BUG: incompatible dtype when creating string column with loc (#55249)

dont raise unnecessary warning

doc/source/whatsnew/v2.1.2.rst

@@ -13,7 +13,7 @@ including other versions of pandas.
 
 Fixed regressions
 ~~~~~~~~~~~~~~~~~
--
+- Fixed bug where PDEP-6 warning about setting an item of an incompatible dtype was being shown when creating a new conditional column (:issue:`55025`)
 -
 
 .. ---------------------------------------------------------------------------

pandas/core/dtypes/missing.py

@@ -624,6 +624,9 @@ def infer_fill_value(val):
             return np.array("NaT", dtype=DT64NS_DTYPE)
         elif dtype in ["timedelta", "timedelta64"]:
             return np.array("NaT", dtype=TD64NS_DTYPE)
+        return np.array(np.nan, dtype=object)
+    elif val.dtype.kind == "U":
+        return np.array(np.nan, dtype=val.dtype)
     return np.nan
 
 

pandas/tests/frame/indexing/test_indexing.py

@@ -1890,6 +1890,21 @@ def test_setitem_dict_and_set_disallowed_multiindex(self, key):
             df.loc[key] = 1
 
 
+def test_adding_new_conditional_column() -> None:
+    # https://github.com/pandas-dev/pandas/issues/55025
+    df = DataFrame({"x": [1]})
+    df.loc[df["x"] == 1, "y"] = "1"
+    expected = DataFrame({"x": [1], "y": ["1"]})
+    tm.assert_frame_equal(df, expected)
+
+    df = DataFrame({"x": [1]})
+    # try inserting something which numpy would store as 'object'
+    value = lambda x: x
+    df.loc[df["x"] == 1, "y"] = value
+    expected = DataFrame({"x": [1], "y": [value]})
+    tm.assert_frame_equal(df, expected)
+
+
 class TestSetitemValidation:
     # This is adapted from pandas/tests/arrays/masked/test_indexing.py
     # but checks for warnings instead of errors.

pandas/tests/frame/indexing/test_set_value.py

@@ -26,10 +26,7 @@ def test_set_value_resize(self, float_frame):
         assert float_frame._get_value("foobar", "qux") == 0
 
         res = float_frame.copy()
-        with tm.assert_produces_warning(
-            FutureWarning, match="Setting an item of incompatible dtype"
-        ):
-            res._set_value("foobar", "baz", "sam")
+        res._set_value("foobar", "baz", "sam")
         assert res["baz"].dtype == np.object_
 
         res = float_frame.copy()