(self, dtype, copy=True):
+        if is_dtype_equal(dtype, self._dtype):
+            if not copy:
+                return self
+        dtype = pandas_dtype(dtype)
+        if isinstance(dtype, type(self.dtype)):
+            return type(self)(self._data, copy=copy, context=dtype.context)
+
+        return super().astype(dtype, copy=copy)
+
+    def __setitem__(self, key, value):
+        if is_list_like(value):
+            if is_scalar(key):
+                raise ValueError("setting an array element with a sequence.")
+            value = [decimal.Decimal(v) for v in value]
+        else:
+            value = decimal.Decimal(value)
+
+        key = check_array_indexer(self, key)
+        self._data[key] = value
+
+    def __len__(self) -> int:
+        return len(self._data)
+
+    def __contains__(self, item) -> bool | np.bool_:
+        if not isinstance(item, decimal.Decimal):
+            return False
+        elif item.is_nan():
+            return self.isna().any()
+        else:
+            return super().__contains__(item)
+
+    @property
+    def nbytes(self) -> int:
+        n = len(self)
+        if n:
+            return n * sys.getsizeof(self[0])
+        return 0
+
+    def isna(self):
+        return np.array([x.is_nan() for x in self._data], dtype=bool)
+
+    @property
+    def _na_value(self):
+        return decimal.Decimal("NaN")
+
+    def _formatter(self, boxed=False):
+        if boxed:
+            return "Decimal: {}".format
+        return repr
+
+    @classmethod
+    def _concat_same_type(cls, to_concat):
+        return cls(np.concatenate([x._data for x in to_concat]))
+
+    def _reduce(self, name: str, *, skipna: bool = True, **kwargs):
+
+        if skipna:
+            # If we don't have any NAs, we can ignore skipna
+            if self.isna().any():
+                other = self[~self.isna()]
+                return other._reduce(name, **kwargs)
+
+        if name == "sum" and len(self) == 0:
+            # GH#29630 avoid returning int 0 or np.bool_(False) on old numpy
+            return decimal.Decimal(0)
+
+        try:
+            op = getattr(self.data, name)
+        except AttributeError as err:
+            raise NotImplementedError(
+                f"decimal does not support the {name} operation"
+            ) from err
+        return op(axis=0)
+
+    def _cmp_method(self, other, op):
+        # For use with OpsMixin
+        def convert_values(param):
+            if isinstance(param, ExtensionArray) or is_list_like(param):
+                ovalues = param
+            else:
+                # Assume it's an object
+                ovalues = [param] * len(self)
+            return ovalues
+
+        lvalues = self
+        rvalues = convert_values(other)
+
+        # If the operator is not defined for the underlying objects,
+        # a TypeError should be raised
+        res = [op(a, b) for (a, b) in zip(lvalues, rvalues)]
+
+        return np.asarray(res, dtype=bool)
+
+    def value_counts(self, dropna: bool = True):
+        from pandas.core.algorithms import value_counts
+
+        return value_counts(self.to_numpy(), dropna=dropna)
+
+
+def to_decimal(values, context=None):
+    return DecimalArray([decimal.Decimal(x) for x in values], context=context)
+
+
+def make_data():
+    return [decimal.Decimal(random.random()) for _ in range(100)]
+
+
+DecimalArray._add_arithmetic_ops()

dist/client/pandas/tests/extension/decimal/test_decimal.py

@@ -0,0 +1,481 @@
+import decimal
+import operator
+
+import numpy as np
+import pytest
+
+import pandas as pd
+import pandas._testing as tm
+from pandas.api.types import infer_dtype
+from pandas.tests.extension import base
+from pandas.tests.extension.decimal.array import (
+    DecimalArray,
+    DecimalDtype,
+    make_data,
+    to_decimal,
+)
+
+
+@pytest.fixture
+def dtype():
+    return DecimalDtype()
+
+
+@pytest.fixture
+def data():
+    return DecimalArray(make_data())
+
+
+@pytest.fixture
+def data_for_twos():
+    return DecimalArray([decimal.Decimal(2) for _ in range(100)])
+
+
+@pytest.fixture
+def data_missing():
+    return DecimalArray([decimal.Decimal("NaN"), decimal.Decimal(1)])
+
+
+@pytest.fixture
+def data_for_sorting():
+    return DecimalArray(
+        [decimal.Decimal("1"), decimal.Decimal("2"), decimal.Decimal("0")]
+    )
+
+
+@pytest.fixture
+def data_missing_for_sorting():
+    return DecimalArray(
+        [decimal.Decimal("1"), decimal.Decimal("NaN"), decimal.Decimal("0")]
+    )
+
+
+@pytest.fixture
+def na_cmp():
+    return lambda x, y: x.is_nan() and y.is_nan()
+
+
+@pytest.fixture
+def na_value():
+    return decimal.Decimal("NaN")
+
+
+@pytest.fixture
+def data_for_grouping():
+    b = decimal.Decimal("1.0")
+    a = decimal.Decimal("0.0")
+    c = decimal.Decimal("2.0")
+    na = decimal.Decimal("NaN")
+    return DecimalArray([b, b, na, na, a, a, b, c])
+
+
+class TestDtype(base.BaseDtypeTests):
+    def test_hashable(self, dtype):
+        pass
+
+    @pytest.mark.parametrize("skipna", [True, False])
+    def test_infer_dtype(self, data, data_missing, skipna):
+        # here overriding base test to ensure we fall back to return
+        # "unknown-array" for an EA pandas doesn't know
+        assert infer_dtype(data, skipna=skipna) == "unknown-array"
+        assert infer_dtype(data_missing, skipna=skipna) == "unknown-array"
+
+
+class TestInterface(base.BaseInterfaceTests):
+    pass
+
+
+class TestConstructors(base.BaseConstructorsTests):
+    pass
+
+
+class TestReshaping(base.BaseReshapingTests):
+    pass
+
+
+class TestGetitem(base.BaseGetitemTests):
+    def test_take_na_value_other_decimal(self):
+        arr = DecimalArray([decimal.Decimal("1.0"), decimal.Decimal("2.0")])
+        result = arr.take([0, -1], allow_fill=True, fill_value=decimal.Decimal("-1.0"))
+        expected = DecimalArray([decimal.Decimal("1.0"), decimal.Decimal("-1.0")])
+        self.assert_extension_array_equal(result, expected)
+
+
+class TestIndex(base.BaseIndexTests):
+    pass
+
+
+class TestMissing(base.BaseMissingTests):
+    pass
+
+
+class Reduce:
+    def check_reduce(self, s, op_name, skipna):
+
+        if op_name in ["median", "skew", "kurt"]:
+            msg = r"decimal does not support the .* operation"
+            with pytest.raises(NotImplementedError, match=msg):
+                getattr(s, op_name)(skipna=skipna)
+
+        else:
+            result = getattr(s, op_name)(skipna=skipna)
+            expected = getattr(np.asarray(s), op_name)()
+            tm.assert_almost_equal(result, expected)
+
+
+class TestNumericReduce(Reduce, base.BaseNumericReduceTests):
+    pass
+
+
+class TestBooleanReduce(Reduce, base.BaseBooleanReduceTests):
+    pass
+
+
+class TestMethods(base.BaseMethodsTests):
+    @pytest.mark.parametrize("dropna", [True, False])
+    def test_value_counts(self, all_data, dropna, request):
+        all_data = all_data[:10]
+        if dropna:
+            other = np.array(all_data[~all_data.isna()])
+        else:
+            other = all_data
+
+        vcs = pd.Series(all_data).value_counts(dropna=dropna)
+        vcs_ex = pd.Series(other).value_counts(dropna=dropna)
+
+        with decimal.localcontext() as ctx:
+            # avoid raising when comparing Decimal("NAN") < Decimal(2)
+            ctx.traps[decimal.InvalidOperation] = False
+
+            result = vcs.sort_index()
+            expected = vcs_ex.sort_index()
+
+        tm.assert_series_equal(result, expected)
+
+
+class TestCasting(base.BaseCastingTests):
+    pass
+
+
+class TestGroupby(base.BaseGroupbyTests):
+    def test_groupby_agg_extension(self, data_for_grouping):
+        super().test_groupby_agg_extension(data_for_grouping)
+
+
+class TestSetitem(base.BaseSetitemTests):
+    pass
+
+
+class TestPrinting(base.BasePrintingTests):
+    def test_series_repr(self, data):
+        # Overriding this base test to explicitly test that
+        # the custom _formatter is used
+        ser = pd.Series(data)
+        assert data.dtype.name in repr(ser)
+        assert "Decimal: " in repr(ser)
+
+
+@pytest.mark.xfail(
+    reason=(
+        "DecimalArray constructor raises bc _from_sequence wants Decimals, not ints."
+        "Easy to fix, just need to do it."
+    ),
+    raises=TypeError,
+)
+def test_series_constructor_coerce_data_to_extension_dtype_raises():
+    xpr = (
+        "Cannot cast data to extension dtype 'decimal'. Pass the "
+        "extension array directly."
+    )
+    with pytest.raises(ValueError, match=xpr):
+        pd.Series([0, 1, 2], dtype=DecimalDtype())
+
+
+def test_series_constructor_with_dtype():
+    arr = DecimalArray([decimal.Decimal("10.0")])
+    result = pd.Series(arr, dtype=DecimalDtype())
+    expected = pd.Series(arr)
+    tm.assert_series_equal(result, expected)
+
+    result = pd.Series(arr, dtype="int64")
+    expected = pd.Series([10])
+    tm.assert_series_equal(result, expected)
+
+
+def test_dataframe_constructor_with_dtype():
+    arr = DecimalArray([decimal.Decimal("10.0")])
+
+    result = pd.DataFrame({"A": arr}, dtype=DecimalDtype())
+    expected = pd.DataFrame({"A": arr})
+    tm.assert_frame_equal(result, expected)
+
+    arr = DecimalArray([decimal.Decimal("10.0")])
+    result = pd.DataFrame({"A": arr}, dtype="int64")
+    expected = pd.DataFrame({"A": [10]})
+    tm.assert_frame_equal(result, expected)
+
+
+@pytest.mark.parametrize("frame", [True, False])
+def test_astype_dispatches(frame):
+    # This is a dtype-specific test that ensures Series[decimal].astype
+    # gets all the way through to ExtensionArray.astype
+    # Designing a reliable smoke test that works for arbitrary data types
+    # is difficult.
+    data = pd.Series(DecimalArray([decimal.Decimal(2)]), name="a")
+    ctx = decimal.Context()
+    ctx.prec = 5
+
+    if frame:
+        data = data.to_frame()
+
+    result = data.astype(DecimalDtype(ctx))
+
+    if frame:
+        result = result["a"]
+
+    assert result.dtype.context.prec == ctx.prec
+
+
+class TestArithmeticOps(base.BaseArithmeticOpsTests):
+    def check_opname(self, s, op_name, other, exc=None):
+        super().check_opname(s, op_name, other, exc=None)
+
+    def test_arith_series_with_array(self, data, all_arithmetic_operators):
+        op_name = all_arithmetic_operators
+        s = pd.Series(data)
+
+        context = decimal.getcontext()
+        divbyzerotrap = context.traps[decimal.DivisionByZero]
+        invalidoptrap = context.traps[decimal.InvalidOperation]
+        context.traps[decimal.DivisionByZero] = 0
+        context.traps[decimal.InvalidOperation] = 0
+
+        # Decimal supports ops with int, but not float
+        other = pd.Series([int(d * 100) for d in data])
+        self.check_opname(s, op_name, other)
+
+        if "mod" not in op_name:
+            self.check_opname(s, op_name, s * 2)
+
+        self.check_opname(s, op_name, 0)
+        self.check_opname(s, op_name, 5)
+        context.traps[decimal.DivisionByZero] = divbyzerotrap
+        context.traps[decimal.InvalidOperation] = invalidoptrap
+
+    def _check_divmod_op(self, s, op, other, exc=NotImplementedError):
+        # We implement divmod
+        super()._check_divmod_op(s, op, other, exc=None)
+
+
+class TestComparisonOps(base.BaseComparisonOpsTests):
+    def test_compare_scalar(self, data, comparison_op):
+        s = pd.Series(data)
+        self._compare_other(s, data, comparison_op, 0.5)
+
+    def test_compare_array(self, data, comparison_op):
+        s = pd.Series(data)
+
+        alter = np.random.choice([-1, 0, 1], len(data))
+        # Randomly double, halve or keep same value
+        other = pd.Series(data) * [decimal.Decimal(pow(2.0, i)) for i in alter]
+        self._compare_other(s, data, comparison_op, other)
+
+
+class DecimalArrayWithoutFromSequence(DecimalArray):
+    """Helper class for testing error handling in _from_sequence."""
+
+    def _from_sequence(cls, scalars, dtype=None, copy=False):
+        raise KeyError("For the test")
+
+
+class DecimalArrayWithoutCoercion(DecimalArrayWithoutFromSequence):
+    @classmethod
+    def _create_arithmetic_method(cls, op):
+        return cls._create_method(op, coerce_to_dtype=False)
+
+
+DecimalArrayWithoutCoercion._add_arithmetic_ops()
+
+
+def test_combine_from_sequence_raises(monkeypatch):
+    # https://github.com/pandas-dev/pandas/issues/22850
+    cls = DecimalArrayWithoutFromSequence
+
+    @classmethod
+    def construct_array_type(cls):
+        return DecimalArrayWithoutFromSequence
+
+    monkeypatch.setattr(DecimalDtype, "construct_array_type", construct_array_type)
+
+    arr = cls([decimal.Decimal("1.0"), decimal.Decimal("2.0")])
+    ser = pd.Series(arr)
+    result = ser.combine(ser, operator.add)
+
+    # note: object dtype
+    expected = pd.Series(
+        [decimal.Decimal("2.0"), decimal.Decimal("4.0")], dtype="object"
+    )
+    tm.assert_series_equal(result, expected)
+
+
+@pytest.mark.parametrize(
+    "class_", [DecimalArrayWithoutFromSequence, DecimalArrayWithoutCoercion]
+)
+def test_scalar_ops_from_sequence_raises(class_):
+    # op(EA, EA) should return an EA, or an ndarray if it's not possible
+    # to return an EA with the return values.
+    arr = class_([decimal.Decimal("1.0"), decimal.Decimal("2.0")])
+    result = arr + arr
+    expected = np.array(
+        [decimal.Decimal("2.0"), decimal.Decimal("4.0")], dtype="object"
+    )
+    tm.assert_numpy_array_equal(result, expected)
+
+
+@pytest.mark.parametrize(
+    "reverse, expected_div, expected_mod",
+    [(False, [0, 1, 1, 2], [1, 0, 1, 0]), (True, [2, 1, 0, 0], [0, 0, 2, 2])],
+)
+def test_divmod_array(reverse, expected_div, expected_mod):
+    # https://github.com/pandas-dev/pandas/issues/22930
+    arr = to_decimal([1, 2, 3, 4])
+    if reverse:
+        div, mod = divmod(2, arr)
+    else:
+        div, mod = divmod(arr, 2)
+    expected_div = to_decimal(expected_div)
+    expected_mod = to_decimal(expected_mod)
+
+    tm.assert_extension_array_equal(div, expected_div)
+    tm.assert_extension_array_equal(mod, expected_mod)
+
+
+def test_ufunc_fallback(data):
+    a = data[:5]
+    s = pd.Series(a, index=range(3, 8))
+    result = np.abs(s)
+    expected = pd.Series(np.abs(a), index=range(3, 8))
+    tm.assert_series_equal(result, expected)
+
+
+def test_array_ufunc():
+    a = to_decimal([1, 2, 3])
+    result = np.exp(a)
+    expected = to_decimal(np.exp(a._data))
+    tm.assert_extension_array_equal(result, expected)
+
+
+def test_array_ufunc_series():
+    a = to_decimal([1, 2, 3])
+    s = pd.Series(a)
+    result = np.exp(s)
+    expected = pd.Series(to_decimal(np.exp(a._data)))
+    tm.assert_series_equal(result, expected)
+
+
+def test_array_ufunc_series_scalar_other():
+    # check _HANDLED_TYPES
+    a = to_decimal([1, 2, 3])
+    s = pd.Series(a)
+    result = np.add(s, decimal.Decimal(1))
+    expected = pd.Series(np.add(a, decimal.Decimal(1)))
+    tm.assert_series_equal(result, expected)
+
+
+def test_array_ufunc_series_defer():
+    a = to_decimal([1, 2, 3])
+    s = pd.Series(a)
+
+    expected = pd.Series(to_decimal([2, 4, 6]))
+    r1 = np.add(s, a)
+    r2 = np.add(a, s)
+
+    tm.assert_series_equal(r1, expected)
+    tm.assert_series_equal(r2, expected)
+
+
+def test_groupby_agg():
+    # Ensure that the result of agg is inferred to be decimal dtype
+    # https://github.com/pandas-dev/pandas/issues/29141
+
+    data = make_data()[:5]
+    df = pd.DataFrame(
+        {"id1": [0, 0, 0, 1, 1], "id2": [0, 1, 0, 1, 1], "decimals": DecimalArray(data)}
+    )
+
+    # single key, selected column
+    expected = pd.Series(to_decimal([data[0], data[3]]))
+    result = df.groupby("id1")["decimals"].agg(lambda x: x.iloc[0])
+    tm.assert_series_equal(result, expected, check_names=False)
+    result = df["decimals"].groupby(df["id1"]).agg(lambda x: x.iloc[0])
+    tm.assert_series_equal(result, expected, check_names=False)
+
+    # multiple keys, selected column
+    expected = pd.Series(
+        to_decimal([data[0], data[1], data[3]]),
+        index=pd.MultiIndex.from_tuples([(0, 0), (0, 1), (1, 1)]),
+    )
+    result = df.groupby(["id1", "id2"])["decimals"].agg(lambda x: x.iloc[0])
+    tm.assert_series_equal(result, expected, check_names=False)
+    result = df["decimals"].groupby([df["id1"], df["id2"]]).agg(lambda x: x.iloc[0])
+    tm.assert_series_equal(result, expected, check_names=False)
+
+    # multiple columns
+    expected = pd.DataFrame({"id2": [0, 1], "decimals": to_decimal([data[0], data[3]])})
+    result = df.groupby("id1").agg(lambda x: x.iloc[0])
+    tm.assert_frame_equal(result, expected, check_names=False)
+
+
+def test_groupby_agg_ea_method(monkeypatch):
+    # Ensure that the result of agg is inferred to be decimal dtype
+    # https://github.com/pandas-dev/pandas/issues/29141
+
+    def DecimalArray__my_sum(self):
+        return np.sum(np.array(self))
+
+    monkeypatch.setattr(DecimalArray, "my_sum", DecimalArray__my_sum, raising=False)
+
+    data = make_data()[:5]
+    df = pd.DataFrame({"id": [0, 0, 0, 1, 1], "decimals": DecimalArray(data)})
+    expected = pd.Series(to_decimal([data[0] + data[1] + data[2], data[3] + data[4]]))
+
+    result = df.groupby("id")["decimals"].agg(lambda x: x.values.my_sum())
+    tm.assert_series_equal(result, expected, check_names=False)
+    s = pd.Series(DecimalArray(data))
+    result = s.groupby(np.array([0, 0, 0, 1, 1])).agg(lambda x: x.values.my_sum())
+    tm.assert_series_equal(result, expected, check_names=False)
+
+
+def test_indexing_no_materialize(monkeypatch):
+    # See https://github.com/pandas-dev/pandas/issues/29708
+    # Ensure that indexing operations do not materialize (convert to a numpy
+    # array) the ExtensionArray unnecessary
+
+    def DecimalArray__array__(self, dtype=None):
+        raise Exception("tried to convert a DecimalArray to a numpy array")
+
+    monkeypatch.setattr(DecimalArray, "__array__", DecimalArray__array__, raising=False)
+
+    data = make_data()
+    s = pd.Series(DecimalArray(data))
+    df = pd.DataFrame({"a": s, "b": range(len(s))})
+
+    # ensure the following operations do not raise an error
+    s[s > 0.5]
+    df[s > 0.5]
+    s.at[0]
+    df.at[0, "a"]
+
+
+def test_to_numpy_keyword():
+    # test the extra keyword
+    values = [decimal.Decimal("1.1111"), decimal.Decimal("2.2222")]
+    expected = np.array(
+        [decimal.Decimal("1.11"), decimal.Decimal("2.22")], dtype="object"
+    )
+    a = pd.array(values, dtype="decimal")
+    result = a.to_numpy(decimals=2)
+    tm.assert_numpy_array_equal(result, expected)
+
+    result = pd.Series(a).to_numpy(decimals=2)
+    tm.assert_numpy_array_equal(result, expected)

