BUG: overflow in astype(td64ns) (#40008)

Author: jbrockmendel

doc/source/whatsnew/v1.3.0.rst

@@ -308,6 +308,7 @@ Timedelta
 - Bug in constructing :class:`Timedelta` from ``np.timedelta64`` objects with non-nanosecond units that are out of bounds for ``timedelta64[ns]`` (:issue:`38965`)
 - Bug in constructing a :class:`TimedeltaIndex` incorrectly accepting ``np.datetime64("NaT")`` objects (:issue:`39462`)
 - Bug in constructing :class:`Timedelta` from input string with only symbols and no digits failed to raise an error (:issue:`39710`)
+- Bug in :class:`TimedeltaIndex` and :func:`to_timedelta` failing to raise when passed non-nanosecond ``timedelta64`` arrays that overflow when converting to ``timedelta64[ns]`` (:issue:`40008`)
 
 Timezones
 ^^^^^^^^^

pandas/_libs/tslibs/conversion.pyx

@@ -239,6 +239,11 @@ def ensure_datetime64ns(arr: ndarray, copy: bool=True):
         return result
 
     unit = get_datetime64_unit(arr.flat[0])
+    if unit == NPY_DATETIMEUNIT.NPY_FR_GENERIC:
+        # without raising explicitly here, we end up with a SystemError
+        # built-in function ensure_datetime64ns returned a result with an error
+        raise ValueError("datetime64/timedelta64 must have a unit specified")
+
     if unit == NPY_FR_ns:
         if copy:
             arr = arr.copy()

pandas/core/arrays/timedeltas.py

@@ -24,7 +24,10 @@
     iNaT,
     to_offset,
 )
-from pandas._libs.tslibs.conversion import precision_from_unit
+from pandas._libs.tslibs.conversion import (
+    ensure_timedelta64ns,
+    precision_from_unit,
+)
 from pandas._libs.tslibs.fields import get_timedelta_field
 from pandas._libs.tslibs.timedeltas import (
     array_to_timedelta64,
@@ -982,8 +985,7 @@ def sequence_to_td64ns(data, copy=False, unit=None, errors="raise"):
     elif is_timedelta64_dtype(data.dtype):
         if data.dtype != TD64NS_DTYPE:
             # non-nano unit
-            # TODO: watch out for overflows
-            data = data.astype(TD64NS_DTYPE)
+            data = ensure_timedelta64ns(data)
             copy = False
 
     else:
@@ -1025,8 +1027,8 @@ def ints_to_td64ns(data, unit="ns"):
         dtype_str = f"timedelta64[{unit}]"
         data = data.view(dtype_str)
 
-        # TODO: watch out for overflows when converting from lower-resolution
-        data = data.astype("timedelta64[ns]")
+        data = ensure_timedelta64ns(data)
+
         # the astype conversion makes a copy, so we can avoid re-copying later
         copy_made = True
 

pandas/tests/dtypes/test_common.py

@@ -765,7 +765,7 @@ def test_astype_datetime64_bad_dtype_raises(from_type, to_type):
 
 @pytest.mark.parametrize("from_type", [np.datetime64, np.timedelta64])
 def test_astype_object_preserves_datetime_na(from_type):
-    arr = np.array([from_type("NaT")])
+    arr = np.array([from_type("NaT", "ns")])
     result = astype_nansafe(arr, dtype=np.dtype("object"))
 
     assert isna(result)[0]

pandas/tests/reshape/test_cut.py

@@ -145,18 +145,28 @@ def test_bins_not_monotonic():
             ),
         ),
         (
-            [np.timedelta64(-1), np.timedelta64(0), np.timedelta64(1)],
+            [
+                np.timedelta64(-1, "ns"),
+                np.timedelta64(0, "ns"),
+                np.timedelta64(1, "ns"),
+            ],
             np.array(
                 [
-                    np.timedelta64(-np.iinfo(np.int64).max),
-                    np.timedelta64(0),
-                    np.timedelta64(np.iinfo(np.int64).max),
+                    np.timedelta64(-np.iinfo(np.int64).max, "ns"),
+                    np.timedelta64(0, "ns"),
+                    np.timedelta64(np.iinfo(np.int64).max, "ns"),
                 ]
             ),
             IntervalIndex.from_tuples(
                 [
-                    (np.timedelta64(-np.iinfo(np.int64).max), np.timedelta64(0)),
-                    (np.timedelta64(0), np.timedelta64(np.iinfo(np.int64).max)),
+                    (
+                        np.timedelta64(-np.iinfo(np.int64).max, "ns"),
+                        np.timedelta64(0, "ns"),
+                    ),
+                    (
+                        np.timedelta64(0, "ns"),
+                        np.timedelta64(np.iinfo(np.int64).max, "ns"),
+                    ),
                 ]
             ),
         ),

pandas/tests/tools/test_to_timedelta.py

@@ -6,6 +6,8 @@
 import numpy as np
 import pytest
 
+from pandas.errors import OutOfBoundsTimedelta
+
 import pandas as pd
 from pandas import (
     Series,
@@ -14,6 +16,7 @@
     to_timedelta,
 )
 import pandas._testing as tm
+from pandas.core.arrays import TimedeltaArray
 
 
 class TestTimedeltas:
@@ -75,6 +78,19 @@ def test_to_timedelta(self):
         expected = TimedeltaIndex([np.timedelta64(1, "D")] * 5)
         tm.assert_index_equal(result, expected)
 
+    def test_to_timedelta_oob_non_nano(self):
+        arr = np.array([pd.NaT.value + 1], dtype="timedelta64[s]")
+
+        msg = r"Out of bounds for nanosecond timedelta64\[s\] -9223372036854775807"
+        with pytest.raises(OutOfBoundsTimedelta, match=msg):
+            to_timedelta(arr)
+
+        with pytest.raises(OutOfBoundsTimedelta, match=msg):
+            TimedeltaIndex(arr)
+
+        with pytest.raises(OutOfBoundsTimedelta, match=msg):
+            TimedeltaArray._from_sequence(arr)
+
     def test_to_timedelta_dataframe(self):
         # GH 11776
         arr = np.arange(10).reshape(2, 5)