Re: ALTER TABLE DDL Triggers?

richard(dot)broersma(at)gmail(dot)com ("Richard Broersma") writes:
> I don't believe that DDL Triggers exist, correct?

That is correct.[1]

The usual point is that you cannot attach triggers to pg_catalog
tables, which would be the "obvious" way of trying to notice DDL
changes. (e.g. - by having triggers that would fire when pg_class or
pg_attribute are changed)

It would presumably not be _completely_ implausible to run a trigger
when a table was ALTERed; the trouble would be in evaluating the
semantics what OLD.* and NEW.* ought to contain, _if anything_.

If you took the [1] TRUNCATE approach, there isn't _anything_ (in
terms of OLD.*/NEW.* values) passed to the function; in that case, it
is sufficient to have something (like a function parameter) to
identify the table, and thus pass that info downstream (e.g. - with
replication, passing the TRUNCATE on to downstream nodes). TRUNCATE
is pretty simple; we know well that all it's doing is to get rid of
all the contents of the table at a point in time.

In contrast, the effect of ALTER TABLE is to make near-arbitrary
alterations to pg_class, pg_attribute, and such, and there is, on the
one hand, no obvious semantic of what data to even imagine passing on,
and, on the other, a grand problem of reconstructing the change if you
*did* have access to those underlying tables. That's usually where
the discussion peters out when people propose DDL triggers.

[1] Or about 99% so. There is a change committed for 8.4 where
TRUNCATE can fire a trigger. But it's somewhat disputable whether
TRUNCATE should properly be considered DDL or not.
--
let name="cbbrowne" and tld="linuxfinances.info" in name ^ "@" ^ tld;;
http://www3.sympatico.ca/cbbrowne/advocacy.html
When aiming for the common denominator, be prepared for the occasional
division by zero.