# The purpose of these tests are to ensure that calling quantities using
# array methods returns quantities with the right units, or raises exceptions.

import numpy as np
import pytest
from numpy.testing import assert_array_equal

from astropy import units as u
from astropy.utils.compat.numpycompat import NUMPY_LT_2_0
from astropy.utils.compat.optional_deps import HAS_ARRAY_API_STRICT


class TestQuantityArrayCopy:
    """
    Test whether arrays are properly copied/used in place
    """

    def test_copy_on_creation(self):
        v = np.arange(1000.0)
        q_nocopy = u.Quantity(v, "km/s", copy=False)
        q_copy = u.Quantity(v, "km/s", copy=True)
        v[0] = -1.0
        assert q_nocopy[0].value == v[0]
        assert q_copy[0].value != v[0]

    def test_to_copies(self):
        q = u.Quantity(np.arange(1.0, 100.0), "km/s")
        q2 = q.to(u.m / u.s)
        assert np.all(q.value != q2.value)
        q3 = q.to(u.km / u.s)
        assert np.all(q.value == q3.value)
        q[0] = -1.0 * u.km / u.s
        assert q[0].value != q3[0].value

    def test_si_copies(self):
        q = u.Quantity(np.arange(100.0), "m/s")
        q2 = q.si
        assert np.all(q.value == q2.value)
        q[0] = -1.0 * u.m / u.s
        assert q[0].value != q2[0].value

    def test_getitem_is_view(self):
        """Check that [keys] work, and that, like ndarray, it returns
        a view, so that changing one changes the other.

        Also test that one can add axes (closes #1422)
        """
        q = u.Quantity(np.arange(100.0), "m/s")
        q_sel = q[10:20]
        q_sel[0] = -1.0 * u.m / u.s
        assert q_sel[0] == q[10]
        # also check that getitem can do new axes
        q2 = q[:, np.newaxis]
        q2[10, 0] = -9 * u.m / u.s
        assert np.all(q2.flatten() == q)

    def test_flat(self):
        q = u.Quantity(np.arange(9.0).reshape(3, 3), "m/s")
        q_flat = q.flat
        # check that a single item is a quantity (with the right value)
        assert q_flat[8] == 8.0 * u.m / u.s
        # and that getting a range works as well
        assert np.all(q_flat[0:2] == np.arange(2.0) * u.m / u.s)
        # as well as getting items via iteration
        q_flat_list = list(q.flat)
        assert np.all(u.Quantity(q_flat_list) == u.Quantity(list(q.value.flat), q.unit))
        # check that flat works like a view of the real array
        q_flat[8] = -1.0 * u.km / u.s
        assert q_flat[8] == -1.0 * u.km / u.s
        assert q[2, 2] == -1.0 * u.km / u.s
        # while if one goes by an iterated item, a copy is made
        q_flat_list[8] = -2 * u.km / u.s
        assert q_flat_list[8] == -2.0 * u.km / u.s
        assert q_flat[8] == -1.0 * u.km / u.s
        assert q[2, 2] == -1.0 * u.km / u.s


class TestQuantityReshapeFuncs:
    """Test different ndarray methods that alter the array shape

    tests: reshape, squeeze, ravel, flatten, transpose, swapaxes
    """

    def test_reshape(self):
        q = np.arange(6.0) * u.m
        q_reshape = q.reshape(3, 2)
        assert isinstance(q_reshape, u.Quantity)
        assert q_reshape.unit == q.unit
        assert np.all(q_reshape.value == q.value.reshape(3, 2))

    def test_squeeze(self):
        q = np.arange(6.0).reshape(6, 1) * u.m
        q_squeeze = q.squeeze()
        assert isinstance(q_squeeze, u.Quantity)
        assert q_squeeze.unit == q.unit
        assert np.all(q_squeeze.value == q.value.squeeze())

