kosiew commented on code in PR #22476:
URL: https://github.com/apache/datafusion/pull/22476#discussion_r3367392171
##########
datafusion/optimizer/src/optimize_projections/mod.rs:
##########
@@ -381,29 +379,12 @@ fn optimize_projections(
// These operators have no inputs, so stop the optimization
process.
return Ok(Transformed::no(plan));
}
- LogicalPlan::RecursiveQuery(recursive) => {
- // Only allow subqueries that reference the current CTE; nested
subqueries are not yet
- // supported for projection pushdown for simplicity.
- // TODO: be able to do projection pushdown on recursive CTEs with
subqueries
- if plan_contains_other_subqueries(
- recursive.static_term.as_ref(),
- &recursive.name,
- ) || plan_contains_other_subqueries(
- recursive.recursive_term.as_ref(),
- &recursive.name,
- ) {
- return Ok(Transformed::no(plan));
- }
-
- plan.inputs()
- .into_iter()
- .map(|input| {
- indices
- .clone()
- .with_projection_beneficial()
- .with_plan_exprs(&plan, input.schema())
- })
- .collect::<Result<Vec<_>>>()?
+ LogicalPlan::RecursiveQuery(_) => {
+ // optimize the static and recursive terms
Review Comment:
Nice addition. This branch is encoding an important invariant: recursive CTE
terms are optimization boundaries and should not inherit projection
requirements from the parent plan.
It may be worth making that a bit more explicit in the comment so the intent
is clear to future readers. Something along the lines of:
`// Treat each recursive CTE term like a standalone subquery: optimize its
internals, but do not push parent required indices through the RecursiveQuery
boundary.`
That extra context could help prevent outer-column pruning from accidentally
being reintroduced here later.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
