Return multiple fields as a record in PostgreSQL with PL/pgSQL
I am writing a SP, using PL/pgSQL.
I want to return a record, comprised of fields from several different tables. Could look something like this:
CREATE OR REPLACE FUNCTION get_object_fields(name text)
RETURNS RECORD AS $$
BEGIN
-- fetch fields f1, f2 and f3 from table t1
-- fetch fields f4, f5 from table t2
-- fetch fields f6, f7 and f8 from table t3
-- return fields f1 ... f8 as a record
END
$$ language plpgsql;
How may I return the fields from different tables as fields in a single record?
[Edit]
I have realized that the example I gave above was slightly too simplistic. Some of the fields I need to be retrieving, will be saved as separate rows in the database table being queried, but I want to return them in the 'flattened' record structure.
The code below should help illustrate further:
CREATE TABLE user (id int, school_id int, name varchar(32));
CREATE TYPE my_type AS (
user1_id int,
user1_name varchar(32),
user2_id int,
user2_name varchar(32)
);
CREATE OR REPLACE FUNCTION get_two_users_from_school(schoolid int)
RETURNS my_type AS $$
DECLARE
result my_type;
temp_result user;
BEGIN
-- for purpose of this question assume 2 rows returned
SELECT id, name INTO temp_result FROM user where school_id = schoolid LIMIT 2;
-- Will the (pseudo)code below work?:
result.user1_id := temp_result[0].id ;
result.user1_name := temp_result[0].name ;
result.user2_id := temp_result[1].id ;
result.user2_name := temp_result[1].name ;
return result ;
END
$$ language plpgsql
Don't use CREATE TYPE to return a polymorphic result. Use and abuse the RECORD type instead. Check it out:
CREATE FUNCTION test_ret(a TEXT, b TEXT) RETURNS RECORD AS $$
DECLARE
ret RECORD;
BEGIN
-- Arbitrary expression to change the first parameter
IF LENGTH(a) < LENGTH(b) THEN
SELECT TRUE, a || b, 'a shorter than b' INTO ret;
ELSE
SELECT FALSE, b || a INTO ret;
END IF;
RETURN ret;
END;$$ LANGUAGE plpgsql;
Pay attention to the fact that it can optionally return two or three columns depending on the input.
test=> SELECT test_ret('foo','barbaz');
test_ret
----------------------------------
(t,foobarbaz,"a shorter than b")
(1 row)
test=> SELECT test_ret('barbaz','foo');
test_ret
----------------------------------
(f,foobarbaz)
(1 row)
This does wreak havoc on code, so do use a consistent number of columns, but it's ridiculously handy for returning optional error messages with the first parameter returning the success of the operation. Rewritten using a consistent number of columns:
CREATE FUNCTION test_ret(a TEXT, b TEXT) RETURNS RECORD AS $$
DECLARE
ret RECORD;
BEGIN
-- Note the CASTING being done for the 2nd and 3rd elements of the RECORD
IF LENGTH(a) < LENGTH(b) THEN
ret := (TRUE, (a || b)::TEXT, 'a shorter than b'::TEXT);
ELSE
ret := (FALSE, (b || a)::TEXT, NULL::TEXT);
END IF;
RETURN ret;
END;$$ LANGUAGE plpgsql;
Almost to epic hotness:
test=> SELECT test_ret('foobar','bar');
test_ret
----------------
(f,barfoobar,)
(1 row)
test=> SELECT test_ret('foo','barbaz');
test_ret
----------------------------------
(t,foobarbaz,"a shorter than b")
(1 row)
But how do you split that out in to multiple rows so that your ORM layer of choice can convert the values in to your language of choice's native data types? The hotness:
test=> SELECT a, b, c FROM test_ret('foo','barbaz') AS (a BOOL, b TEXT, c TEXT);
a | b | c
---+-----------+------------------
t | foobarbaz | a shorter than b
(1 row)
test=> SELECT a, b, c FROM test_ret('foobar','bar') AS (a BOOL, b TEXT, c TEXT);
a | b | c
---+-----------+---
f | barfoobar |
(1 row)
This is one of the coolest and most underused features in PostgreSQL. Please spread the word.
You need to define a new type and define your function to return that type.
