BUG: merge between partial index and index fails when result is empty

doc/source/whatsnew/v1.1.0.rst

@@ -936,6 +936,7 @@ Reshaping
 - Bug in :meth:`DataFrame.replace` casts columns to ``object`` dtype if items in ``to_replace`` not in values (:issue:`32988`)
 - Ensure only named functions can be used in :func:`eval()` (:issue:`32460`)
 - Fixed bug in :func:`melt` where melting MultiIndex columns with ``col_level`` > 0 would raise a ``KeyError`` on ``id_vars`` (:issue:`34129`)
+- Fixed bug in :func:`merge` where an error was raised when performing an inner join with partial index and ``right_index`` when no overlap between indices (:issue:`33814`)
 
 Sparse
 ^^^^^^

pandas/core/reshape/merge.py

@@ -837,7 +837,7 @@ def _maybe_add_join_keys(self, result, left_indexer, right_indexer):
                 # make sure to just use the right values
                 mask = left_indexer == -1
                 if mask.all():
-                    key_col = rvals
+                    key_col = Index(rvals)
                 else:
                     key_col = Index(lvals).where(~mask, rvals)
 

pandas/tests/reshape/merge/test_merge.py

@@ -2227,3 +2227,21 @@ def test_categorical_non_unique_monotonic(n_categories):
         index=left_index,
     )
     tm.assert_frame_equal(expected, result)
+
+
+@pytest.mark.parametrize(
+    ("kwargs", "args"),
+    [
+        ({"left_on": ["b"], "right_index": True}, 1),
+        ({"left_index": True, "right_on": ["b"]}, -1),
+    ],
+)
+def test_merge_empty_right_index_left_on(kwargs, args):
+    # GH 33814
+    df1 = pd.DataFrame({"a": [1], "b": [2]}).set_index(["a", "b"])
+    df2 = pd.DataFrame({"b": [1]}).set_index(["b"])
+    result = pd.merge(*[df1, df2][::args], **kwargs)
+    expected = pd.DataFrame(
+        index=MultiIndex(levels=[[int()], [int()]], codes=[[], []], names=["a", "b"])
+    )
+    tm.assert_frame_equal(result, expected)