avoid reindex with empty multi-index

Author: matiaslindgren

pandas/core/frame.py

@@ -4450,6 +4450,11 @@ def _set_item_frame_value(self, key, value: DataFrame) -> None:
                 loc, (slice, Series, np.ndarray, Index)
             ):
                 cols_droplevel = maybe_droplevels(cols, key)
+                if (
+                    not isinstance(cols_droplevel, MultiIndex)
+                    and not cols_droplevel.any()
+                ):
+                    return
                 if len(cols_droplevel) and not cols_droplevel.equals(value.columns):
                     value = value.reindex(cols_droplevel, axis=1)
 

pandas/tests/indexing/multiindex/test_multiindex.py

@@ -249,3 +249,20 @@ def test_groupyby_rename_categories_operation_with_multiindex(self, operation):
         expected = getattr(a, operation)(b.sort_index(ascending=False))
 
         tm.assert_series_equal(result, expected)
+
+    def test_multiindex_assign_aligns_as_implicit_tuple(self):
+        # GH 61841
+        cols = MultiIndex.from_tuples([("A", "B")])
+        df1 = DataFrame([[i] for i in range(3)], columns=cols)
+        df2 = DataFrame([[i] for i in range(3)], columns=cols)
+        df3 = DataFrame([[i] for i in range(3)], columns=cols)
+        s1 = df1["A"].rolling(2).mean()
+        s2 = df2["A"].rolling(2).mean()
+        s3 = df3["A"].rolling(2).mean()
+        df1["C"] = s1
+        df1["C"] = s1
+        df1["C"] = s1
+        df2["C"] = s2
+        df3[("C", "")] = s3
+        tm.assert_frame_equal(df1, df2)
+        tm.assert_frame_equal(df1, df3)