dist/client/pandas/tests/extension/json/__init__.py

@@ -0,0 +1,7 @@
+from pandas.tests.extension.json.array import (
+    JSONArray,
+    JSONDtype,
+    make_data,
+)
+
+__all__ = ["JSONArray", "JSONDtype", "make_data"]

dist/client/pandas/tests/extension/json/array.py

@@ -0,0 +1,241 @@
+"""
+Test extension array for storing nested data in a pandas container.
+
+The JSONArray stores lists of dictionaries. The storage mechanism is a list,
+not an ndarray.
+
+Note
+----
+We currently store lists of UserDicts. Pandas has a few places
+internally that specifically check for dicts, and does non-scalar things
+in that case. We *want* the dictionaries to be treated as scalars, so we
+hack around pandas by using UserDicts.
+"""
+from __future__ import annotations
+
+from collections import (
+    UserDict,
+    abc,
+)
+import itertools
+import numbers
+import random
+import string
+import sys
+from typing import (
+    Any,
+    Mapping,
+)
+
+import numpy as np
+
+from pandas._typing import type_t
+
+from pandas.core.dtypes.cast import construct_1d_object_array_from_listlike
+from pandas.core.dtypes.common import (
+    is_bool_dtype,
+    is_list_like,
+    pandas_dtype,
+)
+
+import pandas as pd
+from pandas.api.extensions import (
+    ExtensionArray,
+    ExtensionDtype,
+)
+from pandas.core.indexers import unpack_tuple_and_ellipses
+
+
+class JSONDtype(ExtensionDtype):
+    type = abc.Mapping
+    name = "json"
+    na_value: Mapping[str, Any] = UserDict()
+
+    @classmethod
+    def construct_array_type(cls) -> type_t[JSONArray]:
+        """
+        Return the array type associated with this dtype.
+
+        Returns
+        -------
+        type
+        """
+        return JSONArray
+
+
+class JSONArray(ExtensionArray):
+    dtype = JSONDtype()
+    __array_priority__ = 1000
+
+    def __init__(self, values, dtype=None, copy=False):
+        for val in values:
+            if not isinstance(val, self.dtype.type):
+                raise TypeError("All values must be of type " + str(self.dtype.type))
+        self.data = values
+
+        # Some aliases for common attribute names to ensure pandas supports
+        # these
+        self._items = self._data = self.data
+        # those aliases are currently not working due to assumptions
+        # in internal code (GH-20735)
+        # self._values = self.values = self.data
+
+    @classmethod
+    def _from_sequence(cls, scalars, dtype=None, copy=False):
+        return cls(scalars)
+
+    @classmethod
+    def _from_factorized(cls, values, original):
+        return cls([UserDict(x) for x in values if x != ()])
+
+    def __getitem__(self, item):
+        if isinstance(item, tuple):
+            item = unpack_tuple_and_ellipses(item)
+
+        if isinstance(item, numbers.Integral):
+            return self.data[item]
+        elif isinstance(item, slice) and item == slice(None):
+            # Make sure we get a view
+            return type(self)(self.data)
+        elif isinstance(item, slice):
+            # slice
+            return type(self)(self.data[item])
+        elif not is_list_like(item):
+            # e.g. "foo" or 2.5
+            # exception message copied from numpy
+            raise IndexError(
+                r"only integers, slices (`:`), ellipsis (`...`), numpy.newaxis "
+                r"(`None`) and integer or boolean arrays are valid indices"
+            )
+        else:
+            item = pd.api.indexers.check_array_indexer(self, item)
+            if is_bool_dtype(item.dtype):
+                return self._from_sequence([x for x, m in zip(self, item) if m])
+            # integer
+            return type(self)([self.data[i] for i in item])
+
+    def __setitem__(self, key, value):
+        if isinstance(key, numbers.Integral):
+            self.data[key] = value
+        else:
+            if not isinstance(value, (type(self), abc.Sequence)):
+                # broadcast value
+                value = itertools.cycle([value])
+
+            if isinstance(key, np.ndarray) and key.dtype == "bool":
+                # masking
+                for i, (k, v) in enumerate(zip(key, value)):
+                    if k:
+                        assert isinstance(v, self.dtype.type)
+                        self.data[i] = v
+            else:
+                for k, v in zip(key, value):
+                    assert isinstance(v, self.dtype.type)
+                    self.data[k] = v
+
+    def __len__(self) -> int:
+        return len(self.data)
+
+    def __eq__(self, other):
+        return NotImplemented
+
+    def __ne__(self, other):
+        return NotImplemented
+
+    def __array__(self, dtype=None):
+        if dtype is None:
+            dtype = object
+        return np.asarray(self.data, dtype=dtype)
+
+    @property
+    def nbytes(self) -> int:
+        return sys.getsizeof(self.data)
+
+    def isna(self):
+        return np.array([x == self.dtype.na_value for x in self.data], dtype=bool)
+
+    def take(self, indexer, allow_fill=False, fill_value=None):
+        # re-implement here, since NumPy has trouble setting
+        # sized objects like UserDicts into scalar slots of
+        # an ndarary.
+        indexer = np.asarray(indexer)
+        msg = (
+            "Index is out of bounds or cannot do a "
+            "non-empty take from an empty array."
+        )
+
+        if allow_fill:
+            if fill_value is None:
+                fill_value = self.dtype.na_value
+            # bounds check
+            if (indexer < -1).any():
+                raise ValueError
+            try:
+                output = [
+                    self.data[loc] if loc != -1 else fill_value for loc in indexer
+                ]
+            except IndexError as err:
+                raise IndexError(msg) from err
+        else:
+            try:
+                output = [self.data[loc] for loc in indexer]
+            except IndexError as err:
+                raise IndexError(msg) from err
+
+        return self._from_sequence(output)
+
+    def copy(self):
+        return type(self)(self.data[:])
+
+    def astype(self, dtype, copy=True):
+        # NumPy has issues when all the dicts are the same length.
+        # np.array([UserDict(...), UserDict(...)]) fails,
+        # but np.array([{...}, {...}]) works, so cast.
+        from pandas.core.arrays.string_ import StringDtype
+
+        dtype = pandas_dtype(dtype)
+        # needed to add this check for the Series constructor
+        if isinstance(dtype, type(self.dtype)) and dtype == self.dtype:
+            if copy:
+                return self.copy()
+            return self
+        elif isinstance(dtype, StringDtype):
+            value = self.astype(str)  # numpy doesn'y like nested dicts
+            return dtype.construct_array_type()._from_sequence(value, copy=False)
+
+        return np.array([dict(x) for x in self], dtype=dtype, copy=copy)
+
+    def unique(self):
+        # Parent method doesn't work since np.array will try to infer
+        # a 2-dim object.
+        return type(self)([dict(x) for x in {tuple(d.items()) for d in self.data}])
+
+    @classmethod
+    def _concat_same_type(cls, to_concat):
+        data = list(itertools.chain.from_iterable(x.data for x in to_concat))
+        return cls(data)
+
+    def _values_for_factorize(self):
+        frozen = self._values_for_argsort()
+        if len(frozen) == 0:
+            # factorize_array expects 1-d array, this is a len-0 2-d array.
+            frozen = frozen.ravel()
+        return frozen, ()
+
+    def _values_for_argsort(self):
+        # Bypass NumPy's shape inference to get a (N,) array of tuples.
+        frozen = [tuple(x.items()) for x in self]
+        return construct_1d_object_array_from_listlike(frozen)
+
+
+def make_data():
+    # TODO: Use a regular dict. See _NDFrameIndexer._setitem_with_indexer
+    return [
+        UserDict(
+            [
+                (random.choice(string.ascii_letters), random.randint(0, 100))
+                for _ in range(random.randint(0, 10))
+            ]
+        )
+        for _ in range(100)
+    ]

