On Wed, 21 Feb 2007, Gregory Stark wrote:
> "Gavin Sherry" <swm@alcove.com.au> writes:
>
> > The WITH support seems okay. I guess I'd thought it might be represented
> > different internally (not a sub query) but the approach Greg has taken is
> > probably more straight forward (in that you get a lot of proven code for
> > free). It should work fine for recursive queries too, if you just re-seed
> > the param keys for every scan of the 'sub-query'.
>
> I don't think it works for recursive queries. Since you can't have the same
> executor plan in motion for two different sets of parameters simultaneously.
> That's why I was talking about a Memoize node.
Can you elaborate on the 'two different sets of parameters' bit? I'm still
without coffee.
> It is sufficient for the non-recursive case which might make it worthwhile
> putting it in 8.3. But even there user's expectations are probably that the
> reason they're writing it as a cte is precisely to avoid duplicate execution.
I wonder if the planner should decide that?
Thanks,
Gavin