    def test_ravel(self):
        q = np.arange(6.0).reshape(3, 2) * u.m
        q_ravel = q.ravel()
        assert isinstance(q_ravel, u.Quantity)
        assert q_ravel.unit == q.unit
        assert np.all(q_ravel.value == q.value.ravel())

    def test_flatten(self):
        q = np.arange(6.0).reshape(3, 2) * u.m
        q_flatten = q.flatten()
        assert isinstance(q_flatten, u.Quantity)
        assert q_flatten.unit == q.unit
        assert np.all(q_flatten.value == q.value.flatten())

    def test_transpose(self):
        q = np.arange(6.0).reshape(3, 2) * u.m
        q_transpose = q.transpose()
        assert isinstance(q_transpose, u.Quantity)
        assert q_transpose.unit == q.unit
        assert np.all(q_transpose.value == q.value.transpose())

    def test_swapaxes(self):
        q = np.arange(6.0).reshape(3, 1, 2) * u.m
        q_swapaxes = q.swapaxes(0, 2)
        assert isinstance(q_swapaxes, u.Quantity)
        assert q_swapaxes.unit == q.unit
        assert np.all(q_swapaxes.value == q.value.swapaxes(0, 2))

    def test_flat_attributes(self):
        """While ``flat`` doesn't make a copy, it changes the shape."""
        q = np.arange(6.0).reshape(3, 1, 2) * u.m
        qf = q.flat
        # flat shape is same as before reshaping
        assert len(qf) == 6
        # see TestQuantityArrayCopy.test_flat for tests of iteration
        # and slicing and setting. Here we test the properties and methods to
        # match `numpy.ndarray.flatiter`
        assert qf.base is q
        # testing the indices -- flat and full -- into the array
        assert qf.coords == (0, 0, 0)  # to start
        assert qf.index == 0
        # now consume the iterator
        endindices = [(qf.index, qf.coords) for x in qf][-2]  # next() oversteps
        assert endindices[0] == 5
        assert endindices[1] == (2, 0, 1)  # shape of q - 1

        # also check q_flat copies properly
        q_flat_copy = qf.copy()
        assert all(q_flat_copy == q.flatten())
        assert isinstance(q_flat_copy, u.Quantity)
        assert not np.may_share_memory(q_flat_copy, q)


class TestQuantityStatsFuncs:
    """
    Test statistical functions
    """

    def test_mean(self):
        q1 = np.array([1.0, 2.0, 4.0, 5.0, 6.0]) * u.m
        assert_array_equal(np.mean(q1), 3.6 * u.m)
        assert_array_equal(np.mean(q1, keepdims=True), [3.6] * u.m)

    def test_mean_inplace(self):
        q1 = np.array([1.0, 2.0, 4.0, 5.0, 6.0]) * u.m
        qi = 1.5 * u.s
        qi2 = np.mean(q1, out=qi)
        assert qi2 is qi
        assert qi == 3.6 * u.m

    def test_mean_where(self):
        q1 = np.array([1.0, 2.0, 4.0, 5.0, 6.0, 7.0]) * u.m
        assert_array_equal(np.mean(q1, where=q1 < 7 * u.m), 3.6 * u.m)

    def test_std(self):
        q1 = np.array([1.0, 2.0]) * u.m
        assert_array_equal(np.std(q1), 0.5 * u.m)
        assert_array_equal(q1.std(axis=-1, keepdims=True), [0.5] * u.m)

    def test_std_inplace(self):
        q1 = np.array([1.0, 2.0]) * u.m
        qi = 1.5 * u.s
        np.std(q1, out=qi)
        assert qi == 0.5 * u.m

    def test_std_where(self):
        q1 = np.array([1.0, 2.0, 3.0]) * u.m
        assert_array_equal(np.std(q1, where=q1 < 3 * u.m), 0.5 * u.m)

    def test_var(self):
        q1 = np.array([1.0, 2.0]) * u.m
        assert_array_equal(np.var(q1), 0.25 * u.m**2)
        assert_array_equal(q1.var(axis=0, keepdims=True), [0.25] * u.m**2)

    def test_var_inplace(self):
        q1 = np.array([1.0, 2.0]) * u.m
        qi = 1.5 * u.s
        np.var(q1, out=qi)
        assert qi == 0.25 * u.m**2

    def test_var_where(self):
        q1 = np.array([1.0, 2.0, 3.0]) * u.m
        assert_array_equal(np.var(q1, where=q1 < 3 * u.m), 0.25 * u.m**2)

    def test_median(self):
        q1 = np.array([1.0, 2.0, 4.0, 5.0, 6.0]) * u.m
        assert np.median(q1) == 4.0 * u.m

    def test_median_inplace(self):
        q1 = np.array([1.0, 2.0, 4.0, 5.0, 6.0]) * u.m
        qi = 1.5 * u.s
        np.median(q1, out=qi)
        assert qi == 4 * u.m

    def test_min(self):
        q1 = np.array([1.0, 2.0, 4.0, 5.0, 6.0]) * u.m
        assert np.min(q1) == 1.0 * u.m

    def test_min_inplace(self):
        q1 = np.array([1.0, 2.0, 4.0, 5.0, 6.0]) * u.m
        qi = 1.5 * u.s
        np.min(q1, out=qi)
        assert qi == 1.0 * u.m

    def test_min_where(self):
        q1 = np.array([0.0, 1.0, 2.0, 4.0, 5.0, 6.0]) * u.m
        assert np.min(q1, initial=10 * u.m, where=q1 > 0 * u.m) == 1.0 * u.m

    def test_argmin(self):
        q1 = np.array([6.0, 2.0, 4.0, 5.0, 6.0]) * u.m
        assert np.argmin(q1) == 1

    def test_argmin_keepdims(self):
        q1 = np.array([[6.0, 2.0], [4.0, 5.0]]) * u.m
        assert_array_equal(q1.argmin(axis=0, keepdims=True), np.array([[1, 0]]))

    def test_max(self):
        q1 = np.array([1.0, 2.0, 4.0, 5.0, 6.0]) * u.m
        assert np.max(q1) == 6.0 * u.m

    def test_max_inplace(self):
        q1 = np.array([1.0, 2.0, 4.0, 5.0, 6.0]) * u.m
        qi = 1.5 * u.s
        np.max(q1, out=qi)
        assert qi == 6.0 * u.m

    def test_max_where(self):
        q1 = np.array([1.0, 2.0, 4.0, 5.0, 6.0, 7.0]) * u.m
        assert np.max(q1, initial=0 * u.m, where=q1 < 7 * u.m) == 6.0 * u.m

    def test_argmax(self):
        q1 = np.array([5.0, 2.0, 4.0, 5.0, 6.0]) * u.m
        assert np.argmax(q1) == 4

    def test_argmax_keepdims(self):
        q1 = np.array([[6.0, 2.0], [4.0, 5.0]]) * u.m
        assert_array_equal(q1.argmax(axis=0, keepdims=True), np.array([[0, 1]]))

    def test_clip(self):
        q1 = np.array([1.0, 2.0, 4.0, 5.0, 6.0]) * u.km / u.m
        c1 = q1.clip(1500, 5.5 * u.Mm / u.km)
        assert np.all(c1 == np.array([1.5, 2.0, 4.0, 5.0, 5.5]) * u.km / u.m)

    def test_clip_inplace(self):
        q1 = np.array([1.0, 2.0, 4.0, 5.0, 6.0]) * u.km / u.m
        c1 = q1.clip(1500, 5.5 * u.Mm / u.km, out=q1)
        assert np.all(q1 == np.array([1.5, 2.0, 4.0, 5.0, 5.5]) * u.km / u.m)
        c1[0] = 10 * u.Mm / u.mm
        assert np.all(c1.value == q1.value)

