BUG: Do not fail when parsing pydatetime objects in pd.to_datetime (#49893)

* parent 0168e27
author Antonio Ossa Guerra <[email protected]> 1666800993 -0300
committer MarcoGorelli <> 1669293453 +0000

Parse `datetime` properly in `pd.to_datetime`

When applying `pd.to_datetime` on array-like structure that contain a
`datetime.datetime` object, while using the `format` argument, a
`ValueError` is raised because the `datetime.datetime` object does not
match the expected format.

The implemented solution looks for `datetime.datetime` instances in the
`array_strptime` method. If an instance of this type is found, it's
properly handled by the new `_parse_python_datetime_object`, which
returns the expected Numpy datetime object.

Signed-off-by: Antonio Ossa Guerra <[email protected]>

* fixup

* 🏷️ typing

* ignore pylint nitpick

* better naming

* keep use-a-generator check

* use fromisoformat

* change awareness to be UTC

* Revert "use fromisoformat"

This reverts commit af272e1.

* rename input to args

Signed-off-by: Antonio Ossa Guerra <[email protected]>
Co-authored-by: Antonio Ossa Guerra <[email protected]>
Co-authored-by: MarcoGorelli <>

Author: MarcoGorelli

doc/source/whatsnew/v2.0.0.rst

@@ -638,6 +638,7 @@ 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`)
 -
 
 Timedelta

pandas/_libs/tslibs/strptime.pyi

@@ -7,6 +7,7 @@ def array_strptime(
     fmt: str | None,
     exact: bool = ...,
     errors: str = ...,
+    utc: bool = ...,
 ) -> tuple[np.ndarray, np.ndarray]: ...
 
 # first ndarray is M8[ns], second is object ndarray of tzinfo | None

pandas/_libs/tslibs/strptime.pyx

@@ -1,10 +1,14 @@
 """Strptime-related classes and functions.
 """
 from cpython.datetime cimport (
+    PyDateTime_Check,
     date,
+    import_datetime,
     tzinfo,
 )
 
+import_datetime()
+
 from _thread import allocate_lock as _thread_allocate_lock
 
 import numpy as np
@@ -16,6 +20,7 @@ from numpy cimport (
 )
 
 from pandas._libs.missing cimport checknull_with_nat_and_na
+from pandas._libs.tslibs.conversion cimport convert_timezone
 from pandas._libs.tslibs.nattype cimport (
     NPY_NAT,
     c_nat_strings as nat_strings,
@@ -25,7 +30,9 @@ from pandas._libs.tslibs.np_datetime cimport (
     check_dts_bounds,
     npy_datetimestruct,
     npy_datetimestruct_to_datetime,
+    pydatetime_to_dt64,
 )
+from pandas._libs.tslibs.timestamps cimport _Timestamp
 
 
 cdef dict _parse_code_table = {'y': 0,
@@ -53,7 +60,13 @@ cdef dict _parse_code_table = {'y': 0,
                                'u': 22}
 
 
-def array_strptime(ndarray[object] values, str fmt, bint exact=True, errors='raise'):
+def array_strptime(
+    ndarray[object] values,
+    str fmt,
+    bint exact=True,
+    errors='raise',
+    bint utc=False,
+):
     """
     Calculates the datetime structs represented by the passed array of strings
 
@@ -78,6 +91,9 @@ def array_strptime(ndarray[object] values, str fmt, bint exact=True, errors='rai
         bint is_raise = errors=='raise'
         bint is_ignore = errors=='ignore'
         bint is_coerce = errors=='coerce'
+        bint found_naive = False
+        bint found_tz = False
+        tzinfo tz_out = None
 
     assert is_raise or is_ignore or is_coerce
 
@@ -128,12 +144,30 @@ def array_strptime(ndarray[object] values, str fmt, bint exact=True, errors='rai
             if val in nat_strings:
                 iresult[i] = NPY_NAT
                 continue
-        else:
-            if checknull_with_nat_and_na(val):
-                iresult[i] = NPY_NAT
-                continue
+        elif checknull_with_nat_and_na(val):
+            iresult[i] = NPY_NAT
+            continue
+        elif PyDateTime_Check(val):
+            if val.tzinfo is not None:
+                found_tz = True
+            else:
+                found_naive = True
+            tz_out = convert_timezone(
+                val.tzinfo,
+                tz_out,
+                found_naive,
+                found_tz,
+                utc,
+            )
+            if isinstance(val, _Timestamp):
+                iresult[i] = val.tz_localize(None).as_unit("ns").value
             else:
-                val = str(val)
+                iresult[i] = pydatetime_to_dt64(val.replace(tzinfo=None), &dts)
+                check_dts_bounds(&dts)
+            result_timezone[i] = val.tzinfo
+            continue
+        else:
+            val = str(val)
 
