Re: Assert failure in try_nestloop_path()

On Sun, Sep 27, 2026 at 1:00 AM Alexander Lakhin <exclusion(at)gmail(dot)com> wrote:
> I've discovered that:
> echo "geqo_threshold = 2" >/tmp/temp.config
> TEMP_CONFIG=/tmp/temp.config TESTS="partition_join" make -s check-tests
>
> triggers one of the assertions:
> TRAP: failed Assert("no_duplicate_clause_serials(*restrict_clauses)"), File: "relnode.c", Line: 2079, PID: 634394

Thanks for the report. I think the culprit is that during GEQO join
planning, add_child_join_rel_equivalences() will re-add the same child
EC members in each GEQO round, since it has no concept that the
members it wants might be there already. As a result, in the reported
case, the same EC member PHV(COALESCE(t3_p1.c, t4.c)) appears three
times in the same EC. So, in get_joinrel_parampathinfo(), the
generate_join_implied_equalities() call generate two identical
RestrictInfos for "PHV(...) = PHV(...)".

I think we need to do something to avoid generating duplicate EC
members in add_child_join_rel_equivalences().

- Richard