Re: Properly pathify the union planner

On Thu, 28 Mar 2024 at 15:56, Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:
> I haven't studied the underlying problem yet, so I'm not quite
> buying into whether we need this struct at all ...

The problem is, when planning a UNION child query, we want to try and
produce some Paths that would suit the top-level UNION query so that a
Merge Append -> Unique can be used rather than a Append -> Sort ->
Unique or Append -> Hash Aggregate.

The problem is informing the UNION child query about what it is. I
thought I could do root->parent_root->parse->setOperations for a UNION
child to know what it is, but that breaks for a query such as:

WITH q1(x) AS (SELECT 1)
SELECT FROM q1 UNION SELECT FROM q1

as the CTE also has root->parent_root->parse->setOperations set and in
the above case, that's a problem as there's some code that tries to
match the non-resjunk child targetlists up with the SetOperationStmt's
SortGroupClauses, but there's a mismatch for the CTE. The actual
UNION children should have a 1:1 match for non-junk columns.

> but assuming
> we do, I feel like "PlannerContext" is a pretty poor name.
> There's basically nothing to distinguish it from "PlannerInfo",
> not to mention that readers would likely assume it's a memory
> context of some sort.
>
> Perhaps "SubqueryContext" or the like would be better? It
> still has the conflict with memory contexts though.

Maybe something with "Parameters" in the name?

David