Consider the following example for 7.4.3
% cat test.sql
CREATE SCHEMA test_schema;
SET search_path TO test_schema;
CREATE TABLE test_table (
x serial,
s bigint
) WITHOUT OIDS;
CREATE FUNCTION test_update_s() RETURNS trigger AS '
BEGIN
SELECT INTO NEW.s sum(x) FROM test_table;
RETURN NEW;
END' LANGUAGE 'plPgSQL';
CREATE TRIGGER test_update_s_trg BEFORE INSERT OR UPDATE ON test_table
FOR EACH ROW EXECUTE PROCEDURE test_update_s();
SET search_path TO public; -- here's the key
INSERT INTO test_schema.test_table DEFAULT VALUES;
Then:
fduch@~=# \i test.sql
CREATE SCHEMA
SET
psql:test.sql:6: NOTICE: CREATE TABLE will create implicit sequence "test_table_x_seq" for "serial" column
"test_table.x"
CREATE TABLE
CREATE FUNCTION
CREATE TRIGGER
SET
psql:test.sql:18: ERROR: relation "test_table" does not exist
CONTEXT: PL/pgSQL function "test_update_s" line 2 at select into variables
So, the function written WITHIN test_schema and FOR work with
test_schema doesn't see relations inside test_schema until the
search_path points to it?
The same for (at least inlined) SQL functions.
Of course, the workaround is to fully qualify test_schema.test_table
within the function, or to set appropriate search_path when working with
that table/function, but...
But the following works:
fduch@~=# SET search_path TO test_schema;
SET
fduch@~=# INSERT INTO test_table DEFAULT VALUES ;
INSERT 0 1
fduch@~=# SET search_path TO public;
SET
fduch@~=# INSERT INTO test_schema.test_table DEFAULT VALUES ;
INSERT 0 1
AFAIK plPgSQL function, have been compiled once, caches query plans so
it directly refers tables (by oids or something else).
Is that the reason for the last insert to work?
--
Fduch M. Pravking
