Re: BUG #6051: wCTE query fail with wrong error text on a table with rules

On Sat, Jun 04, 2011 at 05:24:22PM -0400, Tom Lane wrote:
> Merlin Moncure <mmoncure(at)gmail(dot)com> writes:
> > On Fri, Jun 3, 2011 at 10:42 AM, Jehan-Guillaume (ioguix) de Rorthais
> > <jgdr(at)dalibo(dot)com> wrote:
> >> test=# CREATE RULE test_ins AS ON INSERT TO test DO INSTEAD INSERT INTO
> >> test2 VALUES (NEW.i);
> >> CREATE RULE
> >> test=# WITH t1 AS (
> >> DELETE FROM ONLY test RETURNING *
> >> )
> >> INSERT INTO test SELECT * FROM t1;
> >> ERROR: could not find CTE "t1"
>
> > IIRC the fact that rules don't play nice with wCTE was brought up
> > several times during the implementation discussions. I'm not saying
> > the error message is great, but you can pretty much add this to the
> > giant pile of reasons not to use rules at all (particularly in 9.1
> > with the view triggers).
>
> There are definitely cases that don't work, but I had thought we at
> least threw an intelligible "not implemented" error for all of them.
> This one seems to be an oversight: specifically, rewriteRuleAction()
> isn't considering the possibility that the rewritten rule action will
> need to make use of CTEs from the original query.
>
> We could paste a copy of the original's cteList into the rule action,
> but there are still issues:
>
> * If there's more than one rule action, we could end up executing
> multiple copies of the same CTE query; which breaks the expectation
> of single evaluation for a CTE.
>
> * If there are CTEs attached to the rule action, as well as to the
> original query, and there is a conflict of CTE names between them,
> we can't handle that AFAICS. (The planner expects to look up entries
> in a cteList by name...)
>
> * Maybe some other things that aren't obvious yet.
>
> I don't particularly mind throwing a not-implemented error for the first
> case (ie, just say multiple rule actions don't mix with CTE queries);

Great :)

> but the second case seems seriously annoying, since there's no way
> for someone to write a CTE-containing rule action without risking a
> conflict. Ideas anybody?

Is there some way to throw "not implemented" there, too? It seems to
me that making accommodations for the user-modifiable part of our
rewrite rules isn't the right direction to go, as we're well on our
way to phasing the user-modifiable part out.

Cheers,
David.
--
David Fetter <david(at)fetter(dot)org> http://fetter.org/
Phone: +1 415 235 3778 AIM: dfetter666 Yahoo!: dfetter
Skype: davidfetter XMPP: david(dot)fetter(at)gmail(dot)com
iCal: webcal://www.tripit.com/feed/ical/people/david74/tripit.ics

Remember to vote!
Consider donating to Postgres: http://www.postgresql.org/about/donate