fix: Mismatch between timezone-aware and naive datetimes in start date and bookmarks is now correctly handled

singer_sdk/streams/core.py

@@ -365,6 +365,23 @@ def _write_replication_key_signpost(
         state = self.get_context_state(context)
         write_replication_key_signpost(state, value)
 
+    def _parse_datetime(self, value: str) -> datetime.datetime:  # noqa: PLR6301
+        """Parse a datetime string.
+
+        Args:
+            value: The datetime string.
+
+        Returns:
+            The parsed datetime, timezone-aware preferred.
+        """
+        result = datetime_fromisoformat(value)
+
+        # Ensure datetime is timezone-aware
+        if not result.tzinfo:
+            result = result.astimezone()
+
+        return result
+
     def compare_start_date(self, value: str, start_date_value: str) -> str:
         """Compare a bookmark value to a start date and return the most recent value.
 
@@ -384,7 +401,7 @@ def compare_start_date(self, value: str, start_date_value: str) -> str:
             The most recent value between the bookmark and start date.
         """
         if self.is_timestamp_replication_key:
-            return max(value, start_date_value, key=datetime_fromisoformat)
+            return max(value, start_date_value, key=self._parse_datetime)
 
         return value
 

tests/core/test_streams.py

@@ -141,6 +141,18 @@ def test_stream_apply_catalog(stream: Stream):
             parse(CONFIG_START_DATE).replace(tzinfo=datetime.timezone.utc),
             id="datetime-repl-key-old-bookmark",
         ),
+        pytest.param(
+            "test",
+            None,
+            "2021-01-02T00:00:00-08:00",
+            datetime.datetime(
+                2021,
+                1,
+                2,
+                tzinfo=datetime.timezone(datetime.timedelta(hours=-8)),
+            ),
+            id="datetime-repl-key-recent-bookmark-tz-aware",
+        ),
         pytest.param(
             "unix_ts",
             None,