On 06/27/2018 09:58 AM, Andreas Joseph Krogh wrote:
> >
> > but we have to settle on a way of controlling it.
>
> +1 from me.
>
> I am running more and more into situations where people consider
> this a bug rather than a feature.
>
> FWIW, I think a GUC that switches between the current (mostly
> unwanted, at least surprising)
> way and one where the CTE is optimized together with the main query
> would suit "most" people.
>
> For sake of compatibility this could default to the current behaviour
>
> +1 from me. The default should be "no fence" for sake of least surprise
> I think. Documenting the change would be sufficient.
> I hope this will be picked up in the comming V12-cycle.
FYI this subject has been discussed in this thread :
https://www.postgresql.org/message-id/5351711493487900%40web53g.yandex.ru
Regards,