BUG: tz_localize(UTC) not making a copy (pandas-dev#46460)

Author: jbrockmendel

doc/source/whatsnew/v1.5.0.rst

@@ -391,6 +391,7 @@ Datetimelike
 - Bug in :class:`Timestamp` construction when passing datetime components as positional arguments and ``tzinfo`` as a keyword argument incorrectly raising (:issue:`31929`)
 - Bug in :meth:`Index.astype` when casting from object dtype to ``timedelta64[ns]`` dtype incorrectly casting ``np.datetime64("NaT")`` values to ``np.timedelta64("NaT")`` instead of raising (:issue:`45722`)
 - Bug in :meth:`SeriesGroupBy.value_counts` index when passing categorical column (:issue:`44324`)
+- Bug in :meth:`DatetimeIndex.tz_localize` localizing to UTC failing to make a copy of the underlying data (:issue:`46460`)
 -
 
 Timedelta

pandas/_libs/tslibs/tzconversion.pyx

@@ -64,8 +64,6 @@ cdef int64_t tz_localize_to_utc_single(
         return val - _tz_localize_using_tzinfo_api(val, tz, to_utc=True)
 
     elif is_fixed_offset(tz):
-        # TODO: in this case we should be able to use get_utcoffset,
-        #  that returns None for e.g. 'dateutil//usr/share/zoneinfo/Etc/GMT-9'
         _, deltas, _ = get_dst_info(tz)
         delta = deltas[0]
         return val - delta
@@ -121,9 +119,10 @@ timedelta-like}
         Py_ssize_t delta_idx_offset, delta_idx, pos_left, pos_right
         int64_t *tdata
         int64_t v, left, right, val, v_left, v_right, new_local, remaining_mins
-        int64_t first_delta
+        int64_t first_delta, delta
         int64_t shift_delta = 0
-        ndarray[int64_t] trans, result, result_a, result_b, dst_hours
+        ndarray[int64_t] trans, result_a, result_b, dst_hours
+        int64_t[::1] result
         npy_datetimestruct dts
         bint infer_dst = False, is_dst = False, fill = False
         bint shift_forward = False, shift_backward = False
@@ -132,7 +131,7 @@ timedelta-like}
 
     # Vectorized version of DstTzInfo.localize
     if is_utc(tz) or tz is None:
-        return vals
+        return vals.copy()
 
     result = np.empty(n, dtype=np.int64)
 
@@ -143,7 +142,18 @@ timedelta-like}
                 result[i] = NPY_NAT
             else:
                 result[i] = v - _tz_localize_using_tzinfo_api(v, tz, to_utc=True)
-        return result
+        return result.base  # to return underlying ndarray
+
+    elif is_fixed_offset(tz):
+        _, deltas, _ = get_dst_info(tz)
+        delta = deltas[0]
+        for i in range(n):
+            v = vals[i]
+            if v == NPY_NAT:
+                result[i] = NPY_NAT
+            else:
+                result[i] = v - delta
+        return result.base  # to return underlying ndarray
 
     # silence false-positive compiler warning
     ambiguous_array = np.empty(0, dtype=bool)
@@ -298,7 +308,7 @@ timedelta-like}
                 stamp = _render_tstamp(val)
                 raise pytz.NonExistentTimeError(stamp)
 
-    return result
+    return result.base  # .base to get underlying ndarray
 
 
 cdef inline Py_ssize_t bisect_right_i8(int64_t *data,

pandas/core/arrays/datetimes.py

@@ -456,9 +456,13 @@ def _generate_range(
             endpoint_tz = start.tz if start is not None else end.tz
 
             if tz is not None and endpoint_tz is None:
-                i8values = tzconversion.tz_localize_to_utc(
-                    i8values, tz, ambiguous=ambiguous, nonexistent=nonexistent
-                )
+
+                if not timezones.is_utc(tz):
+                    # short-circuit tz_localize_to_utc which would make
+                    #  an unnecessary copy with UTC but be a no-op.
+                    i8values = tzconversion.tz_localize_to_utc(
+                        i8values, tz, ambiguous=ambiguous, nonexistent=nonexistent
+                    )
 
                 # i8values is localized datetime64 array -> have to convert
                 # start/end as well to compare
@@ -2126,6 +2130,8 @@ def _sequence_to_dt64ns(
         if tz is not None:
             # Convert tz-naive to UTC
             tz = timezones.maybe_get_tz(tz)
+            # TODO: if tz is UTC, are there situations where we *don't* want a
+            #  copy?  tz_localize_to_utc always makes one.
             data = tzconversion.tz_localize_to_utc(
                 data.view("i8"), tz, ambiguous=ambiguous
             )

pandas/tests/indexes/datetimes/test_timezones.py

@@ -327,6 +327,17 @@ def test_tz_convert_unsorted(self, tzstr):
     # -------------------------------------------------------------
     # DatetimeIndex.tz_localize
 
+    def test_tz_localize_utc_copies(self, utc_fixture):
+        # GH#46460
+        times = ["2015-03-08 01:00", "2015-03-08 02:00", "2015-03-08 03:00"]
+        index = DatetimeIndex(times)
+
+        res = index.tz_localize(utc_fixture)
+        assert not tm.shares_memory(res, index)
+
+        res2 = index._data.tz_localize(utc_fixture)
+        assert not tm.shares_memory(index._data, res2)
+
     def test_dti_tz_localize_nonexistent_raise_coerce(self):
         # GH#13057
         times = ["2015-03-08 01:00", "2015-03-08 02:00", "2015-03-08 03:00"]

pandas/tests/tslibs/test_conversion.py

@@ -50,6 +50,18 @@ def _compare_local_to_utc(tz_didx, naive_didx):
         tm.assert_numpy_array_equal(result, expected)
 
 
+def test_tz_localize_to_utc_copies():
+    # GH#46460
+    arr = np.arange(5, dtype="i8")
+    result = tzconversion.tz_convert_from_utc(arr, tz=UTC)
+    tm.assert_numpy_array_equal(result, arr)
+    assert not np.shares_memory(arr, result)
+
+    result = tzconversion.tz_convert_from_utc(arr, tz=None)
+    tm.assert_numpy_array_equal(result, arr)
+    assert not np.shares_memory(arr, result)
+
+
 def test_tz_convert_single_matches_tz_convert_hourly(tz_aware_fixture):
     tz = tz_aware_fixture
     tz_didx = date_range("2014-03-01", "2015-01-10", freq="H", tz=tz)