BUG: Fix failing hash function for certain non-ns resolution Timedeltas

doc/source/whatsnew/v2.1.0.rst

@@ -393,6 +393,7 @@ Timedelta
 - :meth:`TimedeltaIndex.map` with ``na_action="ignore"`` now works as expected (:issue:`51644`)
 - Bug in :class:`TimedeltaIndex` division or multiplication leading to ``.freq`` of "0 Days" instead of ``None`` (:issue:`51575`)
 - Bug in :class:`Timedelta` with Numpy timedelta64 objects not properly raising ``ValueError`` (:issue:`52806`)
+- Bug in :meth:`Timedelta.__hash__`, raising an ``OutOfBoundsTimedelta`` on certain large values of second resolution (:issue:`54037`)
 - Bug in :meth:`Timedelta.round` with values close to the implementation bounds returning incorrect results instead of raising ``OutOfBoundsTimedelta`` (:issue:`51494`)
 - Bug in :meth:`arrays.TimedeltaArray.map` and :meth:`TimedeltaIndex.map`, where the supplied callable operated array-wise instead of element-wise (:issue:`51977`)
 -

pandas/_libs/tslibs/timedeltas.pyx

@@ -1156,7 +1156,7 @@ cdef class _Timedelta(timedelta):
             #  resolution.
             try:
                 obj = (<_Timedelta>self)._as_creso(<NPY_DATETIMEUNIT>(self._creso + 1))
-            except OverflowError:
+            except OutOfBoundsTimedelta:
                 # Doesn't fit, so we're off the hook
                 return hash(self._value)
             else:

pandas/tests/scalar/timedelta/test_timedelta.py

@@ -305,6 +305,12 @@ def test_resolution(self, td):
         assert result == expected
         assert result._creso == expected._creso
 
+    def test_hash(self) -> None:
+        # GH#54037
+        second_resolution_max = Timedelta(0).as_unit("s").max
+
+        assert hash(second_resolution_max)
+
 
 def test_timedelta_class_min_max_resolution():
     # when accessed on the class (as opposed to an instance), we default