    def test_conj(self):
        q1 = np.array([1.0, 2.0, 4.0, 5.0, 6.0]) * u.km / u.m
        assert np.all(q1.conj() == q1)

    def test_ptp(self):
        q1 = np.array([1.0, 2.0, 4.0, 5.0, 6.0]) * u.m
        assert np.ptp(q1) == 5.0 * u.m

    def test_ptp_inplace(self):
        q1 = np.array([1.0, 2.0, 4.0, 5.0, 6.0]) * u.m
        qi = 1.5 * u.s
        np.ptp(q1, out=qi)
        assert qi == 5.0 * u.m

    def test_round(self):
        q1 = np.array([1.253, 2.253, 3.253]) * u.kg
        assert np.all(np.round(q1) == np.array([1, 2, 3]) * u.kg)
        assert np.all(np.round(q1, decimals=2) == np.round(q1.value, decimals=2) * u.kg)
        assert np.all(q1.round(decimals=2) == q1.value.round(decimals=2) * u.kg)

    def test_round_inplace(self):
        q1 = np.array([1.253, 2.253, 3.253]) * u.kg
        qi = np.zeros(3) * u.s
        a = q1.round(decimals=2, out=qi)
        assert a is qi
        assert np.all(q1.round(decimals=2) == qi)

    def test_sum(self):
        q1 = np.array([1.0, 2.0, 6.0]) * u.m
        assert np.all(q1.sum() == 9.0 * u.m)
        assert np.all(np.sum(q1) == 9.0 * u.m)

        q2 = np.array([[4.0, 5.0, 9.0], [1.0, 1.0, 1.0]]) * u.s
        assert np.all(q2.sum(0) == np.array([5.0, 6.0, 10.0]) * u.s)
        assert np.all(np.sum(q2, 0) == np.array([5.0, 6.0, 10.0]) * u.s)

    def test_sum_inplace(self):
        q1 = np.array([1.0, 2.0, 6.0]) * u.m
        qi = 1.5 * u.s
        np.sum(q1, out=qi)
        assert qi == 9.0 * u.m

    def test_sum_where(self):
        q1 = np.array([1.0, 2.0, 6.0, 7.0]) * u.m
        where = q1 < 7 * u.m
        assert np.all(q1.sum(where=where) == 9.0 * u.m)
        assert np.all(np.sum(q1, where=where) == 9.0 * u.m)

    @pytest.mark.parametrize("initial", [0, 0 * u.m, 1 * u.km])
    def test_sum_initial(self, initial):
        q1 = np.array([1.0, 2.0, 6.0, 7.0]) * u.m
        expected = 16 * u.m + initial
        assert q1.sum(initial=initial) == expected
        assert np.sum(q1, initial=initial) == expected

    def test_sum_dimensionless_initial(self):
        q1 = np.array([1.0, 2.0, 6.0, 7.0]) * u.one
        assert q1.sum(initial=1000) == 1016 * u.one

    @pytest.mark.parametrize("initial", [10, 1 * u.s])
    def test_sum_initial_exception(self, initial):
        q1 = np.array([1.0, 2.0, 6.0, 7.0]) * u.m
        with pytest.raises(u.UnitsError):
            q1.sum(initial=initial)

    def test_cumsum(self):
        q1 = np.array([1, 2, 6]) * u.m
        assert np.all(q1.cumsum() == np.array([1, 3, 9]) * u.m)
        assert np.all(np.cumsum(q1) == np.array([1, 3, 9]) * u.m)

        q2 = np.array([4, 5, 9]) * u.s
        assert np.all(q2.cumsum() == np.array([4, 9, 18]) * u.s)
        assert np.all(np.cumsum(q2) == np.array([4, 9, 18]) * u.s)

    def test_cumsum_inplace(self):
        q1 = np.array([1, 2, 6]) * u.m
        qi = np.ones(3) * u.s
        np.cumsum(q1, out=qi)
        assert np.all(qi == np.array([1, 3, 9]) * u.m)
        q2 = q1
        q1.cumsum(out=q1)
        assert np.all(q2 == qi)

    def test_prod(self):
        q1 = np.array([1, 2, 6]) * u.m
        with pytest.raises(u.UnitsError) as exc:
            q1.prod()
        with pytest.raises(u.UnitsError) as exc:
            np.prod(q1)

        q2 = np.array([3.0, 4.0, 5.0]) * u.Unit(1)
        assert q2.prod() == 60.0 * u.Unit(1)
        assert np.prod(q2) == 60.0 * u.Unit(1)

    def test_cumprod(self):
        q1 = np.array([1, 2, 6]) * u.m
        with pytest.raises(u.UnitsError) as exc:
            q1.cumprod()
        with pytest.raises(u.UnitsError) as exc:
            np.cumprod(q1)

        q2 = np.array([3, 4, 5]) * u.Unit(1)
        assert np.all(q2.cumprod() == np.array([3, 12, 60]) * u.Unit(1))
        assert np.all(np.cumprod(q2) == np.array([3, 12, 60]) * u.Unit(1))

    def test_diff(self):
        q1 = np.array([1.0, 2.0, 4.0, 10.0]) * u.m
        assert np.all(q1.diff() == np.array([1.0, 2.0, 6.0]) * u.m)
        assert np.all(np.diff(q1) == np.array([1.0, 2.0, 6.0]) * u.m)

    def test_ediff1d(self):
        q1 = np.array([1.0, 2.0, 4.0, 10.0]) * u.m
        assert np.all(q1.ediff1d() == np.array([1.0, 2.0, 6.0]) * u.m)
        assert np.all(np.ediff1d(q1) == np.array([1.0, 2.0, 6.0]) * u.m)

    def test_dot_meth(self):
        q1 = np.array([1.0, 2.0, 4.0, 10.0]) * u.m
        q2 = np.array([3.0, 4.0, 5.0, 6.0]) * u.s
        q3 = q1.dot(q2)
        assert q3.value == np.dot(q1.value, q2.value)
        assert q3.unit == u.m * u.s

    def test_trace_func(self):
        q = np.array([[1.0, 2.0], [3.0, 4.0]]) * u.m
        assert np.trace(q) == 5.0 * u.m

    def test_trace_meth(self):
        q1 = np.array([[1.0, 2.0], [3.0, 4.0]]) * u.m
        assert q1.trace() == 5.0 * u.m

        cont = u.Quantity(4.0, u.s)

        q2 = np.array([[3.0, 4.0], [5.0, 6.0]]) * u.m
        q2.trace(out=cont)
        assert cont == 9.0 * u.m

    def test_clip_func(self):
        q = np.arange(10) * u.m
        assert np.all(
            np.clip(q, 3 * u.m, 6 * u.m)
            == np.array([3.0, 3.0, 3.0, 3.0, 4.0, 5.0, 6.0, 6.0, 6.0, 6.0]) * u.m
        )

    def test_clip_meth(self):
        expected = np.array([3.0, 3.0, 3.0, 3.0, 4.0, 5.0, 6.0, 6.0, 6.0, 6.0]) * u.m

        q1 = np.arange(10) * u.m
        q3 = q1.clip(3 * u.m, 6 * u.m)
        assert np.all(q1.clip(3 * u.m, 6 * u.m) == expected)

        cont = np.zeros(10) * u.s

        q1.clip(3 * u.m, 6 * u.m, out=cont)

        assert np.all(cont == expected)


class TestArrayConversion:
    """
    Test array conversion methods
    """

    def test_item(self):
        q1 = u.Quantity(np.array([1, 2, 3]), u.m / u.km, dtype=int)
        assert q1.item(1) == 2 * q1.unit

        q1[1] = 1
        assert q1[1] == 1000 * u.m / u.km
        q1[1] = 100 * u.cm / u.km
        assert q1[1] == 1 * u.m / u.km
        with pytest.raises(TypeError):
            q1[1] = 1.5 * u.m / u.km

