I create the trigger A1 so that an article with a certain type, that is 'Bert' cannot be added more than once and it can have only 1 in the stock.
However, although i create the trigger, i can still add an article with the type 'Bert'. Somehow, the count returns '0' but when i run the same sql statement, it returns the correct number. It also starts counting properly if I drop the trigger and re-add it. Any ideas what might be going wrong?
TRIGGER A1 BEFORE INSERT ON mytable
FOR EACH ROW
DECLARE
l_count NUMBER;
PRAGMA AUTONOMOUS_TRANSACTION;
BEGIN
SELECT COUNT(*) INTO l_count FROM mytable WHERE article = :new.article;
dbms_output.put_line('Count: ' || l_count);
IF l_count >0 THEN
IF(:new.TYPEB = 'Bert') THEN
dbms_output.put_line('article already exists!');
ROLLBACK;
END IF;
ELSIF (:new.TYPEB = 'Bert' AND :new.stock_count>1) THEN
dbms_output.put_line('stock cannot have more than 1 of this article with type Bert');
ROLLBACK;
END IF;
END;
This is the insert statement I use:
INSERT INTO mytable VALUES('Chip',1,9,1,'Bert');
A couple of points. First, you are misusing the autonomous transaction pragma. It is meant for separate transactions you need to commit or rollback independently of the main transaction. You are using it to rollback the main transaction -- and you never commit if there is no error.
And those "unforeseen consequences" someone mentioned? One of them is that your count always returns 0. So remove the pragma both because it is being misused and so the count will return a proper value.
Another thing is don't have commits or rollbacks within triggers. Raise an error and let the controlling code do what it needs to do. I know the rollbacks were because of the pragma. Just don't forget to remove them when you remove the pragma.
The following trigger works for me:
CREATE OR REPLACE TRIGGER trg_mytable_biu
BEFORE INSERT OR UPDATE ON mytable
FOR EACH ROW
WHEN (NEW.TYPEB = 'Bert') -- Don't even execute unless this is Bert
DECLARE
L_COUNT NUMBER;
BEGIN
SELECT COUNT(*) INTO L_COUNT
FROM MYTABLE
WHERE ARTICLE = :NEW.ARTICLE
AND TYPEB = :NEW.TYPEB;
IF L_COUNT > 0 THEN
RAISE_APPLICATION_ERROR( -20001, 'Bert already exists!' );
ELSIF :NEW.STOCK_COUNT > 1 THEN
RAISE_APPLICATION_ERROR( -20001, 'Can''t insert more than one Bert!' );
END IF;
END;
However, it's not a good idea for a trigger on a table to separately access that table. Usually the system won't even allow it -- this trigger won't execute at all if changed to "after". If it is allowed to execute, one can never be sure of the results obtained -- as you already found out. Actually, I'm a little surprised the trigger above works. I would feel uneasy using it in a real database.
The best option when a trigger must access the target table is to hide the table behind a view and write an "instead of" trigger on the view. That trigger can access the table all it wants.
You need to do an AFTER trigger, not a BEFORE trigger. Doing a count(*) "BEFORE" the insert occurs results in zero rows because the data hasn't been inserted yet.
