yujun777 commented on code in PR #70516:
URL: https://github.com/apache/doris/pull/70516#discussion_r4639674645
##########
fe/fe-core/src/main/java/org/apache/doris/nereids/processor/post/RuntimeFilterPushDownVisitor.java:
##########
@@ -308,6 +318,25 @@ && canPushThroughJoinChild(join, false, ctx)) {
}
if (newTarget != null && newTarget.getInputSlots().size() == 1
&& !newTarget.equals(ctx.srcExpr)) {
+ if (expandByOuterJoin) {
Review Comment:
[P1] Stop expanded filters at projections with non-movable functions
The direct-column guard still admits `R.k` when a different expression in
the same projection must run before filtering. For example, consider:
```sql
SELECT l.k, r.checked
FROM L l
LEFT JOIN (SELECT k, assert_true(v > 0, 'bad') AS checked FROM R) r
ON l.k = r.k
JOIN E e ON l.k = e.k
```
With INT keys, `L={1,2}`, `E={2}`, and `R={(1,0),(2,1)}`, keep the join
order and use an IN runtime filter that the scan waits for. Before expansion,
the lower join's build-side projection evaluates `assert_true` on `(1,0)` and
raises `bad`. With expansion, `visitPhysicalProject` pushes `RF({2})` through
the projection to the scan on `R.k`. The scan drops `(1,0)` before the
projection evaluates it, so the query can succeed.
Could we apply the existing `containsNoneMovableFunction()` guard in
`visitPhysicalProject` and extend the row-count regression suite to verify that
both settings of the expansion flag raise `bad`? [Existing predicate pushdown
guard](https://github.com/apache/doris/blob/3a5d3e7f5c142bcba1e52ddbd1cea2c4cad6c12a/fe/fe-core/src/main/java/org/apache/doris/nereids/processor/post/PushDownFilterThroughProject.java#L35-L50)
--
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]