On 4/14/25 08:49, Richard Guo wrote:
On Mon, Apr 7, 2025 at 4:50 PM Richard Guo <guofengli...@gmail.com> wrote:
Hence, I propose the attached patch for the fix.
BTW, there is a XXX comment there saying that maybe we can make the
remaining join quals part of the inner scan's filter instead of the
join filter. I don't think this is possible in all cases. In the
above query, 'coalesce(t2.b) = t3.b' cannot be made part of t3's scan
filter, according to join_clause_is_movable_into. So I removed that
comment in the patch while we're here.
Any thoughts?
Here is an updated patch with a commit message. Regarding
backporting, I'm inclined not to, given the lack of field reports.
Any objections to pushing it?
No objections.
By the way, I think you should be less shy about adding to CC people who
have been involved in the discussion (at least David should have a
chance to know). Also, don't hesitate to use the 'Reviewed-by' tag to
let people know who else may remember the context of the problem, just
in case.
--
regards, Andrei Lepikhov
