On Tue, Nov 17, 2009 at 03:54, Marko Tiikkaja
<marko.tiikkaja@cs.helsinki.fi> wrote:
>> Also, after reading through the previous threads; it was not
>> immediately obvious that you dealt with
>> http://archives.postgresql.org/pgsql-hackers/2009-10/msg00566.php by
>> only allowing selects or values at the top level of with.
>
> This is actually just something missing from the current implementation.
> The relevant posts are in the same thread:
> http://archives.postgresql.org/pgsql-hackers/2009-10/msg00558.php and
> the two follow-ups. The comment in ExecutePlan() is a bit misleading.
Hrm I tried the various forms of:
with x as ( ... ) insert/update/delete
and could not get any of them to work. So I assumed the comment
about only SELECT and values were allowed was correct. Maybe a
function that does an insert or update at the top level could get it
to break?
> What I meant is that we don't call GetCurrentCommandId() in
> standard_ExecutorStart(). Instead we get a new CID for every CTE with
> INSERT/UPDATE/DELETE. That comment tried to point out the fact that
> this strategy could fail if there was a non-SELECT query as the
> top-level statement because we wouldn't increment the CID after the last
> CTE.
Right... Which I thought was more or less the recommendation? Guess
Ill have to go re-read that discussion.
> I did it this way because it works well for the purposes of this
> patch and I didn't see an obvious way to determine whether we need a new
> CID for the top-level statement or not.
>
> I'll send an updated patch in a couple of days.
Peachy.
