From 544f039ae23743169e45ee3c99886f26847698df Mon Sep 17 00:00:00 2001
From: Ashwin Srinath <3190405+shwina@users.noreply.github.com>
Date: Fri, 1 Jul 2022 13:36:54 -0400
Subject: [PATCH] Fix index alignment for Series objects with repeated index
 (#11103)

The logic for aligning the indexes of two Series objects with a repeated value of the index is broken. We previously introduced an optimization that avoids a join operation, but unfortunately, that doesn't work in every situation, as seen in https://github.com/rapidsai/cudf/issues/11094. This PR removes that special-casing.

Closes #11094

Authors:
  - Ashwin Srinath (https://github.com/shwina)

Approvers:
  - GALI PREM SAGAR (https://github.com/galipremsagar)

URL: https://github.com/rapidsai/cudf/pull/11103
---
 python/cudf/cudf/core/series.py       | 26 ++++++++++----------------
 python/cudf/cudf/tests/test_binops.py | 19 +++++++++++++++++++
 2 files changed, 29 insertions(+), 16 deletions(-)

diff --git a/python/cudf/cudf/core/series.py b/python/cudf/cudf/core/series.py
index 0cb63fb99a4..e9e1b175839 100644
--- a/python/cudf/cudf/core/series.py
+++ b/python/cudf/cudf/core/series.py
@@ -4712,22 +4712,16 @@ def _align_indices(series_list, how="outer", allow_non_unique=False):
     if all_index_equal:
         return series_list
 
-    if how == "outer":
-        combined_index = cudf.core.reshape.concat(
-            [sr.index for sr in series_list]
-        ).unique()
-        combined_index.names = new_index_names
-    else:
-        combined_index = series_list[0].index
-        for sr in series_list[1:]:
-            combined_index = (
-                cudf.DataFrame(index=sr.index).join(
-                    cudf.DataFrame(index=combined_index),
-                    sort=True,
-                    how="inner",
-                )
-            ).index
-        combined_index.names = new_index_names
+    combined_index = series_list[0].index
+    for sr in series_list[1:]:
+        combined_index = (
+            cudf.DataFrame(index=sr.index).join(
+                cudf.DataFrame(index=combined_index),
+                sort=True,
+                how=how,
+            )
+        ).index
+    combined_index.names = new_index_names
 
     # align all Series to the combined index
     result = [
diff --git a/python/cudf/cudf/tests/test_binops.py b/python/cudf/cudf/tests/test_binops.py
index 0d1bac6aead..2397dba7f76 100644
--- a/python/cudf/cudf/tests/test_binops.py
+++ b/python/cudf/cudf/tests/test_binops.py
@@ -593,6 +593,14 @@ def test_different_shapes_and_columns_with_unaligned_indices(binop):
     cd_frame["y"] = cd_frame["y"].astype(np.float64)
     utils.assert_eq(cd_frame, pd_frame)
 
+    pdf1 = pd.DataFrame({"x": [1, 1]}, index=["a", "a"])
+    pdf2 = pd.DataFrame({"x": [2]}, index=["a"])
+    gdf1 = cudf.DataFrame.from_pandas(pdf1)
+    gdf2 = cudf.DataFrame.from_pandas(pdf2)
+    pd_frame = binop(pdf1, pdf2)
+    cd_frame = binop(gdf1, gdf2)
+    utils.assert_eq(pd_frame, cd_frame)
+
 
 @pytest.mark.parametrize(
     "df2",
@@ -2962,3 +2970,14 @@ def test_binops_dot(df, other):
     got = df @ other
 
     utils.assert_eq(expected, got)
+
+
+def test_binop_series_with_repeated_index():
+    # GH: #11094
+    psr1 = pd.Series([1, 1], index=["a", "a"])
+    psr2 = pd.Series([1], index=["a"])
+    gsr1 = cudf.from_pandas(psr1)
+    gsr2 = cudf.from_pandas(psr2)
+    expected = psr1 - psr2
+    got = gsr1 - gsr2
+    utils.assert_eq(expected, got)