Here's a new version. Many of the old complaints have been fixed; particularly, the handling of partitioned tables is now much cleaner and straightforward. Amit Langote helped considerably in getting this part to shape -- thanks for that. Amit also helped correct the EvalPlanQual behavior, which wasn't quite up to snuff.
There are a few things that can still be improved here. For one, I need to clean up the interactions with table AM (and thus heapam.c etc). Secondarily, and I'm now not sure that I really want to do it, is change the representation for executor: instead of creating a fake join between target and source, perhaps we should have just source, and give optimizer a separate query to fetch tuples from target.
What I did do is change how the target table is represented from parse analysis to executor. For example, in the original code, there were two RTEs that represented the target table; that is gone. Now the target table is always just the query's resultRelation. This removes a good number of ugly hacks that had been objected to.
I'll park this in the January commitfest now.
-- Álvaro Herrera Valdivia, Chile — https://www.EnterpriseDB.com/ "Cómo ponemos nuestros dedos en la arcilla del otro. Eso es la amistad; jugar al alfarero y ver qué formas se pueden sacar del otro" (C. Halloway en La Feria de las Tinieblas, R. Bradbury)
Hi,
+ skipped_path = total - insert_path - update_path - delete_path;
Should there be an assertion that skipped_path is not negative ?
+ * We maintain separate transaction tables for UPDATE/INSERT/DELETE since + * MERGE can run all three actions in a single statement. Note that UPDATE
+ * needs both old and new transition tables
Should the 'transaction' in the first line be transition ?