Let's have some record stored in a longer format while the current format is shorter. If this record is updated multiple times within a single transaction and then a savepoint cleanup happens (e.g. due to some runtime error related to another data), then a bugcheck can happen. This seems to be caused by the lack of sync between actual record length on disk (new version will have a short format) and rec_length field of new_rpb->rpb_record which may keep the longer value (from some prior operations with another format).