Actually, there is a plpgsql CASE
statement, not to be confused with the SQL CASE
expression:
CREATE OR REPLACE function test(n integer)
RETURNS TABLE (name text, city text) AS
$func$
BEGIN
CASE n
WHEN 1 THEN
RETURN QUERY SELECT t.name, t.city FROM table1 t;
WHEN 2 THEN
RETURN QUERY SELECT t.foo, t.bar FROM table2 t;
WHEN 3 THEN
RETURN QUERY SELECT t.bar, t.bamm FROM view1 t;
END CASE;
END
$func$ LANGUAGE plpgsql;
If you declare your function as
RETURNS TABLE (name text, city text)
, then your SELECT statements should have a column list with matching types.
If on the other hand you want toSELECT *
, declare the function asRETURNS SETOF table1
accordingly.When naming columns in the return type, those variable are visible in the function body. Be sure to table-qualify column names that would conflict with same names. So
t.name
instead of justname
.Column names from queries inside the function are not visible outside. Only the declared return type. So names don't have to match, just data types.
Either way, I suggest to simplify things:
CREATE OR REPLACE function test(n integer)
RETURNS SETOF table1 AS
$func$
SELECT * FROM table1 t WHERE n = 1
UNION ALL
SELECT * FROM table2 t WHERE n = 2
UNION ALL
SELECT * FROM view1 t WHERE n = 3;
$func$ LANGUAGE sql;
Same result. Just as fast. SQL or PL/pgSQL function is a matter of taste and some other details. PL/pgSQL is probably faster for stand-alone calls. SQL can more easily be nested.