BUG: Format mismatch doesn't coerce to NaT

doc/source/whatsnew/v0.24.0.rst

@@ -1548,6 +1548,7 @@ Datetimelike
 - Bug in :meth:`DatetimeIndex.astype`, :meth:`PeriodIndex.astype` and :meth:`TimedeltaIndex.astype` ignoring the sign of the ``dtype`` for unsigned integer dtypes (:issue:`24405`).
 - Fixed bug in :meth:`Series.max` with ``datetime64[ns]``-dtype failing to return ``NaT`` when nulls are present and ``skipna=False`` is passed (:issue:`24265`)
 - Bug in :func:`to_datetime` where arrays of ``datetime`` objects containing both timezone-aware and timezone-naive ``datetimes`` would fail to raise ``ValueError`` (:issue:`24569`)
+- Bug in :func:`to_datetime` with invalid datetime format doesn't coerce input to ``NaT`` even if ``errors='coerce'`` (:issue:`24763`)
 
 Timedelta
 ^^^^^^^^^

pandas/core/tools/datetimes.py

@@ -265,15 +265,25 @@ def _convert_listlike_datetimes(arg, box, format, name=None, tz=None,
                 except tslibs.OutOfBoundsDatetime:
                     if errors == 'raise':
                         raise
-                    result = arg
+                    elif errors == 'coerce':
+                        result = np.empty(arg.shape, dtype='M8[ns]')
+                        iresult = result.view('i8')
+                        iresult.fill(tslibs.iNaT)
+                    else:
+                        result = arg
                 except ValueError:
                     # if format was inferred, try falling back
                     # to array_to_datetime - terminate here
                     # for specified formats
                     if not infer_datetime_format:
                         if errors == 'raise':
                             raise
-                        result = arg
+                        elif errors == 'coerce':
+                            result = np.empty(arg.shape, dtype='M8[ns]')
+                            iresult = result.view('i8')
+                            iresult.fill(tslibs.iNaT)
+                        else:
+                            result = arg
         except ValueError as e:
             # Fallback to try to convert datetime objects if timezone-aware
             #  datetime objects are found without passing `utc=True`

pandas/tests/indexes/datetimes/test_tools.py

@@ -555,6 +555,63 @@ def test_datetime_invalid_datatype(self):
         with pytest.raises(TypeError):
             pd.to_datetime(pd.to_datetime)
 
+    @pytest.mark.parametrize('value', ["a", "00:01:99"])
+    @pytest.mark.parametrize('infer', [True, False])
+    @pytest.mark.parametrize('format', [None, 'H%:M%:S%'])
+    def test_datetime_invalid_scalar(self, value, format, infer):
+        # GH24763
+        res = pd.to_datetime(value, errors='ignore', format=format,
+                             infer_datetime_format=infer)
+        assert res == value
+
+        res = pd.to_datetime(value, errors='coerce', format=format,
+                             infer_datetime_format=infer)
+        assert res is pd.NaT
+
+        with pytest.raises(ValueError):
+            pd.to_datetime(value, errors='raise', format=format,
+                           infer_datetime_format=infer)
+
+    @pytest.mark.parametrize('value', ["3000/12/11 00:00:00"])
+    @pytest.mark.parametrize('infer', [True, False])
+    @pytest.mark.parametrize('format', [None, 'H%:M%:S%'])
+    def test_datetime_outofbounds_scalar(self, value, format, infer):
+        # GH24763
+        res = pd.to_datetime(value, errors='ignore', format=format,
+                             infer_datetime_format=infer)
+        assert res == value
+
+        res = pd.to_datetime(value, errors='coerce', format=format,
+                             infer_datetime_format=infer)
+        assert res is pd.NaT
+
+        if format is not None:
+            with pytest.raises(ValueError):
+                pd.to_datetime(value, errors='raise', format=format,
+                               infer_datetime_format=infer)
+        else:
+            with pytest.raises(OutOfBoundsDatetime):
+                pd.to_datetime(value, errors='raise', format=format,
+                               infer_datetime_format=infer)
+
+    @pytest.mark.parametrize('values', [["a"], ["00:01:99"],
+                                        ["a", "b", "99:00:00"]])
+    @pytest.mark.parametrize('infer', [True, False])
+    @pytest.mark.parametrize('format', [None, 'H%:M%:S%'])
+    def test_datetime_invalid_index(self, values, format, infer):
+        # GH24763
+        res = pd.to_datetime(values, errors='ignore', format=format,
+                             infer_datetime_format=infer)
+        tm.assert_index_equal(res, pd.Index(values))
+
+        res = pd.to_datetime(values, errors='coerce', format=format,
+                             infer_datetime_format=infer)
+        tm.assert_index_equal(res, pd.DatetimeIndex([pd.NaT] * len(values)))
+
+        with pytest.raises(ValueError):
+            pd.to_datetime(values, errors='raise', format=format,
+                           infer_datetime_format=infer)
+
     @pytest.mark.parametrize("utc", [True, None])
     @pytest.mark.parametrize("format", ['%Y%m%d %H:%M:%S', None])
     @pytest.mark.parametrize("box", [True, False])