On Tue, 7 May 2024 at 17:01, Tom Lane <tgl@sss.pgh.pa.us> wrote:
>
> Richard Guo <guofenglinux@gmail.com> writes:
> > Thank you for the report. I looked at this a little bit and I think
> > here is what happened. In deconstruct_distribute_oj_quals we call
> > distribute_quals_to_rels using the uncopied sjinfo->syn_lefthand as
> > outerjoin_nonnullable, which eventually becomes rinfo->outer_relids.
> > Later on, when we remove useless left joins, we modify
> > sjinfo->syn_lefthand using bms_del_member and recycle
> > sjinfo->syn_lefthand. And that causes the rinfo->outer_relids becomes
> > invalid, and finally triggers this issue in join_clause_is_movable_to.
>
> Hmm, the SJE code didn't really touch any of this logic, so why
> didn't we see the failure before?
The bms_free() occurs in remove_rel_from_query() on:
sjinf->syn_lefthand = bms_del_member(sjinf->syn_lefthand, relid);
I've not looked, but I assumed the revert must have removed some
common code that was added and reverted with SJE.
David