dist/client/pandas/tests/extension/json/test_json.py

@@ -0,0 +1,371 @@
+import collections
+import operator
+import sys
+
+import pytest
+
+import pandas as pd
+import pandas._testing as tm
+from pandas.tests.extension import base
+from pandas.tests.extension.json.array import (
+    JSONArray,
+    JSONDtype,
+    make_data,
+)
+
+
+@pytest.fixture
+def dtype():
+    return JSONDtype()
+
+
+@pytest.fixture
+def data():
+    """Length-100 PeriodArray for semantics test."""
+    data = make_data()
+
+    # Why the while loop? NumPy is unable to construct an ndarray from
+    # equal-length ndarrays. Many of our operations involve coercing the
+    # EA to an ndarray of objects. To avoid random test failures, we ensure
+    # that our data is coercible to an ndarray. Several tests deal with only
+    # the first two elements, so that's what we'll check.
+
+    while len(data[0]) == len(data[1]):
+        data = make_data()
+
+    return JSONArray(data)
+
+
+@pytest.fixture
+def data_missing():
+    """Length 2 array with [NA, Valid]"""
+    return JSONArray([{}, {"a": 10}])
+
+
+@pytest.fixture
+def data_for_sorting():
+    return JSONArray([{"b": 1}, {"c": 4}, {"a": 2, "c": 3}])
+
+
+@pytest.fixture
+def data_missing_for_sorting():
+    return JSONArray([{"b": 1}, {}, {"a": 4}])
+
+
+@pytest.fixture
+def na_value(dtype):
+    return dtype.na_value
+
+
+@pytest.fixture
+def na_cmp():
+    return operator.eq
+
+
+@pytest.fixture
+def data_for_grouping():
+    return JSONArray(
+        [
+            {"b": 1},
+            {"b": 1},
+            {},
+            {},
+            {"a": 0, "c": 2},
+            {"a": 0, "c": 2},
+            {"b": 1},
+            {"c": 2},
+        ]
+    )
+
+
+class BaseJSON:
+    # NumPy doesn't handle an array of equal-length UserDicts.
+    # The default assert_series_equal eventually does a
+    # Series.values, which raises. We work around it by
+    # converting the UserDicts to dicts.
+    @classmethod
+    def assert_series_equal(cls, left, right, *args, **kwargs):
+        if left.dtype.name == "json":
+            assert left.dtype == right.dtype
+            left = pd.Series(
+                JSONArray(left.values.astype(object)), index=left.index, name=left.name
+            )
+            right = pd.Series(
+                JSONArray(right.values.astype(object)),
+                index=right.index,
+                name=right.name,
+            )
+        tm.assert_series_equal(left, right, *args, **kwargs)
+
+    @classmethod
+    def assert_frame_equal(cls, left, right, *args, **kwargs):
+        obj_type = kwargs.get("obj", "DataFrame")
+        tm.assert_index_equal(
+            left.columns,
+            right.columns,
+            exact=kwargs.get("check_column_type", "equiv"),
+            check_names=kwargs.get("check_names", True),
+            check_exact=kwargs.get("check_exact", False),
+            check_categorical=kwargs.get("check_categorical", True),
+            obj=f"{obj_type}.columns",
+        )
+
+        jsons = (left.dtypes == "json").index
+
+        for col in jsons:
+            cls.assert_series_equal(left[col], right[col], *args, **kwargs)
+
+        left = left.drop(columns=jsons)
+        right = right.drop(columns=jsons)
+        tm.assert_frame_equal(left, right, *args, **kwargs)
+
+
+class TestDtype(BaseJSON, base.BaseDtypeTests):
+    pass
+
+
+class TestInterface(BaseJSON, base.BaseInterfaceTests):
+    def test_custom_asserts(self):
+        # This would always trigger the KeyError from trying to put
+        # an array of equal-length UserDicts inside an ndarray.
+        data = JSONArray(
+            [
+                collections.UserDict({"a": 1}),
+                collections.UserDict({"b": 2}),
+                collections.UserDict({"c": 3}),
+            ]
+        )
+        a = pd.Series(data)
+        self.assert_series_equal(a, a)
+        self.assert_frame_equal(a.to_frame(), a.to_frame())
+
+        b = pd.Series(data.take([0, 0, 1]))
+        msg = r"ExtensionArray are different"
+        with pytest.raises(AssertionError, match=msg):
+            self.assert_series_equal(a, b)
+
+        with pytest.raises(AssertionError, match=msg):
+            self.assert_frame_equal(a.to_frame(), b.to_frame())
+
+    @pytest.mark.xfail(
+        reason="comparison method not implemented for JSONArray (GH-37867)"
+    )
+    def test_contains(self, data):
+        # GH-37867
+        super().test_contains(data)
+
+
+class TestConstructors(BaseJSON, base.BaseConstructorsTests):
+    @pytest.mark.xfail(reason="not implemented constructor from dtype")
+    def test_from_dtype(self, data):
+        # construct from our dtype & string dtype
+        super(self).test_from_dtype(data)
+
+    @pytest.mark.xfail(reason="RecursionError, GH-33900")
+    def test_series_constructor_no_data_with_index(self, dtype, na_value):
+        # RecursionError: maximum recursion depth exceeded in comparison
+        rec_limit = sys.getrecursionlimit()
+        try:
+            # Limit to avoid stack overflow on Windows CI
+            sys.setrecursionlimit(100)
+            super().test_series_constructor_no_data_with_index(dtype, na_value)
+        finally:
+            sys.setrecursionlimit(rec_limit)
+
+    @pytest.mark.xfail(reason="RecursionError, GH-33900")
+    def test_series_constructor_scalar_na_with_index(self, dtype, na_value):
+        # RecursionError: maximum recursion depth exceeded in comparison
+        rec_limit = sys.getrecursionlimit()
+        try:
+            # Limit to avoid stack overflow on Windows CI
+            sys.setrecursionlimit(100)
+            super().test_series_constructor_scalar_na_with_index(dtype, na_value)
+        finally:
+            sys.setrecursionlimit(rec_limit)
+
+    @pytest.mark.xfail(reason="collection as scalar, GH-33901")
+    def test_series_constructor_scalar_with_index(self, data, dtype):
+        # TypeError: All values must be of type <class 'collections.abc.Mapping'>
+        super().test_series_constructor_scalar_with_index(data, dtype)
+
+
+class TestReshaping(BaseJSON, base.BaseReshapingTests):
+    @pytest.mark.skip(reason="Different definitions of NA")
+    def test_stack(self):
+        """
+        The test does .astype(object).stack(). If we happen to have
+        any missing values in `data`, then we'll end up with different
+        rows since we consider `{}` NA, but `.astype(object)` doesn't.
+        """
+
+    @pytest.mark.xfail(reason="dict for NA")
+    def test_unstack(self, data, index):
+        # The base test has NaN for the expected NA value.
+        # this matches otherwise
+        return super().test_unstack(data, index)
+
+
+class TestGetitem(BaseJSON, base.BaseGetitemTests):
+    pass
+
+
+class TestIndex(BaseJSON, base.BaseIndexTests):
+    pass
+
+
+class TestMissing(BaseJSON, base.BaseMissingTests):
+    @pytest.mark.skip(reason="Setting a dict as a scalar")
+    def test_fillna_series(self):
+        """We treat dictionaries as a mapping in fillna, not a scalar."""
+
+    @pytest.mark.skip(reason="Setting a dict as a scalar")
+    def test_fillna_frame(self):
+        """We treat dictionaries as a mapping in fillna, not a scalar."""
+
+
+unhashable = pytest.mark.skip(reason="Unhashable")
+
+
+class TestReduce(base.BaseNoReduceTests):
+    pass
+
+
+class TestMethods(BaseJSON, base.BaseMethodsTests):
+    @unhashable
+    def test_value_counts(self, all_data, dropna):
+        pass
+
+    @unhashable
+    def test_value_counts_with_normalize(self, data):
+        pass
+
+    @unhashable
+    def test_sort_values_frame(self):
+        # TODO (EA.factorize): see if _values_for_factorize allows this.
+        pass
+
+    @pytest.mark.parametrize("ascending", [True, False])
+    def test_sort_values(self, data_for_sorting, ascending, sort_by_key):
+        super().test_sort_values(data_for_sorting, ascending, sort_by_key)
+
+    @pytest.mark.parametrize("ascending", [True, False])
+    def test_sort_values_missing(
+        self, data_missing_for_sorting, ascending, sort_by_key
+    ):
+        super().test_sort_values_missing(
+            data_missing_for_sorting, ascending, sort_by_key
+        )
+
+    @pytest.mark.skip(reason="combine for JSONArray not supported")
+    def test_combine_le(self, data_repeated):
+        pass
+
+    @pytest.mark.skip(reason="combine for JSONArray not supported")
+    def test_combine_add(self, data_repeated):
+        pass
+
+    @pytest.mark.skip(reason="combine for JSONArray not supported")
+    def test_combine_first(self, data):
+        pass
+
+    @unhashable
+    def test_hash_pandas_object_works(self, data, kind):
+        super().test_hash_pandas_object_works(data, kind)
+
+    @pytest.mark.skip(reason="broadcasting error")
+    def test_where_series(self, data, na_value):
+        # Fails with
+        # *** ValueError: operands could not be broadcast together
+        # with shapes (4,) (4,) (0,)
+        super().test_where_series(data, na_value)
+
+    @pytest.mark.skip(reason="Can't compare dicts.")
+    def test_searchsorted(self, data_for_sorting):
+        super().test_searchsorted(data_for_sorting)
+
+    @pytest.mark.skip(reason="Can't compare dicts.")
+    def test_equals(self, data, na_value, as_series):
+        pass
+
+
+class TestCasting(BaseJSON, base.BaseCastingTests):
+    @pytest.mark.skip(reason="failing on np.array(self, dtype=str)")
+    def test_astype_str(self):
+        """This currently fails in NumPy on np.array(self, dtype=str) with
+
+        *** ValueError: setting an array element with a sequence
+        """
+
+
+# We intentionally don't run base.BaseSetitemTests because pandas'
+# internals has trouble setting sequences of values into scalar positions.
+
+
+class TestGroupby(BaseJSON, base.BaseGroupbyTests):
+    @unhashable
+    def test_groupby_extension_transform(self):
+        """
+        This currently fails in Series.name.setter, since the
+        name must be hashable, but the value is a dictionary.
+        I think this is what we want, i.e. `.name` should be the original
+        values, and not the values for factorization.
+        """
+
+    @unhashable
+    def test_groupby_extension_apply(self):
+        """
+        This fails in Index._do_unique_check with
+
+        >   hash(val)
+        E   TypeError: unhashable type: 'UserDict' with
+
+        I suspect that once we support Index[ExtensionArray],
+        we'll be able to dispatch unique.
+        """
+
+    @unhashable
+    def test_groupby_extension_agg(self):
+        """
+        This fails when we get to tm.assert_series_equal when left.index
+        contains dictionaries, which are not hashable.
+        """
+
+    @unhashable
+    def test_groupby_extension_no_sort(self):
+        """
+        This fails when we get to tm.assert_series_equal when left.index
+        contains dictionaries, which are not hashable.
+        """
+
+    @pytest.mark.xfail(reason="GH#39098: Converts agg result to object")
+    def test_groupby_agg_extension(self, data_for_grouping):
+        super().test_groupby_agg_extension(data_for_grouping)
+
+
+class TestArithmeticOps(BaseJSON, base.BaseArithmeticOpsTests):
+    def test_arith_frame_with_scalar(self, data, all_arithmetic_operators, request):
+        if len(data[0]) != 1:
+            mark = pytest.mark.xfail(reason="raises in coercing to Series")
+            request.node.add_marker(mark)
+        super().test_arith_frame_with_scalar(data, all_arithmetic_operators)
+
+    def test_add_series_with_extension_array(self, data):
+        ser = pd.Series(data)
+        with pytest.raises(TypeError, match="unsupported"):
+            ser + data
+
+    def test_divmod_series_array(self):
+        # GH 23287
+        # skipping because it is not implemented
+        pass
+
+    def _check_divmod_op(self, s, op, other, exc=NotImplementedError):
+        return super()._check_divmod_op(s, op, other, exc=TypeError)
+
+
+class TestComparisonOps(BaseJSON, base.BaseComparisonOpsTests):
+    pass
+
+
+class TestPrinting(BaseJSON, base.BasePrintingTests):
+    pass

