BUG: preserve object-dtype index when accessing DataFrame column / PERF: improve perf of Series fastpath constructor (#42950)

Author: jorisvandenbossche

asv_bench/benchmarks/series_methods.py

@@ -3,6 +3,7 @@
 import numpy as np
 
 from pandas import (
+    Index,
     NaT,
     Series,
     date_range,
@@ -12,20 +13,23 @@
 
 
 class SeriesConstructor:
-
-    params = [None, "dict"]
-    param_names = ["data"]
-
-    def setup(self, data):
+    def setup(self):
         self.idx = date_range(
             start=datetime(2015, 10, 26), end=datetime(2016, 1, 1), freq="50s"
         )
-        dict_data = dict(zip(self.idx, range(len(self.idx))))
-        self.data = None if data is None else dict_data
+        self.data = dict(zip(self.idx, range(len(self.idx))))
+        self.array = np.array([1, 2, 3])
+        self.idx2 = Index(["a", "b", "c"])
 
-    def time_constructor(self, data):
+    def time_constructor_dict(self):
         Series(data=self.data, index=self.idx)
 
+    def time_constructor_no_data(self):
+        Series(data=None, index=self.idx)
+
+    def time_constructor_fastpath(self):
+        Series(self.array, index=self.idx2, name="name", fastpath=True)
+
 
 class ToFrame:
     params = [["int64", "datetime64[ns]", "category", "Int64"], [None, "foo"]]

doc/source/whatsnew/v1.5.0.rst

@@ -263,6 +263,7 @@ Indexing
 - Bug when setting a value too large for a :class:`Series` dtype failing to coerce to a common type (:issue:`26049`, :issue:`32878`)
 - Bug in :meth:`Series.__setitem__` when setting ``boolean`` dtype values containing ``NA`` incorrectly raising instead of casting to ``boolean`` dtype (:issue:`45462`)
 - Bug in :meth:`Series.__setitem__` where setting :attr:`NA` into a numeric-dtpye :class:`Series` would incorrectly upcast to object-dtype rather than treating the value as ``np.nan`` (:issue:`44199`)
+- Bug in getting a column from a DataFrame with an object-dtype row index with datetime-like values: the resulting Series now preserves the exact object-dtype Index from the parent DataFrame (:issue:`42950`)
 -
 
 Missing

pandas/core/series.py

@@ -457,8 +457,12 @@ def __init__(
                     data = SingleArrayManager.from_array(data, index)
 
         NDFrame.__init__(self, data)
-        self.name = name
-        self._set_axis(0, index, fastpath=True)
+        if fastpath:
+            # skips validation of the name
+            object.__setattr__(self, "_name", name)
+        else:
+            self.name = name
+            self._set_axis(0, index)
 
     def _init_dict(
         self, data, index: Index | None = None, dtype: DtypeObj | None = None
@@ -539,15 +543,14 @@ def _constructor_expanddim(self) -> type[DataFrame]:
     def _can_hold_na(self) -> bool:
         return self._mgr._can_hold_na
 
-    def _set_axis(self, axis: int, labels, fastpath: bool = False) -> None:
+    def _set_axis(self, axis: int, labels) -> None:
         """
         Override generic, we want to set the _typ here.
 
         This is called from the cython code when we set the `index` attribute
         directly, e.g. `series.index = [1, 2, 3]`.
         """
-        if not fastpath:
-            labels = ensure_index(labels)
+        labels = ensure_index(labels)
 
         if labels._is_all_dates:
             deep_labels = labels
@@ -559,17 +562,13 @@ def _set_axis(self, axis: int, labels, fastpath: bool = False) -> None:
             ):
                 try:
                     labels = DatetimeIndex(labels)
-                    # need to set here because we changed the index
-                    if fastpath:
-                        self._mgr.set_axis(axis, labels)
                 except (tslibs.OutOfBoundsDatetime, ValueError):
                     # labels may exceeds datetime bounds,
                     # or not be a DatetimeIndex
                     pass
 
-        if not fastpath:
-            # The ensure_index call above ensures we have an Index object
-            self._mgr.set_axis(axis, labels)
+        # The ensure_index call above ensures we have an Index object
+        self._mgr.set_axis(axis, labels)
 
     # ndarray compatibility
     @property

pandas/tests/frame/indexing/test_indexing.py

@@ -1526,6 +1526,21 @@ def test_loc_iloc_setitem_non_categorical_rhs(
         with pytest.raises(TypeError, match=msg1):
             indexer(df)[key] = ["c", "c"]
 
+    @pytest.mark.parametrize("indexer", [tm.getitem, tm.loc, tm.iloc])
+    def test_getitem_preserve_object_index_with_dates(self, indexer):
+        # https://github.com/pandas-dev/pandas/pull/42950 - when selecting a column
+        # from dataframe, don't try to infer object dtype index on Series construction
+        idx = date_range("2012", periods=3).astype(object)
+        df = DataFrame({0: [1, 2, 3]}, index=idx)
+        assert df.index.dtype == object
+
+        if indexer is tm.getitem:
+            ser = indexer(df)[0]
+        else:
+            ser = indexer(df)[:, 0]
+
+        assert ser.index.dtype == object
+
 
 class TestDepreactedIndexers:
     @pytest.mark.parametrize(