Hi, thank you for your valuable feedback. If you have time you can test
it on sqlsmith (https://github.com/anse1/sqlsmith) -
it will give some corner cases that should be observed.
On 11.05.2026 08:20, solaimurugan vellaipandiyan wrote:
Hi Alena,
I did some additional testing on the latest patch and the behavior
looks good in my environment so far.
I verified the original EXISTS pull-up case and confirmed that after
applying the patch the planner generates optimized Semi Join plans
instead of EXISTS(SubPlan).
I also tested several additional cases discussed in the thread, including:
1.NOT EXISTS queries
2. constant qual conditions
3. ANY array conditions
4. LEFT JOIN cases
5. FULL JOIN cases
For the FULL JOIN scenarios, I checked both the execution plans and
actual query results. In one case the planner simplified the query
into a Hash Join plan, but the final query results remained correct.
I additionally ran:
- make check
- make -C contrib check
Both completed successfully in my environment and I did not observe
planner crashes or incorrect query results during testing.
Overall, the patch seems to improve planner behavior consistently for
these EXISTS pull-up scenarios.
Regards,
Solaimurugan V
--
-----------
Best regards,
Alena Rybakina
