BUG: Copy attrs on pd.merge()

This uses the same logic as `pd.concat()`: Copy `attrs` only if all
input `attrs` are identical.

I've refactored the handling in __finalize__ from special-casing based on th the method name (previously only "concat") to handling "other" parameters
that have an `input_objs` attribute. This is a more scalable architecture compared to hard-coding method names in __finalize__.

Tests added for `concat()` and `merge()`.

Closes pandas-dev#60351.

Author: timhoffm

pandas/core/generic.py

@@ -6053,8 +6053,8 @@ def __finalize__(self, other, method: str | None = None, **kwargs) -> Self:
                 assert isinstance(name, str)
                 object.__setattr__(self, name, getattr(other, name, None))
 
-        if method == "concat":
-            objs = other.objs
+        elif hasattr(other, "input_objs"):
+            objs = other.input_objs
             # propagate attrs only if all concat arguments have the same attrs
             if all(bool(obj.attrs) for obj in objs):
                 # all concatenate arguments have non-empty attrs

pandas/core/reshape/concat.py

@@ -545,7 +545,7 @@ def _get_result(
             result = sample._constructor_from_mgr(mgr, axes=mgr.axes)
             result._name = name
             return result.__finalize__(
-                types.SimpleNamespace(objs=objs), method="concat"
+                types.SimpleNamespace(input_objs=objs), method="concat"
             )
 
         # combine as columns in a frame
@@ -566,7 +566,9 @@ def _get_result(
             )
             df = cons(data, index=index, copy=False)
             df.columns = columns
-            return df.__finalize__(types.SimpleNamespace(objs=objs), method="concat")
+            return df.__finalize__(
+                types.SimpleNamespace(input_objs=objs), method="concat"
+            )
 
     # combine block managers
     else:
@@ -605,7 +607,7 @@ def _get_result(
         )
 
         out = sample._constructor_from_mgr(new_data, axes=new_data.axes)
-        return out.__finalize__(types.SimpleNamespace(objs=objs), method="concat")
+        return out.__finalize__(types.SimpleNamespace(input_objs=objs), method="concat")
 
 
 def new_axes(

pandas/core/reshape/merge.py

@@ -10,6 +10,7 @@
 )
 import datetime
 from functools import partial
+import types
 from typing import (
     TYPE_CHECKING,
     Literal,
@@ -1115,7 +1116,9 @@ def get_result(self) -> DataFrame:
 
         self._maybe_restore_index_levels(result)
 
-        return result.__finalize__(self, method="merge")
+        return result.__finalize__(
+            types.SimpleNamespace(input_objs=[self.left, self.right]), method="merge"
+        )
 
     @final
     @cache_readonly

pandas/tests/frame/test_api.py

@@ -315,7 +315,7 @@ def test_attrs(self):
         result = df.rename(columns=str)
         assert result.attrs == {"version": 1}
 
-    def test_attrs_deepcopy(self):
+    def test_attrs_is_deepcopy(self):
         df = DataFrame({"A": [2, 3]})
         assert df.attrs == {}
         df.attrs["tags"] = {"spam", "ham"}
@@ -324,6 +324,30 @@ def test_attrs_deepcopy(self):
         assert result.attrs == df.attrs
         assert result.attrs["tags"] is not df.attrs["tags"]
 
+    def test_attrs_concat(self):
+        # concat propagates attrs if all input attrs are equal
+        df1 = DataFrame({"A": [2, 3]})
+        df1.attrs = {"a": 1, "b": 2}
+        df2 = DataFrame({"A": [4, 5]})
+        df2.attrs = df1.attrs.copy()
+        df3 = DataFrame({"A": [6, 7]})
+        df3.attrs = df1.attrs.copy()
+        assert pd.concat([df1, df2, df3]).attrs == df1.attrs
+        # concat does not propagate attrs if input attrs are different
+        df2.attrs = {"c": 3}
+        assert pd.concat([df1, df2, df3]).attrs == {}
+
+    def test_attrs_merge(self):
+        # merge propagates attrs if all input attrs are equal
+        df1 = DataFrame({"key": ["a", "b"], "val1": [1, 2]})
+        df1.attrs = {"a": 1, "b": 2}
+        df2 = DataFrame({"key": ["a", "b"], "val2": [3, 4]})
+        df2.attrs = df1.attrs.copy()
+        assert pd.merge(df1, df2).attrs == df1.attrs
+        # merge does not propagate attrs if input attrs are different
+        df2.attrs = {"c": 3}
+        assert pd.merge(df1, df2).attrs == {}
+
     @pytest.mark.parametrize("allows_duplicate_labels", [True, False, None])
     def test_set_flags(
         self,