A-ha, interesting !
I think we have some specific use cases where it'd be worth the overhead, I'd need to measure it, though;
Do you think there'd be room to accept a contribution for such functionality with a disabled-by-default pg setting,
or are you skeptical it would ever be worth the trade-off ?
Thanks again,
Danny
On Thu, Aug 4, 2022 at 4:38 PM Tom Lane <tgl@xxxxxxxxxxxxx> wrote:
Danny Shemesh <dany74q@xxxxxxxxx> writes:
> That is of course correct, but what I mean is that, I think that if one
> would explicitly query f(x), and never for x directly, it would've been
> theoretically possible to say that the index is covering for every f(x),
> wouldn't it ?
Theoretically, yeah, but we don't support that: an index-only scan
will only be considered if x itself is available from the index.
There are a couple of reasons for that, but the main one is that
detecting whether an index matches the query would be far more expensive
if it had to consider _expression_ subtrees not just the base Vars.
regards, tom lane
