I like the first style, myself, and in fact it's similar in concept to some ideas I'd sketched out months ago when Neophytos and I started talking about getting rid of the need to laboriously write Oracle and PG DML statements to make use of service contracts. It seems more in the style of today's ad_page_contract.
As Jon says, I don't think there will be much need to build service contract definitions dynamically. Even if it is a pain it's still going to be easier than it would be in today's SQL-based approach. And AFAIK no one's ever attempted to generate SC defs dynamically using the current implementation!
I think Tilmann's idea of extending ad_proc (not ad_form!) with an "-implements" switch is an interesting one ...
