"Karl O. Pinc" <kop@meme.com> writes:
> why can't I put a SELECT rule on a table?
Because then it would be a view.
As for $SUBJECT, the problem is that there will never be an insert into
a view --- not at the level of a physical insert attempt anyway --- and
thus there is nothing for a trigger to do.
The reason there will never be an insertion trigger event is that we
reject any INSERT on a view that isn't rewritten (by an unconditional
DO INSTEAD rule) into something else.
I recall a prior discussion about making it possible to use triggers on
views as a substitute for DO INSTEAD rules, by removing the rewrite-time
check and only erroring out if we actually get to the point of
attempting a physical insert. Then a BEFORE INSERT trigger could do
something appropriate with the data and return NULL to prevent the
error. This seems like a good idea because triggers often are much
easier to work with than rules --- eg, there's no problem with multiple
evaluations of volatile functions, even if you send the data to several
places. However, I'm not sure that the idea scales to cover updates and
deletes; with no concept of physical tuple identity (ctid) for the view
rows, it's not clear that you can write triggers that will reliably do
the right things.
regards, tom lane