dist/client/pandas/tests/extension/list/__init__.py

@@ -0,0 +1,7 @@
+from pandas.tests.extension.list.array import (
+    ListArray,
+    ListDtype,
+    make_data,
+)
+
+__all__ = ["ListArray", "ListDtype", "make_data"]

dist/client/pandas/tests/extension/list/array.py

@@ -0,0 +1,132 @@
+"""
+Test extension array for storing nested data in a pandas container.
+
+The ListArray stores an ndarray of lists.
+"""
+from __future__ import annotations
+
+import numbers
+import random
+import string
+
+import numpy as np
+
+from pandas._typing import type_t
+
+from pandas.core.dtypes.base import ExtensionDtype
+
+import pandas as pd
+from pandas.api.types import (
+    is_object_dtype,
+    is_string_dtype,
+)
+from pandas.core.arrays import ExtensionArray
+
+
+class ListDtype(ExtensionDtype):
+    type = list
+    name = "list"
+    na_value = np.nan
+
+    @classmethod
+    def construct_array_type(cls) -> type_t[ListArray]:
+        """
+        Return the array type associated with this dtype.
+
+        Returns
+        -------
+        type
+        """
+        return ListArray
+
+
+class ListArray(ExtensionArray):
+    dtype = ListDtype()
+    __array_priority__ = 1000
+
+    def __init__(self, values, dtype=None, copy=False):
+        if not isinstance(values, np.ndarray):
+            raise TypeError("Need to pass a numpy array as values")
+        for val in values:
+            if not isinstance(val, self.dtype.type) and not pd.isna(val):
+                raise TypeError("All values must be of type " + str(self.dtype.type))
+        self.data = values
+
+    @classmethod
+    def _from_sequence(cls, scalars, dtype=None, copy=False):
+        data = np.empty(len(scalars), dtype=object)
+        data[:] = scalars
+        return cls(data)
+
+    def __getitem__(self, item):
+        if isinstance(item, numbers.Integral):
+            return self.data[item]
+        else:
+            # slice, list-like, mask
+            return type(self)(self.data[item])
+
+    def __len__(self) -> int:
+        return len(self.data)
+
+    def isna(self):
+        return np.array(
+            [not isinstance(x, list) and np.isnan(x) for x in self.data], dtype=bool
+        )
+
+    def take(self, indexer, allow_fill=False, fill_value=None):
+        # re-implement here, since NumPy has trouble setting
+        # sized objects like UserDicts into scalar slots of
+        # an ndarary.
+        indexer = np.asarray(indexer)
+        msg = (
+            "Index is out of bounds or cannot do a "
+            "non-empty take from an empty array."
+        )
+
+        if allow_fill:
+            if fill_value is None:
+                fill_value = self.dtype.na_value
+            # bounds check
+            if (indexer < -1).any():
+                raise ValueError
+            try:
+                output = [
+                    self.data[loc] if loc != -1 else fill_value for loc in indexer
+                ]
+            except IndexError as err:
+                raise IndexError(msg) from err
+        else:
+            try:
+                output = [self.data[loc] for loc in indexer]
+            except IndexError as err:
+                raise IndexError(msg) from err
+
+        return self._from_sequence(output)
+
+    def copy(self):
+        return type(self)(self.data[:])
+
+    def astype(self, dtype, copy=True):
+        if isinstance(dtype, type(self.dtype)) and dtype == self.dtype:
+            if copy:
+                return self.copy()
+            return self
+        elif is_string_dtype(dtype) and not is_object_dtype(dtype):
+            # numpy has problems with astype(str) for nested elements
+            return np.array([str(x) for x in self.data], dtype=dtype)
+        return np.array(self.data, dtype=dtype, copy=copy)
+
+    @classmethod
+    def _concat_same_type(cls, to_concat):
+        data = np.concatenate([x.data for x in to_concat])
+        return cls(data)
+
+
+def make_data():
+    # TODO: Use a regular dict. See _NDFrameIndexer._setitem_with_indexer
+    data = np.empty(100, dtype=object)
+    data[:] = [
+        [random.choice(string.ascii_letters) for _ in range(random.randint(0, 10))]
+        for _ in range(100)
+    ]
+    return data

dist/client/pandas/tests/extension/list/test_list.py

@@ -0,0 +1,33 @@
+import pytest
+
+import pandas as pd
+from pandas.tests.extension.list.array import (
+    ListArray,
+    ListDtype,
+    make_data,
+)
+
+
+@pytest.fixture
+def dtype():
+    return ListDtype()
+
+
+@pytest.fixture
+def data():
+    """Length-100 ListArray for semantics test."""
+    data = make_data()
+
+    while len(data[0]) == len(data[1]):
+        data = make_data()
+
+    return ListArray(data)
+
+
+def test_to_csv(data):
+    # https://github.com/pandas-dev/pandas/issues/28840
+    # array with list-likes fail when doing astype(str) on the numpy array
+    # which was done in to_native_types
+    df = pd.DataFrame({"a": data})
+    res = df.to_csv()
+    assert str(data[0]) in res

