TST: Add ExtensionArray base tests for ops between pandas objects and EAs

[TomAugspurger]

We should document the expected behavior for binary operations between ExtensionArrays and pandas' own objects (Series, DataFrame, Index). We should update our own EAs to match the docs.

In [4]: integer_array([1, 2]) + pd.Series([1, 2])
Out[4]: IntegerArray([2, 4], dtype='Int64')

Out[4] should be a Series with "Int64" dtype, to match the behavior of ndarray + series.

We've documented this, but need to test between series, index, dataframe, and the array class

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

Lines 83 to 88 in 6bf6cd2

	def test_arith_series_with_array(self, data, all_arithmetic_operators):
	# ndarray & other series
	op_name = all_arithmetic_operators
	s = pd.Series(data)
	self.check_opname(s, op_name, pd.Series([s.iloc[0]] * len(s)),
	exc=self.series_array_exc)

[jorisvandenbossche]

and also the interaction with non-pandas objects (lists, arrays)

[TomAugspurger]

Tangentially related, I noticed today that divmod(ndarray, Series) raises.

In [3]: arr = np.array([1, 2])

In [4]: divmod(arr, pd.Series(arr))
---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
<ipython-input-4-b14c14fd32f3> in <module>
----> 1 divmod(arr, pd.Series(arr))

TypeError: unsupported operand type(s) for divmod(): 'numpy.ndarray' and 'Series'

Given that both division on mod work, divmod probably should as well.

[jbrockmendel]

Tangentially related, I noticed today that divmod(ndarray, Series) raises.

I think this is #19252.

[TomAugspurger]

I think this is sufficiently documented in extending.rst. Closing.

[jorisvandenbossche]

It's indeed written in the docs, but do we actually properly test it in the base tests?

Because eg test_arith_series_with_array

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

Lines 83 to 88 in 6bf6cd2

	def test_arith_series_with_array(self, data, all_arithmetic_operators):
	# ndarray & other series
	op_name = all_arithmetic_operators
	s = pd.Series(data)
	self.check_opname(s, op_name, pd.Series([s.iloc[0]] * len(s)),
	exc=self.series_array_exc)

is actually testing the op with series and series, not series and array.

And we also don't have it with eg Index or DataFrame.

[TomAugspurger]

Ok, will re-purpose for tests.