TST: Add tests for num-bool coercion

[kgoehner]

Boolean dtypes in both Series and DataFrames should be coerced
to the other operand's dtype.

• closes BUG: boolean frames multiplied by floats have dtypes=object #18549
• tests added / passed
• passes black pandas
• passes git diff upstream/master -u -- "*.py" | flake8 --diff
• whatsnew entry

[WillAyd]

@jbrockmendel was this perhaps fixed by some of the _maybe_promote work you've been doing?

[jbrockmendel]

was this perhaps fixed by some of the _maybe_promote work you've been doing?

Doubtful. I don't think maybe_promote is used anywhere in/around ops.

@Kazz47 would the same test be applicable for Index classes? Can you parametrize the test over [Index, Series, DataFrame] and put it somewhere in tests.arithmetic?

[kgoehner]

@jbrockmendel some of the operators (multiply and floordiv) aren't supported by Index. Should these be skipped if we see a TypeError for some operators? If so I can update the PR. Otherwise, do you have a more graceful way of handling this?

    @pytest.mark.parametrize("container_cls_assert", [(pd.Index, tm.assert_index_equal),
                                                      (pd.Series, tm.assert_series_equal),
                                                      (pd.DataFrame, tm.assert_frame_equal)])
    @pytest.mark.parametrize("num", [1.0, 1])
    @pytest.mark.parametrize(
        "op", [operator.add, operator.mul, operator.floordiv, operator.sub]
    )
    def test_array_like_bool_and_num_op_coerce(self, container_cls_assert, num, op):
        # GH 18549
        container_cls = container_cls_assert[0]
        container_assert = container_cls_assert[1]
        expected = container_cls([op(num, num)])
        bool_array = container_cls([True])
        container_assert(expected, op(bool_array, num))
        container_assert(expected, op(num, bool_array))

[jbrockmendel]

some of the operators (multiply and floordiv) aren't supported by Index

If the array you're passing to Index is all numeric, you should get Int64Index or Float64Index, so you'll be OK.

You can use tm.assert_equal instead of container_cls_assert.

In tests.arithmetic we use a box or box_with_array fixture along with calls to tm.box_expected(obj, box) to parametrize tests across classes. If you can adhere to that pattern that'd be ideal

[kgoehner]

If the array you're passing to Index is all numeric, you should get Int64Index or Float64Index, so you'll be OK.

Since we are passing a bool we don't get Int64Index or Float64Index and the numeric operators are being disabled.

pandas/pandas/core/indexes/base.py

Line 5505 in 2931f02

Index._add_numeric_methods_disabled()

Latest version:

class TestNumericArraylikeArithmeticWithBool:
    @pytest.mark.parametrize("num", [1.0, 1])
    @pytest.mark.parametrize(
        "op", [operator.add, operator.mul, operator.floordiv, operator.sub]
    )
    def test_array_like_bool_and_num_op_coerce(self, num, op, box_with_array):
        # GH 18549
        expected = [op(num, num)]
        expected = tm.box_expected(expected, box_with_array)
        bool_box = tm.box_expected([True], box_with_array)
        tm.assert_equal(expected, op(bool_box, num))
        tm.assert_equal(expected, op(num, bool_box))

[jreback]

rather than catching you should return on the operators that we don't want to test, making this more explicit

[WillAyd]

@Kazz47 is this still active? Can you merge master and address comment?

[WillAyd]

Nice PR but needs rebase and I think has gone stale. Ping if you'd like to pick this back up