API/ENH: tz_localize handling of nonexistent times: rename keyword + add shift option

doc/source/timeseries.rst

@@ -2357,6 +2357,31 @@ constructor as well as ``tz_localize``.
    # tz_convert(None) is identical with tz_convert('UTC').tz_localize(None)
    didx.tz_convert('UCT').tz_localize(None)
 
+.. _timeseries.timezone_nonexsistent:
+
+Nonexistent Times when Localizing
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+A DST transition may also shift the local time ahead by 1 hour creating nonexistent
+local times. The behavior of localizing a timeseries with nonexistent times
+can be controlled by the ``nonexistent`` argument. The following options are available:
+
+* ``shift``: Shifts nonexistent times forward to the closest real time
+* ``NaT``: Replaces nonexistent times with ``NaT``
+* ``raise``: Raises a ``pytz.NonExistentTimeError`` (the default behavior)
+
+.. ipython:: python
+    # 2:30 is a nonexistent time
+    dti = date_range(start='2015-03-29 01:30:00', periods=3, freq='H')
+    dti
+    dti.tz_localize('Europe/Warsaw', nonexistent='shift')
+    dti.tz_localize('Europe/Warsaw', nonexistent='NaT')
+
+.. code-block:: ipython
+
+   In [2]: dti.tz_localize('Europe/Warsaw')
+   NonExistentTimeError: 2015-03-29 02:30:00
+
 .. _timeseries.timezone_series:
 
 TZ Aware Dtypes

doc/source/whatsnew/v0.24.0.txt

@@ -194,6 +194,7 @@ Other Enhancements
 - :meth:`Index.to_frame` now supports overriding column name(s) (:issue:`22580`).
 - New attribute :attr:`__git_version__` will return git commit sha of current build (:issue:`21295`).
 - Compatibility with Matplotlib 3.0 (:issue:`22790`).
+- :meth:`Timestamp.tz_localize`, :meth:`DatetimeIndex.tz_localize`, and :meth:`Series.tz_localize` have gained the ``nonexistent`` argument for alternative handling of nonexistent times. See :ref:`timeseries.timezone_nonexsistent` (:issue:`8917`)
 
 .. _whatsnew_0240.api_breaking:
 
@@ -574,6 +575,7 @@ Deprecations
   many ``Series``, ``Index`` or 1-dimensional ``np.ndarray``, or alternatively, only scalar values. (:issue:`21950`)
 - :meth:`FrozenNDArray.searchsorted` has deprecated the ``v`` parameter in favor of ``value`` (:issue:`14645`)
 - :func:`DatetimeIndex.shift` now accepts ``periods`` argument instead of ``n`` for consistency with :func:`Index.shift` and :func:`Series.shift`. Using ``n`` throws a deprecation warning (:issue:`22458`)
+- :meth:`Timestamp.tz_localize`, :meth:`DatetimeIndex.tz_localize`, and :meth:`Series.tz_localize` have deprecated the ``errors` argument in favor of the ``ambiguous`` and ``nonexistent`` arguments (:issue:`8917`)
 
 .. _whatsnew_0240.prior_deprecations:
 

pandas/_libs/tslibs/conversion.pyx

@@ -1,5 +1,4 @@
 # -*- coding: utf-8 -*-
-
 import cython
 from cython import Py_ssize_t
 
@@ -44,6 +43,7 @@ from nattype cimport NPY_NAT, checknull_with_nat
 # Constants
 
 cdef int64_t DAY_NS = 86400000000000LL
+cdef int64_t HOURS_NS = 3600000000000
 NS_DTYPE = np.dtype('M8[ns]')
 TD_DTYPE = np.dtype('m8[ns]')
 
@@ -458,8 +458,7 @@ cdef _TSObject convert_str_to_tsobject(object ts, object tz, object unit,
                 if tz is not None:
                     # shift for localize_tso
                     ts = tz_localize_to_utc(np.array([ts], dtype='i8'), tz,
-                                            ambiguous='raise',
-                                            errors='raise')[0]
+                                            ambiguous='raise')[0]
 
         except OutOfBoundsDatetime:
             # GH#19382 for just-barely-OutOfBounds falling back to dateutil
@@ -826,7 +825,7 @@ def tz_convert(int64_t[:] vals, object tz1, object tz2):
 @cython.boundscheck(False)
 @cython.wraparound(False)
 def tz_localize_to_utc(ndarray[int64_t] vals, object tz, object ambiguous=None,
-                       object errors='raise'):
+                       object nonexistent=None):
     """
     Localize tzinfo-naive i8 to given time zone (using pytz). If
     there are ambiguities in the values, raise AmbiguousTimeError.
@@ -837,7 +836,10 @@ def tz_localize_to_utc(ndarray[int64_t] vals, object tz, object ambiguous=None,
     tz : tzinfo or None
     ambiguous : str, bool, or arraylike
         If arraylike, must have the same length as vals
-    errors : {"raise", "coerce"}, default "raise"
+    nonexistent : str, bool, or arraylike
+        If arraylike, must have the same length as vals
+
+        .. versionadded:: 0.24.0
 
