Allow merging index column with data column using keyword "on"

python/cudf/cudf/core/join/join.py

@@ -196,14 +196,14 @@ def perform_merge(self) -> Frame:
 
     def _compute_join_keys(self):
         # Computes self._keys
+        left_keys = []
+        right_keys = []
         if (
             self.left_index
             or self.right_index
             or self.left_on
             or self.right_on
         ):
-            left_keys = []
-            right_keys = []
             if self.left_index:
                 left_keys.extend(
                     [
@@ -234,14 +234,35 @@ def _compute_join_keys(self):
                         for on in _coerce_to_tuple(self.right_on)
                     ]
                 )
+        elif self.on:
+            on_names = _coerce_to_tuple(self.on)
+            for on in on_names:
+                # If `on` is provided, checks whether merging on
+                # columns, indexes or merging  index with column
+                if on in self.lhs._data and on not in self.rhs._data:
+                    # case1: merge on lhs column with rhs index
+                    left_keys.append(_Indexer(name=on, column=True))
+                    right_keys.append(_Indexer(name=on, index=True))
+
+                elif on not in self.lhs._data and on in self.rhs._data:
+                    # case2: merge on rhs column with lhs index
+                    left_keys.append(_Indexer(name=on, index=True))
+                    right_keys.append(_Indexer(name=on, column=True))
+
+                elif on not in self.lhs._data and on not in self.rhs._data:
+                    # case3: merge on lhs index with rhs index
+                    left_keys.append(_Indexer(name=on, index=True))
+                    right_keys.append(_Indexer(name=on, index=True))
+
+                else:
+                    # case4: merge on lhs column with rhs column
+                    left_keys.append(_Indexer(name=on, column=True))
+                    right_keys.append(_Indexer(name=on, column=True))
+
         else:
-            # Use `on` if provided. Otherwise,
-            # implicitly use identically named columns as the key columns:
-            on_names = (
-                _coerce_to_tuple(self.on)
-                if self.on is not None
-                else set(self.lhs._data) & set(self.rhs._data)
-            )
+            # if `on` not provided and not merging index with column or on
+            # both indexes, then use intersection  of columns in both frames
+            on_names = set(self.lhs._data) & set(self.rhs._data)
             left_keys = [_Indexer(name=on, column=True) for on in on_names]
             right_keys = [_Indexer(name=on, column=True) for on in on_names]
 
@@ -384,12 +405,16 @@ def _validate_merge_params(
         if how not in {"left", "inner", "outer", "leftanti", "leftsemi"}:
             raise NotImplementedError(f"{how} merge not supported yet")
 
-        # Passing 'on' with 'left_on' or 'right_on' is ambiguous
-        if on and (left_on or right_on):
-            raise ValueError(
-                'Can only pass argument "on" OR "left_on" '
-                'and "right_on", not a combination of both.'
-            )
+        if on:
+            if left_on or right_on:
+                # Passing 'on' with 'left_on' or 'right_on' is ambiguous
+                raise ValueError(
+                    'Can only pass argument "on" OR "left_on" '
+                    'and "right_on", not a combination of both.'
+                )
+            else:
+                # the validity of 'on' being checked by _Indexer
+                return
 
         # Can't merge on unnamed Series
         if (isinstance(lhs, cudf.Series) and not lhs.name) or (

python/cudf/cudf/tests/test_joining.py

@@ -1738,3 +1738,58 @@ def test_join_renamed_index():
     )
     got = df.merge(df, left_index=True, right_index=True, how="inner")
     assert_join_results_equal(expect, got, how="inner")
+
+
+@pytest.mark.parametrize(
+    "lhs_col, lhs_idx, rhs_col, rhs_idx, on",
+    [
+        (["A", "B"], "L0", ["B", "C"], "L0", ["B"]),
+        (["A", "B"], "L0", ["B", "C"], "L0", ["L0"]),
+        (["A", "B"], "L0", ["B", "C"], "L0", ["B", "L0"]),
+        (["A", "B"], "L0", ["C", "L0"], "A", ["A"]),
+        (["A", "B"], "L0", ["C", "L0"], "A", ["L0"]),
+        (["A", "B"], "L0", ["C", "L0"], "A", ["A", "L0"]),
+    ],
+)
+@pytest.mark.parametrize(
+    "how", ["left", "inner", "right", "outer", "leftanti", "leftsemi"]
+)
+def test_join_merge_with_on(lhs_col, lhs_idx, rhs_col, rhs_idx, on, how):
+    lhs_data = {col_name: [] for col_name in lhs_col}
+    lhs_index = cudf.Index([], name=lhs_idx)
+
+    rhs_data = {col_name: [] for col_name in rhs_col}
+    rhs_index = cudf.Index([], name=rhs_idx)
+
+    gd_left = cudf.DataFrame(lhs_data, lhs_index)
+    gd_right = cudf.DataFrame(rhs_data, rhs_index)
+    pd_left = gd_left.to_pandas()
+    pd_right = gd_right.to_pandas()
+
+    expect = pd_left.merge(pd_right, on=on)
+    got = gd_left.merge(gd_right, on=on)
+
+    assert_join_results_equal(expect, got, how=how)
+
+
+@pytest.mark.parametrize(
+    "on", ["A", "L0"],
+)
+@pytest.mark.parametrize(
+    "how", ["left", "inner", "right", "outer", "leftanti", "leftsemi"]
+)
+def test_join_merge_invalid_keys(on, how):
+    with pytest.raises(KeyError):
+        gd_left = cudf.DataFrame(
+            {"A": [], "B": []}, index=cudf.Index([], name="C")
+        )
+        gd_right = cudf.DataFrame(
+            {"D": [], "E": []}, index=cudf.Index([], name="F")
+        )
+        pd_left = gd_left.to_pandas()
+        pd_right = gd_right.to_pandas()
+
+        expect = pd_left.merge(pd_right, on=on)
+        got = gd_left.merge(gd_right, on=on)
+
+        assert_join_results_equal(expect, got, how=how)