All previously existing answers are outdated or were inefficient to begin with.
Assuming you want to return three integer columns.
PL/pgSQL function
Here’s how you do it with modern PL/pgSQL (PostgreSQL 8.4 or later):
CREATE OR REPLACE FUNCTION f_foo() -- (open_id numeric) -- parameter not used
RETURNS TABLE (a int, b int, c int) AS
$func$
BEGIN
RETURN QUERY VALUES
(1,2,3)
, (3,4,5)
, (3,4,5)
;
END
$func$ LANGUAGE plpgsql IMMUTABLE ROWS 3;
In Postgres 9.6 or later you can also add PARALLEL SAFE.
Call:
SELECT * FROM f_foo();
Major points
-
Use
RETURNS TABLEto define an ad-hoc row type to return.
OrRETURNS SETOF mytblto use a pre-defined row type. -
Use
RETURN QUERYto return multiple rows with one command. -
Use a
VALUESexpression to enter multiple rows manually. This is standard SQL and has been around for ever. -
If you actually need a parameter, use a parameter name
(open_id numeric)instead ofALIAS, which is discouraged. In the example the parameter wasn’t used and just noise … -
No need for double-quoting perfectly legal identifiers. Double-quotes are only needed to force otherwise illegal names (mixed-case, illegal characters or reserved words).
-
Function volatility can be
IMMUTABLE, since the result never changes. -
ROWS 3is optional, but since we know how many rows are returned, we might as well declare it to Postgres. Can help the query planner to pick the best plan.
Simple SQL
For a simple case like this, you can use a plain SQL statement instead:
VALUES (1,2,3), (3,4,5), (3,4,5)
Or, if you want (or have) to define specific column names and types:
SELECT *
FROM (
VALUES (1::int, 2::int, 3::int)
, (3, 4, 5)
, (3, 4, 5)
) AS t(a, b, c);
SQL function
You can wrap it into a simple SQL function instead:
CREATE OR REPLACE FUNCTION f_foo()
RETURNS TABLE (a int, b int, c int) AS
$func$
VALUES (1, 2, 3)
, (3, 4, 5)
, (3, 4, 5);
$func$ LANGUAGE sql IMMUTABLE ROWS 3;