     Returns
     -------
@@ -849,16 +851,13 @@ def tz_localize_to_utc(ndarray[int64_t] vals, object tz, object ambiguous=None,
         ndarray ambiguous_array
         Py_ssize_t i, idx, pos, ntrans, n = len(vals)
         int64_t *tdata
-        int64_t v, left, right
+        int64_t v, left, right, val, v_left, v_right
         ndarray[int64_t] result, result_a, result_b, dst_hours
         npy_datetimestruct dts
         bint infer_dst = False, is_dst = False, fill = False
-        bint is_coerce = errors == 'coerce', is_raise = errors == 'raise'
+        bint shift = False, fill_nonexist = False
 
     # Vectorized version of DstTzInfo.localize
-
-    assert is_coerce or is_raise
-
     if tz == UTC or tz is None:
         return vals
 
@@ -888,39 +887,43 @@ def tz_localize_to_utc(ndarray[int64_t] vals, object tz, object ambiguous=None,
                              "the same size as vals")
         ambiguous_array = np.asarray(ambiguous)
 
+    if is_string_object(nonexistent):
+        if nonexistent == 'NaT':
+            fill_nonexist = True
+        elif nonexistent == 'shift':
+            shift = True
+
     trans, deltas, typ = get_dst_info(tz)
 
     tdata = <int64_t*> cnp.PyArray_DATA(trans)
     ntrans = len(trans)
 
+    # Determine whether each date lies left of the DST transition (store in
+    # result_a) or right of the DST transition (store in result_b)
     result_a = np.empty(n, dtype=np.int64)
     result_b = np.empty(n, dtype=np.int64)
     result_a.fill(NPY_NAT)
     result_b.fill(NPY_NAT)
 
-    # left side
-    idx_shifted = (np.maximum(0, trans.searchsorted(
+    idx_shifted_left = (np.maximum(0, trans.searchsorted(
         vals - DAY_NS, side='right') - 1)).astype(np.int64)
 
-    for i in range(n):
-        v = vals[i] - deltas[idx_shifted[i]]
-        pos = bisect_right_i8(tdata, v, ntrans) - 1
-
-        # timestamp falls to the left side of the DST transition
-        if v + deltas[pos] == vals[i]:
-            result_a[i] = v
-
-    # right side
-    idx_shifted = (np.maximum(0, trans.searchsorted(
+    idx_shifted_right = (np.maximum(0, trans.searchsorted(
         vals + DAY_NS, side='right') - 1)).astype(np.int64)
 
     for i in range(n):
-        v = vals[i] - deltas[idx_shifted[i]]
-        pos = bisect_right_i8(tdata, v, ntrans) - 1
+        val = vals[i]
+        v_left = val - deltas[idx_shifted_left[i]]
+        pos_left = bisect_right_i8(tdata, v_left, ntrans) - 1
+        # timestamp falls to the left side of the DST transition
+        if v_left + deltas[pos_left] == val:
+            result_a[i] = v_left
 
+        v_right = val - deltas[idx_shifted_right[i]]
+        pos_right = bisect_right_i8(tdata, v_right, ntrans) - 1
         # timestamp falls to the right side of the DST transition
-        if v + deltas[pos] == vals[i]:
-            result_b[i] = v
+        if v_right + deltas[pos_right] == val:
+            result_b[i] = v_right
 
     if infer_dst:
         dst_hours = np.empty(n, dtype=np.int64)
@@ -935,7 +938,7 @@ def tz_localize_to_utc(ndarray[int64_t] vals, object tz, object ambiguous=None,
             stamp = _render_tstamp(vals[trans_idx])
             raise pytz.AmbiguousTimeError(
                 "Cannot infer dst time from %s as there "
-                "are no repeated times" % stamp)
+                "are no repeated times".format(stamp))
         # Split the array into contiguous chunks (where the difference between
         # indices is 1).  These are effectively dst transitions in different
         # years which is useful for checking that there is not an ambiguous
@@ -960,18 +963,19 @@ def tz_localize_to_utc(ndarray[int64_t] vals, object tz, object ambiguous=None,
                 if switch_idx.size > 1:
                     raise pytz.AmbiguousTimeError(
                         "There are %i dst switches when "
-                        "there should only be 1." % switch_idx.size)
+                        "there should only be 1.".format(switch_idx.size))
                 switch_idx = switch_idx[0] + 1
                 # Pull the only index and adjust
                 a_idx = grp[:switch_idx]
                 b_idx = grp[switch_idx:]
                 dst_hours[grp] = np.hstack((result_a[a_idx], result_b[b_idx]))
 