    @pytest.mark.skipif(not NUMPY_LT_2_0, reason="itemset method removed in numpy 2.0")
    def test_itemset(self):
        q1 = u.Quantity(np.array([1, 2, 3]), u.m / u.km, dtype=int)
        assert q1.item(1) == 2 * q1.unit
        q1.itemset(1, 1)
        assert q1.item(1) == 1000 * u.m / u.km
        q1.itemset(1, 100 * u.cm / u.km)
        assert q1.item(1) == 1 * u.m / u.km
        with pytest.raises(TypeError):
            q1.itemset(1, 1.5 * u.m / u.km)
        with pytest.raises(ValueError):
            q1.itemset()

    def test_take_put(self):
        q1 = np.array([1, 2, 3]) * u.m / u.km
        assert q1.take(1) == 2 * u.m / u.km
        assert all(q1.take((0, 2)) == np.array([1, 3]) * u.m / u.km)
        q1.put((1, 2), (3, 4))
        assert np.all(q1.take((1, 2)) == np.array([3000, 4000]) * q1.unit)
        q1.put(0, 500 * u.cm / u.km)
        assert q1.item(0) == 5 * u.m / u.km

    def test_slice(self):
        """Test that setitem changes the unit if needed (or ignores it for
        values where that is allowed; viz., #2695)"""
        q2 = np.array([[1.0, 2.0, 3.0], [4.0, 5.0, 6.0]]) * u.km / u.m
        q1 = q2.copy()
        q2[0, 0] = 10000.0
        assert q2.unit == q1.unit
        assert q2[0, 0].value == 10.0
        q2[0] = 9.0 * u.Mm / u.km
        assert all(q2.flatten()[:3].value == np.array([9.0, 9.0, 9.0]))
        q2[0, :-1] = 8000.0
        assert all(q2.flatten()[:3].value == np.array([8.0, 8.0, 9.0]))
        with pytest.raises(u.UnitsError):
            q2[1, 1] = 10 * u.s
        # just to be sure, repeat with a dimensionfull unit
        q3 = u.Quantity(np.arange(10.0), "m/s")
        q3[5] = 100.0 * u.cm / u.s
        assert q3[5].value == 1.0
        # and check unit is ignored for 0, inf, nan, where that is reasonable
        q3[5] = 0.0
        assert q3[5] == 0.0
        q3[5] = np.inf
        assert np.isinf(q3[5])
        q3[5] = np.nan
        assert np.isnan(q3[5])

    def test_fill(self):
        q1 = np.array([1, 2, 3]) * u.m / u.km
        q1.fill(2)
        assert np.all(q1 == 2000 * u.m / u.km)

    def test_repeat_compress_diagonal(self):
        q1 = np.array([1, 2, 3]) * u.m / u.km
        q2 = q1.repeat(2)
        assert q2.unit == q1.unit
        assert all(q2.value == q1.value.repeat(2))
        q2.sort()
        assert q2.unit == q1.unit
        q2 = q1.compress(np.array([True, True, False, False]))
        assert q2.unit == q1.unit
        assert all(q2.value == q1.value.compress(np.array([True, True, False, False])))
        q1 = np.array([[1, 2], [3, 4]]) * u.m / u.km
        q2 = q1.diagonal()
        assert q2.unit == q1.unit
        assert all(q2.value == q1.value.diagonal())

    def test_view(self):
        q1 = np.array([1, 2, 3], dtype=np.int64) * u.m / u.km
        q2 = q1.view(np.ndarray)
        assert not hasattr(q2, "unit")
        q3 = q2.view(u.Quantity)
        assert q3._unit is None
        # MaskedArray copies and properties assigned in __dict__
        q4 = np.ma.MaskedArray(q1)
        assert q4._unit is q1._unit
        q5 = q4.view(u.Quantity)
        assert q5.unit is q1.unit

