BUG: make tz_localize operate on values rather than categories (#28300)

Author: MarcoGorelli

doc/source/whatsnew/v1.0.0.rst

@@ -288,6 +288,9 @@ Categorical
 - :meth:`Categorical.searchsorted` and :meth:`CategoricalIndex.searchsorted` now work on unordered categoricals also (:issue:`21667`)
 - Added test to assert roundtripping to parquet with :func:`DataFrame.to_parquet` or :func:`read_parquet` will preserve Categorical dtypes for string types (:issue:`27955`)
 - Changed the error message in :meth:`Categorical.remove_categories` to always show the invalid removals as a set (:issue:`28669`)
+- Using date accessors on a categorical dtyped :class:`Series` of datetimes was not returning an object of the
+  same type as if one used the :meth:`.str.` / :meth:`.dt.` on a :class:`Series` of that type. E.g. when accessing :meth:`Series.dt.tz_localize` on a
+  :class:`Categorical` with duplicate entries, the accessor was skipping duplicates (:issue: `27952`)
 
 
 Datetimelike

pandas/core/indexes/accessors.py

@@ -16,7 +16,6 @@
 from pandas.core.dtypes.generic import ABCSeries
 
 from pandas.core.accessor import PandasDelegate, delegate_names
-from pandas.core.algorithms import take_1d
 from pandas.core.arrays import DatetimeArray, PeriodArray, TimedeltaArray
 from pandas.core.base import NoNewAttributesMixin, PandasObject
 from pandas.core.indexes.datetimes import DatetimeIndex
@@ -75,9 +74,7 @@ def _delegate_property_get(self, name):
 
         result = np.asarray(result)
 
-        # blow up if we operate on categories
         if self.orig is not None:
-            result = take_1d(result, self.orig.cat.codes)
             index = self.orig.index
         else:
             index = self._parent.index
@@ -324,7 +321,12 @@ def __new__(cls, data):
 
         orig = data if is_categorical_dtype(data) else None
         if orig is not None:
-            data = Series(orig.values.categories, name=orig.name, copy=False)
+            data = Series(
+                orig.array,
+                name=orig.name,
+                copy=False,
+                dtype=orig.values.categories.dtype,
+            )
 
         if is_datetime64_dtype(data.dtype):
             return DatetimeProperties(data, orig)

pandas/tests/series/test_datetime_values.py

@@ -344,6 +344,39 @@ def test_dt_namespace_accessor_categorical(self):
         expected = Series([2017, 2017, 2018, 2018], name="foo")
         tm.assert_series_equal(result, expected)
 
+    def test_dt_tz_localize_categorical(self, tz_aware_fixture):
+        # GH 27952
+        tz = tz_aware_fixture
+        datetimes = pd.Series(
+            ["2019-01-01", "2019-01-01", "2019-01-02"], dtype="datetime64[ns]"
+        )
+        categorical = datetimes.astype("category")
+        result = categorical.dt.tz_localize(tz)
+        expected = datetimes.dt.tz_localize(tz)
+        tm.assert_series_equal(result, expected)
+
+    def test_dt_tz_convert_categorical(self, tz_aware_fixture):
+        # GH 27952
+        tz = tz_aware_fixture
+        datetimes = pd.Series(
+            ["2019-01-01", "2019-01-01", "2019-01-02"], dtype="datetime64[ns, MET]"
+        )
+        categorical = datetimes.astype("category")
+        result = categorical.dt.tz_convert(tz)
+        expected = datetimes.dt.tz_convert(tz)
+        tm.assert_series_equal(result, expected)
+
+    @pytest.mark.parametrize("accessor", ["year", "month", "day"])
+    def test_dt_other_accessors_categorical(self, accessor):
+        # GH 27952
+        datetimes = pd.Series(
+            ["2018-01-01", "2018-01-01", "2019-01-02"], dtype="datetime64[ns]"
+        )
+        categorical = datetimes.astype("category")
+        result = getattr(categorical.dt, accessor)
+        expected = getattr(datetimes.dt, accessor)
+        tm.assert_series_equal(result, expected)
+
     def test_dt_accessor_no_new_attributes(self):
         # https://github.com/pandas-dev/pandas/issues/10673
         s = Series(date_range("20130101", periods=5, freq="D"))