What do references to OLD evaluate to in the WHEN cause of an Oracle insert trigger?

Question

When writing a row-level trigger in Oracle, I know that you can use the OLD and NEW pseudo-records to reference the old and new state of the row that fired the trigger. I know that in an INSERT trigger OLD doesn't contain any data, but I'm not sure how this affects the evaluation of a WHEN clause for that trigger. For example, if I have the following trigger:

CREATE OR REPLACE TRIGGER mung_row
BEFORE INSERT OR UPDATE ON some_table
FOR EACH ROW
BEGIN
    :NEW.foo = 'some val';
END;

and I want to modify this trigger to only run on an update when foo was previously null, but always run on an insert, I could satisfy the update part of the change by adding a WHERE clause:

CREATE OR REPLACE TRIGGER mung_row
BEFORE INSERT OR UPDATE ON some_table
FOR EACH ROW
WHEN (OLD.foo IS NULL)
BEGIN
    :NEW.foo = 'some val';
END;

Will this cause problems in the insert case? What will OLD.foo evaluate to in the INSERT?

I'm aware that I could split the INSERT and UPDATE triggers or use INSERTING/UPDATING/DELETING in the trigger body, but I'd rather not in the case that inspired this question.

Answer 1

When a record is being inserted, every field of OLD will be NULL, including the fields marked as NOT NULL in the table's definition.

For example, suppose your table has a non-nullable column named id:

CREATE TABLE some_table (
    id NUMBER NOT NULL,
    foo VARCHAR2(100)
)

When a record is inserted into this table, OLD.id will be NULL. However, when a record is updated in this table, OLD.id will not be NULL. Because you only want to change :NEW.foo if a record is being updated, you just have to check to see if OLD.id has a non-null value.

CREATE OR REPLACE TRIGGER mung_row
BEFORE INSERT OR UPDATE ON some_table
FOR EACH ROW
WHEN (OLD.id IS NOT NULL AND OLD.foo IS NULL)
BEGIN
    :NEW.foo = 'some val';
END;