BUG: catch overflow in timestamp + offset operations

xref statsmodels/statsmodels#3374

Author: jreback

doc/source/whatsnew/v0.20.0.txt

@@ -355,6 +355,7 @@ Bug Fixes
 - Bug in ``DataFrame.groupby().describe()`` when grouping on ``Index`` containing tuples (:issue:`14848`)
 - Bug in creating a ``MultiIndex`` with tuples and not passing a list of names; this will now raise ``ValueError`` (:issue:`15110`)
 
+- Bug in catching an overflow in ``Timestamp`` + ``Timedelta/Offset`` operations (:issue:`15126`)
 
 
 

pandas/tseries/offsets.py

@@ -2710,6 +2710,9 @@ def __add__(self, other):
  return self.apply(other)
  except ApplyTypeError:
  return NotImplemented
+ except OverflowError:
+ raise OverflowError("the add operation between {} and {} "
+ "will overflow".format(self, other))
 
  def __eq__(self, other):
  if isinstance(other, compat.string_types):
@@ -2748,14 +2751,20 @@ def nanos(self):
 
  def apply(self, other):
  # Timestamp can handle tz and nano sec, thus no need to use apply_wraps
- if isinstance(other, (datetime, np.datetime64, date)):
+ if isinstance(other, Timestamp):
+ result = as_timestamp(other).__add__(self)
+ if result == NotImplemented:
+ raise OverflowError
+ return result
+ elif isinstance(other, (datetime, np.datetime64, date)):
  return as_timestamp(other) + self
+
  if isinstance(other, timedelta):
  return other + self.delta
  elif isinstance(other, type(self)):
  return type(self)(self.n + other.n)
- else:
-  raise ApplyTypeError('Unhandled type: %s' % type(other).__name__)
+
+ raise ApplyTypeError('Unhandled type: %s' % type(other).__name__)
 
  _prefix = 'undefined'
 

pandas/tseries/tests/test_timedeltas.py

@@ -971,6 +971,11 @@ def test_overflow(self):
  s2 = s[0:1000]
  result = (s2 - s2.min()).sum()
 
+ def test_overflow_on_construction(self):
+ # xref https://github.com/statsmodels/statsmodels/issues/3374
+ value = pd.Timedelta('1day').value * 20169940
+ self.assertRaises(OverflowError, pd.Timedelta, value)
+
  def test_timedelta_ops_scalar(self):
  # GH 6808
  base = pd.to_datetime('20130101 09:01:12.123456')

pandas/tseries/tests/test_timeseries.py

@@ -3320,6 +3320,21 @@ def test_datetime64_with_DateOffset(self):
  # expected = rng + timedelta(2)
  # self.assertTrue(result.equals(expected))
 
+ def test_really_large_offset(self):
+ # xref https://github.com/statsmodels/statsmodels/issues/3374
+ # ends up multiplying really large numbers which overflow
+
+ t = Timestamp('2017-01-13 00:00:00', freq='D')
+ offset = 20169940 * pd.offsets.Day(1)
+
+ def f():
+ t + offset
+ self.assertRaises(OverflowError, f)
+
+ def f():
+ offset + t
+ self.assertRaises(OverflowError, f)
+
  def test_get_duplicates(self):
  idx = DatetimeIndex(['2000-01-01', '2000-01-02', '2000-01-02',
  '2000-01-03', '2000-01-03', '2000-01-04'])

pandas/tslib.pyx

@@ -1220,7 +1220,10 @@ cdef class _Timestamp(datetime):
  return Timestamp((self.freq * other).apply(self), freq=self.freq)
 
  elif isinstance(other, timedelta) or hasattr(other, 'delta'):
- nanos = _delta_to_nanoseconds(other)
+ try:
+ nanos = _delta_to_nanoseconds(other)
+ except OverflowError:
+ return NotImplemented
  result = Timestamp(self.value + nanos,
  tz=self.tzinfo, freq=self.freq)
  if getattr(other, 'normalize', False):
@@ -3125,7 +3128,7 @@ class Timedelta(_Timedelta):
  if not (is_integer_object(other) or is_float_object(other)):
  return NotImplemented
 
- return Timedelta(other *self.value, unit='ns')
+ return Timedelta(other * self.value, unit='ns')
 
  __rmul__ = __mul__