test_compute.py

# Licensed to the Apache Software Foundation (ASF) under one
# or more contributor license agreements.  See the NOTICE file
# distributed with this work for additional information
# regarding copyright ownership.  The ASF licenses this file
# to you under the Apache License, Version 2.0 (the
# "License"); you may not use this file except in compliance
# with the License.  You may obtain a copy of the License at
#
#   http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing,
# software distributed under the License is distributed on an
# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
# KIND, either express or implied.  See the License for the
# specific language governing permissions and limitations
# under the License.

from collections import namedtuple
import datetime
import decimal
from functools import lru_cache, partial
import inspect
import itertools
import math
import os
import pytest
import random
import sys
import textwrap

try:
    import numpy as np
except ImportError:
    np = None

try:
    import pandas as pd
except ImportError:
    pd = None

import pyarrow as pa
import pyarrow.compute as pc
from pyarrow.lib import ArrowNotImplementedError

try:
    import pyarrow.substrait as pas
except ImportError:
    pas = None

exported_functions = [
    func for (name, func) in sorted(pc.__dict__.items())
    if hasattr(func, '__arrow_compute_function__')]

exported_option_classes = [
    cls for (name, cls) in sorted(pc.__dict__.items())
    if (isinstance(cls, type) and
        cls is not pc.FunctionOptions and
        issubclass(cls, pc.FunctionOptions))]

numerical_arrow_types = [
    pa.int8(),
    pa.int16(),
    pa.int64(),
    pa.uint8(),
    pa.uint16(),
    pa.uint64(),
    pa.float32(),
    pa.float64()
]


all_array_types = [
    ('bool', [True, False, False, True, True]),
    ('uint8', range(5)),
    ('int8', range(5)),
    ('uint16', range(5)),
    ('int16', range(5)),
    ('uint32', range(5)),
    ('int32', range(5)),
    ('uint64', range(5, 10)),
    ('int64', range(5, 10)),
    ('float', [0, 0.1, 0.2, 0.3, 0.4]),
    ('double', [0, 0.1, 0.2, 0.3, 0.4]),
    ('string', ['a', 'b', None, 'ddd', 'ee']),
    ('binary', [b'a', b'b', b'c', b'ddd', b'ee']),
    (pa.binary(3), [b'abc', b'bcd', b'cde', b'def', b'efg']),
    (pa.list_(pa.int8()), [[1, 2], [3, 4], [5, 6], None, [9, 16]]),
    (pa.large_list(pa.int16()), [[1], [2, 3, 4], [5, 6], None, [9, 16]]),
    (pa.struct([('a', pa.int8()), ('b', pa.int8())]), [
        {'a': 1, 'b': 2}, None, {'a': 3, 'b': 4}, None, {'a': 5, 'b': 6}]),
]


def test_exported_functions():
    # Check that all exported concrete functions can be called with
    # the right number of arguments.
    # Note that unregistered functions (e.g. with a mismatching name)
    # will raise KeyError.
    functions = exported_functions
    assert len(functions) >= 10
    for func in functions:
        desc = func.__arrow_compute_function__
        if desc['options_required']:
            # Skip this function as it will fail with a different error
            # message if we don't pass an options instance.
            continue
        arity = desc['arity']
        if arity == 0:
            continue
        if arity is Ellipsis:
            args = [object()] * 3
        else:
            args = [object()] * arity
        with pytest.raises(TypeError,
                           match="Got unexpected argument type "
                                 "<class 'object'> for compute function"):
            func(*args)


def test_hash_aggregate_not_exported():
    # Ensure we are not leaking hash aggregate functions
    # which are not callable by themselves.
    for func in exported_functions:
        arrow_f = pc.get_function(func.__arrow_compute_function__["name"])
        assert arrow_f.kind != "hash_aggregate"


def test_exported_option_classes():
    classes = exported_option_classes
    assert len(classes) >= 10
    for cls in classes:
        # Option classes must have an introspectable constructor signature,
        # and that signature should not have any *args or **kwargs.
        sig = inspect.signature(cls)
        for param in sig.parameters.values():
            assert param.kind not in (param.VAR_POSITIONAL,
                                      param.VAR_KEYWORD)


@pytest.mark.filterwarnings(
    "ignore:pyarrow.CumulativeSumOptions is deprecated as of 14.0"
)
def test_option_class_equality(request):
    options = [
        pc.ArraySortOptions(),
        pc.AssumeTimezoneOptions("UTC"),
        pc.CastOptions.safe(pa.int8()),
        pc.CountOptions(),
        pc.DayOfWeekOptions(count_from_zero=False, week_start=0),
        pc.DictionaryEncodeOptions(),
        pc.RunEndEncodeOptions(),
        pc.ElementWiseAggregateOptions(skip_nulls=True),
        pc.ExtractRegexOptions("pattern"),
        pc.FilterOptions(),
        pc.IndexOptions(pa.scalar(1)),
        pc.JoinOptions(),
        pc.ListSliceOptions(0, -1, 1, True),
        pc.ListFlattenOptions(recursive=False),
        pc.MakeStructOptions(["field", "names"],
                             field_nullability=[True, True],
                             field_metadata=[pa.KeyValueMetadata({"a": "1"}),
                                             pa.KeyValueMetadata({"b": "2"})]),
        pc.MapLookupOptions(pa.scalar(1), "first"),
        pc.MatchSubstringOptions("pattern"),
        pc.ModeOptions(),
        pc.NullOptions(),
        pc.PadOptions(5),
        pc.PairwiseOptions(period=1),
        pc.PartitionNthOptions(1, null_placement="at_start"),
        pc.CumulativeOptions(start=None, skip_nulls=False),
        pc.QuantileOptions(),
        pc.RandomOptions(),
        pc.RankOptions(sort_keys="ascending",
                       null_placement="at_start", tiebreaker="max"),
        pc.ReplaceSliceOptions(0, 1, "a"),
        pc.ReplaceSubstringOptions("a", "b"),
        pc.RoundOptions(2, "towards_infinity"),
        pc.RoundBinaryOptions("towards_infinity"),
        pc.RoundTemporalOptions(1, "second", week_starts_monday=True),
        pc.RoundToMultipleOptions(100, "towards_infinity"),
        pc.ScalarAggregateOptions(),
        pc.SelectKOptions(0, sort_keys=[("b", "ascending")]),
        pc.SetLookupOptions(pa.array([1])),
        pc.SliceOptions(0, 1, 1),
        pc.SortOptions([("dummy", "descending")], null_placement="at_start"),
        pc.SplitOptions(),
        pc.SplitPatternOptions("pattern"),
        pc.StrftimeOptions(),
        pc.StrptimeOptions("%Y", "s", True),
        pc.StructFieldOptions(indices=[]),
        pc.TakeOptions(),
        pc.TDigestOptions(),
        pc.TrimOptions(" "),
        pc.Utf8NormalizeOptions("NFKC"),
        pc.VarianceOptions(),
        pc.WeekOptions(week_starts_monday=True, count_from_zero=False,
                       first_week_is_fully_in_year=False),
    ]
    # Timezone database might not be installed on Windows or Emscripten
    if request.config.pyarrow.is_enabled["timezone_data"]:
        options.append(pc.AssumeTimezoneOptions("Europe/Ljubljana"))

    classes = {type(option) for option in options}

    for cls in exported_option_classes:
        # Timezone database might not be installed on Windows or Emscripten
        if (
            cls not in classes
            and (request.config.pyarrow.is_enabled["timezone_data"])
            and cls != pc.AssumeTimezoneOptions
        ):
            try:
                options.append(cls())
            except TypeError:
                pytest.fail(f"Options class is not tested: {cls}")

    for option in options:
        assert option == option
        assert repr(option).startswith(option.__class__.__name__)
        buf = option.serialize()
        deserialized = pc.FunctionOptions.deserialize(buf)
        assert option == deserialized
        # TODO remove the check under the if statement and the filterwarnings
        # mark when the deprecated class CumulativeSumOptions is removed.
        if repr(option).startswith("CumulativeSumOptions"):
            assert repr(deserialized).startswith("CumulativeOptions")
        else:
            assert repr(option) == repr(deserialized)
    for option1, option2 in zip(options, options[1:]):
        assert option1 != option2

    assert repr(pc.IndexOptions(pa.scalar(1))) == "IndexOptions(value=int64:1)"
    assert repr(pc.ArraySortOptions()) == \
        "ArraySortOptions(order=Ascending, null_placement=AtEnd)"


def test_list_functions():
    assert len(pc.list_functions()) > 10
    assert "add" in pc.list_functions()


def _check_get_function(name, expected_func_cls, expected_ker_cls,
                        min_num_kernels=1):
    func = pc.get_function(name)
    assert isinstance(func, expected_func_cls)
    n = func.num_kernels
    assert n >= min_num_kernels
    assert n == len(func.kernels)
    assert all(isinstance(ker, expected_ker_cls) for ker in func.kernels)


def test_get_function_scalar():
    _check_get_function("add", pc.ScalarFunction, pc.ScalarKernel, 8)


def test_get_function_vector():
    _check_get_function("unique", pc.VectorFunction, pc.VectorKernel, 8)


def test_get_function_scalar_aggregate():
    _check_get_function("mean", pc.ScalarAggregateFunction,
                        pc.ScalarAggregateKernel, 8)


def test_get_function_hash_aggregate():
    _check_get_function("hash_sum", pc.HashAggregateFunction,
                        pc.HashAggregateKernel, 1)


@pytest.mark.numpy
def test_call_function_with_memory_pool():
    arr = pa.array(["foo", "bar", "baz"])
    indices = np.array([2, 2, 1])
    result1 = arr.take(indices)
    result2 = pc.call_function('take', [arr, indices],
                               memory_pool=pa.default_memory_pool())
    expected = pa.array(["baz", "baz", "bar"])
    assert result1.equals(expected)
    assert result2.equals(expected)

    result3 = pc.take(arr, indices, memory_pool=pa.default_memory_pool())
    assert result3.equals(expected)


def test_pickle_functions(pickle_module):
    # Pickle registered functions
    for name in pc.list_functions():
        func = pc.get_function(name)
        reconstructed = pickle_module.loads(pickle_module.dumps(func))
        assert type(reconstructed) is type(func)
        assert reconstructed.name == func.name
        assert reconstructed.arity == func.arity
        assert reconstructed.num_kernels == func.num_kernels


def test_pickle_global_functions(pickle_module):
    # Pickle global wrappers (manual or automatic) of registered functions
    for name in pc.list_functions():
        try:
            func = getattr(pc, name)
        except AttributeError:
            # hash_aggregate functions are not exported as callables.
            continue
        reconstructed = pickle_module.loads(pickle_module.dumps(func))
        assert reconstructed is func


def test_function_attributes():
    # Sanity check attributes of registered functions
    for name in pc.list_functions():
        func = pc.get_function(name)
        assert isinstance(func, pc.Function)
        assert func.name == name
        kernels = func.kernels
        assert func.num_kernels == len(kernels)
        assert all(isinstance(ker, pc.Kernel) for ker in kernels)
        repr(func)
        for ker in kernels:
            repr(ker)


def test_input_type_conversion():
    # Automatic array conversion from Python
    arr = pc.add([1, 2], [4, None])
    assert arr.to_pylist() == [5, None]
    # Automatic scalar conversion from Python
    arr = pc.add([1, 2], 4)
    assert arr.to_pylist() == [5, 6]
    # Other scalar type
    assert pc.equal(["foo", "bar", None],
                    "foo").to_pylist() == [True, False, None]


@pytest.mark.parametrize('arrow_type', numerical_arrow_types)
def test_sum_array(arrow_type):
    arr = pa.array([1, 2, 3, 4], type=arrow_type)
    assert arr.sum().as_py() == 10
    assert pc.sum(arr).as_py() == 10

    arr = pa.array([1, 2, 3, 4, None], type=arrow_type)
    assert arr.sum().as_py() == 10
    assert pc.sum(arr).as_py() == 10

    arr = pa.array([None], type=arrow_type)
    assert arr.sum().as_py() is None  # noqa: E711
    assert pc.sum(arr).as_py() is None  # noqa: E711
    assert arr.sum(min_count=0).as_py() == 0
    assert pc.sum(arr, min_count=0).as_py() == 0

    arr = pa.array([], type=arrow_type)
    assert arr.sum().as_py() is None  # noqa: E711
    assert arr.sum(min_count=0).as_py() == 0
    assert pc.sum(arr, min_count=0).as_py() == 0


@pytest.mark.parametrize('arrow_type', numerical_arrow_types)
def test_sum_chunked_array(arrow_type):
    arr = pa.chunked_array([pa.array([1, 2, 3, 4], type=arrow_type)])
    assert pc.sum(arr).as_py() == 10

    arr = pa.chunked_array([
        pa.array([1, 2], type=arrow_type), pa.array([3, 4], type=arrow_type)
    ])
    assert pc.sum(arr).as_py() == 10

    arr = pa.chunked_array([
        pa.array([1, 2], type=arrow_type),
        pa.array([], type=arrow_type),
        pa.array([3, 4], type=arrow_type)
    ])
    assert pc.sum(arr).as_py() == 10

    arr = pa.chunked_array((), type=arrow_type)
    assert arr.num_chunks == 0
    assert pc.sum(arr).as_py() is None  # noqa: E711
    assert pc.sum(arr, min_count=0).as_py() == 0


def test_mode_array():
    # ARROW-9917
    arr = pa.array([1, 1, 3, 4, 3, 5], type='int64')
    mode = pc.mode(arr)
    assert len(mode) == 1
    assert mode[0].as_py() == {"mode": 1, "count": 2}

    mode = pc.mode(arr, n=2)
    assert len(mode) == 2
    assert mode[0].as_py() == {"mode": 1, "count": 2}
    assert mode[1].as_py() == {"mode": 3, "count": 2}

    arr = pa.array([], type='int64')
    assert len(pc.mode(arr)) == 0

    arr = pa.array([1, 1, 3, 4, 3, None], type='int64')
    mode = pc.mode(arr, skip_nulls=False)
    assert len(mode) == 0
    mode = pc.mode(arr, min_count=6)
    assert len(mode) == 0
    mode = pc.mode(arr, skip_nulls=False, min_count=5)
    assert len(mode) == 0

    arr = pa.array([True, False])
    mode = pc.mode(arr, n=2)
    assert len(mode) == 2
    assert mode[0].as_py() == {"mode": False, "count": 1}
    assert mode[1].as_py() == {"mode": True, "count": 1}


def test_mode_chunked_array():
    # ARROW-9917
    arr = pa.chunked_array([pa.array([1, 1, 3, 4, 3, 5], type='int64')])
    mode = pc.mode(arr)
    assert len(mode) == 1
    assert mode[0].as_py() == {"mode": 1, "count": 2}

    mode = pc.mode(arr, n=2)
    assert len(mode) == 2
    assert mode[0].as_py() == {"mode": 1, "count": 2}
    assert mode[1].as_py() == {"mode": 3, "count": 2}

    arr = pa.chunked_array((), type='int64')
    assert arr.num_chunks == 0
    assert len(pc.mode(arr)) == 0


def test_empty_chunked_array():
    msg = "cannot construct ChunkedArray from empty vector and omitted type"
    with pytest.raises(pa.ArrowInvalid, match=msg):
        pa.chunked_array([])

    pa.chunked_array([], type=pa.int8())


def test_variance():
    data = [1, 2, 3, 4, 5, 6, 7, 8]
    assert pc.variance(data).as_py() == 5.25
    assert pc.variance(data, ddof=0).as_py() == 5.25
    assert pc.variance(data, ddof=1).as_py() == 6.0


def test_count_substring():
    for (ty, offset) in [(pa.string(), pa.int32()),
                         (pa.large_string(), pa.int64())]:
        arr = pa.array(["ab", "cab", "abcab", "ba", "AB", None], type=ty)

        result = pc.count_substring(arr, "ab")
        expected = pa.array([1, 1, 2, 0, 0, None], type=offset)
        assert expected == result

        result = pc.count_substring(arr, "ab", ignore_case=True)
        expected = pa.array([1, 1, 2, 0, 1, None], type=offset)
        assert expected == result


def test_count_substring_regex():
    for (ty, offset) in [(pa.string(), pa.int32()),
                         (pa.large_string(), pa.int64())]:
        arr = pa.array(["ab", "cab", "baAacaa", "ba", "AB", None], type=ty)

        result = pc.count_substring_regex(arr, "a+")
        expected = pa.array([1, 1, 3, 1, 0, None], type=offset)
        assert expected.equals(result)

        result = pc.count_substring_regex(arr, "a+", ignore_case=True)
        expected = pa.array([1, 1, 2, 1, 1, None], type=offset)
        assert expected.equals(result)


def test_find_substring():
    for ty in [pa.string(), pa.binary(), pa.large_string(), pa.large_binary()]:
        arr = pa.array(["ab", "cab", "ba", None], type=ty)
        result = pc.find_substring(arr, "ab")
        assert result.to_pylist() == [0, 1, -1, None]

        result = pc.find_substring_regex(arr, "a?b")
        assert result.to_pylist() == [0, 1, 0, None]

        arr = pa.array(["ab*", "cAB*", "ba", "aB?"], type=ty)
        result = pc.find_substring(arr, "aB*", ignore_case=True)
        assert result.to_pylist() == [0, 1, -1, -1]

        result = pc.find_substring_regex(arr, "a?b", ignore_case=True)
        assert result.to_pylist() == [0, 1, 0, 0]


def test_match_like():
    arr = pa.array(["ab", "ba%", "ba", "ca%d", None])
    result = pc.match_like(arr, r"_a\%%")
    expected = pa.array([False, True, False, True, None])
    assert expected.equals(result)

    arr = pa.array(["aB", "bA%", "ba", "ca%d", None])
    result = pc.match_like(arr, r"_a\%%", ignore_case=True)
    expected = pa.array([False, True, False, True, None])
    assert expected.equals(result)
    result = pc.match_like(arr, r"_a\%%", ignore_case=False)
    expected = pa.array([False, False, False, True, None])
    assert expected.equals(result)


def test_match_substring():
    arr = pa.array(["ab", "abc", "ba", None])
    result = pc.match_substring(arr, "ab")
    expected = pa.array([True, True, False, None])
    assert expected.equals(result)

    arr = pa.array(["áB", "Ábc", "ba", None])
    result = pc.match_substring(arr, "áb", ignore_case=True)
    expected = pa.array([True, True, False, None])
    assert expected.equals(result)
    result = pc.match_substring(arr, "áb", ignore_case=False)
    expected = pa.array([False, False, False, None])
    assert expected.equals(result)


def test_match_substring_regex():
    arr = pa.array(["ab", "abc", "ba", "c", None])
    result = pc.match_substring_regex(arr, "^a?b")
    expected = pa.array([True, True, True, False, None])
    assert expected.equals(result)

    arr = pa.array(["aB", "Abc", "BA", "c", None])
    result = pc.match_substring_regex(arr, "^a?b", ignore_case=True)
    expected = pa.array([True, True, True, False, None])
    assert expected.equals(result)
    result = pc.match_substring_regex(arr, "^a?b", ignore_case=False)
    expected = pa.array([False, False, False, False, None])
    assert expected.equals(result)


