Hi Onder,
On Thu, May 30, 2024 at 4:15 PM Önder Kalacı <onderkalaci@gmail.com> wrote:
>> and 2) ORDER BY must already
>> > have been determined to be safe to push down before we get here.
> When I read the code, the decision for that seems to happen in the next line where this patch proposes to modify:
That is not correct; see my previous email.
>> So
>> > in that case, if getting here, we can consider that WITH TIES is also
>> > safe to push down (if the remote is v13 or later).
> See Tom's response here on why it might not be a good idea to pushdown
> WITH TIES: https://www.postgresql.org/message-id/2114796.1715878709%40sss.pgh.pa.us
>
> I think, at least not with this patch, this patch is like a bug-fix. If intended, it should be possible to
> pushdown WITH TIES with a follow-up patch?
With all due respect to him, I think he is missing that the set of
ties is determined according to ORDER BY; as explained in that email,
in that case it is guaranteed that ORDER BY is pushable, so WITH TIES
is also pushable with ORDER BY, I think. We do not currently have a
way to do a remote-version check (without accessing the remote
server), so I agree with you that we should just diable pushing WITH
TIES fow now, though.
Thanks!
Best regards,
Etsuro Fujita