dist/client/pandas/tests/extension/test_boolean.py

@@ -0,0 +1,394 @@
+"""
+This file contains a minimal set of tests for compliance with the extension
+array interface test suite, and should contain no other tests.
+The test suite for the full functionality of the array is located in
+`pandas/tests/arrays/`.
+
+The tests in this file are inherited from the BaseExtensionTests, and only
+minimal tweaks should be applied to get the tests passing (by overwriting a
+parent method).
+
+Additional tests should either be added to one of the BaseExtensionTests
+classes (if they are relevant for the extension interface for all dtypes), or
+be added to the array-specific tests in `pandas/tests/arrays/`.
+
+"""
+import numpy as np
+import pytest
+
+import pandas as pd
+import pandas._testing as tm
+from pandas.core.arrays.boolean import BooleanDtype
+from pandas.tests.extension import base
+
+
+def make_data():
+    return [True, False] * 4 + [np.nan] + [True, False] * 44 + [np.nan] + [True, False]
+
+
+@pytest.fixture
+def dtype():
+    return BooleanDtype()
+
+
+@pytest.fixture
+def data(dtype):
+    return pd.array(make_data(), dtype=dtype)
+
+
+@pytest.fixture
+def data_for_twos(dtype):
+    return pd.array(np.ones(100), dtype=dtype)
+
+
+@pytest.fixture
+def data_missing(dtype):
+    return pd.array([np.nan, True], dtype=dtype)
+
+
+@pytest.fixture
+def data_for_sorting(dtype):
+    return pd.array([True, True, False], dtype=dtype)
+
+
+@pytest.fixture
+def data_missing_for_sorting(dtype):
+    return pd.array([True, np.nan, False], dtype=dtype)
+
+
+@pytest.fixture
+def na_cmp():
+    # we are pd.NA
+    return lambda x, y: x is pd.NA and y is pd.NA
+
+
+@pytest.fixture
+def na_value():
+    return pd.NA
+
+
+@pytest.fixture
+def data_for_grouping(dtype):
+    b = True
+    a = False
+    na = np.nan
+    return pd.array([b, b, na, na, a, a, b], dtype=dtype)
+
+
+class TestDtype(base.BaseDtypeTests):
+    pass
+
+
+class TestInterface(base.BaseInterfaceTests):
+    pass
+
+
+class TestConstructors(base.BaseConstructorsTests):
+    pass
+
+
+class TestGetitem(base.BaseGetitemTests):
+    pass
+
+
+class TestSetitem(base.BaseSetitemTests):
+    pass
+
+
+class TestIndex(base.BaseIndexTests):
+    pass
+
+
+class TestMissing(base.BaseMissingTests):
+    pass
+
+
+class TestArithmeticOps(base.BaseArithmeticOpsTests):
+    implements = {"__sub__", "__rsub__"}
+
+    def check_opname(self, s, op_name, other, exc=None):
+        # overwriting to indicate ops don't raise an error
+        super().check_opname(s, op_name, other, exc=None)
+
+    def _check_op(self, obj, op, other, op_name, exc=NotImplementedError):
+        if exc is None:
+            if op_name in self.implements:
+                msg = r"numpy boolean subtract"
+                with pytest.raises(TypeError, match=msg):
+                    op(obj, other)
+                return
+
+            result = op(obj, other)
+            expected = self._combine(obj, other, op)
+
+            if op_name in (
+                "__floordiv__",
+                "__rfloordiv__",
+                "__pow__",
+                "__rpow__",
+                "__mod__",
+                "__rmod__",
+            ):
+                # combine keeps boolean type
+                expected = expected.astype("Int8")
+            elif op_name in ("__truediv__", "__rtruediv__"):
+                # combine with bools does not generate the correct result
+                #  (numpy behaviour for div is to regard the bools as numeric)
+                expected = self._combine(obj.astype(float), other, op)
+                expected = expected.astype("Float64")
+            if op_name == "__rpow__":
+                # for rpow, combine does not propagate NaN
+                expected[result.isna()] = np.nan
+            self.assert_equal(result, expected)
+        else:
+            with pytest.raises(exc):
+                op(obj, other)
+
+    def _check_divmod_op(self, s, op, other, exc=None):
+        # override to not raise an error
+        super()._check_divmod_op(s, op, other, None)
+
+
+class TestComparisonOps(base.BaseComparisonOpsTests):
+    def check_opname(self, s, op_name, other, exc=None):
+        # overwriting to indicate ops don't raise an error
+        super().check_opname(s, op_name, other, exc=None)
+
+
+class TestReshaping(base.BaseReshapingTests):
+    pass
+
+
+class TestMethods(base.BaseMethodsTests):
+    @pytest.mark.parametrize("na_sentinel", [-1, -2])
+    def test_factorize(self, data_for_grouping, na_sentinel):
+        # override because we only have 2 unique values
+        labels, uniques = pd.factorize(data_for_grouping, na_sentinel=na_sentinel)
+        expected_labels = np.array(
+            [0, 0, na_sentinel, na_sentinel, 1, 1, 0], dtype=np.intp
+        )
+        expected_uniques = data_for_grouping.take([0, 4])
+
+        tm.assert_numpy_array_equal(labels, expected_labels)
+        self.assert_extension_array_equal(uniques, expected_uniques)
+
+    def test_combine_le(self, data_repeated):
+        # override because expected needs to be boolean instead of bool dtype
+        orig_data1, orig_data2 = data_repeated(2)
+        s1 = pd.Series(orig_data1)
+        s2 = pd.Series(orig_data2)
+        result = s1.combine(s2, lambda x1, x2: x1 <= x2)
+        expected = pd.Series(
+            [a <= b for (a, b) in zip(list(orig_data1), list(orig_data2))],
+            dtype="boolean",
+        )
+        self.assert_series_equal(result, expected)
+
+        val = s1.iloc[0]
+        result = s1.combine(val, lambda x1, x2: x1 <= x2)
+        expected = pd.Series([a <= val for a in list(orig_data1)], dtype="boolean")
+        self.assert_series_equal(result, expected)
+
+    def test_searchsorted(self, data_for_sorting, as_series):
+        # override because we only have 2 unique values
+        data_for_sorting = pd.array([True, False], dtype="boolean")
+        b, a = data_for_sorting
+        arr = type(data_for_sorting)._from_sequence([a, b])
+
+        if as_series:
+            arr = pd.Series(arr)
+        assert arr.searchsorted(a) == 0
+        assert arr.searchsorted(a, side="right") == 1
+
+        assert arr.searchsorted(b) == 1
+        assert arr.searchsorted(b, side="right") == 2
+
+        result = arr.searchsorted(arr.take([0, 1]))
+        expected = np.array([0, 1], dtype=np.intp)
+
+        tm.assert_numpy_array_equal(result, expected)
+
+        # sorter
+        sorter = np.array([1, 0])
+        assert data_for_sorting.searchsorted(a, sorter=sorter) == 0
+
+    @pytest.mark.xfail(reason="uses nullable integer")
+    def test_value_counts(self, all_data, dropna):
+        return super().test_value_counts(all_data, dropna)
+
+    @pytest.mark.xfail(reason="uses nullable integer")
+    def test_value_counts_with_normalize(self, data):
+        super().test_value_counts_with_normalize(data)
+
+    def test_argmin_argmax(self, data_for_sorting, data_missing_for_sorting):
+        # override because there are only 2 unique values
+
+        # data_for_sorting -> [B, C, A] with A < B < C -> here True, True, False
+        assert data_for_sorting.argmax() == 0
+        assert data_for_sorting.argmin() == 2
+
+        # with repeated values -> first occurrence
+        data = data_for_sorting.take([2, 0, 0, 1, 1, 2])
+        assert data.argmax() == 1
+        assert data.argmin() == 0
+
+        # with missing values
+        # data_missing_for_sorting -> [B, NA, A] with A < B and NA missing.
+        assert data_missing_for_sorting.argmax() == 0
+        assert data_missing_for_sorting.argmin() == 2
+
+
+class TestCasting(base.BaseCastingTests):
+    pass
+
+
+class TestGroupby(base.BaseGroupbyTests):
+    """
+    Groupby-specific tests are overridden because boolean only has 2
+    unique values, base tests uses 3 groups.
+    """
+
+    def test_grouping_grouper(self, data_for_grouping):
+        df = pd.DataFrame(
+            {"A": ["B", "B", None, None, "A", "A", "B"], "B": data_for_grouping}
+        )
+        gr1 = df.groupby("A").grouper.groupings[0]
+        gr2 = df.groupby("B").grouper.groupings[0]
+
+        tm.assert_numpy_array_equal(gr1.grouping_vector, df.A.values)
+        tm.assert_extension_array_equal(gr2.grouping_vector, data_for_grouping)
+
+    @pytest.mark.parametrize("as_index", [True, False])
+    def test_groupby_extension_agg(self, as_index, data_for_grouping):
+        df = pd.DataFrame({"A": [1, 1, 2, 2, 3, 3, 1], "B": data_for_grouping})
+        result = df.groupby("B", as_index=as_index).A.mean()
+        _, uniques = pd.factorize(data_for_grouping, sort=True)
+
+        if as_index:
+            index = pd.Index(uniques, name="B")
+            expected = pd.Series([3.0, 1.0], index=index, name="A")
+            self.assert_series_equal(result, expected)
+        else:
+            expected = pd.DataFrame({"B": uniques, "A": [3.0, 1.0]})
+            self.assert_frame_equal(result, expected)
+
+    def test_groupby_agg_extension(self, data_for_grouping):
+        # GH#38980 groupby agg on extension type fails for non-numeric types
+        df = pd.DataFrame({"A": [1, 1, 2, 2, 3, 3, 1], "B": data_for_grouping})
+
+        expected = df.iloc[[0, 2, 4]]
+        expected = expected.set_index("A")
+
+        result = df.groupby("A").agg({"B": "first"})
+        self.assert_frame_equal(result, expected)
+
+        result = df.groupby("A").agg("first")
+        self.assert_frame_equal(result, expected)
+
+        result = df.groupby("A").first()
+        self.assert_frame_equal(result, expected)
+
+    def test_groupby_extension_no_sort(self, data_for_grouping):
+        df = pd.DataFrame({"A": [1, 1, 2, 2, 3, 3, 1], "B": data_for_grouping})
+        result = df.groupby("B", sort=False).A.mean()
+        _, index = pd.factorize(data_for_grouping, sort=False)
+
+        index = pd.Index(index, name="B")
+        expected = pd.Series([1.0, 3.0], index=index, name="A")
+        self.assert_series_equal(result, expected)
+
+    def test_groupby_extension_transform(self, data_for_grouping):
+        valid = data_for_grouping[~data_for_grouping.isna()]
+        df = pd.DataFrame({"A": [1, 1, 3, 3, 1], "B": valid})
+
+        result = df.groupby("B").A.transform(len)
+        expected = pd.Series([3, 3, 2, 2, 3], name="A")
+
+        self.assert_series_equal(result, expected)
+
+    def test_groupby_extension_apply(self, data_for_grouping, groupby_apply_op):
+        df = pd.DataFrame({"A": [1, 1, 2, 2, 3, 3, 1], "B": data_for_grouping})
+        df.groupby("B").apply(groupby_apply_op)
+        df.groupby("B").A.apply(groupby_apply_op)
+        df.groupby("A").apply(groupby_apply_op)
+        df.groupby("A").B.apply(groupby_apply_op)
+
+    def test_groupby_apply_identity(self, data_for_grouping):
+        df = pd.DataFrame({"A": [1, 1, 2, 2, 3, 3, 1], "B": data_for_grouping})
+        result = df.groupby("A").B.apply(lambda x: x.array)
+        expected = pd.Series(
+            [
+                df.B.iloc[[0, 1, 6]].array,
+                df.B.iloc[[2, 3]].array,
+                df.B.iloc[[4, 5]].array,
+            ],
+            index=pd.Index([1, 2, 3], name="A"),
+            name="B",
+        )
+        self.assert_series_equal(result, expected)
+
+    def test_in_numeric_groupby(self, data_for_grouping):
+        df = pd.DataFrame(
+            {
+                "A": [1, 1, 2, 2, 3, 3, 1],
+                "B": data_for_grouping,
+                "C": [1, 1, 1, 1, 1, 1, 1],
+            }
+        )
+        result = df.groupby("A").sum().columns
+
+        if data_for_grouping.dtype._is_numeric:
+            expected = pd.Index(["B", "C"])
+        else:
+            expected = pd.Index(["C"])
+
+        tm.assert_index_equal(result, expected)
+
+    @pytest.mark.parametrize("min_count", [0, 10])
+    def test_groupby_sum_mincount(self, data_for_grouping, min_count):
+        df = pd.DataFrame({"A": [1, 1, 2, 2, 3, 3, 1], "B": data_for_grouping})
+        result = df.groupby("A").sum(min_count=min_count)
+        if min_count == 0:
+            expected = pd.DataFrame(
+                {"B": pd.array([3, 0, 0], dtype="Int64")},
+                index=pd.Index([1, 2, 3], name="A"),
+            )
+            tm.assert_frame_equal(result, expected)
+        else:
+            expected = pd.DataFrame(
+                {"B": pd.array([pd.NA] * 3, dtype="Int64")},
+                index=pd.Index([1, 2, 3], name="A"),
+            )
+            tm.assert_frame_equal(result, expected)
+
+
+class TestNumericReduce(base.BaseNumericReduceTests):
+    def check_reduce(self, s, op_name, skipna):
+        result = getattr(s, op_name)(skipna=skipna)
+        expected = getattr(s.astype("float64"), op_name)(skipna=skipna)
+        # override parent function to cast to bool for min/max
+        if np.isnan(expected):
+            expected = pd.NA
+        elif op_name in ("min", "max"):
+            expected = bool(expected)
+        tm.assert_almost_equal(result, expected)
+
+
+class TestBooleanReduce(base.BaseBooleanReduceTests):
+    pass
+
+
+class TestPrinting(base.BasePrintingTests):
+    pass
+
+
+class TestUnaryOps(base.BaseUnaryOpsTests):
+    pass
+
+
+class TestParsing(base.BaseParsingTests):
+    pass
+
+
+class Test2DCompat(base.Dim2CompatTests):
+    pass

