How to do update instead of insert / cancel the insert

Question

Hello,
we are porting an application, written originally for PostgreSQL, to Oracle. Most of the porting is pretty straightforward - change column types, simple trigger syntax changes etc.
But we have a table (more precisely several tables) with a set of triggers, holding a tree-like structure. The point is in PostgreSQL we're using a RETURN NULL; behavior in BEFORE triggers, which means 'do nothing' to simplify thing.
For example imagine tables A and B - and in a trigger on B we could have something like
INSERT INTO A VALUES (id, ...)
without a check on a uniqueness of the id value. And in a BEFORE INSERT trigger on A we have something like
SELECT INTO c id FROM A where id = NEW.id;
IF FOUND THEN
UPDATE A SET .... WHERE id = NEW.id;
-- cancel the insert without an exception
RETURN NULL;
END IF;
which checks for existence of the row, and if it already exists it effectively replaces with UPDATE and cancels the INSERT without an exception.
Is there something like this in Oracle or do I have to completely rewrite / refactor the whole thing?
PS: We do have UNIQUE (more precisely PRIMARY KEY) on the columns - I am aware that this is not sufficient to enforce uniqueness in Oracle.

Answer 1

There are a couple of ways that this could be accomplished in Oracle, although the though of an insert into table b also inserting or updating rows in table a makes me cringe. The easiest way only requires a trigger on table b.
Something along the lines of:
SQL> CREATE TABLE a (ID NUMBER PRIMARY KEY, descr VARCHAR2(10));
Table created.
SQL> CREATE TABLE b (id NUMBER, descr VARCHAR2(10), descr2 VARCHAR2(20));
Table created.
SQL> CREATE TRIGGER b_bi
  2     BEFORE INSERT ON b
  3     FOR EACH ROW
  4  BEGIN
  5     INSERT INTO a VALUES (:new.id, :new.descr);
  6  EXCEPTION
  7     WHEN DUP_VAL_ON_INDEX THEN
  8        UPDATE a SET descr = :new.descr
  9        WHERE id = :new.id;
10  END;
11  /
Trigger created.
SQL> INSERT INTO a VALUES(1, 'One');
1 row created.
SQL> COMMIT;
Commit complete.
SQL> INSERT INTO b VALUES(1, 'NewOne', 'New One');
1 row created.
SQL> SELECT * FROM a;
        ID DESCR
         1 NewOne
SQL> INSERT INTO b VALUES(2, 'Two', 'Old Two');
1 row created.
SQL> SELECT * FROM a;
        ID DESCR
         1 NewOne
         2 Two
SQL> COMMIT;
Commit complete.If you really require the trigger on table a for other reasons, then you could do something like:
SQL> RENAME a to a_table;
Table renamed.
SQL> CREATE VIEW a AS SELECT * FROM a_table;
View created.
SQL> CREATE TRIGGER a_ii
  2     INSTEAD OF INSERT ON a
  3     FOR EACH ROW
  4  BEGIN
  5     INSERT INTO a_table VALUES(:new.id, :new.descr);
  6  EXCEPTION
  7     WHEN DUP_VAL_ON_INDEX THEN
  8        UPDATE a_table SET descr = :new.descr
  9        WHERE id = :new.id;
10  END;
11  /
Trigger created.
SQL> CREATE OR REPLACE TRIGGER b_bi
  2     BEFORE INSERT ON b
  3     FOR EACH ROW
  4  BEGIN
  5     INSERT INTO a VALUES (:new.id, :new.descr);
  6  END;
  7  /
Trigger created.
SQL> INSERT INTO b VALUES(2, 'NewTwo', 'New Two');
1 row created.
SQL> SELECT * FROM a;
        ID DESCR
         1 NewOne
         2 NewTwo
SQL> INSERT INTO b VALUES(3, 'Three','Third');
1 row created.
SQL> SELECT * FROM a;
        ID DESCR
         1 NewOne
         2 NewTwo
         3 ThreeNeither should require any changes to your application code. However, once agan, I think it is a really bad idea for triggers to have side efects on other tables.
John