This is a follow-up for #571. We may need to look into the Spark source code and replicate the type casting logic as UDFs. It's also worth understanding whether this casting behavior is relevant to ...