That was very enlightening Alex. Thanks a lot. I'll keep that post around.

I was confused by Helen's explanation because it missed the "out of
transaction context" and "most pessimistic mode" parts regarding declarative
RI. I was under the impression Helen said that triggers (although they
involve more work) could be use indifferently to reproduce FKs. But maybe I
misunderstood her.

<..>

> So, FK guarantee integrity, neverthelles of opportunity if
> concurrent changes will be rolled back, wisely lurking out of bounds
> of transaction context in case if changes can only potentially lead to
> inconsistence. Trigger, which operates in transaction context, can't
> do this.

I keep wondering why we're not more often alerted against RI in triggers...

--
Jerome