Deprecate replace with categorical columns (#14988)

Matches pandas 2.2 behavior: pandas-dev/pandas#56385

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

Approvers:
  - Michael Wang (https://github.com/isVoid)

URL: #14988

python/cudf/cudf/core/column/categorical.py

@@ -2,6 +2,7 @@
 
 from __future__ import annotations
 
+import warnings
 from collections import abc
 from functools import cached_property
 from typing import TYPE_CHECKING, Any, Mapping, Optional, Sequence, Tuple, cast
@@ -990,14 +991,24 @@ def find_and_replace(
             replaced, to_replace_col, replacement_col
         )
 
-        return column.build_categorical_column(
+        result = column.build_categorical_column(
             categories=new_cats["cats"],
             codes=column.build_column(output.base_data, dtype=output.dtype),
             mask=output.base_mask,
             offset=output.offset,
             size=output.size,
             ordered=self.dtype.ordered,
         )
+        if result.dtype != self.dtype:
+            warnings.warn(
+                "The behavior of replace with "
+                "CategoricalDtype is deprecated. In a future version, replace "
+                "will only be used for cases that preserve the categories. "
+                "To change the categories, use ser.cat.rename_categories "
+                "instead.",
+                FutureWarning,
+            )
+        return result
 
     def isnull(self) -> ColumnBase:
         """

python/cudf/cudf/tests/test_replace.py

@@ -57,13 +57,24 @@ def test_series_replace_all(gsr, to_replace, value):
     else:
         pd_value = value
 
-    actual = gsr.replace(to_replace=gd_to_replace, value=gd_value)
-    if pd_value is None:
-        # TODO: Remove this workaround once cudf
-        # introduces `no_default` values
-        expected = psr.replace(to_replace=pd_to_replace)
-    else:
-        expected = psr.replace(to_replace=pd_to_replace, value=pd_value)
+    with expect_warning_if(
+        isinstance(gsr.dtype, cudf.CategoricalDtype)
+        and isinstance(gd_to_replace, str)
+        and gd_to_replace == "one"
+    ):
+        actual = gsr.replace(to_replace=gd_to_replace, value=gd_value)
+    with expect_warning_if(
+        PANDAS_GE_220
+        and isinstance(gsr.dtype, cudf.CategoricalDtype)
+        and isinstance(gd_to_replace, str)
+        and gd_to_replace == "one"
+    ):
+        if pd_value is None:
+            # TODO: Remove this workaround once cudf
+            # introduces `no_default` values
+            expected = psr.replace(to_replace=pd_to_replace)
+        else:
+            expected = psr.replace(to_replace=pd_to_replace, value=pd_value)
 
     assert_eq(
         expected.sort_values().reset_index(drop=True),
@@ -82,16 +93,19 @@ def test_series_replace():
 
     # Categorical
     psr3 = pd.Series(["one", "two", "three"], dtype="category")
-    psr4 = psr3.replace("one", "two")
+    with expect_warning_if(PANDAS_GE_220):
+        psr4 = psr3.replace("one", "two")
     sr3 = cudf.from_pandas(psr3)
-    sr4 = sr3.replace("one", "two")
+    with pytest.warns(FutureWarning):
+        sr4 = sr3.replace("one", "two")
     assert_eq(
         psr4.sort_values().reset_index(drop=True),
         sr4.sort_values().reset_index(drop=True),
     )
-
-    psr5 = psr3.replace("one", "five")
-    sr5 = sr3.replace("one", "five")
+    with expect_warning_if(PANDAS_GE_220):
+        psr5 = psr3.replace("one", "five")
+    with pytest.warns(FutureWarning):
+        sr5 = sr3.replace("one", "five")
 
     assert_eq(psr5, sr5)
 
@@ -236,11 +250,26 @@ def test_dataframe_replace(df, to_replace, value):
     else:
         gd_to_replace = to_replace
 
-    if pd_value is None:
-        expected = pdf.replace(to_replace=pd_to_replace)
-    else:
-        expected = pdf.replace(to_replace=pd_to_replace, value=pd_value)
-    actual = gdf.replace(to_replace=gd_to_replace, value=gd_value)
+    with expect_warning_if(
+        PANDAS_GE_220
+        and isinstance(df["a"].dtype, cudf.CategoricalDtype)
+        and isinstance(to_replace, str)
+        and to_replace == "two"
+        and isinstance(value, str)
+        and value == "three"
+    ):
+        if pd_value is None:
+            expected = pdf.replace(to_replace=pd_to_replace)
+        else:
+            expected = pdf.replace(to_replace=pd_to_replace, value=pd_value)
+    with expect_warning_if(
+        isinstance(df["a"].dtype, cudf.CategoricalDtype)
+        and isinstance(to_replace, str)
+        and to_replace == "two"
+        and isinstance(value, str)
+        and value == "three"
+    ):
+        actual = gdf.replace(to_replace=gd_to_replace, value=gd_value)
 
     expected_sorted = expected.sort_values(by=list(expected.columns), axis=0)
     actual_sorted = actual.sort_values(by=list(actual.columns), axis=0)
@@ -1342,7 +1371,8 @@ def test_series_replace_errors():
     ],
 )
 def test_replace_nulls(gsr, old, new, expected):
-    actual = gsr.replace(old, new)
+    with expect_warning_if(isinstance(gsr.dtype, cudf.CategoricalDtype)):
+        actual = gsr.replace(old, new)
     assert_eq(
         expected.sort_values().reset_index(drop=True),
         actual.sort_values().reset_index(drop=True),