Re: Teaching planner to short-circuit empty UNION/EXCEPT/INTERSECT inputs

On Fri, 3 Oct 2025 at 04:18, Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:
> 0001's change in is_dummy_rel() seems like a complete hack, especially
> since you didn't bother to change the adjacent comment. Why is that
> necessary?

build_setop_child_paths() wraps the child inputs in SubqueryScanPaths,
so we need to see through those.

An alternative way would be to propagate those during build_setop_child_paths()

--- a/src/backend/optimizer/prep/prepunion.c
+++ b/src/backend/optimizer/prep/prepunion.c
@@ -523,6 +523,13 @@ build_setop_child_paths(PlannerInfo *root, RelOptInfo *rel,
bool is_sorted;
int presorted_keys;

+ /* If the input rel is dummy, propagate that to this query level */
+ if (is_dummy_rel(final_rel))
+ {
+ mark_dummy_rel(rel);
+ continue;
+ }
+

As attached.

> v2 LGTM otherwise.

Thanks

David