In _purge_history_txn, ensure that txn.fetchall has elements before accessing rows

changelog.d/10690.bugfix

@@ -0,0 +1 @@
+Fix a long-standing bug that caused an `AssertionError` when purging history in certain rooms. Contributed by @Kokokokoka.

synapse/storage/databases/main/purge_events.py

@@ -102,15 +102,19 @@ def _purge_history_txn(
             (room_id,),
         )
         rows = txn.fetchall()
-        max_depth = max(row[1] for row in rows)
-
-        if max_depth < token.topological:
-            # We need to ensure we don't delete all the events from the database
-            # otherwise we wouldn't be able to send any events (due to not
-            # having any backwards extremities)
-            raise SynapseError(
-                400, "topological_ordering is greater than forward extremeties"
-            )
+        # if we already have no forwards extremities (for example because they were 
+        # cleared out by the `delete_old_current_state_events` background database 
+        # update), then we may as well carry on.
+        if rows:
+            max_depth = max(row[1] for row in rows)
+
+            if max_depth < token.topological:
+                # We need to ensure we don't delete all the events from the database
+                # otherwise we wouldn't be able to send any events (due to not
+                # having any backwards extremities)
+                raise SynapseError(
+                    400, "topological_ordering is greater than forward extremities"
+                )
 
         logger.info("[purge] looking for events to delete")