def test_trim():
    # \u3000 is unicode whitespace
    arr = pa.array([" foo", None, " \u3000foo bar \t"])
    result = pc.utf8_trim_whitespace(arr)
    expected = pa.array(["foo", None, "foo bar"])
    assert expected.equals(result)

    arr = pa.array([" foo", None, " \u3000foo bar \t"])
    result = pc.ascii_trim_whitespace(arr)
    expected = pa.array(["foo", None, "\u3000foo bar"])
    assert expected.equals(result)

    arr = pa.array([" foo", None, " \u3000foo bar \t"])
    result = pc.utf8_trim(arr, characters=' f\u3000')
    expected = pa.array(["oo", None, "oo bar \t"])
    assert expected.equals(result)
    # Positional option
    result = pc.utf8_trim(arr, ' f\u3000')
    expected = pa.array(["oo", None, "oo bar \t"])
    assert expected.equals(result)


def test_slice_compatibility():
    arr = pa.array(["", "𝑓", "𝑓ö", "𝑓öõ", "𝑓öõḍ", "𝑓öõḍš"])
    for start in range(-6, 6):
        for stop in itertools.chain(range(-6, 6), [None]):
            for step in [-3, -2, -1, 1, 2, 3]:
                expected = pa.array([k.as_py()[start:stop:step]
                                     for k in arr])
                result = pc.utf8_slice_codeunits(
                    arr, start=start, stop=stop, step=step)
                assert expected.equals(result)
                # Positional options
                assert pc.utf8_slice_codeunits(arr,
                                               start, stop, step) == result


def test_binary_slice_compatibility():
    data = [b"", b"a", b"a\xff", b"ab\x00", b"abc\xfb", b"ab\xf2de"]
    arr = pa.array(data)
    for start, stop, step in itertools.product(range(-6, 6),
                                               range(-6, 6),
                                               range(-3, 4)):
        if step == 0:
            continue
        expected = pa.array([k.as_py()[start:stop:step]
                             for k in arr])
        result = pc.binary_slice(
            arr, start=start, stop=stop, step=step)
        assert expected.equals(result)
        # Positional options
        assert pc.binary_slice(arr, start, stop, step) == result
        # Fixed size binary input / output
        for item in data:
            fsb_scalar = pa.scalar(item, type=pa.binary(len(item)))
            expected = item[start:stop:step]
            actual = pc.binary_slice(fsb_scalar, start, stop, step)
            assert actual.type == pa.binary(len(expected))
            assert actual.as_py() == expected


def test_split_pattern():
    arr = pa.array(["-foo---bar--", "---foo---b"])
    result = pc.split_pattern(arr, pattern="---")
    expected = pa.array([["-foo", "bar--"], ["", "foo", "b"]])
    assert expected.equals(result)

    result = pc.split_pattern(arr, "---", max_splits=1)
    expected = pa.array([["-foo", "bar--"], ["", "foo---b"]])
    assert expected.equals(result)

    result = pc.split_pattern(arr, "---", max_splits=1, reverse=True)
    expected = pa.array([["-foo", "bar--"], ["---foo", "b"]])
    assert expected.equals(result)


def test_split_whitespace_utf8():
    arr = pa.array(["foo bar", " foo  \u3000\tb"])
    result = pc.utf8_split_whitespace(arr)
    expected = pa.array([["foo", "bar"], ["", "foo", "b"]])
    assert expected.equals(result)

    result = pc.utf8_split_whitespace(arr, max_splits=1)
    expected = pa.array([["foo", "bar"], ["", "foo  \u3000\tb"]])
    assert expected.equals(result)

    result = pc.utf8_split_whitespace(arr, max_splits=1, reverse=True)
    expected = pa.array([["foo", "bar"], [" foo", "b"]])
    assert expected.equals(result)


def test_split_whitespace_ascii():
    arr = pa.array(["foo bar", " foo  \u3000\tb"])
    result = pc.ascii_split_whitespace(arr)
    expected = pa.array([["foo", "bar"], ["", "foo", "\u3000", "b"]])
    assert expected.equals(result)

    result = pc.ascii_split_whitespace(arr, max_splits=1)
    expected = pa.array([["foo", "bar"], ["", "foo  \u3000\tb"]])
    assert expected.equals(result)

    result = pc.ascii_split_whitespace(arr, max_splits=1, reverse=True)
    expected = pa.array([["foo", "bar"], [" foo  \u3000", "b"]])
    assert expected.equals(result)


def test_split_pattern_regex():
    arr = pa.array(["-foo---bar--", "---foo---b"])
    result = pc.split_pattern_regex(arr, pattern="-+")
    expected = pa.array([["", "foo", "bar", ""], ["", "foo", "b"]])
    assert expected.equals(result)

    result = pc.split_pattern_regex(arr, "-+", max_splits=1)
    expected = pa.array([["", "foo---bar--"], ["", "foo---b"]])
    assert expected.equals(result)

    with pytest.raises(NotImplementedError,
                       match="Cannot split in reverse with regex"):
        result = pc.split_pattern_regex(
            arr, pattern="---", max_splits=1, reverse=True)


def test_min_max():
    # An example generated function wrapper with possible options
    data = [4, 5, 6, None, 1]
    s = pc.min_max(data)
    assert s.as_py() == {'min': 1, 'max': 6}
    s = pc.min_max(data, options=pc.ScalarAggregateOptions())
    assert s.as_py() == {'min': 1, 'max': 6}
    s = pc.min_max(data, options=pc.ScalarAggregateOptions(skip_nulls=True))
    assert s.as_py() == {'min': 1, 'max': 6}
    s = pc.min_max(data, options=pc.ScalarAggregateOptions(skip_nulls=False))
    assert s.as_py() == {'min': None, 'max': None}

    # Options as dict of kwargs
    s = pc.min_max(data, options={'skip_nulls': False})
    assert s.as_py() == {'min': None, 'max': None}
    # Options as named functions arguments
    s = pc.min_max(data, skip_nulls=False)
    assert s.as_py() == {'min': None, 'max': None}

    # Both options and named arguments
    with pytest.raises(TypeError):
        s = pc.min_max(
            data, options=pc.ScalarAggregateOptions(), skip_nulls=False)

    # Wrong options type
    options = pc.TakeOptions()
    with pytest.raises(TypeError):
        s = pc.min_max(data, options=options)

    # Missing argument
    with pytest.raises(TypeError, match="min_max takes 1 positional"):
        s = pc.min_max()


def test_any():
    # ARROW-1846

    options = pc.ScalarAggregateOptions(skip_nulls=False, min_count=0)

    a = pa.array([], type='bool')
    assert pc.any(a).as_py() is None
    assert pc.any(a, min_count=0).as_py() is False
    assert pc.any(a, options=options).as_py() is False

    a = pa.array([False, None, True])
    assert pc.any(a).as_py() is True
    assert pc.any(a, options=options).as_py() is True

    a = pa.array([False, None, False])
    assert pc.any(a).as_py() is False
    assert pc.any(a, options=options).as_py() is None


def test_all():
    # ARROW-10301

    options = pc.ScalarAggregateOptions(skip_nulls=False, min_count=0)

    a = pa.array([], type='bool')
    assert pc.all(a).as_py() is None
    assert pc.all(a, min_count=0).as_py() is True
    assert pc.all(a, options=options).as_py() is True

    a = pa.array([False, True])
    assert pc.all(a).as_py() is False
    assert pc.all(a, options=options).as_py() is False

    a = pa.array([True, None])
    assert pc.all(a).as_py() is True
    assert pc.all(a, options=options).as_py() is None

    a = pa.chunked_array([[True], [True, None]])
    assert pc.all(a).as_py() is True
    assert pc.all(a, options=options).as_py() is None

    a = pa.chunked_array([[True], [False]])
    assert pc.all(a).as_py() is False
    assert pc.all(a, options=options).as_py() is False


def test_is_valid():
    # An example generated function wrapper without options
    data = [4, 5, None]
    assert pc.is_valid(data).to_pylist() == [True, True, False]

    with pytest.raises(TypeError):
        pc.is_valid(data, options=None)


def test_generated_docstrings():
    # With options
    assert pc.min_max.__doc__ == textwrap.dedent("""\
        Compute the minimum and maximum values of a numeric array.

        Null values are ignored by default.
        This can be changed through ScalarAggregateOptions.

        Parameters
        ----------
        array : Array-like
            Argument to compute function.
        skip_nulls : bool, default True
            Whether to skip (ignore) nulls in the input.
            If False, any null in the input forces the output to null.
        min_count : int, default 1
            Minimum number of non-null values in the input.  If the number
            of non-null values is below `min_count`, the output is null.
        options : pyarrow.compute.ScalarAggregateOptions, optional
            Alternative way of passing options.
        memory_pool : pyarrow.MemoryPool, optional
            If not passed, will allocate memory from the default memory pool.
        """)
    # Without options
    assert pc.add.__doc__ == textwrap.dedent("""\
        Add the arguments element-wise.

        Results will wrap around on integer overflow.
        Use function "add_checked" if you want overflow
        to return an error.

        Parameters
        ----------
        x : Array-like or scalar-like
            Argument to compute function.
        y : Array-like or scalar-like
            Argument to compute function.
        memory_pool : pyarrow.MemoryPool, optional
            If not passed, will allocate memory from the default memory pool.
        """)
    # Varargs with options
    assert pc.min_element_wise.__doc__ == textwrap.dedent("""\
        Find the element-wise minimum value.

        Nulls are ignored (by default) or propagated.
        NaN is preferred over null, but not over any valid value.

        Parameters
        ----------
        *args : Array-like or scalar-like
            Argument to compute function.
        skip_nulls : bool, default True
            Whether to skip (ignore) nulls in the input.
            If False, any null in the input forces the output to null.
        options : pyarrow.compute.ElementWiseAggregateOptions, optional
            Alternative way of passing options.
        memory_pool : pyarrow.MemoryPool, optional
            If not passed, will allocate memory from the default memory pool.
        """)
    assert pc.filter.__doc__ == textwrap.dedent("""\
        Filter with a boolean selection filter.

        The output is populated with values from the input at positions
        where the selection filter is non-zero.  Nulls in the selection filter
        are handled based on FilterOptions.

        Parameters
        ----------
        input : Array-like or scalar-like
            Argument to compute function.
        selection_filter : Array-like or scalar-like
            Argument to compute function.
        null_selection_behavior : str, default "drop"
            How to handle nulls in the selection filter.
            Accepted values are "drop", "emit_null".
        options : pyarrow.compute.FilterOptions, optional
            Alternative way of passing options.
        memory_pool : pyarrow.MemoryPool, optional
            If not passed, will allocate memory from the default memory pool.

        Examples
        --------
        >>> import pyarrow as pa
        >>> arr = pa.array(["a", "b", "c", None, "e"])
        >>> mask = pa.array([True, False, None, False, True])
        >>> arr.filter(mask)
        <pyarrow.lib.StringArray object at ...>
        [
          "a",
          "e"
        ]
        >>> arr.filter(mask, null_selection_behavior='emit_null')
        <pyarrow.lib.StringArray object at ...>
        [
          "a",
          null,
          "e"
        ]
        """)


def test_generated_signatures():
    # The self-documentation provided by signatures should show acceptable
    # options and their default values.

    # Without options
    sig = inspect.signature(pc.add)
    assert str(sig) == "(x, y, /, *, memory_pool=None)"
    # With options
    sig = inspect.signature(pc.min_max)
    assert str(sig) == ("(array, /, *, skip_nulls=True, min_count=1, "
                        "options=None, memory_pool=None)")
    # With positional options
    sig = inspect.signature(pc.quantile)
    assert str(sig) == ("(array, /, q=0.5, *, interpolation='linear', "
                        "skip_nulls=True, min_count=0, "
                        "options=None, memory_pool=None)")
    # Varargs with options
    sig = inspect.signature(pc.binary_join_element_wise)
    assert str(sig) == ("(*strings, null_handling='emit_null', "
                        "null_replacement='', options=None, "
                        "memory_pool=None)")
    # Varargs without options
    sig = inspect.signature(pc.choose)
    assert str(sig) == "(indices, /, *values, memory_pool=None)"
    # Nullary with options
    sig = inspect.signature(pc.random)
    assert str(sig) == ("(n, *, initializer='system', "
                        "options=None, memory_pool=None)")


# We use isprintable to find about codepoints that Python doesn't know, but
# utf8proc does (or in a future version of Python the other way around).
# These codepoints cannot be compared between Arrow and the Python
# implementation.
@lru_cache()
def find_new_unicode_codepoints():
    new = set()
    characters = [chr(c) for c in range(0x80, 0x11000)
                  if not (0xD800 <= c < 0xE000)]
    is_printable = pc.utf8_is_printable(pa.array(characters)).to_pylist()
    for i, c in enumerate(characters):
        if is_printable[i] != c.isprintable():
            new.add(ord(c))
    return new


# Python claims there are not alpha, not sure why, they are in
#  gc='Other Letter': https://graphemica.com/%E1%B3%B2
unknown_issue_is_alpha = {0x1cf2, 0x1cf3}
# utf8proc does not know if codepoints are lower case
utf8proc_issue_is_lower = {
    0xaa, 0xba, 0x2b0, 0x2b1, 0x2b2, 0x2b3, 0x2b4,
    0x2b5, 0x2b6, 0x2b7, 0x2b8, 0x2c0, 0x2c1, 0x2e0,
    0x2e1, 0x2e2, 0x2e3, 0x2e4, 0x37a, 0x1d2c, 0x1d2d,
    0x1d2e, 0x1d2f, 0x1d30, 0x1d31, 0x1d32, 0x1d33,
    0x1d34, 0x1d35, 0x1d36, 0x1d37, 0x1d38, 0x1d39,
    0x1d3a, 0x1d3b, 0x1d3c, 0x1d3d, 0x1d3e, 0x1d3f,
    0x1d40, 0x1d41, 0x1d42, 0x1d43, 0x1d44, 0x1d45,
    0x1d46, 0x1d47, 0x1d48, 0x1d49, 0x1d4a, 0x1d4b,
    0x1d4c, 0x1d4d, 0x1d4e, 0x1d4f, 0x1d50, 0x1d51,
    0x1d52, 0x1d53, 0x1d54, 0x1d55, 0x1d56, 0x1d57,
    0x1d58, 0x1d59, 0x1d5a, 0x1d5b, 0x1d5c, 0x1d5d,
    0x1d5e, 0x1d5f, 0x1d60, 0x1d61, 0x1d62, 0x1d63,
    0x1d64, 0x1d65, 0x1d66, 0x1d67, 0x1d68, 0x1d69,
    0x1d6a, 0x1d78, 0x1d9b, 0x1d9c, 0x1d9d, 0x1d9e,
    0x1d9f, 0x1da0, 0x1da1, 0x1da2, 0x1da3, 0x1da4,
    0x1da5, 0x1da6, 0x1da7, 0x1da8, 0x1da9, 0x1daa,
    0x1dab, 0x1dac, 0x1dad, 0x1dae, 0x1daf, 0x1db0,
    0x1db1, 0x1db2, 0x1db3, 0x1db4, 0x1db5, 0x1db6,
    0x1db7, 0x1db8, 0x1db9, 0x1dba, 0x1dbb, 0x1dbc,
    0x1dbd, 0x1dbe, 0x1dbf, 0x2071, 0x207f, 0x2090,
    0x2091, 0x2092, 0x2093, 0x2094, 0x2095, 0x2096,
    0x2097, 0x2098, 0x2099, 0x209a, 0x209b, 0x209c,
    0x2c7c, 0x2c7d, 0xa69c, 0xa69d, 0xa770, 0xa7f8,
    0xa7f9, 0xab5c, 0xab5d, 0xab5e, 0xab5f, }
# utf8proc does not store if a codepoint is numeric
numeric_info_missing = {
    0x3405, 0x3483, 0x382a, 0x3b4d, 0x4e00, 0x4e03,
    0x4e07, 0x4e09, 0x4e5d, 0x4e8c, 0x4e94, 0x4e96,
    0x4ebf, 0x4ec0, 0x4edf, 0x4ee8, 0x4f0d, 0x4f70,
    0x5104, 0x5146, 0x5169, 0x516b, 0x516d, 0x5341,
    0x5343, 0x5344, 0x5345, 0x534c, 0x53c1, 0x53c2,
    0x53c3, 0x53c4, 0x56db, 0x58f1, 0x58f9, 0x5e7a,
    0x5efe, 0x5eff, 0x5f0c, 0x5f0d, 0x5f0e, 0x5f10,
    0x62fe, 0x634c, 0x67d2, 0x6f06, 0x7396, 0x767e,
    0x8086, 0x842c, 0x8cae, 0x8cb3, 0x8d30, 0x9621,
    0x9646, 0x964c, 0x9678, 0x96f6, 0xf96b, 0xf973,
    0xf978, 0xf9b2, 0xf9d1, 0xf9d3, 0xf9fd, 0x10fc5,
    0x10fc6, 0x10fc7, 0x10fc8, 0x10fc9, 0x10fca,
    0x10fcb, }
# utf8proc has no no digit/numeric information
digit_info_missing = {
    0xb2, 0xb3, 0xb9, 0x1369, 0x136a, 0x136b, 0x136c,
    0x136d, 0x136e, 0x136f, 0x1370, 0x1371, 0x19da, 0x2070,
    0x2074, 0x2075, 0x2076, 0x2077, 0x2078, 0x2079, 0x2080,
    0x2081, 0x2082, 0x2083, 0x2084, 0x2085, 0x2086, 0x2087,
    0x2088, 0x2089, 0x2460, 0x2461, 0x2462, 0x2463, 0x2464,
    0x2465, 0x2466, 0x2467, 0x2468, 0x2474, 0x2475, 0x2476,
    0x2477, 0x2478, 0x2479, 0x247a, 0x247b, 0x247c, 0x2488,
    0x2489, 0x248a, 0x248b, 0x248c, 0x248d, 0x248e, 0x248f,
    0x2490, 0x24ea, 0x24f5, 0x24f6, 0x24f7, 0x24f8, 0x24f9,
    0x24fa, 0x24fb, 0x24fc, 0x24fd, 0x24ff, 0x2776, 0x2777,
    0x2778, 0x2779, 0x277a, 0x277b, 0x277c, 0x277d, 0x277e,
    0x2780, 0x2781, 0x2782, 0x2783, 0x2784, 0x2785, 0x2786,
    0x2787, 0x2788, 0x278a, 0x278b, 0x278c, 0x278d, 0x278e,
    0x278f, 0x2790, 0x2791, 0x2792, 0x10a40, 0x10a41,
    0x10a42, 0x10a43, 0x10e60, 0x10e61, 0x10e62, 0x10e63,
    0x10e64, 0x10e65, 0x10e66, 0x10e67, 0x10e68, }
numeric_info_missing = {
    0x3405, 0x3483, 0x382a, 0x3b4d, 0x4e00, 0x4e03,
    0x4e07, 0x4e09, 0x4e5d, 0x4e8c, 0x4e94, 0x4e96,
    0x4ebf, 0x4ec0, 0x4edf, 0x4ee8, 0x4f0d, 0x4f70,
    0x5104, 0x5146, 0x5169, 0x516b, 0x516d, 0x5341,
    0x5343, 0x5344, 0x5345, 0x534c, 0x53c1, 0x53c2,
    0x53c3, 0x53c4, 0x56db, 0x58f1, 0x58f9, 0x5e7a,
    0x5efe, 0x5eff, 0x5f0c, 0x5f0d, 0x5f0e, 0x5f10,
    0x62fe, 0x634c, 0x67d2, 0x6f06, 0x7396, 0x767e,
    0x8086, 0x842c, 0x8cae, 0x8cb3, 0x8d30, 0x9621,
    0x9646, 0x964c, 0x9678, 0x96f6, 0xf96b, 0xf973,
    0xf978, 0xf9b2, 0xf9d1, 0xf9d3, 0xf9fd, }

codepoints_ignore = {
    'is_alnum': numeric_info_missing | digit_info_missing |
    unknown_issue_is_alpha,
    'is_alpha': unknown_issue_is_alpha,
    'is_digit': digit_info_missing,
    'is_numeric': numeric_info_missing,
    'is_lower': utf8proc_issue_is_lower
}


