06.09.2014 19:12, Jan Wieck kirjoitti:
> On 09/06/2014 04:21 AM, Marko Tiikkaja wrote:
>> We wrap these things into (sometimes) simple-looking function so that
>> none of the application developers ever run any SQL. We define an
>> interface between the application and the database, and that interface
>> is implemented using PL/PgSQL functions. Sure, sometimes one function
>> will just fire off a single UPDATE .. RETURNING, or a SELECT, but that
>> doesn't matter. The trick is to be consistent everywhere.
>
> There is precisely your root problem. Instead of educating your
> application developers on how to properly use a relational database
> system, you try to make it foolproof.
There are also other reasons to wrap everything in functions, for
example sharding using pl/proxy which by the way always throws an error
if a SELECT didn't match exactly one row and the function wasn't
declared returning 'SETOF' (although it currently doesn't set any
sqlstate for these errors making it a bit difficult to properly catch them.)
Anyway, I think the discussed feature to make select, update and delete
throw an error if they returned or modified <> 1 row would be more
useful as an extension of the basic sql statements instead of a plpgsql
(2) only feature to make it possible to use it from other languages and
outside functions.
/ Oskari
