BUG: rendering dt64tz values with non-pytz (pandas-dev#49684)

* BUG: rendering dt64tz values with non-pytz

* GH ref

* py38 compat

* compat for no-tzdata

Author: jbrockmendel

doc/source/whatsnew/v2.0.0.rst

@@ -613,6 +613,8 @@ Datetimelike
 - Bug in :class:`DatetimeIndex` constructor failing to raise when ``tz=None`` is explicitly specified in conjunction with timezone-aware ``dtype`` or data (:issue:`48659`)
 - Bug in subtracting a ``datetime`` scalar from :class:`DatetimeIndex` failing to retain the original ``freq`` attribute (:issue:`48818`)
 - Bug in ``pandas.tseries.holiday.Holiday`` where a half-open date interval causes inconsistent return types from :meth:`USFederalHolidayCalendar.holidays` (:issue:`49075`)
+- Bug in rendering :class:`DatetimeIndex` and :class:`Series` and :class:`DataFrame` with timezone-aware dtypes with ``dateutil`` or ``zoneinfo`` timezones near daylight-savings transitions (:issue:`49684`)
+-
 
 Timedelta
 ^^^^^^^^^

pandas/_libs/tslibs/vectorized.pyi

@@ -33,7 +33,6 @@ def get_resolution(
 def ints_to_pydatetime(
     arr: npt.NDArray[np.int64],
     tz: tzinfo | None = ...,
-    fold: bool = ...,
     box: str = ...,
     reso: int = ...,  # NPY_DATETIMEUNIT
 ) -> npt.NDArray[np.object_]: ...

pandas/_libs/tslibs/vectorized.pyx

@@ -94,7 +94,6 @@ def tz_convert_from_utc(ndarray stamps, tzinfo tz, NPY_DATETIMEUNIT reso=NPY_FR_
 def ints_to_pydatetime(
     ndarray stamps,
     tzinfo tz=None,
-    bint fold=False,
     str box="datetime",
     NPY_DATETIMEUNIT reso=NPY_FR_ns,
 ) -> np.ndarray:
@@ -136,6 +135,7 @@ def ints_to_pydatetime(
         tzinfo new_tz
         bint use_date = False, use_ts = False, use_pydt = False
         object res_val
+        bint fold = 0
 
         # Note that `result` (and thus `result_flat`) is C-order and
         #  `it` iterates C-order as well, so the iteration matches
@@ -168,7 +168,7 @@ def ints_to_pydatetime(
 
         else:
 
-            local_val = info.utc_val_to_local_val(utc_val, &pos)
+            local_val = info.utc_val_to_local_val(utc_val, &pos, &fold)
             if info.use_pytz:
                 # find right representation of dst etc in pytz timezone
                 new_tz = tz._tzinfos[tz._transition_info[pos]]

pandas/tests/arrays/test_datetimes.py

@@ -4,6 +4,11 @@
 from datetime import timedelta
 import operator
 
+try:
+    from zoneinfo import ZoneInfo
+except ImportError:
+    ZoneInfo = None
+
 import numpy as np
 import pytest
 
@@ -706,3 +711,38 @@ def test_tz_localize_t2d(self):
 
         roundtrip = expected.tz_localize("US/Pacific")
         tm.assert_datetime_array_equal(roundtrip, dta)
+
+    easts = ["US/Eastern", "dateutil/US/Eastern"]
+    if ZoneInfo is not None:
+        try:
+            easts.append(ZoneInfo("US/Eastern"))
+        except KeyError:
+            # No tzdata
+            pass
+
+    @pytest.mark.parametrize("tz", easts)
+    def test_iter_zoneinfo_fold(self, tz):
+        # GH#49684
+        utc_vals = np.array(
+            [1320552000, 1320555600, 1320559200, 1320562800], dtype=np.int64
+        )
+        utc_vals *= 1_000_000_000
+
+        dta = DatetimeArray(utc_vals).tz_localize("UTC").tz_convert(tz)
+
+        left = dta[2]
+        right = list(dta)[2]
+        assert str(left) == str(right)
+        # previously there was a bug where with non-pytz right would be
+        #  Timestamp('2011-11-06 01:00:00-0400', tz='US/Eastern')
+        # while left would be
+        #  Timestamp('2011-11-06 01:00:00-0500', tz='US/Eastern')
+        # The .value's would match (so they would compare as equal),
+        #  but the folds would not
+        assert left.utcoffset() == right.utcoffset()
+
+        # The same bug in ints_to_pydatetime affected .astype, so we test
+        #  that here.
+        right2 = dta.astype(object)[2]
+        assert str(left) == str(right2)
+        assert left.utcoffset() == right2.utcoffset()