@pytest.mark.parametrize('function_name', ['is_alnum', 'is_alpha',
                                           'is_ascii', 'is_decimal',
                                           'is_digit', 'is_lower',
                                           'is_numeric', 'is_printable',
                                           'is_space', 'is_upper', ])
@pytest.mark.parametrize('variant', ['ascii', 'utf8'])
def test_string_py_compat_boolean(function_name, variant):
    arrow_name = variant + "_" + function_name
    py_name = function_name.replace('_', '')
    ignore = codepoints_ignore.get(function_name, set()) | \
        find_new_unicode_codepoints()
    for i in range(128 if ascii else 0x11000):
        if i in range(0xD800, 0xE000):
            continue  # bug? pyarrow doesn't allow utf16 surrogates
        # the issues we know of, we skip
        if i in ignore:
            continue
        # Compare results with the equivalent Python predicate
        # (except "is_space" where functions are known to be incompatible)
        c = chr(i)
        if hasattr(pc, arrow_name) and function_name != 'is_space':
            ar = pa.array([c])
            arrow_func = getattr(pc, arrow_name)
            assert arrow_func(ar)[0].as_py() == getattr(c, py_name)()


def test_pad():
    arr = pa.array([None, 'a', 'abcd'])
    assert pc.ascii_center(arr, width=3).tolist() == [None, ' a ', 'abcd']
    assert pc.ascii_lpad(arr, width=3).tolist() == [None, '  a', 'abcd']
    assert pc.ascii_rpad(arr, width=3).tolist() == [None, 'a  ', 'abcd']
    assert pc.ascii_center(arr, 3).tolist() == [None, ' a ', 'abcd']
    assert pc.ascii_lpad(arr, 3).tolist() == [None, '  a', 'abcd']
    assert pc.ascii_rpad(arr, 3).tolist() == [None, 'a  ', 'abcd']

    arr = pa.array([None, 'á', 'abcd'])
    assert pc.utf8_center(arr, width=3).tolist() == [None, ' á ', 'abcd']
    assert pc.utf8_lpad(arr, width=3).tolist() == [None, '  á', 'abcd']
    assert pc.utf8_rpad(arr, width=3).tolist() == [None, 'á  ', 'abcd']
    assert pc.utf8_center(arr, 3).tolist() == [None, ' á ', 'abcd']
    assert pc.utf8_lpad(arr, 3).tolist() == [None, '  á', 'abcd']
    assert pc.utf8_rpad(arr, 3).tolist() == [None, 'á  ', 'abcd']


@pytest.mark.pandas
def test_replace_slice():
    offsets = range(-3, 4)

    arr = pa.array([None, '', 'a', 'ab', 'abc', 'abcd', 'abcde'])
    series = arr.to_pandas().astype(object).replace({np.nan: None})
    for start in offsets:
        for stop in offsets:
            expected = series.str.slice_replace(start, stop, 'XX')
            actual = pc.binary_replace_slice(
                arr, start=start, stop=stop, replacement='XX')
            assert actual.tolist() == expected.tolist()
            # Positional options
            assert pc.binary_replace_slice(arr, start, stop, 'XX') == actual

    arr = pa.array([None, '', 'π', 'πb', 'πbθ', 'πbθd', 'πbθde'])
    series = arr.to_pandas().astype(object).replace({np.nan: None})
    for start in offsets:
        for stop in offsets:
            expected = series.str.slice_replace(start, stop, 'XX')
            actual = pc.utf8_replace_slice(
                arr, start=start, stop=stop, replacement='XX')
            assert actual.tolist() == expected.tolist()


def test_replace_plain():
    data = pa.array(['foozfoo', 'food', None])
    ar = pc.replace_substring(data, pattern='foo', replacement='bar')
    assert ar.tolist() == ['barzbar', 'bard', None]
    ar = pc.replace_substring(data, 'foo', 'bar')
    assert ar.tolist() == ['barzbar', 'bard', None]

    ar = pc.replace_substring(data, pattern='foo', replacement='bar',
                              max_replacements=1)
    assert ar.tolist() == ['barzfoo', 'bard', None]
    ar = pc.replace_substring(data, 'foo', 'bar', max_replacements=1)
    assert ar.tolist() == ['barzfoo', 'bard', None]


def test_replace_regex():
    data = pa.array(['foo', 'mood', None])
    expected = ['f00', 'm00d', None]
    ar = pc.replace_substring_regex(data, pattern='(.)oo', replacement=r'\100')
    assert ar.tolist() == expected
    ar = pc.replace_substring_regex(data, '(.)oo', replacement=r'\100')
    assert ar.tolist() == expected
    ar = pc.replace_substring_regex(data, '(.)oo', r'\100')
    assert ar.tolist() == expected


def test_extract_regex():
    ar = pa.array(['a1', 'zb2z'])
    expected = [{'letter': 'a', 'digit': '1'}, {'letter': 'b', 'digit': '2'}]
    struct = pc.extract_regex(ar, pattern=r'(?P<letter>[ab])(?P<digit>\d)')
    assert struct.tolist() == expected
    struct = pc.extract_regex(ar, r'(?P<letter>[ab])(?P<digit>\d)')
    assert struct.tolist() == expected


def test_binary_join():
    ar_list = pa.array([['foo', 'bar'], None, []])
    expected = pa.array(['foo-bar', None, ''])
    assert pc.binary_join(ar_list, '-').equals(expected)

    separator_array = pa.array(['1', '2'], type=pa.binary())
    expected = pa.array(['a1b', 'c2d'], type=pa.binary())
    ar_list = pa.array([['a', 'b'], ['c', 'd']], type=pa.list_(pa.binary()))
    assert pc.binary_join(ar_list, separator_array).equals(expected)


def test_binary_join_element_wise():
    null = pa.scalar(None, type=pa.string())
    arrs = [[None, 'a', 'b'], ['c', None, 'd'], [None, '-', '--']]
    assert pc.binary_join_element_wise(*arrs).to_pylist() == \
        [None, None, 'b--d']
    assert pc.binary_join_element_wise('a', 'b', '-').as_py() == 'a-b'
    assert pc.binary_join_element_wise('a', null, '-').as_py() is None
    assert pc.binary_join_element_wise('a', 'b', null).as_py() is None

    skip = pc.JoinOptions(null_handling='skip')
    assert pc.binary_join_element_wise(*arrs, options=skip).to_pylist() == \
        [None, 'a', 'b--d']
    assert pc.binary_join_element_wise(
        'a', 'b', '-', options=skip).as_py() == 'a-b'
    assert pc.binary_join_element_wise(
        'a', null, '-', options=skip).as_py() == 'a'
    assert pc.binary_join_element_wise(
        'a', 'b', null, options=skip).as_py() is None

    replace = pc.JoinOptions(null_handling='replace', null_replacement='spam')
    assert pc.binary_join_element_wise(*arrs, options=replace).to_pylist() == \
        [None, 'a-spam', 'b--d']
    assert pc.binary_join_element_wise(
        'a', 'b', '-', options=replace).as_py() == 'a-b'
    assert pc.binary_join_element_wise(
        'a', null, '-', options=replace).as_py() == 'a-spam'
    assert pc.binary_join_element_wise(
        'a', 'b', null, options=replace).as_py() is None


@pytest.mark.parametrize(('ty', 'values'), all_array_types)
def test_take(ty, values):
    arr = pa.array(values, type=ty)
    for indices_type in [pa.int8(), pa.int64()]:
        indices = pa.array([0, 4, 2, None], type=indices_type)
        result = arr.take(indices)
        result.validate()
        expected = pa.array([values[0], values[4], values[2], None], type=ty)
        assert result.equals(expected)

        # empty indices
        indices = pa.array([], type=indices_type)
        result = arr.take(indices)
        result.validate()
        expected = pa.array([], type=ty)
        assert result.equals(expected)

    indices = pa.array([2, 5])
    with pytest.raises(IndexError):
        arr.take(indices)

    indices = pa.array([2, -1])
    with pytest.raises(IndexError):
        arr.take(indices)


def test_take_indices_types():
    arr = pa.array(range(5))

    for indices_type in ['uint8', 'int8', 'uint16', 'int16',
                         'uint32', 'int32', 'uint64', 'int64']:
        indices = pa.array([0, 4, 2, None], type=indices_type)
        result = arr.take(indices)
        result.validate()
        expected = pa.array([0, 4, 2, None])
        assert result.equals(expected)

    for indices_type in [pa.float32(), pa.float64()]:
        indices = pa.array([0, 4, 2], type=indices_type)
        with pytest.raises(NotImplementedError):
            arr.take(indices)


def test_take_on_chunked_array():
    # ARROW-9504
    arr = pa.chunked_array([
        [
            "a",
            "b",
            "c",
            "d",
            "e"
        ],
        [
            "f",
            "g",
            "h",
            "i",
            "j"
        ]
    ])

    indices = pa.array([0, 5, 1, 6, 9, 2])
    result = arr.take(indices)
    expected = pa.chunked_array([["a", "f", "b", "g", "j", "c"]])
    assert result.equals(expected)

    indices = pa.chunked_array([[1], [9, 2]])
    result = arr.take(indices)
    expected = pa.chunked_array([
        [
            "b"
        ],
        [
            "j",
            "c"
        ]
    ])
    assert result.equals(expected)


@pytest.mark.parametrize('ordered', [False, True])
def test_take_dictionary(ordered):
    arr = pa.DictionaryArray.from_arrays([0, 1, 2, 0, 1, 2], ['a', 'b', 'c'],
                                         ordered=ordered)
    result = arr.take(pa.array([0, 1, 3]))
    result.validate()
    assert result.to_pylist() == ['a', 'b', 'a']
    assert result.dictionary.to_pylist() == ['a', 'b', 'c']
    assert result.type.ordered is ordered


def test_take_null_type():
    # ARROW-10027
    arr = pa.array([None] * 10)
    chunked_arr = pa.chunked_array([[None] * 5] * 2)
    batch = pa.record_batch([arr], names=['a'])
    table = pa.table({'a': arr})

    indices = pa.array([1, 3, 7, None])
    assert len(arr.take(indices)) == 4
    assert len(chunked_arr.take(indices)) == 4
    assert len(batch.take(indices).column(0)) == 4
    assert len(table.take(indices).column(0)) == 4


@pytest.mark.parametrize(('ty', 'values'), all_array_types)
def test_drop_null(ty, values):
    arr = pa.array(values, type=ty)
    result = arr.drop_null()
    result.validate(full=True)
    indices = [i for i in range(len(arr)) if arr[i].is_valid]
    expected = arr.take(pa.array(indices))
    assert result.equals(expected)


def test_drop_null_chunked_array():
    arr = pa.chunked_array([["a", None], ["c", "d", None], [None], []])
    expected_drop = pa.chunked_array([["a"], ["c", "d"], [], []])

    result = arr.drop_null()
    assert result.equals(expected_drop)


def test_drop_null_record_batch():
    batch = pa.record_batch(
        [pa.array(["a", None, "c", "d", None])], names=["a'"])
    result = batch.drop_null()
    expected = pa.record_batch([pa.array(["a", "c", "d"])], names=["a'"])
    assert result.equals(expected)

    batch = pa.record_batch(
        [pa.array(["a", None, "c", "d", None]),
         pa.array([None, None, "c", None, "e"])], names=["a'", "b'"])

    result = batch.drop_null()
    expected = pa.record_batch(
        [pa.array(["c"]), pa.array(["c"])], names=["a'", "b'"])
    assert result.equals(expected)


def test_drop_null_table():
    table = pa.table([pa.array(["a", None, "c", "d", None])], names=["a"])
    expected = pa.table([pa.array(["a", "c", "d"])], names=["a"])
    result = table.drop_null()
    assert result.equals(expected)

    table = pa.table([pa.chunked_array([["a", None], ["c", "d", None]]),
                      pa.chunked_array([["a", None], [None, "d", None]]),
                      pa.chunked_array([["a"], ["b"], [None], ["d", None]])],
                     names=["a", "b", "c"])
    expected = pa.table([pa.array(["a", "d"]),
                         pa.array(["a", "d"]),
                         pa.array(["a", "d"])],
                        names=["a", "b", "c"])
    result = table.drop_null()
    assert result.equals(expected)

    table = pa.table([pa.chunked_array([["a", "b"], ["c", "d", "e"]]),
                      pa.chunked_array([["A"], ["B"], [None], ["D", None]]),
                      pa.chunked_array([["a`", None], ["c`", "d`", None]])],
                     names=["a", "b", "c"])
    expected = pa.table([pa.array(["a", "d"]),
                         pa.array(["A", "D"]),
                         pa.array(["a`", "d`"])],
                        names=["a", "b", "c"])
    result = table.drop_null()
    assert result.equals(expected)


def test_drop_null_null_type():
    arr = pa.array([None] * 10)
    chunked_arr = pa.chunked_array([[None] * 5] * 2)
    batch = pa.record_batch([arr], names=['a'])
    table = pa.table({'a': arr})

    assert len(arr.drop_null()) == 0
    assert len(chunked_arr.drop_null()) == 0
    assert len(batch.drop_null().column(0)) == 0
    assert len(table.drop_null().column(0)) == 0


@pytest.mark.parametrize(('ty', 'values'), all_array_types)
def test_filter(ty, values):
    arr = pa.array(values, type=ty)

    mask = pa.array([True, False, False, True, None])
    result = arr.filter(mask, null_selection_behavior='drop')
    result.validate()
    assert result.equals(pa.array([values[0], values[3]], type=ty))
    result = arr.filter(mask, null_selection_behavior='emit_null')
    result.validate()
    assert result.equals(pa.array([values[0], values[3], None], type=ty))

    # non-boolean dtype
    mask = pa.array([0, 1, 0, 1, 0])
    with pytest.raises(NotImplementedError):
        arr.filter(mask)

    # wrong length
    mask = pa.array([True, False, True])
    with pytest.raises(ValueError, match="must all be the same length"):
        arr.filter(mask)


@pytest.mark.numpy
@pytest.mark.parametrize(('ty', 'values'), all_array_types)
def test_filter_numpy_array_mask(ty, values):
    arr = pa.array(values, type=ty)
    # same test as test_filter with different array type
    mask = np.array([True, False, False, True, None])
    result = arr.filter(mask, null_selection_behavior='drop')
    result.validate()
    assert result.equals(pa.array([values[0], values[3]], type=ty))


def test_filter_chunked_array():
    arr = pa.chunked_array([["a", None], ["c", "d", "e"]])
    expected_drop = pa.chunked_array([["a"], ["e"]])
    expected_null = pa.chunked_array([["a"], [None, "e"]])

    for mask in [
        # mask is array
        pa.array([True, False, None, False, True]),
        # mask is chunked array
        pa.chunked_array([[True, False, None], [False, True]]),
        # mask is python object
        [True, False, None, False, True]
    ]:
        result = arr.filter(mask)
        assert result.equals(expected_drop)
        result = arr.filter(mask, null_selection_behavior="emit_null")
        assert result.equals(expected_null)


def test_filter_record_batch():
    batch = pa.record_batch(
        [pa.array(["a", None, "c", "d", "e"])], names=["a'"])

    # mask is array
    mask = pa.array([True, False, None, False, True])
    result = batch.filter(mask)
    expected = pa.record_batch([pa.array(["a", "e"])], names=["a'"])
    assert result.equals(expected)

    # GH-38770: mask is chunked array
    chunked_mask = pa.chunked_array([[True, False], [None], [False, True]])
    result = batch.filter(chunked_mask)
    assert result.equals(expected)

    result = batch.filter(mask, null_selection_behavior="emit_null")
    expected = pa.record_batch([pa.array(["a", None, "e"])], names=["a'"])
    assert result.equals(expected)


def test_filter_table():
    table = pa.table([pa.array(["a", None, "c", "d", "e"])], names=["a"])
    expected_drop = pa.table([pa.array(["a", "e"])], names=["a"])
    expected_null = pa.table([pa.array(["a", None, "e"])], names=["a"])

    for mask in [
        # mask is array
        pa.array([True, False, None, False, True]),
        # mask is chunked array
        pa.chunked_array([[True, False], [None, False, True]]),
        # mask is python object
        [True, False, None, False, True]
    ]:
        result = table.filter(mask)
        assert result.equals(expected_drop)
        result = table.filter(mask, null_selection_behavior="emit_null")
        assert result.equals(expected_null)


def test_filter_errors():
    arr = pa.chunked_array([["a", None], ["c", "d", "e"]])
    batch = pa.record_batch(
        [pa.array(["a", None, "c", "d", "e"])], names=["a'"])
    table = pa.table([pa.array(["a", None, "c", "d", "e"])], names=["a"])

    for obj in [arr, batch, table]:
        # non-boolean dtype
        mask = pa.array([0, 1, 0, 1, 0])
        with pytest.raises(NotImplementedError):
            obj.filter(mask)

        # wrong length
        mask = pa.array([True, False, True])
        with pytest.raises(pa.ArrowInvalid,
                           match="must all be the same length"):
            obj.filter(mask)

    scalar = pa.scalar(True)
    for filt in [batch, table, scalar]:
        with pytest.raises(TypeError):
            table.filter(filt)


def test_filter_null_type():
    # ARROW-10027
    arr = pa.array([None] * 10)
    chunked_arr = pa.chunked_array([[None] * 5] * 2)
    batch = pa.record_batch([arr], names=['a'])
    table = pa.table({'a': arr})

    mask = pa.array([True, False] * 5)
    assert len(arr.filter(mask)) == 5
    assert len(chunked_arr.filter(mask)) == 5
    assert len(batch.filter(mask).column(0)) == 5
    assert len(table.filter(mask).column(0)) == 5


@pytest.mark.parametrize("typ", ["array", "chunked_array"])
def test_compare_array(typ):
    if typ == "array":
        def con(values):
            return pa.array(values)
    else:
        def con(values):
            return pa.chunked_array([values])

    arr1 = con([1, 2, 3, 4, None])
    arr2 = con([1, 1, 4, None, 4])

    result = pc.equal(arr1, arr2)
    assert result.equals(con([True, False, False, None, None]))

    result = pc.not_equal(arr1, arr2)
    assert result.equals(con([False, True, True, None, None]))

    result = pc.less(arr1, arr2)
    assert result.equals(con([False, False, True, None, None]))

    result = pc.less_equal(arr1, arr2)
    assert result.equals(con([True, False, True, None, None]))

    result = pc.greater(arr1, arr2)
    assert result.equals(con([False, True, False, None, None]))

    result = pc.greater_equal(arr1, arr2)
    assert result.equals(con([True, True, False, None, None]))


@pytest.mark.parametrize("typ", ["array", "chunked_array"])
def test_compare_string_scalar(typ):
    if typ == "array":
        def con(values):
            return pa.array(values)
    else:
        def con(values):
            return pa.chunked_array([values])

    arr = con(['a', 'b', 'c', None])
    scalar = pa.scalar('b')

    result = pc.equal(arr, scalar)
    assert result.equals(con([False, True, False, None]))

    if typ == "array":
        nascalar = pa.scalar(None, type="string")
        result = pc.equal(arr, nascalar)
        isnull = pc.is_null(result)
        assert isnull.equals(con([True, True, True, True]))

    result = pc.not_equal(arr, scalar)
    assert result.equals(con([True, False, True, None]))

    result = pc.less(arr, scalar)
    assert result.equals(con([True, False, False, None]))

    result = pc.less_equal(arr, scalar)
    assert result.equals(con([True, True, False, None]))

    result = pc.greater(arr, scalar)
    assert result.equals(con([False, False, True, None]))

    result = pc.greater_equal(arr, scalar)
    assert result.equals(con([False, True, True, None]))


@pytest.mark.parametrize("typ", ["array", "chunked_array"])
def test_compare_scalar(typ):
    if typ == "array":
        def con(values):
            return pa.array(values)
    else:
        def con(values):
            return pa.chunked_array([values])

    arr = con([1, 2, 3, None])
    scalar = pa.scalar(2)

    result = pc.equal(arr, scalar)
    assert result.equals(con([False, True, False, None]))

    if typ == "array":
        nascalar = pa.scalar(None, type="int64")
        result = pc.equal(arr, nascalar)
        assert result.to_pylist() == [None, None, None, None]

    result = pc.not_equal(arr, scalar)
    assert result.equals(con([True, False, True, None]))

    result = pc.less(arr, scalar)
    assert result.equals(con([True, False, False, None]))

    result = pc.less_equal(arr, scalar)
    assert result.equals(con([True, True, False, None]))

    result = pc.greater(arr, scalar)
    assert result.equals(con([False, False, True, None]))

    result = pc.greater_equal(arr, scalar)
    assert result.equals(con([False, True, True, None]))


def test_compare_chunked_array_mixed():
    arr = pa.array([1, 2, 3, 4, None])
    arr_chunked = pa.chunked_array([[1, 2, 3], [4, None]])
    arr_chunked2 = pa.chunked_array([[1, 2], [3, 4, None]])

    expected = pa.chunked_array([[True, True, True, True, None]])

    for left, right in [
        (arr, arr_chunked),
        (arr_chunked, arr),
        (arr_chunked, arr_chunked2),
    ]:
        result = pc.equal(left, right)
        assert result.equals(expected)


def test_arithmetic_add():
    left = pa.array([1, 2, 3, 4, 5])
    right = pa.array([0, -1, 1, 2, 3])
    result = pc.add(left, right)
    expected = pa.array([1, 1, 4, 6, 8])
    assert result.equals(expected)


def test_arithmetic_subtract():
    left = pa.array([1, 2, 3, 4, 5])
    right = pa.array([0, -1, 1, 2, 3])
    result = pc.subtract(left, right)
    expected = pa.array([1, 3, 2, 2, 2])
    assert result.equals(expected)


def test_arithmetic_multiply():
    left = pa.array([1, 2, 3, 4, 5])
    right = pa.array([0, -1, 1, 2, 3])
    result = pc.multiply(left, right)
    expected = pa.array([0, -2, 3, 8, 15])
    assert result.equals(expected)


@pytest.mark.parametrize("ty", ["round", "round_to_multiple"])
def test_round_to_integer(ty):
    if ty == "round":
        round = pc.round
        RoundOptions = partial(pc.RoundOptions, ndigits=0)
    elif ty == "round_to_multiple":
        round = pc.round_to_multiple
        RoundOptions = partial(pc.RoundToMultipleOptions, multiple=1)

    values = [3.2, 3.5, 3.7, 4.5, -3.2, -3.5, -3.7, None]
    rmode_and_expected = {
        "down": [3, 3, 3, 4, -4, -4, -4, None],
        "up": [4, 4, 4, 5, -3, -3, -3, None],
        "towards_zero": [3, 3, 3, 4, -3, -3, -3, None],
        "towards_infinity": [4, 4, 4, 5, -4, -4, -4, None],
        "half_down": [3, 3, 4, 4, -3, -4, -4, None],
        "half_up": [3, 4, 4, 5, -3, -3, -4, None],
        "half_towards_zero": [3, 3, 4, 4, -3, -3, -4, None],
        "half_towards_infinity": [3, 4, 4, 5, -3, -4, -4, None],
        "half_to_even": [3, 4, 4, 4, -3, -4, -4, None],
        "half_to_odd": [3, 3, 4, 5, -3, -3, -4, None],
    }
    for round_mode, expected in rmode_and_expected.items():
        options = RoundOptions(round_mode=round_mode)
        result = round(values, options=options)
        expected_array = pa.array(expected, type=pa.float64())
        assert expected_array.equals(result)


@pytest.mark.numpy
def test_round():
    values = [320, 3.5, 3.075, 4.5, -3.212, -35.1234, -3.045, None]
    ndigits_and_expected = {
        -2: [300, 0, 0, 0, -0, -0, -0, None],
        -1: [320, 0, 0, 0, -0, -40, -0, None],
        0: [320, 4, 3, 5, -3, -35, -3, None],
        1: [320, 3.5, 3.1, 4.5, -3.2, -35.1, -3, None],
        2: [320, 3.5, 3.08, 4.5, -3.21, -35.12, -3.05, None],
    }
    for ndigits, expected in ndigits_and_expected.items():
        options = pc.RoundOptions(ndigits, "half_towards_infinity")
        result = pc.round(values, options=options)
        np.testing.assert_allclose(result, pa.array(expected), equal_nan=True)
        assert pc.round(values, ndigits,
                        round_mode="half_towards_infinity") == result
        assert pc.round(values, ndigits, "half_towards_infinity") == result


@pytest.mark.numpy
def test_round_to_multiple():
    values = [320, 3.5, 3.075, 4.5, -3.212, -35.1234, -3.045, None]
    multiple_and_expected = {
        0.05: [320, 3.5, 3.1, 4.5, -3.2, -35.1, -3.05, None],
        pa.scalar(0.1): [320, 3.5, 3.1, 4.5, -3.2, -35.1, -3, None],
        2: [320, 4, 4, 4, -4, -36, -4, None],
        10: [320, 0, 0, 0, -0, -40, -0, None],
        pa.scalar(100, type=pa.decimal256(10, 4)):
            [300, 0, 0, 0, -0, -0, -0, None],
    }
    for multiple, expected in multiple_and_expected.items():
        options = pc.RoundToMultipleOptions(multiple, "half_towards_infinity")
        result = pc.round_to_multiple(values, options=options)
        np.testing.assert_allclose(result, pa.array(expected), equal_nan=True)
        assert pc.round_to_multiple(values, multiple,
                                    "half_towards_infinity") == result

    for multiple in [0, -2, pa.scalar(-10.4)]:
        with pytest.raises(pa.ArrowInvalid,
                           match="Rounding multiple must be positive"):
            pc.round_to_multiple(values, multiple=multiple)

    for multiple in [object, 99999999999999999999999]:
        with pytest.raises(TypeError, match="is not a valid multiple type"):
            pc.round_to_multiple(values, multiple=multiple)


def test_round_binary():
    values = [123.456, 234.567, 345.678, 456.789, 123.456, 234.567, 345.678]
    scales = pa.array([-3, -2, -1, 0, 1, 2, 3], pa.int32())
    expected = pa.array(
        [0, 200, 350, 457, 123.5, 234.57, 345.678], pa.float64())
    assert pc.round_binary(values, scales) == expected

    expect_zero = pa.scalar(0, pa.float64())
    expect_inf = pa.scalar(10, pa.float64())
    scale = pa.scalar(-1, pa.int32())

    assert pc.round_binary(
        5.0, scale, round_mode="half_towards_zero") == expect_zero
    assert pc.round_binary(
        5.0, scale, round_mode="half_towards_infinity") == expect_inf


def test_is_null():
    arr = pa.array([1, 2, 3, None])
    result = arr.is_null()
    expected = pa.array([False, False, False, True])
    assert result.equals(expected)
    assert result.equals(pc.is_null(arr))
    result = arr.is_valid()
    expected = pa.array([True, True, True, False])
    assert result.equals(expected)
    assert result.equals(pc.is_valid(arr))

    arr = pa.chunked_array([[1, 2], [3, None]])
    result = arr.is_null()
    expected = pa.chunked_array([[False, False], [False, True]])
    assert result.equals(expected)
    result = arr.is_valid()
    expected = pa.chunked_array([[True, True], [True, False]])
    assert result.equals(expected)

    arr = pa.array([1, 2, 3, None, float("nan")])
    result = arr.is_null()
    expected = pa.array([False, False, False, True, False])
    assert result.equals(expected)

    result = arr.is_null(nan_is_null=True)
    expected = pa.array([False, False, False, True, True])
    assert result.equals(expected)


def test_is_nan():
    arr = pa.array([1, 2, 3, None, float("nan")])
    result = arr.is_nan()
    expected = pa.array([False, False, False, None, True])
    assert result.equals(expected)

    arr = pa.array(["1", "2", None], type=pa.string())
    with pytest.raises(
            ArrowNotImplementedError, match="has no kernel matching input types"):
        _ = arr.is_nan()

    with pytest.raises(
            ArrowNotImplementedError, match="has no kernel matching input types"):
        arr = pa.array([b'a', b'bb', None], type=pa.large_binary())
        _ = arr.is_nan()


def test_fill_null():
    arr = pa.array([1, 2, None, 4], type=pa.int8())
    fill_value = pa.array([5], type=pa.int8())
    with pytest.raises(pa.ArrowInvalid,
                       match="Array arguments must all be the same length"):
        arr.fill_null(fill_value)

    arr = pa.array([None, None, None, None], type=pa.null())
    fill_value = pa.scalar(None, type=pa.null())
    result = arr.fill_null(fill_value)
    expected = pa.array([None, None, None, None])
    assert result.equals(expected)

    arr = pa.array(['a', 'bb', None])
    result = arr.fill_null('ccc')
    expected = pa.array(['a', 'bb', 'ccc'])
    assert result.equals(expected)

    arr = pa.array([b'a', b'bb', None], type=pa.large_binary())
    result = arr.fill_null('ccc')
    expected = pa.array([b'a', b'bb', b'ccc'], type=pa.large_binary())
    assert result.equals(expected)

    arr = pa.array(['a', 'bb', None])
    result = arr.fill_null(None)
    expected = pa.array(['a', 'bb', None])
    assert result.equals(expected)


@pytest.mark.parametrize('arrow_type', numerical_arrow_types)
def test_fill_null_array(arrow_type):
    arr = pa.array([1, 2, None, 4], type=arrow_type)
    fill_value = pa.scalar(5, type=arrow_type)
    result = arr.fill_null(fill_value)
    expected = pa.array([1, 2, 5, 4], type=arrow_type)
    assert result.equals(expected)

    # Implicit conversions
    result = arr.fill_null(5)
    assert result.equals(expected)

    # ARROW-9451: Unsigned integers allow this for some reason
    if not pa.types.is_unsigned_integer(arr.type):
        with pytest.raises((ValueError, TypeError)):
            arr.fill_null('5')

    result = arr.fill_null(pa.scalar(5, type='int8'))
    assert result.equals(expected)


@pytest.mark.parametrize('arrow_type', numerical_arrow_types)
def test_fill_null_chunked_array(arrow_type):
    fill_value = pa.scalar(5, type=arrow_type)
    arr = pa.chunked_array([pa.array([None, 2, 3, 4], type=arrow_type)])
    result = arr.fill_null(fill_value)
    expected = pa.chunked_array([pa.array([5, 2, 3, 4], type=arrow_type)])
    assert result.equals(expected)

    arr = pa.chunked_array([
        pa.array([1, 2], type=arrow_type),
        pa.array([], type=arrow_type),
        pa.array([None, 4], type=arrow_type)
    ])
    expected = pa.chunked_array([
        pa.array([1, 2], type=arrow_type),
        pa.array([], type=arrow_type),
        pa.array([5, 4], type=arrow_type)
    ])
    result = arr.fill_null(fill_value)
    assert result.equals(expected)

    # Implicit conversions
    result = arr.fill_null(5)
    assert result.equals(expected)

    result = arr.fill_null(pa.scalar(5, type='int8'))
    assert result.equals(expected)


def test_logical():
    a = pa.array([True, False, False, None])
    b = pa.array([True, True, False, True])

    assert pc.and_(a, b) == pa.array([True, False, False, None])
    assert pc.and_kleene(a, b) == pa.array([True, False, False, None])

    assert pc.or_(a, b) == pa.array([True, True, False, None])
    assert pc.or_kleene(a, b) == pa.array([True, True, False, True])

    assert pc.xor(a, b) == pa.array([False, True, False, None])

    assert pc.invert(a) == pa.array([False, True, True, None])


def test_dictionary_decode():
    array = pa.array(["a", "a", "b", "c", "b"])
    dictionary_array = array.dictionary_encode()
    dictionary_array_decode = pc.dictionary_decode(dictionary_array)

    assert array != dictionary_array

    assert array == dictionary_array_decode
    assert array == pc.dictionary_decode(array)
    assert pc.dictionary_encode(dictionary_array) == dictionary_array


def test_cast():
    arr = pa.array([1, 2, 3, 4], type='int64')
    options = pc.CastOptions(pa.int8())

    with pytest.raises(TypeError):
        pc.cast(arr, target_type=None)

    with pytest.raises(ValueError):
        pc.cast(arr, 'int32', options=options)

    with pytest.raises(ValueError):
        pc.cast(arr, safe=True, options=options)

    assert pc.cast(arr, options=options) == pa.array(
        [1, 2, 3, 4], type='int8')

    arr = pa.array([2 ** 63 - 1], type='int64')
    allow_overflow_options = pc.CastOptions(
        pa.int32(), allow_int_overflow=True)

    with pytest.raises(pa.ArrowInvalid):
        pc.cast(arr, 'int32')

    assert pc.cast(arr, 'int32', safe=False) == pa.array([-1], type='int32')

    assert pc.cast(arr, options=allow_overflow_options) == pa.array(
        [-1], type='int32')

    arr = pa.array(
        [datetime.datetime(2010, 1, 1), datetime.datetime(2015, 1, 1)])
    expected = pa.array([1262304000000, 1420070400000], type='timestamp[ms]')
    assert pc.cast(arr, 'timestamp[ms]') == expected

    arr = pa.array([[1, 2], [3, 4, 5]], type=pa.large_list(pa.int8()))
    expected = pa.array([["1", "2"], ["3", "4", "5"]],
                        type=pa.list_(pa.utf8()))
    assert pc.cast(arr, expected.type) == expected


@pytest.mark.parametrize('value_type', [pa.date32(), pa.date64()])
def test_identity_cast_dates(value_type):
    dt = datetime.date(1990, 3, 1)

    arr = pa.array([dt], type=value_type)
    assert pc.cast(arr, value_type) == arr


@pytest.mark.parametrize('value_type', numerical_arrow_types)
def test_fsl_to_fsl_cast(value_type):
    # Different field name and different type.
    cast_type = pa.list_(pa.field("element", value_type), 2)

    dtype = pa.int32()
    type = pa.list_(pa.field("values", dtype), 2)

    fsl = pa.FixedSizeListArray.from_arrays(
        pa.array([1, 2, 3, 4, 5, 6], type=dtype), type=type)
    assert cast_type == fsl.cast(cast_type).type

    # Different field name and different type (with null values).
    fsl = pa.FixedSizeListArray.from_arrays(
        pa.array([1, None, None, 4, 5, 6], type=dtype), type=type)
    assert cast_type == fsl.cast(cast_type).type

    # Null FSL type.
    dtype = pa.null()
    type = pa.list_(pa.field("values", dtype), 2)
    fsl = pa.FixedSizeListArray.from_arrays(
        pa.array([None, None, None, None, None, None], type=dtype), type=type)
    assert cast_type == fsl.cast(cast_type).type

    # Different sized FSL
    cast_type = pa.list_(pa.field("element", value_type), 3)
    err_msg = 'Size of FixedSizeList is not the same.'
    with pytest.raises(pa.lib.ArrowTypeError, match=err_msg):
        fsl.cast(cast_type)


DecimalTypeTraits = namedtuple('DecimalTypeTraits',
                               ('name', 'factory', 'max_precision'))

