Fix #13306: Hour overflow in tz-aware datetime conversions.

closes #13306

Author: Uwe Hoffmann <[email protected]>

Closes #13313 from uwedeportivo/master and squashes the following commits:

be3ed90 [Uwe Hoffmann] whatsnew entry for issue #13306
1f5f7a5 [Uwe Hoffmann] Code Review jreback
82f263a [Uwe Hoffmann] Use vectorized searchsorted and tests.
a1ed5a5 [Uwe Hoffmann] Fix #13306: Hour overflow in tz-aware datetime conversions.

Author: uwedeportivo

doc/source/whatsnew/v0.18.2.txt

@@ -338,7 +338,7 @@ Bug Fixes
 - Bug in ``.resample(..)`` with a ``PeriodIndex`` not changing its ``freq`` appropriately when empty (:issue:`13067`)
 - Bug in ``.resample(..)`` with a ``PeriodIndex`` not retaining its type or name with an empty ``DataFrame``appropriately when empty (:issue:`13212`)
 - Bug in ``groupby(..).resample(..)`` where passing some keywords would raise an exception (:issue:`13235`)
-
+- Bug in ``.tz_convert`` on a tz-aware ``DateTimeIndex`` that relied on index being sorted for correct results (:issue: `13306`)
 
 
 

pandas/tseries/tests/test_timezones.py

@@ -902,6 +902,88 @@ def test_utc_with_system_utc(self):
         # check that the time hasn't changed.
         self.assertEqual(ts, ts.tz_convert(dateutil.tz.tzutc()))
 
+    def test_tz_convert_hour_overflow_dst(self):
+        # Regression test for:
+        # https://github.com/pydata/pandas/issues/13306
+
+        # sorted case US/Eastern -> UTC
+        ts = ['2008-05-12 09:50:00',
+              '2008-12-12 09:50:35',
+              '2009-05-12 09:50:32']
+        tt = to_datetime(ts).tz_localize('US/Eastern')
+        ut = tt.tz_convert('UTC')
+        expected = np.array([13, 14, 13], dtype=np.int32)
+        self.assert_numpy_array_equal(ut.hour, expected)
+
+        # sorted case UTC -> US/Eastern
+        ts = ['2008-05-12 13:50:00',
+              '2008-12-12 14:50:35',
+              '2009-05-12 13:50:32']
+        tt = to_datetime(ts).tz_localize('UTC')
+        ut = tt.tz_convert('US/Eastern')
+        expected = np.array([9, 9, 9], dtype=np.int32)
+        self.assert_numpy_array_equal(ut.hour, expected)
+
+        # unsorted case US/Eastern -> UTC
+        ts = ['2008-05-12 09:50:00',
+              '2008-12-12 09:50:35',
+              '2008-05-12 09:50:32']
+        tt = to_datetime(ts).tz_localize('US/Eastern')
+        ut = tt.tz_convert('UTC')
+        expected = np.array([13, 14, 13], dtype=np.int32)
+        self.assert_numpy_array_equal(ut.hour, expected)
+
+        # unsorted case UTC -> US/Eastern
+        ts = ['2008-05-12 13:50:00',
+              '2008-12-12 14:50:35',
+              '2008-05-12 13:50:32']
+        tt = to_datetime(ts).tz_localize('UTC')
+        ut = tt.tz_convert('US/Eastern')
+        expected = np.array([9, 9, 9], dtype=np.int32)
+        self.assert_numpy_array_equal(ut.hour, expected)
+
+    def test_tz_convert_hour_overflow_dst_timestamps(self):
+        # Regression test for:
+        # https://github.com/pydata/pandas/issues/13306
+
+        tz = self.tzstr('US/Eastern')
+
+        # sorted case US/Eastern -> UTC
+        ts = [Timestamp('2008-05-12 09:50:00', tz=tz),
+              Timestamp('2008-12-12 09:50:35', tz=tz),
+              Timestamp('2009-05-12 09:50:32', tz=tz)]
+        tt = to_datetime(ts)
+        ut = tt.tz_convert('UTC')
+        expected = np.array([13, 14, 13], dtype=np.int32)
+        self.assert_numpy_array_equal(ut.hour, expected)
+
+        # sorted case UTC -> US/Eastern
+        ts = [Timestamp('2008-05-12 13:50:00', tz='UTC'),
+              Timestamp('2008-12-12 14:50:35', tz='UTC'),
+              Timestamp('2009-05-12 13:50:32', tz='UTC')]
+        tt = to_datetime(ts)
+        ut = tt.tz_convert('US/Eastern')
+        expected = np.array([9, 9, 9], dtype=np.int32)
+        self.assert_numpy_array_equal(ut.hour, expected)
+
+        # unsorted case US/Eastern -> UTC
+        ts = [Timestamp('2008-05-12 09:50:00', tz=tz),
+              Timestamp('2008-12-12 09:50:35', tz=tz),
+              Timestamp('2008-05-12 09:50:32', tz=tz)]
+        tt = to_datetime(ts)
+        ut = tt.tz_convert('UTC')
+        expected = np.array([13, 14, 13], dtype=np.int32)
+        self.assert_numpy_array_equal(ut.hour, expected)
+
+        # unsorted case UTC -> US/Eastern
+        ts = [Timestamp('2008-05-12 13:50:00', tz='UTC'),
+              Timestamp('2008-12-12 14:50:35', tz='UTC'),
+              Timestamp('2008-05-12 13:50:32', tz='UTC')]
+        tt = to_datetime(ts)
+        ut = tt.tz_convert('US/Eastern')
+        expected = np.array([9, 9, 9], dtype=np.int32)
+        self.assert_numpy_array_equal(ut.hour, expected)
+
     def test_tslib_tz_convert_trans_pos_plus_1__bug(self):
         # Regression test for tslib.tz_convert(vals, tz1, tz2).
         # See https://github.com/pydata/pandas/issues/4496 for details.

