Whats is the best workaround to inherit unique contraints in Postgres? - sql

Question

In Postgres I am trying to inherit an unique attribute from a parent to its children. The parent table is an abstract table, with no entries in it. The names of all children should be unique. Referring to the following little (stupid) example: There should be no apple with the same name as a banana (or of course other apple).
Minimalistic Postgres example for this scenario:
CREATE TABLE fruit(CONSTRAINT fruit_uniq_name UNIQUE (name))
CREATE TABLE banana(name text, length integer) INHERITS (fruit)
CREATE TABLE apple(name, diameter integer NOT NULL,) INHERITS (fruit)
After having read many posts on this problem. All of them came to the conclusion that this scenario is impossible to master only with Postgres' inheritance, I'd like to know if there is a best practice workaround, e.g. using triggers and functions, for this problem?
I would be very happy for every little snippet of code, that could help me out of this annoying trap.

Answer 1

I followed the advice of Laurenz Albe and I think I solved the problem by using triggers on the tables apple and banana and a trigger function tgf_name_exists() that tests the uniqueness.
This is the trigger function that tests for the uniqueness of the children names:
CREATE OR REPLACE FUNCTION tgf_name_exits()
RETURNS trigger
LANGUAGE 'plpgsql'
VOLATILE
COST 100
AS $BODY$
declare
count_apple integer;
count_banana integer;
name text;
schema text;
error_text text;
BEGIN
-- Variables
error_text = '';
schema = TG_TABLE_SCHEMA; -- actual schema
name = NEW.name; --- actual attribute name
-- Check
EXECUTE format('SELECT count(*) FROM %s.apple apl WHERE apl.name=%L', schema, name) INTO count_apple;
EXECUTE format('SELECT count(*) FROM %s.banana ban WHERE ban.name=%L', schema, name) INTO count_banana;
-- Info
RAISE NOTICE 'Schema: %', schema;
RAISE NOTICE 'Name: %', name;
RAISE NOTICE 'Count: %', count_apple;
RAISE NOTICE 'Count: %', count_banana;
IF count_apple > 0 OR count_banana > 0 THEN
-- Name ist already used
if count_apple > 0 then
error_text = error_text || "apple "
end if;
if count_banana > 0 then
error_text = error_text || "banana "
end if;
RAISE EXCEPTION 'Name % already existing in table %', name, error_text;
ELSE
-- Name is unused -> OK
RETURN NEW;
END IF;
END;
$BODY$;
These are the triggers for the tables apple and banana
CREATE TRIGGER tg_apple_name_instert_update
BEFORE INSERT OR UPDATE
ON apple
FOR EACH ROW
EXECUTE PROCEDURE tgf_name_exits();
CREATE TRIGGER tg_banana_name_uniq
BEFORE INSERT OR UPDATE
ON banana
FOR EACH ROW
EXECUTE PROCEDURE tgf_name_exits();
It would be very kind if someone could check this. From here it looks like it works.

Answer 2

maybe i am thinking too simplistically about this, but it seems like this functionality can be recreated like this :
CREATE FUNCTION fruit_check ( _name_ text )
RETURNS
boolean AS
$$ SELECT _name_ NOT IN ( SELECT name FROM fruit )
$$
STABLE LANGUAGE SQL ;
ALTER TABLE fruit ADD CHECK ( fruit_check ( name ) ) ;
CREATE TABLE banana ( LIKE fruit INCLUDING ALL , length int ) INHERITS ( fruit ) ;
CREATE TABLE apple ( LIKE fruit INCLUDING ALL , diameter int NOT NULL ) INHERITS ( fruit ) ;
-- now test ...
INSERT INTO banana VALUES ( 'ripe' , 21 ) ; -- INSERT 0 1
INSERT INTO apple VALUES ( 'rosie' , 21 ) ; -- INSERT 0 1
INSERT INTO banana VALUES ( 'rosie' , 21 ) ;
--
-- ERROR: new row for relation "banana" violates check constraint "fruit_name_check"
INSERT INTO apple VALUES ( 'ripe' , 21 ) ;
--
-- ERROR: new row for relation "apple" violates check constraint "fruit_name_check"
UPDATE apple set name = 'ripe' where name = 'rosie' ;
--
-- ERROR: new row for relation "apple" violates check constraint "fruit_name_check"