BUG: DataFrame.replace with out of bound datetime causing RecursionEr…

…ror (pandas-dev#22108)

doc/source/whatsnew/v0.24.0.txt

@@ -654,7 +654,7 @@ Reshaping
 - Bug in :meth:`Series.combine_first` with ``datetime64[ns, tz]`` dtype which would return tz-naive result (:issue:`21469`)
 - Bug in :meth:`Series.where` and :meth:`DataFrame.where` with ``datetime64[ns, tz]`` dtype (:issue:`21546`)
 - Bug in :meth:`Series.mask` and :meth:`DataFrame.mask` with ``list`` conditionals (:issue:`21891`)
--
+- Bug in :meth:`DataFrame.replace` raises RecursionError when converting OutOfBounds ``datetime64[ns, tz]`` (:issue:`20380`)
 -
 
 Build Changes

pandas/core/dtypes/cast.py

@@ -6,7 +6,7 @@
 import warnings
 
 from pandas._libs import tslib, lib, tslibs
-from pandas._libs.tslibs import iNaT
+from pandas._libs.tslibs import iNaT, OutOfBoundsDatetime
 from pandas.compat import string_types, text_type, PY3
 from .common import (ensure_object, is_bool, is_integer, is_float,
                      is_complex, is_datetimetz, is_categorical_dtype,
@@ -838,7 +838,13 @@ def soft_convert_objects(values, datetime=True, numeric=True, timedelta=True,
 
     # Soft conversions
     if datetime:
-        values = lib.maybe_convert_objects(values, convert_datetime=datetime)
+        # GH 20380, when datetime is beyond year 2262, hence outside
+        # bound of nanosecond-resolution 64-bit integers.
+        try:
+            values = lib.maybe_convert_objects(values,
+                                               convert_datetime=datetime)
+        except OutOfBoundsDatetime:
+            pass
 
     if timedelta and is_object_dtype(values.dtype):
         # Object check to ensure only run if previous did not convert

pandas/core/internals/blocks.py

@@ -802,12 +802,14 @@ def replace(self, to_replace, value, inplace=False, filter=None,
                                     copy=not inplace) for b in blocks]
             return blocks
         except (TypeError, ValueError):
-
             # try again with a compatible block
             block = self.astype(object)
-            return block.replace(
-                to_replace=original_to_replace, value=value, inplace=inplace,
-                filter=filter, regex=regex, convert=convert)
+            return block.replace(to_replace=original_to_replace,
+                                 value=value,
+                                 inplace=inplace,
+                                 filter=filter,
+                                 regex=regex,
+                                 convert=convert)
 
     def _replace_single(self, *args, **kwargs):
         """ no-op on a non-ObjectBlock """

pandas/tests/frame/test_replace.py

@@ -755,40 +755,37 @@ def test_replace_for_new_dtypes(self):
         result = tsframe.fillna(method='bfill')
         assert_frame_equal(result, tsframe.fillna(method='bfill'))
 
-    def test_replace_dtypes(self):
-        # int
-        df = DataFrame({'ints': [1, 2, 3]})
-        result = df.replace(1, 0)
-        expected = DataFrame({'ints': [0, 2, 3]})
-        assert_frame_equal(result, expected)
-
-        df = DataFrame({'ints': [1, 2, 3]}, dtype=np.int32)
-        result = df.replace(1, 0)
-        expected = DataFrame({'ints': [0, 2, 3]}, dtype=np.int32)
-        assert_frame_equal(result, expected)
-
-        df = DataFrame({'ints': [1, 2, 3]}, dtype=np.int16)
-        result = df.replace(1, 0)
-        expected = DataFrame({'ints': [0, 2, 3]}, dtype=np.int16)
-        assert_frame_equal(result, expected)
-
-        # bools
-        df = DataFrame({'bools': [True, False, True]})
-        result = df.replace(False, True)
-        assert result.values.all()
-
-        # complex blocks
-        df = DataFrame({'complex': [1j, 2j, 3j]})
-        result = df.replace(1j, 0j)
-        expected = DataFrame({'complex': [0j, 2j, 3j]})
-        assert_frame_equal(result, expected)
-
-        # datetime blocks
-        prev = datetime.today()
-        now = datetime.today()
-        df = DataFrame({'datetime64': Index([prev, now, prev])})
-        result = df.replace(prev, now)
-        expected = DataFrame({'datetime64': Index([now] * 3)})
+    @pytest.mark.parametrize('frame, to_replace, value, expected', [
+        (DataFrame({'ints': [1, 2, 3]}), 1, 0,
+         DataFrame({'ints': [0, 2, 3]})),
+        (DataFrame({'ints': [1, 2, 3]}, dtype=np.int32), 1, 0,
+         DataFrame({'ints': [0, 2, 3]}, dtype=np.int32)),
+        (DataFrame({'ints': [1, 2, 3]}, dtype=np.int16), 1, 0,
+         DataFrame({'ints': [0, 2, 3]}, dtype=np.int16)),
+        (DataFrame({'bools': [True, False, True]}), False, True,
+         DataFrame({'bools': [True, True, True]})),
+        (DataFrame({'complex': [1j, 2j, 3j]}), 1j, 0,
+         DataFrame({'complex': [0j, 2j, 3j]})),
+        (DataFrame({'datetime64': Index([datetime(2018, 5, 28),
+                                         datetime(2018, 7, 28),
+                                         datetime(2018, 5, 28)])}),
+         datetime(2018, 5, 28), datetime(2018, 7, 28),
+         DataFrame({'datetime64': Index([datetime(2018, 7, 28)] * 3)})),
+        # GH 20380
+        (DataFrame({'dt': [datetime(3017, 12, 20)], 'str': ['foo']}),
+         'foo', 'bar',
+         DataFrame({'dt': [datetime(3017, 12, 20)], 'str': ['bar']})),
+        (DataFrame({'A': date_range('20130101', periods=3, tz='US/Eastern'),
+                    'B': [0, np.nan, 2]}),
+         Timestamp('20130102', tz='US/Eastern'),
+         Timestamp('20130104', tz='US/Eastern'),
+         DataFrame({'A': [Timestamp('20130101', tz='US/Eastern'),
+                          Timestamp('20130104', tz='US/Eastern'),
+                          Timestamp('20130103', tz='US/Eastern')],
+                    'B': [0, np.nan, 2]}))
+    ])
+    def test_replace_dtypes(self, frame, to_replace, value, expected):
+        result = getattr(frame, 'replace')(to_replace, value)
         assert_frame_equal(result, expected)
 
     def test_replace_input_formats_listlike(self):