handle DST appropriately in Timestamp.replace (pandas-dev#18618)

Author: jbrockmendel

doc/source/whatsnew/v0.23.0.txt

@@ -428,3 +428,4 @@ Other
 ^^^^^
 
 - Improved error message when attempting to use a Python keyword as an identifier in a ``numexpr`` backed query (:issue:`18221`)
+- :func:`Timestamp.replace` will now handle Daylight Savings transitions gracefully (:issue:`18319`)

pandas/_libs/tslibs/timestamps.pyx

@@ -33,7 +33,7 @@ from np_datetime cimport (reverse_ops, cmp_scalar, check_dts_bounds,
                           is_leapyear)
 from timedeltas import Timedelta
 from timedeltas cimport delta_to_nanoseconds
-from timezones cimport get_timezone, is_utc, maybe_get_tz
+from timezones cimport get_timezone, is_utc, maybe_get_tz, treat_tz_as_pytz
 
 # ----------------------------------------------------------------------
 # Constants
@@ -922,8 +922,18 @@ class Timestamp(_Timestamp):
             _tzinfo = tzinfo
 
         # reconstruct & check bounds
-        ts_input = datetime(dts.year, dts.month, dts.day, dts.hour, dts.min,
-                            dts.sec, dts.us, tzinfo=_tzinfo)
+        if _tzinfo is not None and treat_tz_as_pytz(_tzinfo):
+            # replacing across a DST boundary may induce a new tzinfo object
+            # see GH#18319
+            ts_input = _tzinfo.localize(datetime(dts.year, dts.month, dts.day,
+                                                 dts.hour, dts.min, dts.sec,
+                                                 dts.us))
+            _tzinfo = ts_input.tzinfo
+        else:
+            ts_input = datetime(dts.year, dts.month, dts.day,
+                                dts.hour, dts.min, dts.sec, dts.us,
+                                tzinfo=_tzinfo)
+
         ts = convert_datetime_to_tsobject(ts_input, _tzinfo)
         value = ts.value + (dts.ps // 1000)
         if value != NPY_NAT:

pandas/tests/tseries/test_timezones.py

@@ -61,6 +61,10 @@ def tzstr(self, tz):
     def localize(self, tz, x):
         return tz.localize(x)
 
+    def normalize(self, ts):
+        tzinfo = ts.tzinfo
+        return tzinfo.normalize(ts)
+
     def cmptz(self, tz1, tz2):
         # Compare two timezones. Overridden in subclass to parameterize
         # tests.
@@ -935,6 +939,27 @@ def test_datetimeindex_tz_nat(self):
         assert isna(idx[1])
         assert idx[0].tzinfo is not None
 
+    def test_replace_across_dst(self):
+        # GH#18319 check that 1) timezone is correctly normalized and
+        # 2) that hour is not incorrectly changed by this normalization
+        tz = self.tz('US/Eastern')
+
+        ts_naive = Timestamp('2017-12-03 16:03:30')
+        ts_aware = self.localize(tz, ts_naive)
+
+        # Preliminary sanity-check
+        assert ts_aware == self.normalize(ts_aware)
+
+        # Replace across DST boundary
+        ts2 = ts_aware.replace(month=6)
+
+        # Check that `replace` preserves hour literal
+        assert (ts2.hour, ts2.minute) == (ts_aware.hour, ts_aware.minute)
+
+        # Check that post-replace object is appropriately normalized
+        ts2b = self.normalize(ts2)
+        assert ts2 == ts2b
+
 
 class TestTimeZoneSupportDateutil(TestTimeZoneSupportPytz):
 
@@ -959,6 +984,10 @@ def cmptz(self, tz1, tz2):
     def localize(self, tz, x):
         return x.replace(tzinfo=tz)
 
+    def normalize(self, ts):
+        # no-op for dateutil
+        return ts
+
     @td.skip_if_windows
     def test_utc_with_system_utc(self):
         from pandas._libs.tslibs.timezones import maybe_get_tz