FloatToDecimalCase = namedtuple('FloatToDecimalCase',
                                ('precision', 'scale', 'float_val'))

decimal_type_traits = [DecimalTypeTraits('decimal32', pa.decimal32, 9),
                       DecimalTypeTraits('decimal64', pa.decimal64, 18),
                       DecimalTypeTraits('decimal128', pa.decimal128, 38),
                       DecimalTypeTraits('decimal256', pa.decimal256, 76)]


def largest_scaled_float_not_above(val, scale):
    """
    Find the largest float f such as `f * 10**scale <= val`
    """
    assert val >= 0
    assert scale >= 0
    float_val = float(val) / 10**scale
    if float_val * 10**scale > val:
        # Take the float just below... it *should* satisfy
        float_val = np.nextafter(float_val, 0.0)
        if float_val * 10**scale > val:
            float_val = np.nextafter(float_val, 0.0)
    assert float_val * 10**scale <= val
    return float_val


def scaled_float(int_val, scale):
    """
    Return a float representation (possibly approximate) of `int_val**-scale`
    """
    assert isinstance(int_val, int)
    unscaled = decimal.Decimal(int_val)
    scaled = unscaled.scaleb(-scale)
    float_val = float(scaled)
    return float_val


def integral_float_to_decimal_cast_cases(float_ty, max_precision):
    """
    Return FloatToDecimalCase instances with integral values.
    """
    mantissa_digits = 16
    for precision in range(1, max_precision, 3):
        for scale in range(0, precision, 2):
            yield FloatToDecimalCase(precision, scale, 0.0)
            yield FloatToDecimalCase(precision, scale, 1.0)
            epsilon = 10**max(precision - mantissa_digits, scale)
            abs_maxval = largest_scaled_float_not_above(
                10**precision - epsilon, scale)
            yield FloatToDecimalCase(precision, scale, abs_maxval)


def real_float_to_decimal_cast_cases(float_ty, max_precision):
    """
    Return FloatToDecimalCase instances with real values.
    """
    mantissa_digits = 16
    for precision in range(1, max_precision, 3):
        for scale in range(0, precision, 2):
            epsilon = 2 * 10**max(precision - mantissa_digits, 0)
            abs_minval = largest_scaled_float_not_above(epsilon, scale)
            abs_maxval = largest_scaled_float_not_above(
                10**precision - epsilon, scale)
            yield FloatToDecimalCase(precision, scale, abs_minval)
            yield FloatToDecimalCase(precision, scale, abs_maxval)


def random_float_to_decimal_cast_cases(float_ty, max_precision):
    """
    Return random-generated FloatToDecimalCase instances.
    """
    r = random.Random(42)
    for precision in range(1, max_precision, 6):
        for scale in range(0, precision, 4):
            for i in range(20):
                unscaled = r.randrange(0, 10**precision)
                float_val = scaled_float(unscaled, scale)
                assert float_val * 10**scale < 10**precision
                yield FloatToDecimalCase(precision, scale, float_val)


def check_cast_float_to_decimal(float_ty, float_val, decimal_ty, decimal_ctx,
                                max_precision):
    # Use the Python decimal module to build the expected result
    # using the right precision
    decimal_ctx.prec = decimal_ty.precision
    decimal_ctx.rounding = decimal.ROUND_HALF_EVEN
    expected = decimal_ctx.create_decimal_from_float(float_val)
    # Round `expected` to `scale` digits after the decimal point
    expected = expected.quantize(decimal.Decimal(1).scaleb(-decimal_ty.scale))
    s = pa.scalar(float_val, type=float_ty)
    actual = pc.cast(s, decimal_ty).as_py()
    if actual != expected:
        # Allow the last digit to vary. The tolerance is higher for
        # very high precisions as rounding errors can accumulate in
        # the iterative algorithm (GH-35576).
        diff_digits = abs(actual - expected) * 10**decimal_ty.scale
        limit = 2 if decimal_ty.precision < max_precision - 2 else 4
        assert diff_digits <= limit, (
            f"float_val = {float_val!r}, precision={decimal_ty.precision}, "
            f"expected = {expected!r}, actual = {actual!r}, "
            f"diff_digits = {diff_digits!r}")


# Cannot test float32 as case generators above assume float64
@pytest.mark.numpy
@pytest.mark.parametrize('float_ty', [pa.float64()], ids=str)
@pytest.mark.parametrize('decimal_ty', decimal_type_traits,
                         ids=lambda v: v.name)
@pytest.mark.parametrize('case_generator',
                         [integral_float_to_decimal_cast_cases,
                          real_float_to_decimal_cast_cases,
                          random_float_to_decimal_cast_cases],
                         ids=['integrals', 'reals', 'random'])
def test_cast_float_to_decimal(float_ty, decimal_ty, case_generator):
    with decimal.localcontext() as ctx:
        for case in case_generator(float_ty, decimal_ty.max_precision):
            check_cast_float_to_decimal(
                float_ty, case.float_val,
                decimal_ty.factory(case.precision, case.scale),
                ctx, decimal_ty.max_precision)


@pytest.mark.numpy
@pytest.mark.parametrize('float_ty', [pa.float32(), pa.float64()], ids=str)
@pytest.mark.parametrize('decimal_traits', decimal_type_traits,
                         ids=lambda v: v.name)
def test_cast_float_to_decimal_random(float_ty, decimal_traits):
    """
    Test float-to-decimal conversion against exactly generated values.
    """
    r = random.Random(43)
    np_float_ty = {
        pa.float32(): np.float32,
        pa.float64(): np.float64,
    }[float_ty]
    mantissa_bits = {
        pa.float32(): 24,
        pa.float64(): 53,
    }[float_ty]
    float_exp_min, float_exp_max = {
        pa.float32(): (-126, 127),
        pa.float64(): (-1022, 1023),
    }[float_ty]
    mantissa_digits = math.floor(math.log10(2**mantissa_bits))
    max_precision = decimal_traits.max_precision

    # For example, decimal32 <-> float64
    if max_precision < mantissa_digits:
        mantissa_bits = math.floor(math.log2(10**max_precision))
        mantissa_digits = math.floor(math.log10(2**mantissa_bits))

    with decimal.localcontext() as ctx:
        precision = mantissa_digits
        ctx.prec = precision
        # The scale must be chosen so as
        # 1) it's within bounds for the decimal type
        # 2) the floating point exponent is within bounds
        min_scale = max(-max_precision,
                        precision + math.ceil(math.log10(2**float_exp_min)))
        max_scale = min(max_precision,
                        math.floor(math.log10(2**float_exp_max)))
        for scale in range(min_scale, max_scale):
            decimal_ty = decimal_traits.factory(precision, scale)
            # We want to random-generate a float from its mantissa bits
            # and exponent, and compute the expected value in the
            # decimal domain. The float exponent has to ensure the
            # expected value doesn't overflow and doesn't lose precision.
            float_exp = (-mantissa_bits +
                         math.floor(math.log2(10**(precision - scale))))
            assert float_exp_min <= float_exp <= float_exp_max
            for i in range(5):
                mantissa = r.randrange(0, 2**mantissa_bits)
                float_val = np.ldexp(np_float_ty(mantissa), float_exp)
                assert isinstance(float_val, np_float_ty)
                # Make sure we compute the exact expected value and
                # round by half-to-even when converting to the expected precision.
                if float_exp >= 0:
                    expected = decimal.Decimal(mantissa) * 2**float_exp
                else:
                    expected = decimal.Decimal(mantissa) / 2**-float_exp
                expected_as_int = round(expected.scaleb(scale))
                actual = pc.cast(
                    pa.scalar(float_val, type=float_ty), decimal_ty).as_py()
                actual_as_int = round(actual.scaleb(scale))
                # We allow for a minor rounding error between expected and actual
                assert abs(actual_as_int - expected_as_int) <= 1


def test_strptime():
    arr = pa.array(["5/1/2020", None, "12/13/1900"])

    got = pc.strptime(arr, format='%m/%d/%Y', unit='s')
    expected = pa.array(
        [datetime.datetime(2020, 5, 1), None, datetime.datetime(1900, 12, 13)],
        type=pa.timestamp('s'))
    assert got == expected
    # Positional format
    assert pc.strptime(arr, '%m/%d/%Y', unit='s') == got

    expected = pa.array([datetime.datetime(2020, 1, 5), None, None],
                        type=pa.timestamp('s'))
    got = pc.strptime(arr, format='%d/%m/%Y', unit='s', error_is_null=True)
    assert got == expected

    with pytest.raises(pa.ArrowInvalid,
                       match="Failed to parse string: '5/1/2020'"):
        pc.strptime(arr, format='%Y-%m-%d', unit='s', error_is_null=False)

    with pytest.raises(pa.ArrowInvalid,
                       match="Failed to parse string: '5/1/2020'"):
        pc.strptime(arr, format='%Y-%m-%d', unit='s')

    got = pc.strptime(arr, format='%Y-%m-%d', unit='s', error_is_null=True)
    assert got == pa.array([None, None, None], type=pa.timestamp('s'))


@pytest.mark.pandas
@pytest.mark.timezone_data
def test_strftime():
    times = ["2018-03-10 09:00", "2038-01-31 12:23", None]
    timezones = ["CET", "UTC", "Europe/Ljubljana"]

    formats = ["%a", "%A", "%w", "%d", "%b", "%B", "%m", "%y", "%Y", "%H", "%I",
               "%p", "%M", "%z", "%Z", "%j", "%U", "%W", "%%", "%G", "%V", "%u"]
    if sys.platform != "win32":
        # Locale-dependent formats don't match on Windows
        formats.extend(["%c", "%x", "%X"])

    for timezone in timezones:
        ts = pd.to_datetime(times).tz_localize(timezone)
        for unit in ["s", "ms", "us", "ns"]:
            tsa = pa.array(ts, type=pa.timestamp(unit, timezone))
            for fmt in formats:
                options = pc.StrftimeOptions(fmt)
                result = pc.strftime(tsa, options=options)
                # cast to the same type as result to ignore string vs large_string
                expected = pa.array(ts.strftime(fmt)).cast(result.type)
                assert result.equals(expected)

        fmt = "%Y-%m-%dT%H:%M:%S"

        # Default format
        tsa = pa.array(ts, type=pa.timestamp("s", timezone))
        result = pc.strftime(tsa, options=pc.StrftimeOptions())
        expected = pa.array(ts.strftime(fmt)).cast(result.type)
        assert result.equals(expected)

        # Default format plus timezone
        tsa = pa.array(ts, type=pa.timestamp("s", timezone))
        result = pc.strftime(tsa, options=pc.StrftimeOptions(fmt + "%Z"))
        expected = pa.array(ts.strftime(fmt + "%Z")).cast(result.type)
        assert result.equals(expected)

        # Pandas %S is equivalent to %S in arrow for unit="s"
        tsa = pa.array(ts, type=pa.timestamp("s", timezone))
        options = pc.StrftimeOptions("%S")
        result = pc.strftime(tsa, options=options)
        expected = pa.array(ts.strftime("%S")).cast(result.type)
        assert result.equals(expected)

        # Pandas %S.%f is equivalent to %S in arrow for unit="us"
        tsa = pa.array(ts, type=pa.timestamp("us", timezone))
        options = pc.StrftimeOptions("%S")
        result = pc.strftime(tsa, options=options)
        expected = pa.array(ts.strftime("%S.%f")).cast(result.type)
        assert result.equals(expected)

        # Test setting locale
        tsa = pa.array(ts, type=pa.timestamp("s", timezone))
        options = pc.StrftimeOptions(fmt, locale="C")
        result = pc.strftime(tsa, options=options)
        expected = pa.array(ts.strftime(fmt)).cast(result.type)
        assert result.equals(expected)

    # Test timestamps without timezone
    fmt = "%Y-%m-%dT%H:%M:%S"
    ts = pd.to_datetime(times)
    tsa = pa.array(ts, type=pa.timestamp("s"))
    result = pc.strftime(tsa, options=pc.StrftimeOptions(fmt))
    expected = pa.array(ts.strftime(fmt)).cast(result.type)

    # Positional format
    assert pc.strftime(tsa, fmt) == result

    assert result.equals(expected)
    with pytest.raises(pa.ArrowInvalid,
                       match="Timezone not present, cannot convert to string"):
        pc.strftime(tsa, options=pc.StrftimeOptions(fmt + "%Z"))
    with pytest.raises(pa.ArrowInvalid,
                       match="Timezone not present, cannot convert to string"):
        pc.strftime(tsa, options=pc.StrftimeOptions(fmt + "%z"))


def _check_datetime_components(timestamps, timezone=None):
    from pyarrow.vendored.version import Version

    ts = pd.to_datetime(timestamps).tz_localize(
        "UTC").tz_convert(timezone).to_series()
    tsa = pa.array(ts, pa.timestamp("ns", tz=timezone))

    subseconds = ((ts.dt.microsecond * 10 ** 3 +
                   ts.dt.nanosecond) * 10 ** -9).round(9)
    iso_calendar_fields = [
        pa.field('iso_year', pa.int64()),
        pa.field('iso_week', pa.int64()),
        pa.field('iso_day_of_week', pa.int64())
    ]

    if Version(pd.__version__) < Version("1.1.0"):
        # https://github.com/pandas-dev/pandas/issues/33206
        iso_year = ts.map(lambda x: x.isocalendar()[0]).astype("int64")
        iso_week = ts.map(lambda x: x.isocalendar()[1]).astype("int64")
        iso_day = ts.map(lambda x: x.isocalendar()[2]).astype("int64")
    else:
        # Casting is required because pandas isocalendar returns int32
        # while arrow isocalendar returns int64.
        iso_year = ts.dt.isocalendar()["year"].astype("int64")
        iso_week = ts.dt.isocalendar()["week"].astype("int64")
        iso_day = ts.dt.isocalendar()["day"].astype("int64")

    iso_calendar = pa.StructArray.from_arrays(
        [iso_year, iso_week, iso_day],
        fields=iso_calendar_fields)

    # Casting is required because pandas with 2.0.0 various numeric
    # date/time attributes have dtype int32 (previously int64)
    year = ts.dt.year.astype("int64")
    month = ts.dt.month.astype("int64")
    day = ts.dt.day.astype("int64")
    dayofweek = ts.dt.dayofweek.astype("int64")
    dayofyear = ts.dt.dayofyear.astype("int64")
    quarter = ts.dt.quarter.astype("int64")
    hour = ts.dt.hour.astype("int64")
    minute = ts.dt.minute.astype("int64")
    second = ts.dt.second.values.astype("int64")
    microsecond = ts.dt.microsecond.astype("int64")
    nanosecond = ts.dt.nanosecond.astype("int64")

    assert pc.year(tsa).equals(pa.array(year))
    assert pc.is_leap_year(tsa).equals(pa.array(ts.dt.is_leap_year))
    assert pc.month(tsa).equals(pa.array(month))
    assert pc.day(tsa).equals(pa.array(day))
    assert pc.day_of_week(tsa).equals(pa.array(dayofweek))
    assert pc.day_of_year(tsa).equals(pa.array(dayofyear))
    assert pc.iso_year(tsa).equals(pa.array(iso_year))
    assert pc.iso_week(tsa).equals(pa.array(iso_week))
    assert pc.iso_calendar(tsa).equals(iso_calendar)
    assert pc.quarter(tsa).equals(pa.array(quarter))
    assert pc.hour(tsa).equals(pa.array(hour))
    assert pc.minute(tsa).equals(pa.array(minute))
    assert pc.second(tsa).equals(pa.array(second))
    assert pc.millisecond(tsa).equals(pa.array(microsecond // 10 ** 3))
    assert pc.microsecond(tsa).equals(pa.array(microsecond % 10 ** 3))
    assert pc.nanosecond(tsa).equals(pa.array(nanosecond))
    assert pc.subsecond(tsa).equals(pa.array(subseconds))
    assert pc.local_timestamp(tsa).equals(pa.array(ts.dt.tz_localize(None)))

    if ts.dt.tz:
        if ts.dt.tz is datetime.timezone.utc:
            # datetime with utc returns None for dst()
            is_dst = [False] * len(ts)
        else:
            is_dst = ts.apply(lambda x: x.dst().seconds > 0)
        assert pc.is_dst(tsa).equals(pa.array(is_dst))

    day_of_week_options = pc.DayOfWeekOptions(
        count_from_zero=False, week_start=1)
    assert pc.day_of_week(tsa, options=day_of_week_options).equals(
        pa.array(dayofweek + 1))

    week_options = pc.WeekOptions(
        week_starts_monday=True, count_from_zero=False,
        first_week_is_fully_in_year=False)
    assert pc.week(tsa, options=week_options).equals(pa.array(iso_week))


@pytest.mark.pandas
def test_extract_datetime_components(request):
    timestamps = ["1970-01-01T00:00:59.123456789",
                  "2000-02-29T23:23:23.999999999",
                  "2033-05-18T03:33:20.000000000",
                  "2020-01-01T01:05:05.001",
                  "2019-12-31T02:10:10.002",
                  "2019-12-30T03:15:15.003",
                  "2009-12-31T04:20:20.004132",
                  "2010-01-01T05:25:25.005321",
                  "2010-01-03T06:30:30.006163",
                  "2010-01-04T07:35:35.0",
                  "2006-01-01T08:40:40.0",
                  "2005-12-31T09:45:45.0",
                  "2008-12-28T00:00:00.0",
                  "2008-12-29T00:00:00.0",
                  "2012-01-01T01:02:03.0"]
    timezones = ["UTC", "US/Central", "Asia/Kolkata",
                 "Etc/GMT-4", "Etc/GMT+4", "Australia/Broken_Hill"]

    # Test timezone naive timestamp array
    _check_datetime_components(timestamps)

    # Test timezone aware timestamp array
    if not request.config.pyarrow.is_enabled["timezone_data"]:
        pytest.skip('Timezone database is not installed on Windows')
    else:
        for timezone in timezones:
            _check_datetime_components(timestamps, timezone)


@pytest.mark.parametrize("unit", ["s", "ms", "us", "ns"])
def test_iso_calendar_longer_array(unit):
    # https://github.com/apache/arrow/issues/38655
    # ensure correct result for array length > 32
    arr = pa.array([datetime.datetime(2022, 1, 2, 9)]*50, pa.timestamp(unit))
    result = pc.iso_calendar(arr)
    expected = pa.StructArray.from_arrays(
        [[2021]*50, [52]*50, [7]*50],
        names=['iso_year', 'iso_week', 'iso_day_of_week']
    )
    assert result.equals(expected)