CREATE TYPE my_type AS (f1 varchar(10), f2 varchar(10) /* , ... */ );
CREATE OR REPLACE FUNCTION get_object_fields(name text)
RETURNS my_type
AS
$$
DECLARE
result_record my_type;
BEGIN
SELECT f1, f2, f3
INTO result_record.f1, result_record.f2, result_record.f3
FROM table1
WHERE pk_col = 42;
SELECT f3
INTO result_record.f3
FROM table2
WHERE pk_col = 24;
RETURN result_record;
END
$$ LANGUAGE plpgsql;
If you want to return more than one record you need to define the function as returns setof my_type
Update
Another option is to use RETURNS TABLE() instead of creating a TYPE which was introduced in Postgres 8.4
CREATE OR REPLACE FUNCTION get_object_fields(name text)
RETURNS TABLE (f1 varchar(10), f2 varchar(10) /* , ... */ )
...
To return a single row
Simpler with OUT parameters:
CREATE OR REPLACE FUNCTION get_object_fields(_school_id int
, OUT user1_id int
, OUT user1_name varchar(32)
, OUT user2_id int
, OUT user2_name varchar(32)) AS
$func$
BEGIN
SELECT INTO user1_id, user1_name
u.id, u.name
FROM users u
WHERE u.school_id = _school_id
LIMIT 1; -- make sure query returns 1 row - better in a more deterministic way?
user2_id := user1_id + 1; -- some calculation
SELECT INTO user2_name
u.name
FROM users u
WHERE u.id = user2_id;
END
$func$ LANGUAGE plpgsql;
Call:
SELECT * FROM get_object_fields(1);
You don't need to create a type just for the sake of this plpgsql function. It may be useful if you want to bind multiple functions to the same composite type. Else,
OUTparameters do the job.There is no
RETURNstatement.OUTparameters are returned automatically with this form that returns a single row.RETURNis optional.Since
OUTparameters are visible everywhere inside the function body (and can be used just like any other variable), make sure to table-qualify columns of the same name to avoid naming conflicts! (Better yet, use distinct names to begin with.)
Simpler yet - also to return 0-n rows
Typically, this can be simpler and faster if queries in the function body can be combined. And you can use RETURNS TABLE() (since Postgres 8.4, long before the question was asked) to return 0-n rows.
The example from above can be written as:
CREATE OR REPLACE FUNCTION get_object_fields2(_school_id int)
RETURNS TABLE (user1_id int
, user1_name varchar(32)
, user2_id int
, user2_name varchar(32)) AS
$func$
BEGIN
RETURN QUERY
SELECT u1.id, u1.name, u2.id, u2.name
FROM users u1
JOIN users u2 ON u2.id = u1.id + 1
WHERE u1.school_id = _school_id
LIMIT 1; -- may be optional
END
$func$ LANGUAGE plpgsql;
Call:
SELECT * FROM get_object_fields2(1);
RETURNS TABLEis effectively the same as having a bunch ofOUTparameters combined withRETURNS SETOF record, just shorter.The major difference: this function can return 0, 1 or many rows, while the first version always returns 1 row.
AddLIMIT 1like demonstrated to only allow 0 or 1 row.RETURN QUERYis simple way to return results from a query directly.
You can use multiple instances in a single function to add more rows to the output.
db<>fiddle here (demonstrating both)
Varying row-type
If your function is supposed to dynamically return results with a different row-type depending on the input, read more here:
- Refactor a PL/pgSQL function to return the output of various SELECT queries
If you have a table with this exact record layout, use its name as a type, otherwise you will have to declare the type explicitly:
CREATE OR REPLACE FUNCTION get_object_fields
(
name text
)
RETURNS mytable
AS
$$
DECLARE f1 INT;
DECLARE f2 INT;
…
DECLARE f8 INT;
DECLARE retval mytable;
BEGIN
-- fetch fields f1, f2 and f3 from table t1
-- fetch fields f4, f5 from table t2
-- fetch fields f6, f7 and f8 from table t3
retval := (f1, f2, …, f8);
RETURN retval;
END
$$ language plpgsql;
You can achieve this by using simply as a returns set of records using return query.
CREATE OR REPLACE FUNCTION schemaName.get_two_users_from_school(schoolid bigint)
RETURNS SETOF record
LANGUAGE plpgsql
AS $function$
begin
return query
SELECT id, name FROM schemaName.user where school_id = schoolid;
end;
$function$
And call this function as : select * from schemaName.get_two_users_from_school(schoolid) as x(a bigint, b varchar);