BUG: to_datetime raising on invalid offsets with errors=coerce and infer_datetime_format

Author: MarcoGorelli

doc/source/whatsnew/v1.6.0.rst

@@ -162,6 +162,7 @@ Categorical
 Datetimelike
 ^^^^^^^^^^^^
 - Bug in :func:`pandas.infer_freq`, raising ``TypeError`` when inferred on :class:`RangeIndex` (:issue:`47084`)
+- Bug in :func:`to_datetime` was raising on invalid offsets with ``errors='coerce'`` and ``infer_datetime_format=True`` (:issue:`48633`)
 -
 
 Timedelta

pandas/_libs/tslibs/parsing.pyx

@@ -943,7 +943,7 @@ def format_is_iso(f: str) -> bint:
     return False
 
 
-def guess_datetime_format(dt_str, bint dayfirst=False):
+def guess_datetime_format(dt_str: str, bint dayfirst=False) -> str | None:
     """
     Guess the datetime format of a given datetime string.
 
@@ -1026,7 +1026,12 @@ def guess_datetime_format(dt_str, bint dayfirst=False):
             # This separation will prevent subsequent processing
             # from correctly parsing the time zone format.
             # So in addition to the format nomalization, we rejoin them here.
-            tokens[offset_index] = parsed_datetime.strftime("%z")
+            try:
+                tokens[offset_index] = parsed_datetime.strftime("%z")
+            except ValueError:
+                # Invalid offset might not have raised in du_parse
+                # https://github.com/dateutil/dateutil/issues/188
+                return None
             tokens = tokens[:offset_index + 1 or None]
 
     format_guess = [None] * len(tokens)

pandas/tests/tools/test_to_datetime.py

@@ -1143,10 +1143,13 @@ def test_to_datetime_coerce(self):
         )
         tm.assert_index_equal(result, expected)
 
-    def test_to_datetime_coerce_malformed(self):
+    @pytest.mark.parametrize("infer_datetime_format", [True, False])
+    def test_to_datetime_coerce_malformed(self, infer_datetime_format):
         # GH 28299
         ts_strings = ["200622-12-31", "111111-24-11"]
-        result = to_datetime(ts_strings, errors="coerce")
+        result = to_datetime(
+            ts_strings, errors="coerce", infer_datetime_format=infer_datetime_format
+        )
         expected = Index([NaT, NaT])
         tm.assert_index_equal(result, expected)