It seems curious that you have code in make_signed_type to max precision at 2 * HOST_BITS_PER_WIDE_INT, but you do not have such a limit in the nearly identical code in make_unsigned_type. Shouldn't it be in both places? Jim