BUG: to_datetime fails with np.datetime64 and non-ISO format (#50038)

Co-authored-by: MarcoGorelli <>

Author: MarcoGorelli

doc/source/whatsnew/v2.0.0.rst

@@ -658,7 +658,7 @@ Datetimelike
 - Bug in subtracting a ``datetime`` scalar from :class:`DatetimeIndex` failing to retain the original ``freq`` attribute (:issue:`48818`)
 - Bug in ``pandas.tseries.holiday.Holiday`` where a half-open date interval causes inconsistent return types from :meth:`USFederalHolidayCalendar.holidays` (:issue:`49075`)
 - Bug in rendering :class:`DatetimeIndex` and :class:`Series` and :class:`DataFrame` with timezone-aware dtypes with ``dateutil`` or ``zoneinfo`` timezones near daylight-savings transitions (:issue:`49684`)
-- Bug in :func:`to_datetime` was raising ``ValueError`` when parsing :class:`Timestamp` or ``datetime`` objects with non-ISO8601 ``format`` (:issue:`49298`)
+- Bug in :func:`to_datetime` was raising ``ValueError`` when parsing :class:`Timestamp`, ``datetime``, or ``np.datetime64`` objects with non-ISO8601 ``format`` (:issue:`49298`, :issue:`50036`)
 -
 
 Timedelta

pandas/_libs/tslibs/strptime.pyx

@@ -14,13 +14,17 @@ from _thread import allocate_lock as _thread_allocate_lock
 import numpy as np
 import pytz
 
+cimport numpy as cnp
 from numpy cimport (
     int64_t,
     ndarray,
 )
 
 from pandas._libs.missing cimport checknull_with_nat_and_na
-from pandas._libs.tslibs.conversion cimport convert_timezone
+from pandas._libs.tslibs.conversion cimport (
+    convert_timezone,
+    get_datetime64_nanos,
+)
 from pandas._libs.tslibs.nattype cimport (
     NPY_NAT,
     c_nat_strings as nat_strings,
@@ -33,6 +37,9 @@ from pandas._libs.tslibs.np_datetime cimport (
     pydatetime_to_dt64,
 )
 from pandas._libs.tslibs.timestamps cimport _Timestamp
+from pandas._libs.util cimport is_datetime64_object
+
+cnp.import_array()
 
 
 cdef dict _parse_code_table = {"y": 0,
@@ -166,6 +173,9 @@ def array_strptime(
                 check_dts_bounds(&dts)
             result_timezone[i] = val.tzinfo
             continue
+        elif is_datetime64_object(val):
+            iresult[i] = get_datetime64_nanos(val, NPY_FR_ns)
+            continue
         else:
             val = str(val)
 

pandas/tests/tools/test_to_datetime.py

@@ -758,6 +758,19 @@ def test_to_datetime_today_now_unicode_bytes(self, arg):
     def test_to_datetime_dt64s(self, cache, dt):
         assert to_datetime(dt, cache=cache) == Timestamp(dt)
 
+    @pytest.mark.parametrize(
+        "arg, format",
+        [
+            ("2001-01-01", "%Y-%m-%d"),
+            ("01-01-2001", "%d-%m-%Y"),
+        ],
+    )
+    def test_to_datetime_dt64s_and_str(self, arg, format):
+        # https://github.com/pandas-dev/pandas/issues/50036
+        result = to_datetime([arg, np.datetime64("2020-01-01")], format=format)
+        expected = DatetimeIndex(["2001-01-01", "2020-01-01"])
+        tm.assert_index_equal(result, expected)
+
     @pytest.mark.parametrize(
         "dt", [np.datetime64("1000-01-01"), np.datetime64("5000-01-02")]
     )