Fix joining on indexes with duplicate level names

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

@@ -208,7 +208,7 @@ def _compute_join_keys(self):
                 left_keys.extend(
                     [
                         _Indexer(name=on, index=True)
-                        for on in self.lhs.index.names
+                        for on in self.lhs.index._data.names
                     ]
                 )
             if self.left_on:
@@ -223,7 +223,7 @@ def _compute_join_keys(self):
                 right_keys.extend(
                     [
                         _Indexer(name=on, index=True)
-                        for on in self.rhs.index.names
+                        for on in self.rhs.index._data.names
                     ]
                 )
             if self.right_on:

python/cudf/cudf/core/multiindex.py

@@ -1524,9 +1524,17 @@ def from_pandas(cls, multiindex, nan_as_null=None):
         if not isinstance(multiindex, pd.MultiIndex):
             raise TypeError("not a pandas.MultiIndex")
 
+        # if `multiindex` has two or more levels that
+        # have the same name, then `multiindex.to_frame()`
+        # results in a DataFrame containing only one of those
+        # levels. Thus, set `names` to some tuple of unique values
+        # and then call `multiindex.to_frame(name=names)`,
+        # which preserves all levels of `multiindex`.
+        names = tuple(range(len(multiindex.names)))
+
         mi = cls(
             names=multiindex.names,
-            source_data=multiindex.to_frame(),
+            source_data=multiindex.to_frame(name=names),
             nan_as_null=nan_as_null,
         )
 

python/cudf/cudf/tests/test_joining.py

@@ -2112,3 +2112,23 @@ def test_string_join_values_nulls():
     got = got.sort_values(by=["a", "b", "c"]).reset_index(drop=True)
 
     assert_join_results_equal(expect, got, how="left")
+
+
+def test_join_on_index_with_duplicate_names():
+    # although index levels with duplicate names are poorly supported
+    # overall, we *should* be able to join on them:
+    lhs = pd.DataFrame({"a": [1, 2, 3]})
+    rhs = pd.DataFrame({"b": [1, 2, 3]})
+    lhs.index = pd.MultiIndex.from_tuples(
+        [(1, 1), (1, 2), (2, 1)], names=["x", "x"]
+    )
+    rhs.index = pd.MultiIndex.from_tuples(
+        [(1, 1), (1, 3), (2, 1)], names=["x", "x"]
+    )
+    expect = lhs.join(rhs, how="inner")
+
+    lhs = cudf.from_pandas(lhs)
+    rhs = cudf.from_pandas(rhs)
+    got = lhs.join(rhs, how="inner")
+
+    assert_join_results_equal(expect, got, how="inner")