On Wed, Feb 12, 2025 at 10:04 PM Ashutosh Bapat
<ashutosh.bapat.oss@gmail.com> wrote:
> Hmm. The code there assumes that all the Relids will at least have one
> parent each of the children involved. For some reason
> sjinfo->min_lefthand has only relid 1 but not 2 or 5. 2 and 5 are
> actually the parent relids of the children passed in respectively. The
> join is between 2 and 5, then why is 1 appearing in the min_lefthand.
> It might be legitimate, but we need to find the reason. If it's
> legitimate, I think we need to copy the Relids which haven't undergone
> any translation so as to keep them isolated from the parent relids.
Yes, it's legitimate. For the semijoin, its join clause only
references {1} and {5}, with no other ordering restrictions.
Therefore, the minimum LHS for this join consists only of {1}.
Instead of copying the untranslated Relids and freeing them later, I
think it might be better to modify free_child_join_sjinfo() to avoid
freeing the untranslated members of child_sjinfo.
Thanks
Richard
