david.g.johnston@gmail.com wrote:
On Monday, December 13, 2021, Bryn Llewellyn <bryn@yugabyte.com> wrote:
There must be a reason to prefer a “language sql” procedure over a “language plpgsql” procedure—otherwise the former wouldn’t be supported.
I would say that is true for functions. I wouldn’t assume that for procedures—it’s probable that because sql already worked for functions we got that feature for free when implementing procedures.
Interesting. That’s exactly the kind of historical insight I was after. Thanks.
It’s very tempting to think that “language sql” is meaningful only as a performance feature and in that connection only for a stored function because only a function can be inlined in a surrounding regular SQL statement. (You can invoke a procedure only as a singleton in the dedicated “call” statement.) In other words there can be no inlining benefit for a stored procedure.
It’s certainly no problem for the coder to bracket what would have been the body of a “language sql” DO block with a single “begin… end;”.
I should save any of you the effort of telling me this: a DO block is an anonymous, ephemeral procedure. It’s certainly not an anonymous function.