BUG: Fix DataFrame logical ops Series inconsistency (pandas-dev#28741)

Author: jbrockmendel

doc/source/whatsnew/v1.0.0.rst

@@ -211,6 +211,7 @@ Numeric
 ^^^^^^^
 - Bug in :meth:`DataFrame.quantile` with zero-column :class:`DataFrame` incorrectly raising (:issue:`23925`)
 - :class:`DataFrame` inequality comparisons with object-dtype and ``complex`` entries failing to raise ``TypeError`` like their :class:`Series` counterparts (:issue:`28079`)
+- Bug in :class:`DataFrame` logical operations (`&`, `|`, `^`) not matching :class:`Series` behavior by filling NA values (:issue:`28741`)
 -
 
 Conversion

pandas/core/frame.py

@@ -5286,7 +5286,7 @@ def _arith_op(left, right):
     def _combine_match_index(self, other, func):
         # at this point we have `self.index.equals(other.index)`
 
-        if self._is_mixed_type or other._is_mixed_type:
+        if ops.should_series_dispatch(self, other, func):
             # operate column-wise; avoid costly object-casting in `.values`
             new_data = ops.dispatch_to_series(self, other, func)
         else:

pandas/core/ops/__init__.py

@@ -692,6 +692,7 @@ def _arith_method_FRAME(cls, op, special):
     default_axis = _get_frame_op_default_axis(op_name)
 
     na_op = define_na_arithmetic_op(op, str_rep, eval_kwargs)
+    is_logical = str_rep in ["&", "|", "^"]
 
     if op_name in _op_descriptions:
         # i.e. include "add" but not "__add__"
@@ -707,11 +708,13 @@ def f(self, other, axis=default_axis, level=None, fill_value=None):
         if isinstance(other, ABCDataFrame):
             # Another DataFrame
             pass_op = op if should_series_dispatch(self, other, op) else na_op
+            pass_op = pass_op if not is_logical else op
             return self._combine_frame(other, pass_op, fill_value, level)
         elif isinstance(other, ABCSeries):
             # For these values of `axis`, we end up dispatching to Series op,
             # so do not want the masked op.
             pass_op = op if axis in [0, "columns", None] else na_op
+            pass_op = pass_op if not is_logical else op
             return _combine_series_frame(
                 self, other, pass_op, fill_value=fill_value, axis=axis, level=level
             )

pandas/core/ops/dispatch.py

@@ -56,7 +56,7 @@ def should_series_dispatch(left, right, op):
     Parameters
     ----------
     left : DataFrame
-    right : DataFrame
+    right : DataFrame or Series
     op : binary operator
 
     Returns
@@ -66,6 +66,16 @@ def should_series_dispatch(left, right, op):
     if left._is_mixed_type or right._is_mixed_type:
         return True
 
+    if op.__name__.strip("_") in ["and", "or", "xor", "rand", "ror", "rxor"]:
+        # TODO: GH references for what this fixes
+        # Note: this check must come before the check for nonempty columns.
+        return True
+
+    if right.ndim == 1:
+        # operating with Series, short-circuit checks that would fail
+        #  with AttributeError.
+        return False
+
     if not len(left.columns) or not len(right.columns):
         # ensure obj.dtypes[0] exists for each obj
         return False

pandas/tests/frame/test_operators.py

@@ -127,7 +127,8 @@ def test_logical_ops_empty_frame(self):
         dfa = DataFrame(index=[1], columns=["A"])
 
         result = dfa & dfa
-        assert_frame_equal(result, dfa)
+        expected = DataFrame(False, index=[1], columns=["A"])
+        assert_frame_equal(result, expected)
 
     def test_logical_ops_bool_frame(self):
         # GH#5808
@@ -145,7 +146,11 @@ def test_logical_ops_int_frame(self):
         df1a_bool = DataFrame(True, index=[1], columns=["A"])
 
         result = df1a_int | df1a_bool
-        assert_frame_equal(result, df1a_int)
+        assert_frame_equal(result, df1a_bool)
+
+        # Check that this matches Series behavior
+        res_ser = df1a_int["A"] | df1a_bool["A"]
+        assert_series_equal(res_ser, df1a_bool["A"])
 
     def test_logical_ops_invalid(self):
         # GH#5808

pandas/tests/series/test_operators.py

@@ -250,12 +250,20 @@ def test_scalar_na_logical_ops_corners(self):
 
         with pytest.raises(TypeError):
             d.__and__(s, axis="columns")
+        with pytest.raises(TypeError):
+            d.__and__(s, axis=1)
 
         with pytest.raises(TypeError):
             s & d
+        with pytest.raises(TypeError):
+            d & s
+
+        expected = (s & s).to_frame("A")
+        result = d.__and__(s, axis="index")
+        tm.assert_frame_equal(result, expected)
 
-        # this is wrong as its not a boolean result
-        # result = d.__and__(s,axis='index')
+        result = d.__and__(s, axis=0)
+        tm.assert_frame_equal(result, expected)
 
     @pytest.mark.parametrize("op", [operator.and_, operator.or_, operator.xor])
     def test_logical_ops_with_index(self, op):
@@ -436,20 +444,19 @@ def test_logical_ops_df_compat(self):
         assert_series_equal(s2 & s1, exp)
 
         # True | np.nan => True
-        exp = pd.Series([True, True, True, False], index=list("ABCD"), name="x")
-        assert_series_equal(s1 | s2, exp)
+        exp_or1 = pd.Series([True, True, True, False], index=list("ABCD"), name="x")
+        assert_series_equal(s1 | s2, exp_or1)
         # np.nan | True => np.nan, filled with False
-        exp = pd.Series([True, True, False, False], index=list("ABCD"), name="x")
-        assert_series_equal(s2 | s1, exp)
+        exp_or = pd.Series([True, True, False, False], index=list("ABCD"), name="x")
+        assert_series_equal(s2 | s1, exp_or)
 
         # DataFrame doesn't fill nan with False
-        exp = pd.DataFrame({"x": [True, False, np.nan, np.nan]}, index=list("ABCD"))
-        assert_frame_equal(s1.to_frame() & s2.to_frame(), exp)
-        assert_frame_equal(s2.to_frame() & s1.to_frame(), exp)
+        assert_frame_equal(s1.to_frame() & s2.to_frame(), exp.to_frame())
+        assert_frame_equal(s2.to_frame() & s1.to_frame(), exp.to_frame())
 
         exp = pd.DataFrame({"x": [True, True, np.nan, np.nan]}, index=list("ABCD"))
-        assert_frame_equal(s1.to_frame() | s2.to_frame(), exp)
-        assert_frame_equal(s2.to_frame() | s1.to_frame(), exp)
+        assert_frame_equal(s1.to_frame() | s2.to_frame(), exp_or1.to_frame())
+        assert_frame_equal(s2.to_frame() | s1.to_frame(), exp_or.to_frame())
 
         # different length
         s3 = pd.Series([True, False, True], index=list("ABC"), name="x")
@@ -460,19 +467,17 @@ def test_logical_ops_df_compat(self):
         assert_series_equal(s4 & s3, exp)
 
         # np.nan | True => np.nan, filled with False
-        exp = pd.Series([True, True, True, False], index=list("ABCD"), name="x")
-        assert_series_equal(s3 | s4, exp)
+        exp_or1 = pd.Series([True, True, True, False], index=list("ABCD"), name="x")
+        assert_series_equal(s3 | s4, exp_or1)
         # True | np.nan => True
-        exp = pd.Series([True, True, True, True], index=list("ABCD"), name="x")
-        assert_series_equal(s4 | s3, exp)
+        exp_or = pd.Series([True, True, True, True], index=list("ABCD"), name="x")
+        assert_series_equal(s4 | s3, exp_or)
 
-        exp = pd.DataFrame({"x": [True, False, True, np.nan]}, index=list("ABCD"))
-        assert_frame_equal(s3.to_frame() & s4.to_frame(), exp)
-        assert_frame_equal(s4.to_frame() & s3.to_frame(), exp)
+        assert_frame_equal(s3.to_frame() & s4.to_frame(), exp.to_frame())
+        assert_frame_equal(s4.to_frame() & s3.to_frame(), exp.to_frame())
 
-        exp = pd.DataFrame({"x": [True, True, True, np.nan]}, index=list("ABCD"))
-        assert_frame_equal(s3.to_frame() | s4.to_frame(), exp)
-        assert_frame_equal(s4.to_frame() | s3.to_frame(), exp)
+        assert_frame_equal(s3.to_frame() | s4.to_frame(), exp_or1.to_frame())
+        assert_frame_equal(s4.to_frame() | s3.to_frame(), exp_or.to_frame())
 
 
 class TestSeriesComparisons: