Hi,
On 2019-02-22 16:27:28 -0500, Regina Obe wrote:
> > I think there are probably other ways of fixing this query that won't have
> > such dramatic effects; it doesn't really seem to need to use WITH, and I bet
> > you could also tweak the WITH query to prevent inlining.
>
> Yes I know I can change THIS QUERY. I've changed other ones to work around this.
> Normally I just use a LATERAL for this.
>
> My point is lots of people use CTEs intentionally for this kind of thing because they know they are materialized.
>
> It's going to make a lot of people hesitant to upgrade if they think they need to revisit every CTE (that they
intentionallywrote cause they thought it would be materialized) to throw in a MATERIALIZED keyword.
This was extensively discussed, in several threads about inlining
CTEs. But realistically, it doesn't actually solve the problem to offer
a GUC, because we'd not be able to remove it anytime soon. I see
benefit in discussing how we can address regressions like your example
query here, but I don't feel there's any benefit in re-opening the whole
discussion about GUCs, defaults, and whatnot.
Greetings,
Andres Freund