BUG: loc casting to object for multi block case when setting with list indexer (#49161)

* BUG: loc casting to object for multi block case when setting with list indexer

* Remove xpasses

* Fix windows

* Adress comments

* Remove condition

Author: phofl

doc/source/whatsnew/v2.0.0.rst

@@ -234,6 +234,7 @@ Interval
 Indexing
 ^^^^^^^^
 - Bug in :meth:`DataFrame.reindex` filling with wrong values when indexing columns and index for ``uint`` dtypes (:issue:`48184`)
+- Bug in :meth:`DataFrame.loc` coercing dtypes when setting values with a list indexer (:issue:`49159`)
 - Bug in :meth:`DataFrame.__setitem__` raising ``ValueError`` when right hand side is :class:`DataFrame` with :class:`MultiIndex` columns (:issue:`49121`)
 - Bug in :meth:`DataFrame.reindex` casting dtype to ``object`` when :class:`DataFrame` has single extension array column when re-indexing ``columns`` and ``index`` (:issue:`48190`)
 - Bug in :func:`~DataFrame.describe` when formatting percentiles in the resulting index showed more decimals than needed (:issue:`46362`)

pandas/core/indexing.py

@@ -1898,16 +1898,20 @@ def _setitem_with_indexer_2d_value(self, indexer, value):
 
         ilocs = self._ensure_iterable_column_indexer(indexer[1])
 
-        # GH#7551 Note that this coerces the dtype if we are mixed
-        value = np.array(value, dtype=object)
+        if not is_array_like(value):
+            # cast lists to array
+            value = np.array(value, dtype=object)
         if len(ilocs) != value.shape[1]:
             raise ValueError(
                 "Must have equal len keys and value when setting with an ndarray"
             )
 
         for i, loc in enumerate(ilocs):
-            # setting with a list, re-coerces
-            self._setitem_single_column(loc, value[:, i].tolist(), pi)
+            value_col = value[:, i]
+            if is_object_dtype(value_col.dtype):
+                # casting to list so that we do type inference in setitem_single_column
+                value_col = value_col.tolist()
+            self._setitem_single_column(loc, value_col, pi)
 
     def _setitem_with_indexer_frame_value(self, indexer, value: DataFrame, name: str):
         ilocs = self._ensure_iterable_column_indexer(indexer[1])

pandas/tests/frame/indexing/test_coercion.py

@@ -20,7 +20,6 @@
 
 
 class TestDataFrameSetitemCoercion:
-    @pytest.mark.xfail(reason="Unnecessary cast.")
     @pytest.mark.parametrize("consolidate", [True, False])
     def test_loc_setitem_multiindex_columns(self, consolidate):
         # GH#18415 Setting values in a single column preserves dtype,
@@ -36,9 +35,7 @@ def test_loc_setitem_multiindex_columns(self, consolidate):
         A.loc[2:3, (1, slice(2, 3))] = np.ones((2, 2), dtype=np.float32)
         assert (A.dtypes == np.float32).all()
 
-        msg = "will attempt to set the values inplace instead"
-        with tm.assert_produces_warning(FutureWarning, match=msg):
-            A.loc[0:5, (1, slice(2, 3))] = np.ones((6, 2), dtype=np.float32)
+        A.loc[0:5, (1, slice(2, 3))] = np.ones((6, 2), dtype=np.float32)
 
         assert (A.dtypes == np.float32).all()
 
@@ -119,7 +116,6 @@ def test_15231():
     tm.assert_series_equal(df.dtypes, exp_dtypes)
 
 
-@pytest.mark.xfail(reason="Unnecessarily upcasts to float64")
 def test_iloc_setitem_unnecesssary_float_upcasting():
     # GH#12255
     df = DataFrame(
@@ -132,10 +128,7 @@ def test_iloc_setitem_unnecesssary_float_upcasting():
     orig = df.copy()
 
     values = df[0].values.reshape(2, 1)
-
-    msg = "will attempt to set the values inplace instead"
-    with tm.assert_produces_warning(FutureWarning, match=msg):
-        df.iloc[:, 0:1] = values
+    df.iloc[:, 0:1] = values
 
     tm.assert_frame_equal(df, orig)
 

pandas/tests/frame/indexing/test_indexing.py

@@ -1405,6 +1405,19 @@ def test_loc_named_tuple_for_midx(self):
         )
         tm.assert_frame_equal(result, expected)
 
+    @pytest.mark.parametrize("indexer", [["a"], "a"])
+    @pytest.mark.parametrize("col", [{}, {"b": 1}])
+    def test_set_2d_casting_date_to_int(self, col, indexer):
+        # GH#49159
+        df = DataFrame(
+            {"a": [Timestamp("2022-12-29"), Timestamp("2022-12-30")], **col},
+        )
+        df.loc[[1], indexer] = df["a"] + pd.Timedelta(days=1)
+        expected = DataFrame(
+            {"a": [Timestamp("2022-12-29"), Timestamp("2022-12-31")], **col},
+        )
+        tm.assert_frame_equal(df, expected)
+
     @pytest.mark.parametrize("col", [{}, {"name": "a"}])
     def test_loc_setitem_reordering_with_all_true_indexer(self, col):
         # GH#48701

pandas/tests/indexing/test_indexing.py

@@ -478,7 +478,7 @@ def test_multi_assign(self):
             {
                 "FC": ["a", np.nan, "a", "b", "a", "b"],
                 "PF": [0, 0, 0, 0, 1, 1],
-                "col1": [0.0, 1.0, 4.0, 6.0, 8.0, 10.0],
+                "col1": [0, 1, 4, 6, 8, 10],
                 "col2": [12, 7, 16, np.nan, 20, 22],
             }
         )

pandas/tests/indexing/test_loc.py

@@ -1500,12 +1500,12 @@ def test_loc_setitem_unsorted_multiindex_columns(self, key):
         mi = MultiIndex.from_tuples([("A", 4), ("B", "3"), ("A", "2")])
         df = DataFrame([[1, 2, 3], [4, 5, 6]], columns=mi)
         obj = df.copy()
-        obj.loc[:, key] = np.zeros((2, 2), dtype=int)
+        obj.loc[:, key] = np.zeros((2, 2), dtype="int64")
         expected = DataFrame([[0, 2, 0], [0, 5, 0]], columns=mi)
         tm.assert_frame_equal(obj, expected)
 
         df = df.sort_index(axis=1)
-        df.loc[:, key] = np.zeros((2, 2), dtype=int)
+        df.loc[:, key] = np.zeros((2, 2), dtype="int64")
         expected = expected.sort_index(axis=1)
         tm.assert_frame_equal(df, expected)