From fc4b3d3ecbf95ee9afdcd509554bbeb5367a3059 Mon Sep 17 00:00:00 2001
From: Matthew Roeschke <10647082+mroeschke@users.noreply.github.com>
Date: Tue, 18 Jun 2024 09:02:05 -1000
Subject: [PATCH] Reduce deep copies in Index ops (#16054)

1. Changed `Index.rename(inplace=False)` to shallow copy which matches pandas behavior. Let me know if there's a reason why we should deep copy here.
2. Made `RangeIndex.unique` return a shallow copy like pandas.
3. Made `Index.dropna` with no NA's shallow copy like pandas.

Authors:
  - Matthew Roeschke (https://github.com/mroeschke)

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

URL: https://github.com/rapidsai/cudf/pull/16054
---
 python/cudf/cudf/core/_base_index.py |  6 +++---
 python/cudf/cudf/core/index.py       |  5 +++--
 python/cudf/cudf/tests/test_index.py | 25 +++++++++++++++++++++++--
 3 files changed, 29 insertions(+), 7 deletions(-)

diff --git a/python/cudf/cudf/core/_base_index.py b/python/cudf/cudf/core/_base_index.py
index ad73cd57f7d..caf07b286cd 100644
--- a/python/cudf/cudf/core/_base_index.py
+++ b/python/cudf/cudf/core/_base_index.py
@@ -1120,7 +1120,7 @@ def difference(self, other, sort=None):
         res_name = _get_result_name(self.name, other.name)
 
         if is_mixed_with_object_dtype(self, other) or len(other) == 0:
-            difference = self.copy().unique()
+            difference = self.unique()
             difference.name = res_name
             if sort is True:
                 return difference.sort_values()
@@ -1744,7 +1744,7 @@ def rename(self, name, inplace=False):
             self.name = name
             return None
         else:
-            out = self.copy(deep=True)
+            out = self.copy(deep=False)
             out.name = name
             return out
 
@@ -2068,7 +2068,7 @@ def dropna(self, how="any"):
             raise ValueError(f"{how=} must be 'any' or 'all'")
         try:
             if not self.hasnans:
-                return self.copy()
+                return self.copy(deep=False)
         except NotImplementedError:
             pass
         # This is to be consistent with IndexedFrame.dropna to handle nans
diff --git a/python/cudf/cudf/core/index.py b/python/cudf/cudf/core/index.py
index 1c5d05d2d87..71658695b80 100644
--- a/python/cudf/cudf/core/index.py
+++ b/python/cudf/cudf/core/index.py
@@ -528,7 +528,7 @@ def memory_usage(self, deep: bool = False) -> int:
 
     def unique(self) -> Self:
         # RangeIndex always has unique values
-        return self
+        return self.copy()
 
     @_cudf_nvtx_annotate
     def __mul__(self, other):
@@ -3197,7 +3197,8 @@ def _get_nearest_indexer(
     )
     right_indexer = _get_indexer_basic(
         index=index,
-        positions=positions.copy(deep=True),
+        # positions no longer used so don't copy
+        positions=positions,
         method="backfill",
         target_col=target_col,
         tolerance=tolerance,
diff --git a/python/cudf/cudf/tests/test_index.py b/python/cudf/cudf/tests/test_index.py
index 3d6c71ebc1b..a59836df5ba 100644
--- a/python/cudf/cudf/tests/test_index.py
+++ b/python/cudf/cudf/tests/test_index.py
@@ -252,10 +252,10 @@ def test_index_rename_inplace():
     pds = pd.Index([1, 2, 3], name="asdf")
     gds = Index(pds)
 
-    # inplace=False should yield a deep copy
+    # inplace=False should yield a shallow copy
     gds_renamed_deep = gds.rename("new_name", inplace=False)
 
-    assert gds_renamed_deep._values.data_ptr != gds._values.data_ptr
+    assert gds_renamed_deep._values.data_ptr == gds._values.data_ptr
 
     # inplace=True returns none
     expected_ptr = gds._values.data_ptr
@@ -3214,6 +3214,27 @@ def test_rangeindex_dropna():
     assert_eq(result, expected)
 
 
+def test_rangeindex_unique_shallow_copy():
+    ri_pandas = pd.RangeIndex(1)
+    result = ri_pandas.unique()
+    assert result is not ri_pandas
+
+    ri_cudf = cudf.RangeIndex(1)
+    result = ri_cudf.unique()
+    assert result is not ri_cudf
+    assert_eq(result, ri_cudf)
+
+
+def test_rename_shallow_copy():
+    idx = pd.Index([1])
+    result = idx.rename("a")
+    assert idx.to_numpy(copy=False) is result.to_numpy(copy=False)
+
+    idx = cudf.Index([1])
+    result = idx.rename("a")
+    assert idx._column is result._column
+
+
 @pytest.mark.parametrize("data", [range(2), [10, 11, 12]])
 def test_index_contains_hashable(data):
     gidx = cudf.Index(data)