dist/client/pandas/tests/extension/test_categorical.py

@@ -0,0 +1,310 @@
+"""
+This file contains a minimal set of tests for compliance with the extension
+array interface test suite, and should contain no other tests.
+The test suite for the full functionality of the array is located in
+`pandas/tests/arrays/`.
+
+The tests in this file are inherited from the BaseExtensionTests, and only
+minimal tweaks should be applied to get the tests passing (by overwriting a
+parent method).
+
+Additional tests should either be added to one of the BaseExtensionTests
+classes (if they are relevant for the extension interface for all dtypes), or
+be added to the array-specific tests in `pandas/tests/arrays/`.
+
+"""
+import string
+
+import numpy as np
+import pytest
+
+import pandas as pd
+from pandas import (
+    Categorical,
+    CategoricalIndex,
+    Timestamp,
+)
+import pandas._testing as tm
+from pandas.api.types import CategoricalDtype
+from pandas.tests.extension import base
+
+
+def make_data():
+    while True:
+        values = np.random.choice(list(string.ascii_letters), size=100)
+        # ensure we meet the requirements
+        # 1. first two not null
+        # 2. first and second are different
+        if values[0] != values[1]:
+            break
+    return values
+
+
+@pytest.fixture
+def dtype():
+    return CategoricalDtype()
+
+
+@pytest.fixture
+def data():
+    """Length-100 array for this type.
+
+    * data[0] and data[1] should both be non missing
+    * data[0] and data[1] should not be equal
+    """
+    return Categorical(make_data())
+
+
+@pytest.fixture
+def data_missing():
+    """Length 2 array with [NA, Valid]"""
+    return Categorical([np.nan, "A"])
+
+
+@pytest.fixture
+def data_for_sorting():
+    return Categorical(["A", "B", "C"], categories=["C", "A", "B"], ordered=True)
+
+
+@pytest.fixture
+def data_missing_for_sorting():
+    return Categorical(["A", None, "B"], categories=["B", "A"], ordered=True)
+
+
+@pytest.fixture
+def na_value():
+    return np.nan
+
+
+@pytest.fixture
+def data_for_grouping():
+    return Categorical(["a", "a", None, None, "b", "b", "a", "c"])
+
+
+class TestDtype(base.BaseDtypeTests):
+    pass
+
+
+class TestInterface(base.BaseInterfaceTests):
+    @pytest.mark.xfail(reason="Memory usage doesn't match")
+    def test_memory_usage(self, data):
+        # Is this deliberate?
+        super().test_memory_usage(data)
+
+    def test_contains(self, data, data_missing):
+        # GH-37867
+        # na value handling in Categorical.__contains__ is deprecated.
+        # See base.BaseInterFaceTests.test_contains for more details.
+
+        na_value = data.dtype.na_value
+        # ensure data without missing values
+        data = data[~data.isna()]
+
+        # first elements are non-missing
+        assert data[0] in data
+        assert data_missing[0] in data_missing
+
+        # check the presence of na_value
+        assert na_value in data_missing
+        assert na_value not in data
+
+        # Categoricals can contain other nan-likes than na_value
+        for na_value_obj in tm.NULL_OBJECTS:
+            if na_value_obj is na_value:
+                continue
+            assert na_value_obj not in data
+            assert na_value_obj in data_missing  # this line differs from super method
+
+
+class TestConstructors(base.BaseConstructorsTests):
+    def test_empty(self, dtype):
+        cls = dtype.construct_array_type()
+        result = cls._empty((4,), dtype=dtype)
+
+        assert isinstance(result, cls)
+        # the dtype we passed is not initialized, so will not match the
+        #  dtype on our result.
+        assert result.dtype == CategoricalDtype([])
+
+
+class TestReshaping(base.BaseReshapingTests):
+    pass
+
+
+class TestGetitem(base.BaseGetitemTests):
+    @pytest.mark.skip(reason="Backwards compatibility")
+    def test_getitem_scalar(self, data):
+        # CategoricalDtype.type isn't "correct" since it should
+        # be a parent of the elements (object). But don't want
+        # to break things by changing.
+        super().test_getitem_scalar(data)
+
+
+class TestSetitem(base.BaseSetitemTests):
+    pass
+
+
+class TestIndex(base.BaseIndexTests):
+    pass
+
+
+class TestMissing(base.BaseMissingTests):
+    pass
+
+
+class TestReduce(base.BaseNoReduceTests):
+    pass
+
+
+class TestMethods(base.BaseMethodsTests):
+    @pytest.mark.xfail(reason="Unobserved categories included")
+    def test_value_counts(self, all_data, dropna):
+        return super().test_value_counts(all_data, dropna)
+
+    def test_combine_add(self, data_repeated):
+        # GH 20825
+        # When adding categoricals in combine, result is a string
+        orig_data1, orig_data2 = data_repeated(2)
+        s1 = pd.Series(orig_data1)
+        s2 = pd.Series(orig_data2)
+        result = s1.combine(s2, lambda x1, x2: x1 + x2)
+        expected = pd.Series(
+            [a + b for (a, b) in zip(list(orig_data1), list(orig_data2))]
+        )
+        self.assert_series_equal(result, expected)
+
+        val = s1.iloc[0]
+        result = s1.combine(val, lambda x1, x2: x1 + x2)
+        expected = pd.Series([a + val for a in list(orig_data1)])
+        self.assert_series_equal(result, expected)
+
+
+class TestCasting(base.BaseCastingTests):
+    @pytest.mark.parametrize("cls", [Categorical, CategoricalIndex])
+    @pytest.mark.parametrize("values", [[1, np.nan], [Timestamp("2000"), pd.NaT]])
+    def test_cast_nan_to_int(self, cls, values):
+        # GH 28406
+        s = cls(values)
+
+        msg = "Cannot (cast|convert)"
+        with pytest.raises((ValueError, TypeError), match=msg):
+            s.astype(int)
+
+    @pytest.mark.parametrize(
+        "expected",
+        [
+            pd.Series(["2019", "2020"], dtype="datetime64[ns, UTC]"),
+            pd.Series([0, 0], dtype="timedelta64[ns]"),
+            pd.Series([pd.Period("2019"), pd.Period("2020")], dtype="period[A-DEC]"),
+            pd.Series([pd.Interval(0, 1), pd.Interval(1, 2)], dtype="interval"),
+            pd.Series([1, np.nan], dtype="Int64"),
+        ],
+    )
+    def test_cast_category_to_extension_dtype(self, expected):
+        # GH 28668
+        result = expected.astype("category").astype(expected.dtype)
+
+        tm.assert_series_equal(result, expected)
+
+    @pytest.mark.parametrize(
+        "dtype, expected",
+        [
+            (
+                "datetime64[ns]",
+                np.array(["2015-01-01T00:00:00.000000000"], dtype="datetime64[ns]"),
+            ),
+            (
+                "datetime64[ns, MET]",
+                pd.DatetimeIndex(
+                    [Timestamp("2015-01-01 00:00:00+0100", tz="MET")]
+                ).array,
+            ),
+        ],
+    )
+    def test_consistent_casting(self, dtype, expected):
+        # GH 28448
+        result = Categorical(["2015-01-01"]).astype(dtype)
+        assert result == expected
+
+
+class TestArithmeticOps(base.BaseArithmeticOpsTests):
+    def test_arith_frame_with_scalar(self, data, all_arithmetic_operators, request):
+        # frame & scalar
+        op_name = all_arithmetic_operators
+        if op_name == "__rmod__":
+            request.node.add_marker(
+                pytest.mark.xfail(
+                    reason="rmod never called when string is first argument"
+                )
+            )
+        super().test_arith_frame_with_scalar(data, op_name)
+
+    def test_arith_series_with_scalar(self, data, all_arithmetic_operators, request):
+        op_name = all_arithmetic_operators
+        if op_name == "__rmod__":
+            request.node.add_marker(
+                pytest.mark.xfail(
+                    reason="rmod never called when string is first argument"
+                )
+            )
+        super().test_arith_series_with_scalar(data, op_name)
+
+    def test_add_series_with_extension_array(self, data):
+        ser = pd.Series(data)
+        with pytest.raises(TypeError, match="cannot perform|unsupported operand"):
+            ser + data
+
+    def test_divmod_series_array(self):
+        # GH 23287
+        # skipping because it is not implemented
+        pass
+
+    def _check_divmod_op(self, s, op, other, exc=NotImplementedError):
+        return super()._check_divmod_op(s, op, other, exc=TypeError)
+
+
+class TestComparisonOps(base.BaseComparisonOpsTests):
+    def _compare_other(self, s, data, op, other):
+        op_name = f"__{op.__name__}__"
+        if op_name == "__eq__":
+            result = op(s, other)
+            expected = s.combine(other, lambda x, y: x == y)
+            assert (result == expected).all()
+
+        elif op_name == "__ne__":
+            result = op(s, other)
+            expected = s.combine(other, lambda x, y: x != y)
+            assert (result == expected).all()
+
+        else:
+            msg = "Unordered Categoricals can only compare equality or not"
+            with pytest.raises(TypeError, match=msg):
+                op(data, other)
+
+    @pytest.mark.parametrize(
+        "categories",
+        [["a", "b"], [0, 1], [Timestamp("2019"), Timestamp("2020")]],
+    )
+    def test_not_equal_with_na(self, categories):
+        # https://github.com/pandas-dev/pandas/issues/32276
+        c1 = Categorical.from_codes([-1, 0], categories=categories)
+        c2 = Categorical.from_codes([0, 1], categories=categories)
+
+        result = c1 != c2
+
+        assert result.all()
+
+
+class TestParsing(base.BaseParsingTests):
+    pass
+
+
+class Test2DCompat(base.NDArrayBacked2DTests):
+    def test_repr_2d(self, data):
+        # Categorical __repr__ doesn't include "Categorical", so we need
+        #  to special-case
+        res = repr(data.reshape(1, -1))
+        assert res.count("\nCategories") == 1
+
+        res = repr(data.reshape(-1, 1))
+        assert res.count("\nCategories") == 1

dist/client/pandas/tests/extension/test_common.py

@@ -0,0 +1,81 @@
+import numpy as np
+import pytest
+
+from pandas.core.dtypes import dtypes
+from pandas.core.dtypes.common import is_extension_array_dtype
+
+import pandas as pd
+import pandas._testing as tm
+from pandas.core.arrays import ExtensionArray
+
+
+class DummyDtype(dtypes.ExtensionDtype):
+    pass
+
+
+class DummyArray(ExtensionArray):
+    def __init__(self, data):
+        self.data = data
+
+    def __array__(self, dtype):
+        return self.data
+
+    @property
+    def dtype(self):
+        return DummyDtype()
+
+    def astype(self, dtype, copy=True):
+        # we don't support anything but a single dtype
+        if isinstance(dtype, DummyDtype):
+            if copy:
+                return type(self)(self.data)
+            return self
+
+        return np.array(self, dtype=dtype, copy=copy)
+
+
+class TestExtensionArrayDtype:
+    @pytest.mark.parametrize(
+        "values",
+        [
+            pd.Categorical([]),
+            pd.Categorical([]).dtype,
+            pd.Series(pd.Categorical([])),
+            DummyDtype(),
+            DummyArray(np.array([1, 2])),
+        ],
+    )
+    def test_is_extension_array_dtype(self, values):
+        assert is_extension_array_dtype(values)
+
+    @pytest.mark.parametrize("values", [np.array([]), pd.Series(np.array([]))])
+    def test_is_not_extension_array_dtype(self, values):
+        assert not is_extension_array_dtype(values)
+
+
+def test_astype():
+
+    arr = DummyArray(np.array([1, 2, 3]))
+    expected = np.array([1, 2, 3], dtype=object)
+
+    result = arr.astype(object)
+    tm.assert_numpy_array_equal(result, expected)
+
+    result = arr.astype("object")
+    tm.assert_numpy_array_equal(result, expected)
+
+
+def test_astype_no_copy():
+    arr = DummyArray(np.array([1, 2, 3], dtype=np.int64))
+    result = arr.astype(arr.dtype, copy=False)
+
+    assert arr is result
+
+    result = arr.astype(arr.dtype)
+    assert arr is not result
+
+
+@pytest.mark.parametrize("dtype", [dtypes.CategoricalDtype(), dtypes.IntervalDtype()])
+def test_is_extension_array_dtype(dtype):
+    assert isinstance(dtype, dtypes.ExtensionDtype)
+    assert is_extension_array_dtype(dtype)

dist/client/pandas/tests/extension/test_datetime.py

@@ -0,0 +1,194 @@
+"""
+This file contains a minimal set of tests for compliance with the extension
+array interface test suite, and should contain no other tests.
+The test suite for the full functionality of the array is located in
+`pandas/tests/arrays/`.
+
+The tests in this file are inherited from the BaseExtensionTests, and only
+minimal tweaks should be applied to get the tests passing (by overwriting a
+parent method).
+
+Additional tests should either be added to one of the BaseExtensionTests
+classes (if they are relevant for the extension interface for all dtypes), or
+be added to the array-specific tests in `pandas/tests/arrays/`.
+
+"""
+import numpy as np
+import pytest
+
+from pandas.core.dtypes.dtypes import DatetimeTZDtype
+
+import pandas as pd
+from pandas.core.arrays import DatetimeArray
+from pandas.tests.extension import base
+
+
+@pytest.fixture(params=["US/Central"])
+def dtype(request):
+    return DatetimeTZDtype(unit="ns", tz=request.param)
+
+
+@pytest.fixture
+def data(dtype):
+    data = DatetimeArray(pd.date_range("2000", periods=100, tz=dtype.tz), dtype=dtype)
+    return data
+
+
+@pytest.fixture
+def data_missing(dtype):
+    return DatetimeArray(
+        np.array(["NaT", "2000-01-01"], dtype="datetime64[ns]"), dtype=dtype
+    )
+
+
+@pytest.fixture
+def data_for_sorting(dtype):
+    a = pd.Timestamp("2000-01-01")
+    b = pd.Timestamp("2000-01-02")
+    c = pd.Timestamp("2000-01-03")
+    return DatetimeArray(np.array([b, c, a], dtype="datetime64[ns]"), dtype=dtype)
+
+
+@pytest.fixture
+def data_missing_for_sorting(dtype):
+    a = pd.Timestamp("2000-01-01")
+    b = pd.Timestamp("2000-01-02")
+    return DatetimeArray(np.array([b, "NaT", a], dtype="datetime64[ns]"), dtype=dtype)
+
+
+@pytest.fixture
+def data_for_grouping(dtype):
+    """
+    Expected to be like [B, B, NA, NA, A, A, B, C]
+
+    Where A < B < C and NA is missing
+    """
+    a = pd.Timestamp("2000-01-01")
+    b = pd.Timestamp("2000-01-02")
+    c = pd.Timestamp("2000-01-03")
+    na = "NaT"
+    return DatetimeArray(
+        np.array([b, b, na, na, a, a, b, c], dtype="datetime64[ns]"), dtype=dtype
+    )
+
+
+@pytest.fixture
+def na_cmp():
+    def cmp(a, b):
+        return a is pd.NaT and a is b
+
+    return cmp
+
+
+@pytest.fixture
+def na_value():
+    return pd.NaT
+
+
+# ----------------------------------------------------------------------------
+class BaseDatetimeTests:
+    pass
+
+
+# ----------------------------------------------------------------------------
+# Tests
+class TestDatetimeDtype(BaseDatetimeTests, base.BaseDtypeTests):
+    pass
+
+
+class TestConstructors(BaseDatetimeTests, base.BaseConstructorsTests):
+    def test_series_constructor(self, data):
+        # Series construction drops any .freq attr
+        data = data._with_freq(None)
+        super().test_series_constructor(data)
+
+
+class TestGetitem(BaseDatetimeTests, base.BaseGetitemTests):
+    pass
+
+
+class TestIndex(base.BaseIndexTests):
+    pass
+
+
+class TestMethods(BaseDatetimeTests, base.BaseMethodsTests):
+    def test_combine_add(self, data_repeated):
+        # Timestamp.__add__(Timestamp) not defined
+        pass
+
+
+class TestInterface(BaseDatetimeTests, base.BaseInterfaceTests):
+    pass
+
+
+class TestArithmeticOps(BaseDatetimeTests, base.BaseArithmeticOpsTests):
+    implements = {"__sub__", "__rsub__"}
+
+    def test_arith_frame_with_scalar(self, data, all_arithmetic_operators):
+        # frame & scalar
+        if all_arithmetic_operators in self.implements:
+            df = pd.DataFrame({"A": data})
+            self.check_opname(df, all_arithmetic_operators, data[0], exc=None)
+        else:
+            # ... but not the rest.
+            super().test_arith_frame_with_scalar(data, all_arithmetic_operators)
+
+    def test_arith_series_with_scalar(self, data, all_arithmetic_operators):
+        if all_arithmetic_operators in self.implements:
+            ser = pd.Series(data)
+            self.check_opname(ser, all_arithmetic_operators, ser.iloc[0], exc=None)
+        else:
+            # ... but not the rest.
+            super().test_arith_series_with_scalar(data, all_arithmetic_operators)
+
+    def test_add_series_with_extension_array(self, data):
+        # Datetime + Datetime not implemented
+        ser = pd.Series(data)
+        msg = "cannot add DatetimeArray and DatetimeArray"
+        with pytest.raises(TypeError, match=msg):
+            ser + data
+
+    def test_arith_series_with_array(self, data, all_arithmetic_operators):
+        if all_arithmetic_operators in self.implements:
+            ser = pd.Series(data)
+            self.check_opname(ser, all_arithmetic_operators, ser.iloc[0], exc=None)
+        else:
+            # ... but not the rest.
+            super().test_arith_series_with_scalar(data, all_arithmetic_operators)
+
+    def test_divmod_series_array(self):
+        # GH 23287
+        # skipping because it is not implemented
+        pass
+
+
+class TestCasting(BaseDatetimeTests, base.BaseCastingTests):
+    pass
+
+
+class TestComparisonOps(BaseDatetimeTests, base.BaseComparisonOpsTests):
+    pass
+
+
+class TestMissing(BaseDatetimeTests, base.BaseMissingTests):
+    pass
+
+
+class TestReshaping(BaseDatetimeTests, base.BaseReshapingTests):
+    pass
+
+
+class TestSetitem(BaseDatetimeTests, base.BaseSetitemTests):
+    pass
+
+
+class TestGroupby(BaseDatetimeTests, base.BaseGroupbyTests):
+    pass
+
+
+class TestPrinting(BaseDatetimeTests, base.BasePrintingTests):
+    pass
+
+
+class Test2DCompat(BaseDatetimeTests, base.NDArrayBacked2DTests):
+    pass

