I've been looking at this, and my feeling is that we should drop the PROARGMODE_TABLE business and just define RETURNS TABLE(x int, y int) as exactly equivalent to RETURNS SETOF RECORD with x and y treated as OUT parameters. There isn't any advantage to distinguishing the cases that outweighs breaking client code that looks at pg_proc.proargmodes. I don't believe that the SQL spec prevents us from exposing those parameter names to PL functions, especially since none of our PLs are in the standard at all.
Reason for PROARGMODE_TABLE was protection before name's collision, and x, and y are table attributies (not variables) and then we are protected before collision. It's shortcut for
create function foo() returns setof record as ... select * from foo() as (x int, y int); Regards Pavel Stehule _________________________________________________________________ Najdete si svou lasku a nove pratele na Match.com. http://www.msn.cz/ ---------------------------(end of broadcast)--------------------------- TIP 4: Have you searched our list archives? http://archives.postgresql.org
