PostgreSQL: Using schema and table names from other tables?

You need dynamic SQL - perhaps something like this:

create role stack;
create schema authorization stack;
set role stack;

create or replace function f(p_schema in text, p_table in text)
                           returns setof integer language plpgsql immutable as $$
begin
  return query execute 'select value from '||p_schema||'.'||p_table;
end;$$;

create table t1(value integer);
insert into t1(value) values (1);
insert into t1(value) values (2);

create table t2(value integer);
insert into t2(value) values (1);

create table schemas(schema_name text);
insert into schemas(schema_name) values ('stack');

create table tables(table_name text);
insert into tables(table_name) values ('t1');
insert into tables(table_name) values ('t2');
insert into tables(table_name) values ('t1');
insert into tables(table_name) values ('t2');

select f(schema_name, table_name) from schemas cross join tables;
 f
---
 1
 2
 1
(3 rows)

I'm assuming that every table is present in every schema as the question implies


Jack has demonstrated the way to go. However, I feel there is room for improvement.

I place everything in schema x for convenient testing. Test setup:

DROP SCHEMA x CASCADE;
CREATE SCHEMA x;

-- meta tables for schema and table name    
CREATE TABLE x.schma(schma_id int, schma text);
INSERT INTO  x.schma VALUES (1, 'x');

CREATE TABLE x.tbl(tbl_id int, tbl text);
INSERT INTO  x.tbl VALUES (1, 't1'), (2, 't2');

-- dummy tables to be used in example query:
CREATE TABLE x.t1(id int);
INSERT INTO  x.t1 VALUES (1),(2);

CREATE TABLE x.t2(foo text);
INSERT INTO  x.t2 VALUES ('some text'), ('some more text');

Old function (original answer):

CREATE OR REPLACE FUNCTION x.f_dynaquery_old(int, int, _col text, _type anyelement, OUT col anyelement)
  RETURNS SETOF anyelement AS
$func$
BEGIN
   RETURN QUERY EXECUTE '
   SELECT ' || quote_ident(_col) || '
   FROM   ' || (
       (SELECT schma FROM schma WHERE schma_id = $1) || '.' ||
       (SELECT tbl   FROM tbl   WHERE tbl_id   = $2))::regclass;
END
$func$  LANGUAGE plpgsql;

Cleaner version with format() (update 2017):

CREATE OR REPLACE FUNCTION x.f_dynaquery(_schma_id int, _tbl_id int
                                       , _col text, _type anyelement)
  RETURNS TABLE(col anyelement) AS
$func$
BEGIN
   RETURN QUERY EXECUTE format(
      'SELECT %I FROM %I.%I'
    , _col
    , (SELECT schma FROM schma WHERE schma_id = _schma_id)
    , (SELECT tbl   FROM tbl   WHERE tbl_id   = _tbl_id)
   );
END
$func$  LANGUAGE plpgsql;

COMMENT ON FUNCTION x.f_dynaquery(int, int, text, anyelement)
IS 'Query any column from a dynamically assembled tablename.
$1 .. id of schema
$2 .. id of table
$3 .. name of column
$4 .. type of column (only data type matters, not the value)';

Call:

SELECT col FROM x.f_dynaquery(1, 1, 'id', NULL::int);
  col
-----
   1
   2


SELECT col FROM x.f_dynaquery(1, 2, 'foo', NULL::text);
  col
----------------
 some text
 some more text

Major points

  • Function can return any column of any type. Read in the manual about Polymorphic Types and Declaring Function Parameters.

  • Defend against SQL injection by using quote_ident() and casting to regclass in the old version or with format() in the new one. Related:

    • PL/pgSQL regclass quoting of table named like keyword
  • Retrieve the tablename by query as OP requested. Example is query by id but anything is possible.

  • Name the returned column, so it is easier to reference.