dist/client/pandas/tests/extension/test_extension.py

@@ -0,0 +1,26 @@
+"""
+Tests for behavior if an author does *not* implement EA methods.
+"""
+import numpy as np
+import pytest
+
+from pandas.core.arrays import ExtensionArray
+
+
+class MyEA(ExtensionArray):
+    def __init__(self, values):
+        self._values = values
+
+
+@pytest.fixture
+def data():
+    arr = np.arange(10)
+    return MyEA(arr)
+
+
+class TestExtensionArray:
+    def test_errors(self, data, all_arithmetic_operators):
+        # invalid ops
+        op_name = all_arithmetic_operators
+        with pytest.raises(AttributeError):
+            getattr(data, op_name)

dist/client/pandas/tests/extension/test_external_block.py

@@ -0,0 +1,40 @@
+import numpy as np
+import pytest
+
+from pandas._libs.internals import BlockPlacement
+import pandas.util._test_decorators as td
+
+import pandas as pd
+from pandas.core.internals import BlockManager
+from pandas.core.internals.blocks import ExtensionBlock
+
+pytestmark = td.skip_array_manager_invalid_test
+
+
+class CustomBlock(ExtensionBlock):
+
+    _holder = np.ndarray
+
+    # Cannot override final attribute "_can_hold_na"
+    @property  # type: ignore[misc]
+    def _can_hold_na(self) -> bool:
+        return False
+
+
+@pytest.fixture
+def df():
+    df1 = pd.DataFrame({"a": [1, 2, 3]})
+    blocks = df1._mgr.blocks
+    values = np.arange(3, dtype="int64")
+    bp = BlockPlacement(slice(1, 2))
+    custom_block = CustomBlock(values, placement=bp, ndim=2)
+    blocks = blocks + (custom_block,)
+    block_manager = BlockManager(blocks, [pd.Index(["a", "b"]), df1.index])
+    return pd.DataFrame(block_manager)
+
+
+def test_concat_axis1(df):
+    # GH17954
+    df2 = pd.DataFrame({"c": [0.1, 0.2, 0.3]})
+    res = pd.concat([df, df2], axis=1)
+    assert isinstance(res._mgr.blocks[1], CustomBlock)

dist/client/pandas/tests/extension/test_floating.py

@@ -0,0 +1,232 @@
+"""
+This file contains a minimal set of tests for compliance with the extension
+array interface test suite, and should contain no other tests.
+The test suite for the full functionality of the array is located in
+`pandas/tests/arrays/`.
+
+The tests in this file are inherited from the BaseExtensionTests, and only
+minimal tweaks should be applied to get the tests passing (by overwriting a
+parent method).
+
+Additional tests should either be added to one of the BaseExtensionTests
+classes (if they are relevant for the extension interface for all dtypes), or
+be added to the array-specific tests in `pandas/tests/arrays/`.
+
+"""
+import numpy as np
+import pytest
+
+from pandas.core.dtypes.common import is_extension_array_dtype
+
+import pandas as pd
+import pandas._testing as tm
+from pandas.api.types import is_float_dtype
+from pandas.core.arrays.floating import (
+    Float32Dtype,
+    Float64Dtype,
+)
+from pandas.tests.extension import base
+
+
+def make_data():
+    return (
+        list(np.arange(0.1, 0.9, 0.1))
+        + [pd.NA]
+        + list(np.arange(1, 9.8, 0.1))
+        + [pd.NA]
+        + [9.9, 10.0]
+    )
+
+
+@pytest.fixture(params=[Float32Dtype, Float64Dtype])
+def dtype(request):
+    return request.param()
+
+
+@pytest.fixture
+def data(dtype):
+    return pd.array(make_data(), dtype=dtype)
+
+
+@pytest.fixture
+def data_for_twos(dtype):
+    return pd.array(np.ones(100) * 2, dtype=dtype)
+
+
+@pytest.fixture
+def data_missing(dtype):
+    return pd.array([pd.NA, 0.1], dtype=dtype)
+
+
+@pytest.fixture
+def data_for_sorting(dtype):
+    return pd.array([0.1, 0.2, 0.0], dtype=dtype)
+
+
+@pytest.fixture
+def data_missing_for_sorting(dtype):
+    return pd.array([0.1, pd.NA, 0.0], dtype=dtype)
+
+
+@pytest.fixture
+def na_cmp():
+    # we are pd.NA
+    return lambda x, y: x is pd.NA and y is pd.NA
+
+
+@pytest.fixture
+def na_value():
+    return pd.NA
+
+
+@pytest.fixture
+def data_for_grouping(dtype):
+    b = 0.1
+    a = 0.0
+    c = 0.2
+    na = pd.NA
+    return pd.array([b, b, na, na, a, a, b, c], dtype=dtype)
+
+
+class TestDtype(base.BaseDtypeTests):
+    pass
+
+
+class TestArithmeticOps(base.BaseArithmeticOpsTests):
+    def check_opname(self, s, op_name, other, exc=None):
+        # overwriting to indicate ops don't raise an error
+        super().check_opname(s, op_name, other, exc=None)
+
+    def _check_op(self, s, op, other, op_name, exc=NotImplementedError):
+        if exc is None:
+            sdtype = tm.get_dtype(s)
+            if (
+                hasattr(other, "dtype")
+                and not is_extension_array_dtype(other.dtype)
+                and is_float_dtype(other.dtype)
+            ):
+                # other is np.float64 and would therefore always result in
+                # upcasting, so keeping other as same numpy_dtype
+                other = other.astype(sdtype.numpy_dtype)
+
+            result = op(s, other)
+            expected = self._combine(s, other, op)
+
+            # combine method result in 'biggest' (float64) dtype
+            expected = expected.astype(sdtype)
+
+            self.assert_equal(result, expected)
+        else:
+            with pytest.raises(exc):
+                op(s, other)
+
+    def _check_divmod_op(self, s, op, other, exc=None):
+        super()._check_divmod_op(s, op, other, None)
+
+
+class TestComparisonOps(base.BaseComparisonOpsTests):
+    # TODO: share with IntegerArray?
+    def _check_op(self, s, op, other, op_name, exc=NotImplementedError):
+        if exc is None:
+            result = op(s, other)
+            # Override to do the astype to boolean
+            expected = s.combine(other, op).astype("boolean")
+            self.assert_series_equal(result, expected)
+        else:
+            with pytest.raises(exc):
+                op(s, other)
+
+    def check_opname(self, s, op_name, other, exc=None):
+        super().check_opname(s, op_name, other, exc=None)
+
+    def _compare_other(self, s, data, op, other):
+        op_name = f"__{op.__name__}__"
+        self.check_opname(s, op_name, other)
+
+
+class TestInterface(base.BaseInterfaceTests):
+    pass
+
+
+class TestConstructors(base.BaseConstructorsTests):
+    pass
+
+
+class TestReshaping(base.BaseReshapingTests):
+    pass
+
+
+class TestGetitem(base.BaseGetitemTests):
+    pass
+
+
+class TestSetitem(base.BaseSetitemTests):
+    pass
+
+
+class TestIndex(base.BaseIndexTests):
+    pass
+
+
+class TestMissing(base.BaseMissingTests):
+    pass
+
+
+class TestMethods(base.BaseMethodsTests):
+    @pytest.mark.parametrize("dropna", [True, False])
+    def test_value_counts(self, all_data, dropna):
+        all_data = all_data[:10]
+        if dropna:
+            other = np.array(all_data[~all_data.isna()])
+        else:
+            other = all_data
+
+        result = pd.Series(all_data).value_counts(dropna=dropna).sort_index()
+        expected = pd.Series(other).value_counts(dropna=dropna).sort_index()
+        expected = expected.astype("Int64")
+        expected.index = expected.index.astype(all_data.dtype)
+
+        self.assert_series_equal(result, expected)
+
+    @pytest.mark.xfail(reason="uses nullable integer")
+    def test_value_counts_with_normalize(self, data):
+        super().test_value_counts_with_normalize(data)
+
+
+class TestCasting(base.BaseCastingTests):
+    pass
+
+
+class TestGroupby(base.BaseGroupbyTests):
+    pass
+
+
+class TestNumericReduce(base.BaseNumericReduceTests):
+    def check_reduce(self, s, op_name, skipna):
+        # overwrite to ensure pd.NA is tested instead of np.nan
+        # https://github.com/pandas-dev/pandas/issues/30958
+        result = getattr(s, op_name)(skipna=skipna)
+        if not skipna and s.isna().any():
+            expected = pd.NA
+        else:
+            expected = getattr(s.dropna().astype(s.dtype.numpy_dtype), op_name)(
+                skipna=skipna
+            )
+        tm.assert_almost_equal(result, expected)
+
+
+@pytest.mark.skip(reason="Tested in tests/reductions/test_reductions.py")
+class TestBooleanReduce(base.BaseBooleanReduceTests):
+    pass
+
+
+class TestPrinting(base.BasePrintingTests):
+    pass
+
+
+class TestParsing(base.BaseParsingTests):
+    pass
+
+
+class Test2DCompat(base.Dim2CompatTests):
+    pass