     for i in range(n):
+        val = vals[i]
         left = result_a[i]
         right = result_b[i]
-        if vals[i] == NPY_NAT:
-            result[i] = vals[i]
+        if val == NPY_NAT:
+            result[i] = val
         elif left != NPY_NAT and right != NPY_NAT:
             if left == right:
                 result[i] = left
@@ -986,19 +990,27 @@ def tz_localize_to_utc(ndarray[int64_t] vals, object tz, object ambiguous=None,
                 elif fill:
                     result[i] = NPY_NAT
                 else:
-                    stamp = _render_tstamp(vals[i])
+                    stamp = _render_tstamp(val)
                     raise pytz.AmbiguousTimeError(
                         "Cannot infer dst time from %r, try using the "
-                        "'ambiguous' argument" % stamp)
+                        "'ambiguous' argument".format(stamp))
         elif left != NPY_NAT:
             result[i] = left
         elif right != NPY_NAT:
             result[i] = right
         else:
-            if is_coerce:
+            # Handle nonexistent times
+            if shift:
+                # Shift the nonexistent time forward to the closest existing
+                # time
+                remaining_minutes = val % HOURS_NS
+                new_local = val + (HOURS_NS - remaining_minutes)
+                delta_idx = trans.searchsorted(new_local, side='right') - 1
+                result[i] = new_local - deltas[delta_idx]
+            elif fill_nonexist:
                 result[i] = NPY_NAT
             else:
-                stamp = _render_tstamp(vals[i])
+                stamp = _render_tstamp(val)
                 raise pytz.NonExistentTimeError(stamp)
 
     return result

pandas/_libs/tslibs/nattype.pyx

@@ -564,14 +564,23 @@ class NaTType(_NaT):
             - 'NaT' will return NaT for an ambiguous time
             - 'raise' will raise an AmbiguousTimeError for an ambiguous time
 
-        errors : 'raise', 'coerce', default 'raise'
+        nonexistent : 'shift', 'NaT', default 'raise'
+            - 'shift' will shift the nonexistent time forward to the closest
+              existing time
+            - 'NaT' will return NaT where there are nonexistent times
+            - 'raise' will raise an NonExistentTimeError if there are
+              nonexistent times
+
+            .. versionadded:: 0.24.0
+
+        errors : 'raise', 'coerce', default None
             - 'raise' will raise a NonExistentTimeError if a timestamp is not
                valid in the specified timezone (e.g. due to a transition from
                or to DST time)
             - 'coerce' will return NaT if the timestamp can not be converted
               into the specified timezone
 
-              .. versionadded:: 0.19.0
+              .. deprecated:: 0.24.0
 
         Returns
         -------

pandas/_libs/tslibs/timestamps.pyx

@@ -961,7 +961,8 @@ class Timestamp(_Timestamp):
     def is_leap_year(self):
         return bool(ccalendar.is_leapyear(self.year))
 
-    def tz_localize(self, tz, ambiguous='raise', errors='raise'):
+    def tz_localize(self, tz, ambiguous='raise', nonexistent='raise',
+                    errors=None):
         """
         Convert naive Timestamp to local time zone, or remove
         timezone from tz-aware Timestamp.
@@ -978,14 +979,23 @@ class Timestamp(_Timestamp):
             - 'NaT' will return NaT for an ambiguous time
             - 'raise' will raise an AmbiguousTimeError for an ambiguous time
 
-        errors : 'raise', 'coerce', default 'raise'
+        nonexistent : 'shift', 'NaT', default 'raise'
+            - 'shift' will shift the nonexistent time forward to the closest
+              existing time
+            - 'NaT' will return NaT where there are nonexistent times
+            - 'raise' will raise an NonExistentTimeError if there are
+              nonexistent times
+
+            .. versionadded:: 0.24.0
+
+        errors : 'raise', 'coerce', default None
             - 'raise' will raise a NonExistentTimeError if a timestamp is not
                valid in the specified timezone (e.g. due to a transition from
                or to DST time)
             - 'coerce' will return NaT if the timestamp can not be converted
               into the specified timezone
 
-              .. versionadded:: 0.19.0
+              .. deprecated:: 0.24.0
 
         Returns
         -------
@@ -999,13 +1009,26 @@ class Timestamp(_Timestamp):
         if ambiguous == 'infer':
             raise ValueError('Cannot infer offset with only one time.')
 
+        if errors is not None:
+            warnings.warn("The errors argument is deprecated and will be "
+                          "removed in a future release. Use the ambiguous or "
+                          "nonexistent argument instead.", FutureWarning,
+                          stacklevel=2)
+            if errors == 'coerce':
+                nonexistent = 'NaT'
+            elif errors == 'raise':
+                nonexistent = 'raise'
+            else:
+                raise ValueError("The errors argument must be either coerce "
+                                 "or raise.")
         if self.tzinfo is None:
             # tz naive, localize
             tz = maybe_get_tz(tz)
             if not is_string_object(ambiguous):
                 ambiguous = [ambiguous]
             value = tz_localize_to_utc(np.array([self.value], dtype='i8'), tz,
-                                       ambiguous=ambiguous, errors=errors)[0]
+                                       ambiguous=ambiguous,
+                                       nonexistent=nonexistent)[0]
             return Timestamp(value, tz=tz)
         else:
             if tz is None: