fix: compare UDF transformations by udf_string when present

Source-first from_proto rebuilds callables whose bytecode differs from the live repo function; requiring co_code equality made no-op feast apply rewrite ODFVs and broke universal CLI integration tests. Signed-off-by: Aniket Paluskar <apaluska@redhat.com> Co-authored-by: Cursor <cursoragent@cursor.com>
ntkathole · Aug 17, 2026 · Jul 29, 2026 · Jul 29, 2026 · Jul 30, 2026 · Jul 30, 2026
commit 8485ae7a9d9a1bc238e267ee9bf6d9ffbafbaf85
@@ -133,13 +133,16 @@ def __eq__(self, other):
         if not isinstance(other, PandasTransformation):
             return False
 
-        if (
-            self.udf_string != other.udf_string
-            or self.udf.__code__.co_code != other.udf.__code__.co_code
-        ):
-            return False
-
-        return True
+        # udf_string is the canonical diff identity. Source-first from_proto
+        # rebuilds a new callable (strip+exec) whose bytecode differs from the
+        # live repo function even when the source is unchanged — do not require
+        # co_code equality when both sides have source text.
+        left = self.udf_string or ""
+        right = other.udf_string or ""
+        if left and right:
+            return left == right
+
+        return self.udf.__code__.co_code == other.udf.__code__.co_code
 
     @classmethod
     def from_proto(cls, user_defined_function_proto: UserDefinedFunctionProto):

@@ -144,13 +144,16 @@ def __eq__(self, other):
         if not isinstance(other, PythonTransformation):
             return False
 
-        if (
-            self.udf_string != other.udf_string
-            or self.udf.__code__.co_code != other.udf.__code__.co_code
-        ):
-            return False
-
-        return True
+        # udf_string is the canonical diff identity. Source-first from_proto
+        # rebuilds a new callable (strip+exec) whose bytecode differs from the
+        # live repo function even when the source is unchanged — do not require
+        # co_code equality when both sides have source text.
+        left = self.udf_string or ""
+        right = other.udf_string or ""
+        if left and right:
+            return left == right
+
+        return self.udf.__code__.co_code == other.udf.__code__.co_code
 
     def __reduce__(self):
         """Support for pickle/dill serialization."""

@@ -1,6 +1,8 @@
 import pandas as pd
 
 from feast.transformation.pandas_transformation import PandasTransformation
+from feast.transformation.python_transformation import PythonTransformation
+from feast.transformation.udf_rehydrate import resolve_udf
 
 
 def pandas_udf(features_df: pd.DataFrame) -> pd.DataFrame:
@@ -16,3 +18,32 @@ def test_init_pandas_transformation():
     transformed_df = transformation.transform(features_df)
     assert transformed_df["output1"].values[0] == 1
     assert transformed_df["output2"].values[1] == 3
+
+
+_UDF_SOURCE = '''def pandas_udf(features_df):
+    df = __import__("pandas").DataFrame()
+    df["output1"] = features_df["feature1"]
+    df["output2"] = features_df["feature2"]
+    return df
+'''
+
+
+def test_pandas_transformation_eq_uses_udf_string_not_bytecode():
+    """Repo vs source-rehydrated callables must compare equal when source matches."""
+    rehydrated = resolve_udf(udf_string=_UDF_SOURCE, preferred_name="pandas_udf")
+    assert pandas_udf.__code__.co_code != rehydrated.__code__.co_code
+
+    left = PandasTransformation(udf=pandas_udf, udf_string=_UDF_SOURCE)
+    right = PandasTransformation(udf=rehydrated, udf_string=_UDF_SOURCE)
+    assert left == right
+
+    different = PandasTransformation(udf=rehydrated, udf_string=_UDF_SOURCE + "\n")
+    assert left != different
+
+
+def test_python_transformation_eq_uses_udf_string_not_bytecode():
+    rehydrated = resolve_udf(udf_string=_UDF_SOURCE, preferred_name="pandas_udf")
+    left = PythonTransformation(udf=pandas_udf, udf_string=_UDF_SOURCE)
+    right = PythonTransformation(udf=rehydrated, udf_string=_UDF_SOURCE)
+    assert left == right
+    assert left != PythonTransformation(udf=rehydrated, udf_string="other")