On Saturday, February 21, 2026, Marcos Pegoraro <[email protected]> wrote: > > Is it related only to parallel or any other optimization ? > > Yes, see create aggregate for what is required when setting partial aggregation to true. The explanation on [0] (aggregate function reference page) regarding why json creation functions show false is also informative.
> > And all aggregate functions can have FILTER ? > Yes, the executor handles filter and simply omits invoking the aggregate if the row does not pass the filter. The aggregates have no awareness they have been filtered. Yes, one must reason their way to this, it isn’t spelled out anywhere that I recall. > I'm not exactly sure how to explain this case better, but I think it needs > better wording. > I would agree. Or a cross-reference at least. It’s also another instance of weasel-wording that seems to just invite doubt/questions. But it does say optimizations which means it does not affect the rows returned, which filter does. > > Additionally, on this page there is no mention to FILTER (WHERE ...). > Wouldn't it be better to explain something like it is on [2]. > We already cross-reference to the authoritative sections for this. Could entertain a rephrasing of those cross-references but I would not go and make a big thing of it here. That said, maybe a separate “since all aggregates accept a filter clause its syntax has been o tied here for brevity” sentence would work and match up with the existing one for order by. David J.
