Hi,
While developing some improvements for TPC-DS queries I found out that with
UNION ALL partial paths are not emitted. Whilst fixing that I also came across
the subquery costing which does not seem to consider parallelism when doing
the costing.
I added a simplified testcase in pg-regress to show this goes wrong, and
attached also a before and after explain output of tpc-ds SF100 query 5
based on version 12.4.
I hope I followed all etiquette and these kind of improvements are welcome.
Kind regards,
Luc
Swarm64