Thus we choose "4 * number_of_columns" as a more reasonable value (until someone comes up with something better). This pretends to properly address PR#11989.