Fix loc-getitem ordering when index contains duplicate labels

If the index has duplicate labels then it is not sufficient to sort
with the ordering induced by gathering the requested keys since the
duplicate keys arrive in non-deterministic order from the join in the
table that is being looked up from. To fix this, sort additionally by
a secondary key that is the ordering induced by gathering from the
donor table.

- Closes rapidsai#13658

python/cudf/cudf/core/dataframe.py

@@ -291,12 +291,14 @@ def _getitem_tuple_arg(self, arg):
                         {tmp_col_name: column.arange(len(tmp_arg[0]))},
                         index=as_index(tmp_arg[0]),
                     )
+                    cantor_name = "_" + "_".join(columns_df.columns)
+                    columns_df[cantor_name] = column.arange(len(columns_df))
                     df = other_df.join(columns_df, how="inner")
                     # as join is not assigning any names to index,
                     # update it over here
                     df.index.name = columns_df.index.name
-                    df = df.sort_values(tmp_col_name)
-                    df.drop(columns=[tmp_col_name], inplace=True)
+                    df = df.sort_values(by=[tmp_col_name, cantor_name])
+                    df.drop(columns=[tmp_col_name, cantor_name], inplace=True)
                     # There were no indices found
                     if len(df) == 0:
                         raise KeyError(arg)

python/cudf/cudf/core/indexed_frame.py

@@ -190,7 +190,7 @@ def _indices_from_labels(obj, labels):
     rhs = cudf.DataFrame(
         {"_": cudf.core.column.arange(len(obj))}, index=obj.index
     )
-    return lhs.join(rhs).sort_values("__")["_"]
+    return lhs.join(rhs).sort_values(by=["__", "_"])["_"]
 
 
 def _get_label_range_or_mask(index, start, stop, step):

python/cudf/cudf/tests/test_indexing.py

@@ -1996,6 +1996,20 @@ def test_loc_missing_label_keyerror_issue_13379(index):
         cdf.loc[[0, 5]]
 
 
+@pytest.mark.parametrize("series", [True, False], ids=["Series", "DataFrame"])
+def test_loc_repeated_label_ordering_issue_13658(series):
+    # https://github.com/rapidsai/cudf/issues/13658
+    values = range(2048)
+    index = [1 for _ in values]
+    if series:
+        frame = cudf.Series(values, index=index)
+    else:
+        frame = cudf.DataFrame({"a": values}, index=index)
+    expect = frame.to_pandas().loc[[1]]
+    actual = frame.loc[[1]]
+    assert_eq(actual, expect)
+
+
 class TestLocIndexWithOrder:
     # https://github.com/rapidsai/cudf/issues/12833
     @pytest.fixture(params=["increasing", "decreasing", "neither"])