@pytest.mark.pandas
@pytest.mark.timezone_data
def test_assume_timezone():
    ts_type = pa.timestamp("ns")
    timestamps = pd.to_datetime(["1970-01-01T00:00:59.123456789",
                                 "2000-02-29T23:23:23.999999999",
                                 "2033-05-18T03:33:20.000000000",
                                 "2020-01-01T01:05:05.001",
                                 "2019-12-31T02:10:10.002",
                                 "2019-12-30T03:15:15.003",
                                 "2009-12-31T04:20:20.004132",
                                 "2010-01-01T05:25:25.005321",
                                 "2010-01-03T06:30:30.006163",
                                 "2010-01-04T07:35:35.0",
                                 "2006-01-01T08:40:40.0",
                                 "2005-12-31T09:45:45.0",
                                 "2008-12-28T00:00:00.0",
                                 "2008-12-29T00:00:00.0",
                                 "2012-01-01T01:02:03.0"])
    nonexistent = pd.to_datetime(["2015-03-29 02:30:00",
                                  "2015-03-29 03:30:00"])
    ambiguous = pd.to_datetime(["2018-10-28 01:20:00",
                                "2018-10-28 02:36:00",
                                "2018-10-28 03:46:00"])
    ambiguous_array = pa.array(ambiguous, type=ts_type)
    nonexistent_array = pa.array(nonexistent, type=ts_type)

    for timezone in ["UTC", "US/Central", "Asia/Kolkata"]:
        options = pc.AssumeTimezoneOptions(timezone)
        ta = pa.array(timestamps, type=ts_type)
        expected = timestamps.tz_localize(timezone)
        result = pc.assume_timezone(ta, options=options)
        assert result.equals(pa.array(expected))
        result = pc.assume_timezone(ta, timezone)  # Positional option
        assert result.equals(pa.array(expected))

        ta_zoned = pa.array(timestamps, type=pa.timestamp("ns", timezone))
        with pytest.raises(pa.ArrowInvalid, match="already have a timezone:"):
            pc.assume_timezone(ta_zoned, options=options)

    invalid_options = pc.AssumeTimezoneOptions("Europe/Brusselsss")
    with pytest.raises(ValueError, match="not found in timezone database"):
        pc.assume_timezone(ta, options=invalid_options)

    timezone = "Europe/Brussels"

    options_nonexistent_raise = pc.AssumeTimezoneOptions(timezone)
    options_nonexistent_earliest = pc.AssumeTimezoneOptions(
        timezone, ambiguous="raise", nonexistent="earliest")
    options_nonexistent_latest = pc.AssumeTimezoneOptions(
        timezone, ambiguous="raise", nonexistent="latest")

    with pytest.raises(ValueError,
                       match="Timestamp doesn't exist in "
                       f"timezone '{timezone}'"):
        pc.assume_timezone(nonexistent_array,
                           options=options_nonexistent_raise)

    expected = pa.array(nonexistent.tz_localize(
        timezone, nonexistent="shift_forward"))
    result = pc.assume_timezone(
        nonexistent_array, options=options_nonexistent_latest)
    expected.equals(result)

    expected = pa.array(nonexistent.tz_localize(
        timezone, nonexistent="shift_backward"))
    result = pc.assume_timezone(
        nonexistent_array, options=options_nonexistent_earliest)
    expected.equals(result)

    options_ambiguous_raise = pc.AssumeTimezoneOptions(timezone)
    options_ambiguous_latest = pc.AssumeTimezoneOptions(
        timezone, ambiguous="latest", nonexistent="raise")
    options_ambiguous_earliest = pc.AssumeTimezoneOptions(
        timezone, ambiguous="earliest", nonexistent="raise")

    with pytest.raises(ValueError,
                       match="Timestamp is ambiguous in "
                             f"timezone '{timezone}'"):
        pc.assume_timezone(ambiguous_array, options=options_ambiguous_raise)

    expected = ambiguous.tz_localize(timezone, ambiguous=[True, True, True])
    result = pc.assume_timezone(
        ambiguous_array, options=options_ambiguous_earliest)
    result.equals(pa.array(expected))

    expected = ambiguous.tz_localize(timezone, ambiguous=[False, False, False])
    result = pc.assume_timezone(
        ambiguous_array, options=options_ambiguous_latest)
    result.equals(pa.array(expected))


def _check_temporal_rounding(ts, values, unit):
    unit_shorthand = {
        "nanosecond": "ns",
        "microsecond": "us",
        "millisecond": "ms",
        "second": "s",
        "minute": "min",
        "hour": "h",
        "day": "D"
    }
    greater_unit = {
        "nanosecond": "us",
        "microsecond": "ms",
        "millisecond": "s",
        "second": "min",
        "minute": "h",
        "hour": "D",
    }
    ta = pa.array(ts)

    for value in values:
        frequency = str(value) + unit_shorthand[unit]
        options = pc.RoundTemporalOptions(value, unit)

        result = pc.ceil_temporal(ta, options=options).to_pandas()
        expected = ts.dt.ceil(frequency)
        np.testing.assert_array_equal(result, expected)

        result = pc.floor_temporal(ta, options=options).to_pandas()
        expected = ts.dt.floor(frequency)
        np.testing.assert_array_equal(result, expected)

        result = pc.round_temporal(ta, options=options).to_pandas()
        expected = ts.dt.round(frequency)
        np.testing.assert_array_equal(result, expected)

        # Check rounding with calendar_based_origin=True.
        # Note: rounding to month is not supported in Pandas so we can't
        # approximate this functionality and exclude unit == "day".
        if unit != "day":
            options = pc.RoundTemporalOptions(
                value, unit, calendar_based_origin=True)
            origin = ts.dt.floor(greater_unit[unit])

            if ta.type.tz is None:
                result = pc.ceil_temporal(ta, options=options).to_pandas()
                expected = (ts - origin).dt.ceil(frequency) + origin
                np.testing.assert_array_equal(result, expected)

            result = pc.floor_temporal(ta, options=options).to_pandas()
            expected = (ts - origin).dt.floor(frequency) + origin
            np.testing.assert_array_equal(result, expected)

            result = pc.round_temporal(ta, options=options).to_pandas()
            expected = (ts - origin).dt.round(frequency) + origin
            np.testing.assert_array_equal(result, expected)

        # Check RoundTemporalOptions partial defaults
        if unit == "day":
            result = pc.ceil_temporal(ta, multiple=value).to_pandas()
            expected = ts.dt.ceil(frequency)
            np.testing.assert_array_equal(result, expected)

            result = pc.floor_temporal(ta, multiple=value).to_pandas()
            expected = ts.dt.floor(frequency)
            np.testing.assert_array_equal(result, expected)

            result = pc.round_temporal(ta, multiple=value).to_pandas()
            expected = ts.dt.round(frequency)
            np.testing.assert_array_equal(result, expected)

    # We naively test ceil_is_strictly_greater by adding time unit multiple
    # to regular ceiled timestamp if it is equal to the original timestamp.
    # This does not work if timestamp is zoned since our logic will not
    # account for DST jumps.
    if ta.type.tz is None:
        options = pc.RoundTemporalOptions(
            value, unit, ceil_is_strictly_greater=True)
        result = pc.ceil_temporal(ta, options=options)
        expected = ts.dt.ceil(frequency)

        expected = np.where(
            expected == ts,
            expected + pd.Timedelta(value, unit_shorthand[unit]),
            expected)
        np.testing.assert_array_equal(result, expected)

    # Check RoundTemporalOptions defaults
    if unit == "day":
        frequency = "1D"

        result = pc.ceil_temporal(ta).to_pandas()
        expected = ts.dt.ceil(frequency)
        np.testing.assert_array_equal(result, expected)

        result = pc.floor_temporal(ta).to_pandas()
        expected = ts.dt.floor(frequency)
        np.testing.assert_array_equal(result, expected)

        result = pc.round_temporal(ta).to_pandas()
        expected = ts.dt.round(frequency)
        np.testing.assert_array_equal(result, expected)


@pytest.mark.timezone_data
@pytest.mark.parametrize('unit', ("nanosecond", "microsecond", "millisecond",
                                  "second", "minute", "hour", "day"))
@pytest.mark.pandas
def test_round_temporal(unit):
    values = (1, 2, 3, 4, 5, 6, 7, 10, 15, 24, 60, 250, 500, 750)
    timestamps = [
        "1923-07-07 08:52:35.203790336",
        "1931-03-17 10:45:00.641559040",
        "1932-06-16 01:16:42.911994368",
        "1941-05-27 11:46:43.822831872",
        "1943-12-14 07:32:05.424766464",
        "1954-04-12 04:31:50.699881472",
        "1966-02-12 17:41:28.693282560",
        "1967-02-26 05:56:46.922376960",
        "1975-11-01 10:55:37.016146432",
        "1982-01-21 18:43:44.517366784",
        "1992-01-01 00:00:00.100000000",
        "1999-12-04 05:55:34.794991104",
        "2026-10-26 08:39:00.316686848"]
    ts = pd.Series([pd.Timestamp(x, unit="ns") for x in timestamps])
    _check_temporal_rounding(ts, values, unit)

    timezones = ["Asia/Kolkata", "America/New_York", "Etc/GMT-4", "Etc/GMT+4",
                 "Europe/Brussels", "Pacific/Marquesas", "US/Central", "UTC"]

    for timezone in timezones:
        ts_zoned = ts.dt.tz_localize("UTC").dt.tz_convert(timezone)
        _check_temporal_rounding(ts_zoned, values, unit)


def test_count():
    arr = pa.array([1, 2, 3, None, None])
    assert pc.count(arr).as_py() == 3
    assert pc.count(arr, mode='only_valid').as_py() == 3
    assert pc.count(arr, mode='only_null').as_py() == 2
    assert pc.count(arr, mode='all').as_py() == 5
    assert pc.count(arr, 'all').as_py() == 5

    with pytest.raises(ValueError,
                       match='"something else" is not a valid count mode'):
        pc.count(arr, 'something else')


def test_index():
    arr = pa.array([0, 1, None, 3, 4], type=pa.int64())
    assert pc.index(arr, pa.scalar(0)).as_py() == 0
    assert pc.index(arr, pa.scalar(2, type=pa.int8())).as_py() == -1
    assert pc.index(arr, 4).as_py() == 4
    assert arr.index(3, start=2).as_py() == 3
    assert arr.index(None).as_py() == -1

    arr = pa.chunked_array([[1, 2], [1, 3]], type=pa.int64())
    assert arr.index(1).as_py() == 0
    assert arr.index(1, start=2).as_py() == 2
    assert arr.index(1, start=1, end=2).as_py() == -1


def check_partition_nth(data, indices, pivot, null_placement):
    indices = indices.to_pylist()
    assert len(indices) == len(data)
    assert sorted(indices) == list(range(len(data)))
    until_pivot = [data[indices[i]] for i in range(pivot)]
    after_pivot = [data[indices[i]] for i in range(pivot, len(data))]
    p = data[indices[pivot]]
    if p is None:
        if null_placement == "at_start":
            assert all(v is None for v in until_pivot)
        else:
            assert all(v is None for v in after_pivot)
    else:
        if null_placement == "at_start":
            assert all(v is None or v <= p for v in until_pivot)
            assert all(v >= p for v in after_pivot)
        else:
            assert all(v <= p for v in until_pivot)
            assert all(v is None or v >= p for v in after_pivot)


def test_partition_nth():
    data = list(range(100, 140))
    random.shuffle(data)
    pivot = 10
    indices = pc.partition_nth_indices(data, pivot=pivot)
    check_partition_nth(data, indices, pivot, "at_end")
    # Positional pivot argument
    assert pc.partition_nth_indices(data, pivot) == indices

    with pytest.raises(
            ValueError,
            match="'partition_nth_indices' cannot be called without options"):
        pc.partition_nth_indices(data)


def test_partition_nth_null_placement():
    data = list(range(10)) + [None] * 10
    random.shuffle(data)

    for pivot in (0, 7, 13, 19):
        for null_placement in ("at_start", "at_end"):
            indices = pc.partition_nth_indices(data, pivot=pivot,
                                               null_placement=null_placement)
            check_partition_nth(data, indices, pivot, null_placement)


def test_select_k_array():
    def validate_select_k(select_k_indices, arr, order, stable_sort=False):
        sorted_indices = pc.sort_indices(arr, sort_keys=[("dummy", order)])
        head_k_indices = sorted_indices.slice(0, len(select_k_indices))
        if stable_sort:
            assert select_k_indices == head_k_indices
        else:
            expected = pc.take(arr, head_k_indices)
            actual = pc.take(arr, select_k_indices)
            assert actual == expected

    arr = pa.array([1, 2, None, 0])
    for k in [0, 2, 4]:
        for order in ["descending", "ascending"]:
            result = pc.select_k_unstable(
                arr, k=k, sort_keys=[("dummy", order)])
            validate_select_k(result, arr, order)

        result = pc.top_k_unstable(arr, k=k)
        validate_select_k(result, arr, "descending")

        result = pc.bottom_k_unstable(arr, k=k)
        validate_select_k(result, arr, "ascending")

    result = pc.select_k_unstable(
        arr, options=pc.SelectKOptions(
            k=2, sort_keys=[("dummy", "descending")])
    )
    validate_select_k(result, arr, "descending")

    result = pc.select_k_unstable(
        arr, options=pc.SelectKOptions(k=2, sort_keys=[("dummy", "ascending")])
    )
    validate_select_k(result, arr, "ascending")

    # Position options
    assert pc.select_k_unstable(arr, 2,
                                sort_keys=[("dummy", "ascending")]) == result
    assert pc.select_k_unstable(arr, 2, [("dummy", "ascending")]) == result


def test_select_k_table():
    def validate_select_k(select_k_indices, tbl, sort_keys, stable_sort=False):
        sorted_indices = pc.sort_indices(tbl, sort_keys=sort_keys)
        head_k_indices = sorted_indices.slice(0, len(select_k_indices))
        if stable_sort:
            assert select_k_indices == head_k_indices
        else:
            expected = pc.take(tbl, head_k_indices)
            actual = pc.take(tbl, select_k_indices)
            assert actual == expected

    table = pa.table({"a": [1, 2, 0], "b": [1, 0, 1]})
    for k in [0, 2, 4]:
        result = pc.select_k_unstable(
            table, k=k, sort_keys=[("a", "ascending")])
        validate_select_k(result, table, sort_keys=[("a", "ascending")])

        result = pc.select_k_unstable(
            table, k=k, sort_keys=[(pc.field("a"), "ascending"), ("b", "ascending")])
        validate_select_k(
            result, table, sort_keys=[("a", "ascending"), ("b", "ascending")])

        result = pc.top_k_unstable(table, k=k, sort_keys=["a"])
        validate_select_k(result, table, sort_keys=[("a", "descending")])

        result = pc.bottom_k_unstable(table, k=k, sort_keys=["a", "b"])
        validate_select_k(
            result, table, sort_keys=[("a", "ascending"), ("b", "ascending")])

    with pytest.raises(
            ValueError,
            match="'select_k_unstable' cannot be called without options"):
        pc.select_k_unstable(table)

    with pytest.raises(ValueError,
                       match="select_k_unstable requires a nonnegative `k`"):
        pc.select_k_unstable(table, k=-1, sort_keys=[("a", "ascending")])

    with pytest.raises(ValueError,
                       match="select_k_unstable requires a "
                             "non-empty `sort_keys`"):
        pc.select_k_unstable(table, k=2, sort_keys=[])

    with pytest.raises(ValueError, match="not a valid sort order"):
        pc.select_k_unstable(table, k=k, sort_keys=[("a", "nonscending")])

    with pytest.raises(ValueError,
                       match="Invalid sort key column: No match for.*unknown"):
        pc.select_k_unstable(table, k=k, sort_keys=[("unknown", "ascending")])


def test_array_sort_indices():
    arr = pa.array([1, 2, None, 0])
    result = pc.array_sort_indices(arr)
    assert result.to_pylist() == [3, 0, 1, 2]
    result = pc.array_sort_indices(arr, order="ascending")
    assert result.to_pylist() == [3, 0, 1, 2]
    result = pc.array_sort_indices(arr, order="descending")
    assert result.to_pylist() == [1, 0, 3, 2]
    result = pc.array_sort_indices(arr, order="descending",
                                   null_placement="at_start")
    assert result.to_pylist() == [2, 1, 0, 3]
    result = pc.array_sort_indices(arr, "descending",
                                   null_placement="at_start")
    assert result.to_pylist() == [2, 1, 0, 3]

    with pytest.raises(ValueError, match="not a valid sort order"):
        pc.array_sort_indices(arr, order="nonscending")


def test_sort_indices_array():
    arr = pa.array([1, 2, None, 0])
    result = pc.sort_indices(arr)
    assert result.to_pylist() == [3, 0, 1, 2]
    result = pc.sort_indices(arr, sort_keys=[("dummy", "ascending")])
    assert result.to_pylist() == [3, 0, 1, 2]
    result = pc.sort_indices(arr, sort_keys=[("dummy", "descending")])
    assert result.to_pylist() == [1, 0, 3, 2]
    result = pc.sort_indices(arr, sort_keys=[("dummy", "descending")],
                             null_placement="at_start")
    assert result.to_pylist() == [2, 1, 0, 3]
    # Positional `sort_keys`
    result = pc.sort_indices(arr, [("dummy", "descending")],
                             null_placement="at_start")
    assert result.to_pylist() == [2, 1, 0, 3]
    # Using SortOptions
    result = pc.sort_indices(
        arr, options=pc.SortOptions(sort_keys=[("dummy", "descending")])
    )
    assert result.to_pylist() == [1, 0, 3, 2]
    result = pc.sort_indices(
        arr, options=pc.SortOptions(sort_keys=[("dummy", "descending")],
                                    null_placement="at_start")
    )
    assert result.to_pylist() == [2, 1, 0, 3]


def test_sort_indices_table():
    table = pa.table({"a": [1, 1, None, 0], "b": [1, 0, 0, 1]})

    result = pc.sort_indices(table, sort_keys=[("a", "ascending")])
    assert result.to_pylist() == [3, 0, 1, 2]
    result = pc.sort_indices(table, sort_keys=[(pc.field("a"), "ascending")],
                             null_placement="at_start")
    assert result.to_pylist() == [2, 3, 0, 1]

    result = pc.sort_indices(
        table, sort_keys=[("a", "descending"), ("b", "ascending")]
    )
    assert result.to_pylist() == [1, 0, 3, 2]
    result = pc.sort_indices(
        table, sort_keys=[("a", "descending"), ("b", "ascending")],
        null_placement="at_start"
    )
    assert result.to_pylist() == [2, 1, 0, 3]
    # Positional `sort_keys`
    result = pc.sort_indices(
        table, [("a", "descending"), ("b", "ascending")],
        null_placement="at_start"
    )
    assert result.to_pylist() == [2, 1, 0, 3]

    with pytest.raises(ValueError, match="Must specify one or more sort keys"):
        pc.sort_indices(table)

    with pytest.raises(ValueError,
                       match="Invalid sort key column: No match for.*unknown"):
        pc.sort_indices(table, sort_keys=[("unknown", "ascending")])

    with pytest.raises(ValueError, match="not a valid sort order"):
        pc.sort_indices(table, sort_keys=[("a", "nonscending")])


def test_is_in():
    arr = pa.array([1, 2, None, 1, 2, 3])

    result = pc.is_in(arr, value_set=pa.array([1, 3, None]))
    assert result.to_pylist() == [True, False, True, True, False, True]

    result = pc.is_in(arr, value_set=pa.array([1, 3, None]), skip_nulls=True)
    assert result.to_pylist() == [True, False, False, True, False, True]

    result = pc.is_in(arr, value_set=pa.array([1, 3]))
    assert result.to_pylist() == [True, False, False, True, False, True]

    result = pc.is_in(arr, value_set=pa.array([1, 3]), skip_nulls=True)
    assert result.to_pylist() == [True, False, False, True, False, True]


def test_index_in():
    arr = pa.array([1, 2, None, 1, 2, 3])

    result = pc.index_in(arr, value_set=pa.array([1, 3, None]))
    assert result.to_pylist() == [0, None, 2, 0, None, 1]

    result = pc.index_in(arr, value_set=pa.array([1, 3, None]),
                         skip_nulls=True)
    assert result.to_pylist() == [0, None, None, 0, None, 1]

    result = pc.index_in(arr, value_set=pa.array([1, 3]))
    assert result.to_pylist() == [0, None, None, 0, None, 1]

    result = pc.index_in(arr, value_set=pa.array([1, 3]), skip_nulls=True)
    assert result.to_pylist() == [0, None, None, 0, None, 1]

    # Positional value_set
    result = pc.index_in(arr, pa.array([1, 3]), skip_nulls=True)
    assert result.to_pylist() == [0, None, None, 0, None, 1]


def test_quantile():
    arr = pa.array([1, 2, 3, 4])

    result = pc.quantile(arr)
    assert result.to_pylist() == [2.5]

    result = pc.quantile(arr, interpolation='lower')
    assert result.to_pylist() == [2]
    result = pc.quantile(arr, interpolation='higher')
    assert result.to_pylist() == [3]
    result = pc.quantile(arr, interpolation='nearest')
    assert result.to_pylist() == [3]
    result = pc.quantile(arr, interpolation='midpoint')
    assert result.to_pylist() == [2.5]
    result = pc.quantile(arr, interpolation='linear')
    assert result.to_pylist() == [2.5]

    arr = pa.array([1, 2])

    result = pc.quantile(arr, q=[0.25, 0.5, 0.75])
    assert result.to_pylist() == [1.25, 1.5, 1.75]

    result = pc.quantile(arr, q=[0.25, 0.5, 0.75], interpolation='lower')
    assert result.to_pylist() == [1, 1, 1]
    result = pc.quantile(arr, q=[0.25, 0.5, 0.75], interpolation='higher')
    assert result.to_pylist() == [2, 2, 2]
    result = pc.quantile(arr, q=[0.25, 0.5, 0.75], interpolation='midpoint')
    assert result.to_pylist() == [1.5, 1.5, 1.5]
    result = pc.quantile(arr, q=[0.25, 0.5, 0.75], interpolation='nearest')
    assert result.to_pylist() == [1, 1, 2]
    result = pc.quantile(arr, q=[0.25, 0.5, 0.75], interpolation='linear')
    assert result.to_pylist() == [1.25, 1.5, 1.75]

    # Positional `q`
    result = pc.quantile(arr, [0.25, 0.5, 0.75], interpolation='linear')
    assert result.to_pylist() == [1.25, 1.5, 1.75]

    with pytest.raises(ValueError, match="Quantile must be between 0 and 1"):
        pc.quantile(arr, q=1.1)
    with pytest.raises(ValueError, match="not a valid quantile interpolation"):
        pc.quantile(arr, interpolation='zzz')


def test_tdigest():
    arr = pa.array([1, 2, 3, 4])
    result = pc.tdigest(arr)
    assert result.to_pylist() == [2.5]

    arr = pa.chunked_array([pa.array([1, 2]), pa.array([3, 4])])
    result = pc.tdigest(arr)
    assert result.to_pylist() == [2.5]

    arr = pa.array([1, 2, 3, 4])
    result = pc.tdigest(arr, q=[0, 0.5, 1])
    assert result.to_pylist() == [1, 2.5, 4]

    arr = pa.chunked_array([pa.array([1, 2]), pa.array([3, 4])])
    result = pc.tdigest(arr, [0, 0.5, 1])  # positional `q`
    assert result.to_pylist() == [1, 2.5, 4]


def test_fill_null_segfault():
    # ARROW-12672
    arr = pa.array([None], pa.bool_()).fill_null(False)
    result = arr.cast(pa.int8())
    assert result == pa.array([0], pa.int8())


def test_min_max_element_wise():
    arr1 = pa.array([1, 2, 3])
    arr2 = pa.array([3, 1, 2])
    arr3 = pa.array([2, 3, None])

    result = pc.max_element_wise(arr1, arr2)
    assert result == pa.array([3, 2, 3])
    result = pc.min_element_wise(arr1, arr2)
    assert result == pa.array([1, 1, 2])

    result = pc.max_element_wise(arr1, arr2, arr3)
    assert result == pa.array([3, 3, 3])
    result = pc.min_element_wise(arr1, arr2, arr3)
    assert result == pa.array([1, 1, 2])

    # with specifying the option
    result = pc.max_element_wise(arr1, arr3, skip_nulls=True)
    assert result == pa.array([2, 3, 3])
    result = pc.min_element_wise(arr1, arr3, skip_nulls=True)
    assert result == pa.array([1, 2, 3])
    result = pc.max_element_wise(
        arr1, arr3, options=pc.ElementWiseAggregateOptions())
    assert result == pa.array([2, 3, 3])
    result = pc.min_element_wise(
        arr1, arr3, options=pc.ElementWiseAggregateOptions())
    assert result == pa.array([1, 2, 3])

    # not skipping nulls
    result = pc.max_element_wise(arr1, arr3, skip_nulls=False)
    assert result == pa.array([2, 3, None])
    result = pc.min_element_wise(arr1, arr3, skip_nulls=False)
    assert result == pa.array([1, 2, None])


@pytest.mark.numpy
@pytest.mark.parametrize('start', (1.25, 10.5, -10.5))
@pytest.mark.parametrize('skip_nulls', (True, False))
def test_cumulative_sum(start, skip_nulls):
    # Exact tests (e.g., integral types)
    start_int = int(start)
    starts = [None, start_int, pa.scalar(start_int, type=pa.int8()),
              pa.scalar(start_int, type=pa.int64())]
    for strt in starts:
        arrays = [
            pa.array([1, 2, 3]),
            pa.array([0, None, 20, 30]),
            pa.chunked_array([[0, None], [20, 30]])
        ]
        expected_arrays = [
            pa.array([1, 3, 6]),
            pa.array([0, None, 20, 50])
            if skip_nulls else pa.array([0, None, None, None]),
            pa.chunked_array([[0, None, 20, 50]])
            if skip_nulls else pa.chunked_array([[0, None, None, None]])
        ]
        for i, arr in enumerate(arrays):
            result = pc.cumulative_sum(arr, start=strt, skip_nulls=skip_nulls)
            # Add `start` offset to expected array before comparing
            expected = pc.add(expected_arrays[i], strt if strt is not None
                              else 0)
            assert result.equals(expected)

    starts = [None, start, pa.scalar(start, type=pa.float32()),
              pa.scalar(start, type=pa.float64())]
    for strt in starts:
        arrays = [
            pa.array([1.125, 2.25, 3.03125]),
            pa.array([1, np.nan, 2, -3, 4, 5]),
            pa.array([1, np.nan, None, 3, None, 5])
        ]
        expected_arrays = [
            np.array([1.125, 3.375, 6.40625]),
            np.array([1, np.nan, np.nan, np.nan, np.nan, np.nan]),
            np.array([1, np.nan, None, np.nan, None, np.nan])
            if skip_nulls else np.array([1, np.nan, None, None, None, None])
        ]
        for i, arr in enumerate(arrays):
            result = pc.cumulative_sum(arr, start=strt, skip_nulls=skip_nulls)
            # Add `start` offset to expected array before comparing
            expected = pc.add(expected_arrays[i], strt if strt is not None
                              else 0)
            np.testing.assert_array_almost_equal(result.to_numpy(
                zero_copy_only=False), expected.to_numpy(zero_copy_only=False))

    for strt in ['a', pa.scalar('arrow'), 1.1]:
        with pytest.raises(pa.ArrowInvalid):
            pc.cumulative_sum([1, 2, 3], start=strt)


@pytest.mark.numpy
@pytest.mark.parametrize('start', (1.25, 10.5, -10.5))
@pytest.mark.parametrize('skip_nulls', (True, False))
def test_cumulative_prod(start, skip_nulls):
    # Exact tests (e.g., integral types)
    start_int = int(start)
    starts = [None, start_int, pa.scalar(start_int, type=pa.int8()),
              pa.scalar(start_int, type=pa.int64())]
    for strt in starts:
        arrays = [
            pa.array([1, 2, 3]),
            pa.array([1, None, 20, 5]),
            pa.chunked_array([[1, None], [20, 5]])
        ]
        expected_arrays = [
            pa.array([1, 2, 6]),
            pa.array([1, None, 20, 100])
            if skip_nulls else pa.array([1, None, None, None]),
            pa.chunked_array([[1, None, 20, 100]])
            if skip_nulls else pa.chunked_array([[1, None, None, None]])
        ]
        for i, arr in enumerate(arrays):
            result = pc.cumulative_prod(arr, start=strt, skip_nulls=skip_nulls)
            # Multiply `start` offset to expected array before comparing
            expected = pc.multiply(expected_arrays[i], strt if strt is not None
                                   else 1)
            assert result.equals(expected)

    starts = [None, start, pa.scalar(start, type=pa.float32()),
              pa.scalar(start, type=pa.float64())]
    for strt in starts:
        arrays = [
            pa.array([1.5, 2.5, 3.5]),
            pa.array([1, np.nan, 2, -3, 4, 5]),
            pa.array([1, np.nan, None, 3, None, 5])
        ]
        expected_arrays = [
            np.array([1.5, 3.75, 13.125]),
            np.array([1, np.nan, np.nan, np.nan, np.nan, np.nan]),
            np.array([1, np.nan, None, np.nan, None, np.nan])
            if skip_nulls else np.array([1, np.nan, None, None, None, None])
        ]
        for i, arr in enumerate(arrays):
            result = pc.cumulative_prod(arr, start=strt, skip_nulls=skip_nulls)
            # Multiply `start` offset to expected array before comparing
            expected = pc.multiply(expected_arrays[i], strt if strt is not None
                                   else 1)
            np.testing.assert_array_almost_equal(result.to_numpy(
                zero_copy_only=False), expected.to_numpy(zero_copy_only=False))

    for strt in ['a', pa.scalar('arrow'), 1.1]:
        with pytest.raises(pa.ArrowInvalid):
            pc.cumulative_prod([1, 2, 3], start=strt)


@pytest.mark.numpy
@pytest.mark.parametrize('start', (0.5, 3.5, 6.5))
@pytest.mark.parametrize('skip_nulls', (True, False))
def test_cumulative_max(start, skip_nulls):
    # Exact tests (e.g., integral types)
    start_int = int(start)
    starts = [None, start_int, pa.scalar(start_int, type=pa.int8()),
              pa.scalar(start_int, type=pa.int64())]
    for strt in starts:
        arrays = [
            pa.array([2, 1, 3, 5, 4, 6]),
            pa.array([2, 1, None, 5, 4, None]),
            pa.chunked_array([[2, 1, None], [5, 4, None]])
        ]
        expected_arrays = [
            pa.array([2, 2, 3, 5, 5, 6]),
            pa.array([2, 2, None, 5, 5, None])
            if skip_nulls else pa.array([2, 2, None, None, None, None]),
            pa.chunked_array([[2, 2, None, 5, 5, None]])
            if skip_nulls else
            pa.chunked_array([[2, 2, None, None, None, None]])
        ]
        for i, arr in enumerate(arrays):
            result = pc.cumulative_max(arr, start=strt, skip_nulls=skip_nulls)
            # Max `start` offset with expected array before comparing
            expected = pc.max_element_wise(
                expected_arrays[i], strt if strt is not None else int(-1e9),
                skip_nulls=False)
            assert result.equals(expected)

    starts = [None, start, pa.scalar(start, type=pa.float32()),
              pa.scalar(start, type=pa.float64())]
    for strt in starts:
        arrays = [
            pa.array([2.5, 1.3, 3.7, 5.1, 4.9, 6.2]),
            pa.array([2.5, 1.3, 3.7, np.nan, 4.9, 6.2]),
            pa.array([2.5, 1.3, None, np.nan, 4.9, None])
        ]
        expected_arrays = [
            np.array([2.5, 2.5, 3.7, 5.1, 5.1, 6.2]),
            np.array([2.5, 2.5, 3.7, 3.7, 4.9, 6.2]),
            np.array([2.5, 2.5, None, 2.5, 4.9, None])
            if skip_nulls else np.array([2.5, 2.5, None, None, None, None])
        ]
        for i, arr in enumerate(arrays):
            result = pc.cumulative_max(arr, start=strt, skip_nulls=skip_nulls)
            # Max `start` offset with expected array before comparing
            expected = pc.max_element_wise(
                expected_arrays[i], strt if strt is not None else -1e9,
                skip_nulls=False)
            np.testing.assert_array_almost_equal(result.to_numpy(
                zero_copy_only=False), expected.to_numpy(zero_copy_only=False))

    for strt in ['a', pa.scalar('arrow'), 1.1]:
        with pytest.raises(pa.ArrowInvalid):
            pc.cumulative_max([1, 2, 3], start=strt)


@pytest.mark.numpy
@pytest.mark.parametrize('start', (0.5, 3.5, 6.5))
@pytest.mark.parametrize('skip_nulls', (True, False))
def test_cumulative_min(start, skip_nulls):
    # Exact tests (e.g., integral types)
    start_int = int(start)
    starts = [None, start_int, pa.scalar(start_int, type=pa.int8()),
              pa.scalar(start_int, type=pa.int64())]
    for strt in starts:
        arrays = [
            pa.array([5, 6, 4, 2, 3, 1]),
            pa.array([5, 6, None, 2, 3, None]),
            pa.chunked_array([[5, 6, None], [2, 3, None]])
        ]
        expected_arrays = [
            pa.array([5, 5, 4, 2, 2, 1]),
            pa.array([5, 5, None, 2, 2, None])
            if skip_nulls else pa.array([5, 5, None, None, None, None]),
            pa.chunked_array([[5, 5, None, 2, 2, None]])
            if skip_nulls else
            pa.chunked_array([[5, 5, None, None, None, None]])
        ]
        for i, arr in enumerate(arrays):
            result = pc.cumulative_min(arr, start=strt, skip_nulls=skip_nulls)
            # Min `start` offset with expected array before comparing
            expected = pc.min_element_wise(
                expected_arrays[i], strt if strt is not None else int(1e9),
                skip_nulls=False)
            assert result.equals(expected)

    starts = [None, start, pa.scalar(start, type=pa.float32()),
              pa.scalar(start, type=pa.float64())]
    for strt in starts:
        arrays = [
            pa.array([5.5, 6.3, 4.7, 2.1, 3.9, 1.2]),
            pa.array([5.5, 6.3, 4.7, np.nan, 3.9, 1.2]),
            pa.array([5.5, 6.3, None, np.nan, 3.9, None])
        ]
        expected_arrays = [
            np.array([5.5, 5.5, 4.7, 2.1, 2.1, 1.2]),
            np.array([5.5, 5.5, 4.7, 4.7, 3.9, 1.2]),
            np.array([5.5, 5.5, None, 5.5, 3.9, None])
            if skip_nulls else np.array([5.5, 5.5, None, None, None, None])
        ]
        for i, arr in enumerate(arrays):
            result = pc.cumulative_min(arr, start=strt, skip_nulls=skip_nulls)
            # Min `start` offset with expected array before comparing
            expected = pc.min_element_wise(
                expected_arrays[i], strt if strt is not None else 1e9,
                skip_nulls=False)
            np.testing.assert_array_almost_equal(result.to_numpy(
                zero_copy_only=False), expected.to_numpy(zero_copy_only=False))

    for strt in ['a', pa.scalar('arrow'), 1.1]:
        with pytest.raises(pa.ArrowInvalid):
            pc.cumulative_max([1, 2, 3], start=strt)


def test_make_struct():
    assert pc.make_struct(1, 'a').as_py() == {'0': 1, '1': 'a'}

    assert pc.make_struct(1, 'a', field_names=['i', 's']).as_py() == {
        'i': 1, 's': 'a'}

    assert pc.make_struct([1, 2, 3],
                          "a b c".split()) == pa.StructArray.from_arrays([
                              [1, 2, 3],
                              "a b c".split()], names='0 1'.split())

    with pytest.raises(ValueError,
                       match="Array arguments must all be the same length"):
        pc.make_struct([1, 2, 3, 4], "a b c".split())

    with pytest.raises(ValueError, match="0 arguments but 2 field names"):
        pc.make_struct(field_names=['one', 'two'])


def test_map_lookup():
    ty = pa.map_(pa.utf8(), pa.int32())
    arr = pa.array([[('one', 1), ('two', 2)], [('none', 3)],
                    [], [('one', 5), ('one', 7)], None], type=ty)
    result_first = pa.array([1, None, None, 5, None], type=pa.int32())
    result_last = pa.array([1, None, None, 7, None], type=pa.int32())
    result_all = pa.array([[1], None, None, [5, 7], None],
                          type=pa.list_(pa.int32()))

    assert pc.map_lookup(arr, 'one', 'first') == result_first
    assert pc.map_lookup(arr, pa.scalar(
        'one', type=pa.utf8()), 'first') == result_first
    assert pc.map_lookup(arr, pa.scalar(
        'one', type=pa.utf8()), 'last') == result_last
    assert pc.map_lookup(arr, pa.scalar(
        'one', type=pa.utf8()), 'all') == result_all


def test_struct_fields_options():
    a = pa.array([4, 5, 6], type=pa.int64())
    b = pa.array(["bar", None, ""])
    c = pa.StructArray.from_arrays([a, b], ["a", "b"])
    arr = pa.StructArray.from_arrays([a, c], ["a", "c"])

    assert pc.struct_field(arr, '.c.b') == b
    assert pc.struct_field(arr, b'.c.b') == b
    assert pc.struct_field(arr, ['c', 'b']) == b
    assert pc.struct_field(arr, [1, 'b']) == b
    assert pc.struct_field(arr, (b'c', 'b')) == b
    assert pc.struct_field(arr, pc.field(('c', 'b'))) == b

    assert pc.struct_field(arr, '.a') == a
    assert pc.struct_field(arr, ['a']) == a
    assert pc.struct_field(arr, 'a') == a
    assert pc.struct_field(arr, pc.field(('a',))) == a

    assert pc.struct_field(arr, indices=[1, 1]) == b
    assert pc.struct_field(arr, (1, 1)) == b
    assert pc.struct_field(arr, [0]) == a
    assert pc.struct_field(arr, []) == arr

    with pytest.raises(pa.ArrowInvalid, match="No match for FieldRef"):
        pc.struct_field(arr, 'foo')

    with pytest.raises(pa.ArrowInvalid, match="No match for FieldRef"):
        pc.struct_field(arr, '.c.foo')

    # drill into a non-struct array and continue to ask for a field
    with pytest.raises(pa.ArrowInvalid, match="No match for FieldRef"):
        pc.struct_field(arr, '.a.foo')