         # exact matching
         if exact:

pandas/core/tools/datetimes.py

@@ -318,7 +318,14 @@ def _return_parsed_timezone_results(
     )
     if utc:
         # Convert to the same tz
-        tz_results = np.array([tz_result.tz_convert("utc") for tz_result in tz_results])
+        tz_results = np.array(
+            [
+                tz_result.tz_convert("utc")
+                if tz_result.tzinfo is not None
+                else tz_result.tz_localize("utc")
+                for tz_result in tz_results
+            ]
+        )
 
     return Index(tz_results, name=name)
 
@@ -468,7 +475,9 @@ def _array_strptime_with_fallback(
     Call array_strptime, with fallback behavior depending on 'errors'.
     """
     try:
-        result, timezones = array_strptime(arg, fmt, exact=exact, errors=errors)
+        result, timezones = array_strptime(
+            arg, fmt, exact=exact, errors=errors, utc=utc
+        )
     except OutOfBoundsDatetime:
         if errors == "raise":
             raise
@@ -495,7 +504,7 @@ def _array_strptime_with_fallback(
             # Indicates to the caller to fallback to objects_to_datetime64ns
             return None
     else:
-        if "%Z" in fmt or "%z" in fmt:
+        if any(tz is not None for tz in timezones):
             return _return_parsed_timezone_results(result, timezones, utc, name)
 
     return _box_as_indexlike(result, utc=utc, name=name)

pandas/tests/tools/test_to_datetime.py

@@ -469,6 +469,88 @@ def test_to_datetime_mixed_datetime_and_string(self):
         expected = to_datetime([d1, d2]).tz_convert(pytz.FixedOffset(-60))
         tm.assert_index_equal(res, expected)
 
+    @pytest.mark.parametrize(
+        "fmt",
+        ["%Y-%d-%m %H:%M:%S%z", "%Y-%m-%d %H:%M:%S%z"],
+        ids=["non-ISO8601 format", "ISO8601 format"],
+    )
+    @pytest.mark.parametrize(
+        "utc, args, expected",
+        [
+            pytest.param(
+                True,
+                ["2000-01-01 01:00:00-08:00", "2000-01-01 02:00:00-08:00"],
+                DatetimeIndex(
+                    ["2000-01-01 09:00:00+00:00", "2000-01-01 10:00:00+00:00"],
+                    dtype="datetime64[ns, UTC]",
+                ),
+                id="all tz-aware, with utc",
+            ),
+            pytest.param(
+                False,
+                ["2000-01-01 01:00:00+00:00", "2000-01-01 02:00:00+00:00"],
+                DatetimeIndex(
+                    ["2000-01-01 01:00:00+00:00", "2000-01-01 02:00:00+00:00"],
+                ),
+                id="all tz-aware, without utc",
+            ),
+            pytest.param(
+                True,
+                ["2000-01-01 01:00:00-08:00", "2000-01-01 02:00:00+00:00"],
+                DatetimeIndex(
+                    ["2000-01-01 09:00:00+00:00", "2000-01-01 02:00:00+00:00"],
+                    dtype="datetime64[ns, UTC]",
+                ),
+                id="all tz-aware, mixed offsets, with utc",
+            ),
+        ],
+    )
+    @pytest.mark.parametrize(
+        "constructor",
+        [Timestamp, lambda x: Timestamp(x).to_pydatetime()],
+    )
+    def test_to_datetime_mixed_datetime_and_string_with_format(
+        self, fmt, utc, args, expected, constructor
+    ):
+        # https://github.com/pandas-dev/pandas/issues/49298
+        # note: ISO8601 formats go down a fastpath, so we need to check both
+        # a ISO8601 format and a non-ISO8601 one
+        ts1 = constructor(args[0])
+        ts2 = args[1]
+        result = to_datetime([ts1, ts2], format=fmt, utc=utc)
+        tm.assert_index_equal(result, expected)
+
+    @pytest.mark.parametrize(
+        "fmt",
+        ["%Y-%d-%m %H:%M:%S%z", "%Y-%m-%d %H:%M:%S%z"],
+        ids=["non-ISO8601 format", "ISO8601 format"],
+    )
+    @pytest.mark.parametrize(
+        "args",
+        [
+            pytest.param(
+                ["2000-01-01 01:00:00-08:00", "2000-01-01 02:00:00-07:00"],
+                id="all tz-aware, mixed timezones, without utc",
+            ),
+        ],
+    )
+    @pytest.mark.parametrize(
+        "constructor",
+        [Timestamp, lambda x: Timestamp(x).to_pydatetime()],
+    )
+    def test_to_datetime_mixed_datetime_and_string_with_format_raises(
+        self, fmt, args, constructor
+    ):
+        # https://github.com/pandas-dev/pandas/issues/49298
+        # note: ISO8601 formats go down a fastpath, so we need to check both
+        # a ISO8601 format and a non-ISO8601 one
+        ts1 = constructor(args[0])
+        ts2 = constructor(args[1])
+        with pytest.raises(
+            ValueError, match="cannot be converted to datetime64 unless utc=True"
+        ):
+            to_datetime([ts1, ts2], format=fmt, utc=False)
+
     @pytest.mark.parametrize("infer_datetime_format", [True, False])
     def test_to_datetime_np_str(self, infer_datetime_format):
         # GH#32264