dist/client/pandas/tests/extension/test_integer.py

@@ -0,0 +1,253 @@
+"""
+This file contains a minimal set of tests for compliance with the extension
+array interface test suite, and should contain no other tests.
+The test suite for the full functionality of the array is located in
+`pandas/tests/arrays/`.
+
+The tests in this file are inherited from the BaseExtensionTests, and only
+minimal tweaks should be applied to get the tests passing (by overwriting a
+parent method).
+
+Additional tests should either be added to one of the BaseExtensionTests
+classes (if they are relevant for the extension interface for all dtypes), or
+be added to the array-specific tests in `pandas/tests/arrays/`.
+
+"""
+import numpy as np
+import pytest
+
+import pandas as pd
+import pandas._testing as tm
+from pandas.api.types import (
+    is_extension_array_dtype,
+    is_integer_dtype,
+)
+from pandas.core.arrays.integer import (
+    Int8Dtype,
+    Int16Dtype,
+    Int32Dtype,
+    Int64Dtype,
+    UInt8Dtype,
+    UInt16Dtype,
+    UInt32Dtype,
+    UInt64Dtype,
+)
+from pandas.tests.extension import base
+
+
+def make_data():
+    return list(range(1, 9)) + [pd.NA] + list(range(10, 98)) + [pd.NA] + [99, 100]
+
+
+@pytest.fixture(
+    params=[
+        Int8Dtype,
+        Int16Dtype,
+        Int32Dtype,
+        Int64Dtype,
+        UInt8Dtype,
+        UInt16Dtype,
+        UInt32Dtype,
+        UInt64Dtype,
+    ]
+)
+def dtype(request):
+    return request.param()
+
+
+@pytest.fixture
+def data(dtype):
+    return pd.array(make_data(), dtype=dtype)
+
+
+@pytest.fixture
+def data_for_twos(dtype):
+    return pd.array(np.ones(100) * 2, dtype=dtype)
+
+
+@pytest.fixture
+def data_missing(dtype):
+    return pd.array([pd.NA, 1], dtype=dtype)
+
+
+@pytest.fixture
+def data_for_sorting(dtype):
+    return pd.array([1, 2, 0], dtype=dtype)
+
+
+@pytest.fixture
+def data_missing_for_sorting(dtype):
+    return pd.array([1, pd.NA, 0], dtype=dtype)
+
+
+@pytest.fixture
+def na_cmp():
+    # we are pd.NA
+    return lambda x, y: x is pd.NA and y is pd.NA
+
+
+@pytest.fixture
+def na_value():
+    return pd.NA
+
+
+@pytest.fixture
+def data_for_grouping(dtype):
+    b = 1
+    a = 0
+    c = 2
+    na = pd.NA
+    return pd.array([b, b, na, na, a, a, b, c], dtype=dtype)
+
+
+class TestDtype(base.BaseDtypeTests):
+    pass
+
+
+class TestArithmeticOps(base.BaseArithmeticOpsTests):
+    def check_opname(self, s, op_name, other, exc=None):
+        # overwriting to indicate ops don't raise an error
+        super().check_opname(s, op_name, other, exc=None)
+
+    def _check_op(self, s, op, other, op_name, exc=NotImplementedError):
+        if exc is None:
+            sdtype = tm.get_dtype(s)
+
+            if (
+                hasattr(other, "dtype")
+                and not is_extension_array_dtype(other.dtype)
+                and is_integer_dtype(other.dtype)
+                and sdtype.is_unsigned_integer
+            ):
+                # TODO: comment below is inaccurate; other can be int8, int16, ...
+                #  and the trouble is that e.g. if s is UInt8 and other is int8,
+                #  then result is UInt16
+                # other is np.int64 and would therefore always result in
+                # upcasting, so keeping other as same numpy_dtype
+                other = other.astype(sdtype.numpy_dtype)
+
+            result = op(s, other)
+            expected = self._combine(s, other, op)
+
+            if op_name in ("__rtruediv__", "__truediv__", "__div__"):
+                expected = expected.fillna(np.nan).astype("Float64")
+            else:
+                # combine method result in 'biggest' (int64) dtype
+                expected = expected.astype(sdtype)
+
+            self.assert_equal(result, expected)
+        else:
+            with pytest.raises(exc):
+                op(s, other)
+
+    def _check_divmod_op(self, s, op, other, exc=None):
+        super()._check_divmod_op(s, op, other, None)
+
+
+class TestComparisonOps(base.BaseComparisonOpsTests):
+    def _check_op(self, s, op, other, op_name, exc=NotImplementedError):
+        if exc is None:
+            result = op(s, other)
+            # Override to do the astype to boolean
+            expected = s.combine(other, op).astype("boolean")
+            self.assert_series_equal(result, expected)
+        else:
+            with pytest.raises(exc):
+                op(s, other)
+
+    def check_opname(self, s, op_name, other, exc=None):
+        super().check_opname(s, op_name, other, exc=None)
+
+    def _compare_other(self, s, data, op, other):
+        op_name = f"__{op.__name__}__"
+        self.check_opname(s, op_name, other)
+
+
+class TestInterface(base.BaseInterfaceTests):
+    pass
+
+
+class TestConstructors(base.BaseConstructorsTests):
+    pass
+
+
+class TestReshaping(base.BaseReshapingTests):
+    pass
+
+    # for test_concat_mixed_dtypes test
+    # concat of an Integer and Int coerces to object dtype
+    # TODO(jreback) once integrated this would
+
+
+class TestGetitem(base.BaseGetitemTests):
+    pass
+
+
+class TestSetitem(base.BaseSetitemTests):
+    pass
+
+
+class TestIndex(base.BaseIndexTests):
+    pass
+
+
+class TestMissing(base.BaseMissingTests):
+    pass
+
+
+class TestMethods(base.BaseMethodsTests):
+    @pytest.mark.parametrize("dropna", [True, False])
+    def test_value_counts(self, all_data, dropna):
+        all_data = all_data[:10]
+        if dropna:
+            other = np.array(all_data[~all_data.isna()])
+        else:
+            other = all_data
+
+        result = pd.Series(all_data).value_counts(dropna=dropna).sort_index()
+        expected = pd.Series(other).value_counts(dropna=dropna).sort_index()
+        expected = expected.astype("Int64")
+        expected.index = expected.index.astype(all_data.dtype)
+
+        self.assert_series_equal(result, expected)
+
+    @pytest.mark.xfail(reason="uses nullable integer")
+    def test_value_counts_with_normalize(self, data):
+        super().test_value_counts_with_normalize(data)
+
+
+class TestCasting(base.BaseCastingTests):
+    pass
+
+
+class TestGroupby(base.BaseGroupbyTests):
+    pass
+
+
+class TestNumericReduce(base.BaseNumericReduceTests):
+    def check_reduce(self, s, op_name, skipna):
+        # overwrite to ensure pd.NA is tested instead of np.nan
+        # https://github.com/pandas-dev/pandas/issues/30958
+        result = getattr(s, op_name)(skipna=skipna)
+        if not skipna and s.isna().any():
+            expected = pd.NA
+        else:
+            expected = getattr(s.dropna().astype("int64"), op_name)(skipna=skipna)
+        tm.assert_almost_equal(result, expected)
+
+
+@pytest.mark.skip(reason="Tested in tests/reductions/test_reductions.py")
+class TestBooleanReduce(base.BaseBooleanReduceTests):
+    pass
+
+
+class TestPrinting(base.BasePrintingTests):
+    pass
+
+
+class TestParsing(base.BaseParsingTests):
+    pass
+
+
+class Test2DCompat(base.Dim2CompatTests):
+    pass

dist/client/pandas/tests/extension/test_interval.py

@@ -0,0 +1,188 @@
+"""
+This file contains a minimal set of tests for compliance with the extension
+array interface test suite, and should contain no other tests.
+The test suite for the full functionality of the array is located in
+`pandas/tests/arrays/`.
+
+The tests in this file are inherited from the BaseExtensionTests, and only
+minimal tweaks should be applied to get the tests passing (by overwriting a
+parent method).
+
+Additional tests should either be added to one of the BaseExtensionTests
+classes (if they are relevant for the extension interface for all dtypes), or
+be added to the array-specific tests in `pandas/tests/arrays/`.
+
+"""
+import numpy as np
+import pytest
+
+from pandas.core.dtypes.dtypes import IntervalDtype
+
+from pandas import (
+    Interval,
+    Series,
+)
+from pandas.core.arrays import IntervalArray
+from pandas.tests.extension import base
+
+
+def make_data():
+    N = 100
+    left_array = np.random.uniform(size=N).cumsum()
+    right_array = left_array + np.random.uniform(size=N)
+    return [Interval(left, right) for left, right in zip(left_array, right_array)]
+
+
+@pytest.fixture
+def dtype():
+    return IntervalDtype()
+
+
+@pytest.fixture
+def data():
+    """Length-100 PeriodArray for semantics test."""
+    return IntervalArray(make_data())
+
+
+@pytest.fixture
+def data_missing():
+    """Length 2 array with [NA, Valid]"""
+    return IntervalArray.from_tuples([None, (0, 1)])
+
+
+@pytest.fixture
+def data_for_sorting():
+    return IntervalArray.from_tuples([(1, 2), (2, 3), (0, 1)])
+
+
+@pytest.fixture
+def data_missing_for_sorting():
+    return IntervalArray.from_tuples([(1, 2), None, (0, 1)])
+
+
+@pytest.fixture
+def na_value():
+    return np.nan
+
+
+@pytest.fixture
+def data_for_grouping():
+    a = (0, 1)
+    b = (1, 2)
+    c = (2, 3)
+    return IntervalArray.from_tuples([b, b, None, None, a, a, b, c])
+
+
+class BaseInterval:
+    pass
+
+
+class TestDtype(BaseInterval, base.BaseDtypeTests):
+    pass
+
+
+class TestCasting(BaseInterval, base.BaseCastingTests):
+    pass
+
+
+class TestConstructors(BaseInterval, base.BaseConstructorsTests):
+    pass
+
+
+class TestGetitem(BaseInterval, base.BaseGetitemTests):
+    pass
+
+
+class TestIndex(base.BaseIndexTests):
+    pass
+
+
+class TestGrouping(BaseInterval, base.BaseGroupbyTests):
+    pass
+
+
+class TestInterface(BaseInterval, base.BaseInterfaceTests):
+    pass
+
+
+class TestReduce(base.BaseNoReduceTests):
+    @pytest.mark.parametrize("skipna", [True, False])
+    def test_reduce_series_numeric(self, data, all_numeric_reductions, skipna):
+        op_name = all_numeric_reductions
+        ser = Series(data)
+
+        if op_name in ["min", "max"]:
+            # IntervalArray *does* implement these
+            assert getattr(ser, op_name)(skipna=skipna) in data
+            assert getattr(data, op_name)(skipna=skipna) in data
+            return
+
+        super().test_reduce_series_numeric(data, all_numeric_reductions, skipna)
+
+
+class TestMethods(BaseInterval, base.BaseMethodsTests):
+    @pytest.mark.xfail(reason="addition is not defined for intervals")
+    def test_combine_add(self, data_repeated):
+        super().test_combine_add(data_repeated)
+
+    @pytest.mark.xfail(
+        reason="Raises with incorrect message bc it disallows *all* listlikes "
+        "instead of just wrong-length listlikes"
+    )
+    def test_fillna_length_mismatch(self, data_missing):
+        super().test_fillna_length_mismatch(data_missing)
+
+
+class TestMissing(BaseInterval, base.BaseMissingTests):
+    # Index.fillna only accepts scalar `value`, so we have to xfail all
+    # non-scalar fill tests.
+    unsupported_fill = pytest.mark.xfail(
+        reason="Unsupported fillna option for Interval."
+    )
+
+    @unsupported_fill
+    def test_fillna_limit_pad(self):
+        super().test_fillna_limit_pad()
+
+    @unsupported_fill
+    def test_fillna_series_method(self):
+        super().test_fillna_series_method()
+
+    @unsupported_fill
+    def test_fillna_limit_backfill(self):
+        super().test_fillna_limit_backfill()
+
+    @unsupported_fill
+    def test_fillna_no_op_returns_copy(self):
+        super().test_fillna_no_op_returns_copy()
+
+    @unsupported_fill
+    def test_fillna_series(self):
+        super().test_fillna_series()
+
+    def test_fillna_non_scalar_raises(self, data_missing):
+        msg = "can only insert Interval objects and NA into an IntervalArray"
+        with pytest.raises(TypeError, match=msg):
+            data_missing.fillna([1, 1])
+
+
+class TestReshaping(BaseInterval, base.BaseReshapingTests):
+    pass
+
+
+class TestSetitem(BaseInterval, base.BaseSetitemTests):
+    pass
+
+
+class TestPrinting(BaseInterval, base.BasePrintingTests):
+    @pytest.mark.xfail(reason="Interval has custom repr")
+    def test_array_repr(self, data, size):
+        super().test_array_repr()
+
+
+class TestParsing(BaseInterval, base.BaseParsingTests):
+    @pytest.mark.parametrize("engine", ["c", "python"])
+    def test_EA_types(self, engine, data):
+        expected_msg = r".*must implement _from_sequence_of_strings.*"
+        with pytest.raises(NotImplementedError, match=expected_msg):
+            super().test_EA_types(engine, data)