pandas/tslib.pyx

@@ -3754,8 +3754,8 @@ except:
 
 def tz_convert(ndarray[int64_t] vals, object tz1, object tz2):
     cdef:
-        ndarray[int64_t] utc_dates, tt, result, trans, deltas
-        Py_ssize_t i, pos, n = len(vals)
+        ndarray[int64_t] utc_dates, tt, result, trans, deltas, posn
+        Py_ssize_t i, j, pos, n = len(vals)
         int64_t v, offset
         pandas_datetimestruct dts
         Py_ssize_t trans_len
@@ -3791,19 +3791,18 @@ def tz_convert(ndarray[int64_t] vals, object tz1, object tz2):
                 return vals
 
             trans_len = len(trans)
-            pos = trans.searchsorted(tt[0]) - 1
-            if pos < 0:
-                raise ValueError('First time before start of DST info')
-
-            offset = deltas[pos]
+            posn = trans.searchsorted(tt, side='right')
+            j = 0
             for i in range(n):
                 v = vals[i]
                 if v == NPY_NAT:
                     utc_dates[i] = NPY_NAT
                 else:
-                    while pos + 1 < trans_len and v >= trans[pos + 1]:
-                        pos += 1
-                        offset = deltas[pos]
+                    pos = posn[j] - 1
+                    j = j + 1
+                    if pos < 0:
+                        raise ValueError('First time before start of DST info')
+                    offset = deltas[pos]
                     utc_dates[i] = v - offset
     else:
         utc_dates = vals
@@ -3838,20 +3837,18 @@ def tz_convert(ndarray[int64_t] vals, object tz1, object tz2):
     if (result==NPY_NAT).all():
         return result
 
-    pos = trans.searchsorted(utc_dates[utc_dates!=NPY_NAT][0]) - 1
-    if pos < 0:
-        raise ValueError('First time before start of DST info')
-
-    # TODO: this assumed sortedness :/
-    offset = deltas[pos]
+    posn = trans.searchsorted(utc_dates[utc_dates!=NPY_NAT], side='right')
+    j = 0
     for i in range(n):
         v = utc_dates[i]
         if vals[i] == NPY_NAT:
             result[i] = vals[i]
         else:
-            while pos + 1 < trans_len and v >= trans[pos + 1]:
-                pos += 1
-                offset = deltas[pos]
+            pos = posn[j] - 1
+            j = j + 1
+            if pos < 0:
+                raise ValueError('First time before start of DST info')
+            offset = deltas[pos]
             result[i] = v + offset
     return result