morrySnow opened a new pull request, #67668:
URL: https://github.com/apache/doris/pull/67668
[fix](fe) Preserve false aggregate arguments in not-null inference
### What problem does this PR solve?
Issue Number: None
Related PR: None
Problem Summary: A global aggregate over a left outer join returned an
incorrect result for expressions such as `COUNT(r.v IS NOT NULL)`.
`InferAggNotNull` reused predicate-style not-null inference, where replacing
`r.v` with NULL and folding the expression to FALSE is considered
null-rejecting. FALSE rejects a filter predicate, but it is a valid non-NULL
aggregate argument that COUNT must count. The generated `r.v IS NOT NULL`
filter therefore enabled outer join elimination and discarded null-extended
rows. Add aggregate-specific inference that only treats a folded SQL NULL
result as discardable, while preserving the existing FALSE-or-NULL behavior for
predicate inference. After the fix, the boolean COUNT keeps the left outer join
and returns all three rows, while `COUNT(r.v)` still infers not-null, uses an
inner join, and returns one row.
### Release note
Fix incorrect aggregate results when an argument evaluates to FALSE for a
NULL input from an outer join.
### Check List (For Author)
- Test:
- Regression test
- Unit Test
- Behavior changed: Yes. Preserve FALSE aggregate arguments and prevent
invalid outer join elimination.
- Does this need documentation: No
--
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]