    # TODO: https://issues.apache.org/jira/browse/ARROW-14853
    # assert pc.struct_field(arr) == arr


def test_case_when():
    assert pc.case_when(pc.make_struct([True, False, None],
                                       [False, True, None]),
                        [1, 2, 3],
                        [11, 12, 13]) == pa.array([1, 12, None])


def test_list_element():
    element_type = pa.struct([('a', pa.float64()), ('b', pa.int8())])
    list_type = pa.list_(element_type)
    l1 = [{'a': .4, 'b': 2}, None, {'a': .2, 'b': 4}, None, {'a': 5.6, 'b': 6}]
    l2 = [None, {'a': .52, 'b': 3}, {'a': .7, 'b': 4}, None, {'a': .6, 'b': 8}]
    lists = pa.array([l1, l2], list_type)

    index = 1
    result = pa.compute.list_element(lists, index)
    expected = pa.array([None, {'a': 0.52, 'b': 3}], element_type)
    assert result.equals(expected)

    index = 4
    result = pa.compute.list_element(lists, index)
    expected = pa.array([{'a': 5.6, 'b': 6}, {'a': .6, 'b': 8}], element_type)
    assert result.equals(expected)


def test_count_distinct():
    samples = [datetime.datetime(year=y, month=1, day=1) for y in range(1992, 2092)]
    arr = pa.array(samples, pa.timestamp("ns"))
    assert pc.count_distinct(arr) == pa.scalar(len(samples), type=pa.int64())


def test_count_distinct_options():
    arr = pa.array([1, 2, 3, None, None])
    assert pc.count_distinct(arr).as_py() == 3
    assert pc.count_distinct(arr, mode='only_valid').as_py() == 3
    assert pc.count_distinct(arr, mode='only_null').as_py() == 1
    assert pc.count_distinct(arr, mode='all').as_py() == 4
    assert pc.count_distinct(arr, 'all').as_py() == 4


def test_utf8_normalize():
    arr = pa.array(["01²3"])
    assert pc.utf8_normalize(arr, form="NFC") == arr
    assert pc.utf8_normalize(arr, form="NFKC") == pa.array(["0123"])
    assert pc.utf8_normalize(arr, "NFD") == arr
    assert pc.utf8_normalize(arr, "NFKD") == pa.array(["0123"])
    with pytest.raises(
            ValueError,
            match='"NFZ" is not a valid Unicode normalization form'):
        pc.utf8_normalize(arr, form="NFZ")


def test_random():
    # (note negative integer initializers are accepted)
    for initializer in ['system', 42, -42, b"abcdef"]:
        assert pc.random(0, initializer=initializer) == \
            pa.array([], type=pa.float64())

    # System random initialization => outputs all distinct
    arrays = [tuple(pc.random(100).to_pylist()) for i in range(10)]
    assert len(set(arrays)) == len(arrays)

    arrays = [tuple(pc.random(100, initializer=i % 7).to_pylist())
              for i in range(0, 100)]
    assert len(set(arrays)) == 7

    # Arbitrary hashable objects can be given as initializer
    initializers = [object(), (4, 5, 6), "foo"]
    initializers.extend(os.urandom(10) for i in range(10))
    arrays = [tuple(pc.random(100, initializer=i).to_pylist())
              for i in initializers]
    assert len(set(arrays)) == len(arrays)

    with pytest.raises(TypeError,
                       match=r"initializer should be 'system', an integer, "
                             r"or a hashable object; got \[\]"):
        pc.random(100, initializer=[])


@pytest.mark.parametrize(
    "tiebreaker,expected_values",
    [("min", [3, 1, 4, 6, 4, 6, 1]),
     ("max", [3, 2, 5, 7, 5, 7, 2]),
     ("first", [3, 1, 4, 6, 5, 7, 2]),
     ("dense", [2, 1, 3, 4, 3, 4, 1])]
)
def test_rank_options_tiebreaker(tiebreaker, expected_values):
    arr = pa.array([1.2, 0.0, 5.3, None, 5.3, None, 0.0])
    rank_options = pc.RankOptions(sort_keys="ascending",
                                  null_placement="at_end",
                                  tiebreaker=tiebreaker)
    result = pc.rank(arr, options=rank_options)
    expected = pa.array(expected_values, type=pa.uint64())
    assert result.equals(expected)


def test_rank_options():
    arr = pa.array([1.2, 0.0, 5.3, None, 5.3, None, 0.0])
    expected = pa.array([3, 1, 4, 6, 5, 7, 2], type=pa.uint64())

    # Ensure rank can be called without specifying options
    result = pc.rank(arr)
    assert result.equals(expected)

    # Ensure default RankOptions
    result = pc.rank(arr, options=pc.RankOptions())
    assert result.equals(expected)

    # Ensure sort_keys tuple usage
    result = pc.rank(arr, options=pc.RankOptions(
        sort_keys=[("b", "ascending")])
    )
    assert result.equals(expected)

    result = pc.rank(arr, null_placement="at_start")
    expected_at_start = pa.array([5, 3, 6, 1, 7, 2, 4], type=pa.uint64())
    assert result.equals(expected_at_start)

    result = pc.rank(arr, sort_keys="descending")
    expected_descending = pa.array([3, 4, 1, 6, 2, 7, 5], type=pa.uint64())
    assert result.equals(expected_descending)

    with pytest.raises(ValueError,
                       match=r'"NonExisting" is not a valid tiebreaker'):
        pc.RankOptions(sort_keys="descending",
                       null_placement="at_end",
                       tiebreaker="NonExisting")


def create_sample_expressions():
    # We need a schema for substrait conversion
    schema = pa.schema([pa.field("i64", pa.int64()), pa.field(
        "foo", pa.struct([pa.field("bar", pa.string())]))])

    # Creates a bunch of sample expressions for testing
    # serialization and deserialization. The expressions are categorized
    # to reflect certain nuances in Substrait conversion.
    a = pc.scalar(1)
    b = pc.scalar(1.1)
    c = pc.scalar(True)
    d = pc.scalar("string")
    e = pc.scalar(None)
    f = pc.scalar({'a': 1})
    g = pc.scalar(pa.scalar(1))
    h = pc.scalar(np.int64(2))
    j = pc.scalar(False)
    k = pc.scalar(0)

    # These expression consist entirely of literals
    literal_exprs = [a, b, c, d, e, g, h, j, k]

    # These expressions include at least one function call
    exprs_with_call = [a == b, a != b, a > b, c & j, c | j, ~c, d.is_valid(),
                       a + b, a - b, a * b, a / b, pc.negate(a),
                       pc.add(a, b), pc.subtract(a, b), pc.divide(a, b),
                       pc.multiply(a, b), pc.power(a, a), pc.sqrt(a),
                       pc.exp(b), pc.cos(b), pc.sin(b), pc.tan(b),
                       pc.acos(b), pc.atan(b), pc.asin(b), pc.atan2(b, b),
                       pc.sinh(a), pc.cosh(a), pc.tanh(a),
                       pc.asinh(a), pc.acosh(b), pc.atanh(k),
                       pc.abs(b), pc.sign(a), pc.bit_wise_not(a),
                       pc.bit_wise_and(a, a), pc.bit_wise_or(a, a),
                       pc.bit_wise_xor(a, a), pc.is_nan(b), pc.is_finite(b),
                       pc.coalesce(a, b),
                       a.cast(pa.int32(), safe=False)]

    # These expressions test out various reference styles and may include function
    # calls.  Named references are used here.
    exprs_with_ref = [pc.field('i64') > 5, pc.field('i64') == 5,
                      pc.field('i64') == 7,
                      pc.field(('foo', 'bar')) == 'value',
                      pc.field('foo', 'bar') == 'value']

    # Similar to above but these use numeric references instead of string refs
    exprs_with_numeric_refs = [pc.field(0) > 5, pc.field(0) == 5,
                               pc.field(0) == 7,
                               pc.field((1, 0)) == 'value',
                               pc.field(1, 0) == 'value']

    # Expressions that behave uniquely when converting to/from substrait
    special_cases = [
        f,  # Struct literals lose their field names
        a.isin([1, 2, 3]),  # isin converts to an or list
        pc.field('i64').is_null()  # pyarrow always specifies a FunctionOptions
                                   # for is_null which, being the default, is
                                   # dropped on serialization
    ]

    all_exprs = literal_exprs.copy()
    all_exprs += exprs_with_call
    all_exprs += exprs_with_ref
    all_exprs += special_cases

    return {
        "all": all_exprs,
        "literals": literal_exprs,
        "calls": exprs_with_call,
        "refs": exprs_with_ref,
        "numeric_refs": exprs_with_numeric_refs,
        "special": special_cases,
        "schema": schema
    }

# Tests the Arrow-specific serialization mechanism


@pytest.mark.numpy
def test_expression_serialization_arrow(pickle_module):
    for expr in create_sample_expressions()["all"]:
        assert isinstance(expr, pc.Expression)
        restored = pickle_module.loads(pickle_module.dumps(expr))
        assert expr.equals(restored)


@pytest.mark.numpy
@pytest.mark.substrait
def test_expression_serialization_substrait():

    exprs = create_sample_expressions()
    schema = exprs["schema"]

    # Basic literals don't change on binding and so they will round
    # trip without any change
    for expr in exprs["literals"]:
        serialized = expr.to_substrait(schema)
        deserialized = pc.Expression.from_substrait(serialized)
        assert expr.equals(deserialized)

    # Expressions are bound when they get serialized.  Since bound
    # expressions are not equal to their unbound variants we cannot
    # compare the round tripped with the original
    for expr in exprs["calls"]:
        serialized = expr.to_substrait(schema)
        deserialized = pc.Expression.from_substrait(serialized)
        # We can't compare the expressions themselves because of the bound
        # unbound difference. But we can compare the string representation
        assert str(deserialized) == str(expr)
        serialized_again = deserialized.to_substrait(schema)
        deserialized_again = pc.Expression.from_substrait(serialized_again)
        assert deserialized.equals(deserialized_again)

    for expr, expr_norm in zip(exprs["refs"], exprs["numeric_refs"]):
        serialized = expr.to_substrait(schema)
        deserialized = pc.Expression.from_substrait(serialized)
        assert str(deserialized) == str(expr_norm)
        serialized_again = deserialized.to_substrait(schema)
        deserialized_again = pc.Expression.from_substrait(serialized_again)
        assert deserialized.equals(deserialized_again)

    # For the special cases we get various wrinkles in serialization but we
    # should always get the same thing from round tripping twice
    for expr in exprs["special"]:
        serialized = expr.to_substrait(schema)
        deserialized = pc.Expression.from_substrait(serialized)
        serialized_again = deserialized.to_substrait(schema)
        deserialized_again = pc.Expression.from_substrait(serialized_again)
        assert deserialized.equals(deserialized_again)

    # Special case, we lose the field names of struct literals
    f = exprs["special"][0]
    serialized = f.to_substrait(schema)
    deserialized = pc.Expression.from_substrait(serialized)
    assert deserialized.equals(pc.scalar({'': 1}))

    # Special case, is_in converts to a == opt[0] || a == opt[1] ...
    a = pc.scalar(1)
    expr = a.isin([1, 2, 3])
    target = (a == 1) | (a == 2) | (a == 3)
    serialized = expr.to_substrait(schema)
    deserialized = pc.Expression.from_substrait(serialized)
    # Compare str's here to bypass the bound/unbound difference
    assert str(target) == str(deserialized)
    serialized_again = deserialized.to_substrait(schema)
    deserialized_again = pc.Expression.from_substrait(serialized_again)
    assert deserialized.equals(deserialized_again)


def test_expression_construction():
    zero = pc.scalar(0)
    one = pc.scalar(1)
    true = pc.scalar(True)
    false = pc.scalar(False)
    string = pc.scalar("string")
    field = pc.field("field")
    nested_mixed_types = pc.field(b"a", 1, "b")
    nested_field = pc.field(("nested", "field"))
    nested_field2 = pc.field("nested", "field")

    zero | one == string
    ~true == false
    for typ in ("bool", pa.bool_()):
        field.cast(typ) == true

    field.isin([1, 2])
    nested_mixed_types.isin(["foo", "bar"])
    nested_field.isin(["foo", "bar"])
    nested_field2.isin(["foo", "bar"])

    with pytest.raises(TypeError):
        field.isin(1)

    with pytest.raises(pa.ArrowInvalid):
        field != object()


def test_expression_boolean_operators():
    # https://issues.apache.org/jira/browse/ARROW-11412
    true = pc.scalar(True)
    false = pc.scalar(False)

    with pytest.raises(ValueError, match="cannot be evaluated to python True"):
        true and false

    with pytest.raises(ValueError, match="cannot be evaluated to python True"):
        true or false

    with pytest.raises(ValueError, match="cannot be evaluated to python True"):
        bool(true)

    with pytest.raises(ValueError, match="cannot be evaluated to python True"):
        not true


def test_expression_call_function():
    field = pc.field("field")

    # no options
    assert str(pc.hour(field)) == "hour(field)"

    # default options
    assert str(pc.round(field)) == "round(field)"
    # specified options
    assert str(pc.round(field, ndigits=1)) == \
        "round(field, {ndigits=1, round_mode=HALF_TO_EVEN})"

    # Will convert non-expression arguments if possible
    assert str(pc.add(field, 1)) == "add(field, 1)"
    assert str(pc.add(field, pa.scalar(1))) == "add(field, 1)"

    # Invalid pc.scalar input gives original error message
    msg = "only other expressions allowed as arguments"
    with pytest.raises(TypeError, match=msg):
        pc.add(field, object)


def test_cast_table_raises():
    table = pa.table({'a': [1, 2]})

    with pytest.raises(pa.lib.ArrowTypeError):
        pc.cast(table, pa.int64())


@pytest.mark.parametrize("start,stop,expected", (
    (0, None, [[1, 2, 3], [4, 5, None], [6, None, None], None]),
    (0, 1, [[1], [4], [6], None]),
    (0, 2, [[1, 2], [4, 5], [6, None], None]),
    (1, 2, [[2], [5], [None], None]),
    (2, 4, [[3, None], [None, None], [None, None], None])
))
@pytest.mark.parametrize("step", (1, 2))
@pytest.mark.parametrize("value_type", (pa.string, pa.int16, pa.float64))
@pytest.mark.parametrize("list_type", (pa.list_, pa.large_list, "fixed"))
def test_list_slice_output_fixed(start, stop, step, expected, value_type,
                                 list_type):
    if list_type == "fixed":
        arr = pa.array([[1, 2, 3], [4, 5, None], [6, None, None], None],
                       pa.list_(pa.int8(), 3)).cast(pa.list_(value_type(), 3))
    else:
        arr = pa.array([[1, 2, 3], [4, 5], [6], None],
                       pa.list_(pa.int8())).cast(list_type(value_type()))

    args = arr, start, stop, step, True
    if stop is None and list_type != "fixed":
        msg = ("Unable to produce FixedSizeListArray from "
               "non-FixedSizeListArray without `stop` being set.")
        with pytest.raises(pa.ArrowInvalid, match=msg):
            pc.list_slice(*args)
    else:
        result = pc.list_slice(*args)
        pylist = result.cast(pa.list_(pa.int8(),
                             result.type.list_size)).to_pylist()
        assert pylist == [e[::step] if e else e for e in expected]


@pytest.mark.parametrize("start,stop", (
    (0, None,),
    (0, 1,),
    (0, 2,),
    (1, 2,),
    (2, 4,)
))
@pytest.mark.parametrize("step", (1, 2))
@pytest.mark.parametrize("value_type", (pa.string, pa.int16, pa.float64))
@pytest.mark.parametrize("list_type", (pa.list_, pa.large_list, "fixed"))
def test_list_slice_output_variable(start, stop, step, value_type, list_type):
    if list_type == "fixed":
        data = [[1, 2, 3], [4, 5, None], [6, None, None], None]
        arr = pa.array(
            data,
            pa.list_(pa.int8(), 3)).cast(pa.list_(value_type(), 3))
    else:
        data = [[1, 2, 3], [4, 5], [6], None]
        arr = pa.array(data,
                       pa.list_(pa.int8())).cast(list_type(value_type()))

    # Gets same list type (ListArray vs LargeList)
    if list_type == "fixed":
        list_type = pa.list_  # non fixed output type

    result = pc.list_slice(arr, start, stop, step,
                           return_fixed_size_list=False)
    assert result.type == list_type(value_type())

    pylist = result.cast(pa.list_(pa.int8())).to_pylist()

    # Variable output slicing follows Python's slice semantics
    expected = [d[start:stop:step] if d is not None else None for d in data]
    assert pylist == expected


@pytest.mark.parametrize("return_fixed_size", (True, False, None))
@pytest.mark.parametrize("type", (
    lambda: pa.list_(pa.field('col', pa.int8())),
    lambda: pa.list_(pa.field('col', pa.int8()), 1),
    lambda: pa.large_list(pa.field('col', pa.int8()))))
def test_list_slice_field_names_retained(return_fixed_size, type):
    arr = pa.array([[1]], type())
    out = pc.list_slice(arr, 0, 1, return_fixed_size_list=return_fixed_size)
    assert arr.type.field(0).name == out.type.field(0).name

    # Verify out type matches in type if return_fixed_size_list==None
    if return_fixed_size is None:
        assert arr.type == out.type


def test_list_slice_bad_parameters():
    arr = pa.array([[1]], pa.list_(pa.int8(), 1))
    msg = r"`start`(.*) should be greater than 0 and smaller than `stop`(.*)"
    with pytest.raises(pa.ArrowInvalid, match=msg):
        pc.list_slice(arr, -1, 1)  # negative start?
    with pytest.raises(pa.ArrowInvalid, match=msg):
        pc.list_slice(arr, 2, 1)  # start > stop?

    # TODO(ARROW-18281): start==stop -> empty lists
    with pytest.raises(pa.ArrowInvalid, match=msg):
        pc.list_slice(arr, 0, 0)  # start == stop?

    # Step not >= 1
    msg = "`step` must be >= 1, got: "
    with pytest.raises(pa.ArrowInvalid, match=msg + "0"):
        pc.list_slice(arr, 0, 1, step=0)
    with pytest.raises(pa.ArrowInvalid, match=msg + "-1"):
        pc.list_slice(arr, 0, 1, step=-1)


def check_run_end_encode_decode(run_end_encode_opts=None):
    arr = pa.array([1, 1, 1, 2, 2, 1, 1, 1, 1, 1, 3, 3, 3, 3, 3, 3, 3, 3, 3])
    encoded = pc.run_end_encode(arr, options=run_end_encode_opts)
    decoded = pc.run_end_decode(encoded)
    assert decoded.type == arr.type
    assert decoded.equals(arr)


def test_run_end_encode():
    check_run_end_encode_decode()
    check_run_end_encode_decode(pc.RunEndEncodeOptions(pa.int16()))
    check_run_end_encode_decode(pc.RunEndEncodeOptions('int32'))
    check_run_end_encode_decode(pc.RunEndEncodeOptions(pa.int64()))


def test_pairwise_diff():
    arr = pa.array([1, 2, 3, None, 4, 5])
    expected = pa.array([None, 1, 1, None, None, 1])
    result = pa.compute.pairwise_diff(arr, period=1)
    assert result.equals(expected)

    arr = pa.array([1, 2, 3, None, 4, 5])
    expected = pa.array([None, None, 2, None, 1, None])
    result = pa.compute.pairwise_diff(arr, period=2)
    assert result.equals(expected)

    # negative period
    arr = pa.array([1, 2, 3, None, 4, 5], type=pa.int8())
    expected = pa.array([-1, -1, None, None, -1, None], type=pa.int8())
    result = pa.compute.pairwise_diff(arr, period=-1)
    assert result.equals(expected)

    # wrap around overflow
    arr = pa.array([1, 2, 3, None, 4, 5], type=pa.uint8())
    expected = pa.array([255, 255, None, None, 255, None], type=pa.uint8())
    result = pa.compute.pairwise_diff(arr, period=-1)
    assert result.equals(expected)

    # fail on overflow
    arr = pa.array([1, 2, 3, None, 4, 5], type=pa.uint8())
    with pytest.raises(pa.ArrowInvalid,
                       match="overflow"):
        pa.compute.pairwise_diff_checked(arr, period=-1)