    def test_slice_to_quantity(self):
        """
        Regression test for https://github.com/astropy/astropy/issues/2003
        """

        a = np.random.uniform(size=(10, 8))
        x, y, z = a[:, 1:4].T * u.km / u.s
        total = np.sum(a[:, 1] * u.km / u.s - x)

        assert isinstance(total, u.Quantity)
        assert total == (0.0 * u.km / u.s)

    def test_byte_type_view_field_changes(self):
        q1 = np.array([1, 2, 3], dtype=np.int64) * u.m / u.km
        q2 = q1.byteswap()
        assert q2.unit == q1.unit
        assert all(q2.value == q1.value.byteswap())
        q2 = q1.astype(np.float64)
        assert all(q2 == q1)
        assert q2.dtype == np.float64
        q2a = q1.getfield(np.int32, offset=0)
        q2b = q1.byteswap().getfield(np.int32, offset=4)
        assert q2a.unit == q1.unit
        assert all(q2b.byteswap() == q2a)

    def test_sort(self):
        q1 = np.array([1.0, 5.0, 2.0, 4.0]) * u.km / u.m
        i = q1.argsort()
        assert not hasattr(i, "unit")
        q1.sort()
        i = q1.searchsorted([1500, 2500])
        assert not hasattr(i, "unit")
        assert all(
            i == q1.to(u.dimensionless_unscaled).value.searchsorted([1500, 2500])
        )

    def test_not_implemented(self):
        q1 = np.array([1, 2, 3]) * u.m / u.km

        with pytest.raises(NotImplementedError):
            q1.choose([0, 0, 1])

        with pytest.raises(NotImplementedError):
            q1.tolist()
        with pytest.raises(NotImplementedError):
            q1.tostring()
        with pytest.raises(NotImplementedError):
            q1.tobytes()
        with pytest.raises(NotImplementedError):
            q1.tofile(0)
        with pytest.raises(NotImplementedError):
            q1.dump("a.a")
        with pytest.raises(NotImplementedError):
            q1.dumps()


class TestStructuredArray:
    """Structured arrays are not specifically supported, but we should not
    prevent their use unnecessarily.

    Note that these tests use simple units.  Now that structured units are
    supported, it may make sense to deprecate this.
    """

    def setup_method(self):
        self.ra = (
            np.array(np.arange(12.0).reshape(4, 3)).view(dtype="f8,f8,f8").squeeze()
        )

    def test_creation(self):
        qra = u.Quantity(self.ra, u.m)
        assert np.all(qra[:2].value == self.ra[:2])

    def test_equality(self):
        qra = u.Quantity(self.ra, u.m)
        qra[1] = qra[2]
        assert qra[1] == qra[2]

    def test_assignment_with_non_structured(self):
        qra = u.Quantity(self.ra, u.m)
        qra[1] = 0
        assert qra[1] == np.zeros(3).view(qra.dtype)

    def test_assignment_with_different_names(self):
        qra = u.Quantity(self.ra, u.m)
        dtype = np.dtype([("x", "f8"), ("y", "f8"), ("z", "f8")])
        value = np.array((-1.0, -2.0, -3.0), dtype) << u.km
        qra[1] = value
        assert qra[1] == value
        assert qra[1].value == np.array((-1000.0, -2000.0, -3000.0), qra.dtype)
        # Ensure we do not override dtype names of value.
        assert value.dtype.names == ("x", "y", "z")


@pytest.mark.skipif(not HAS_ARRAY_API_STRICT, reason="requires array_api_strict")
def test_array_api_init():
    import array_api_strict as xp

    array = xp.asarray([1, 2, 3])
    quantity_array = u.Quantity(array, u.m)
    assert type(quantity_array) is u.Quantity
    assert_array_equal(quantity_array, [1, 2, 3] * u.m)