github-actions[bot] commented on code in PR #67919:
URL: https://github.com/apache/doris/pull/67919#discussion_r4043491356
##########
fe/fe-core/src/main/java/org/apache/doris/nereids/rules/rewrite/InferPredicateByReplace.java:
##########
@@ -210,7 +236,9 @@ public static Set<Expression> infer(Set<Expression> inputs)
{
}
Map<Expression, Set<Expression>> exprPredicates = new HashMap<>();
for (Expression input : inputs) {
- if (input.anyMatch(expr -> !((ExpressionTrait)
expr).isDeterministic())
+ // Inference can evaluate a predicate on rows that never reach its
original filter.
+ if (input.anyMatch(expr -> expr instanceof NoneMovableFunction
Review Comment:
[P1] Fence deterministic expressions that can fail on extra rows
This check only covers functions already marked `NoneMovableFunction`, but
deterministic functions can also expose row-dependent errors when inference
evaluates them on the other join input. For example, the reduced plan
```text
InnerJoin(l.a = r.b)
Filter(bitmap_count(to_bitmap_with_check(l.a)) > 0)
Scan L(a={1})
Scan R(b={-1,1})
```
indexes the unary predicate under `l.a`; exact-BIGINT replacement then
creates the same predicate on `r.b`, and `InferPredicates` installs it above R.
The original plan returns the `1=1` match, but the inferred filter evaluates
the unmatched `b=-1`, for which BE's `ToBitmapWithCheck` returns
`InvalidArgument`. This is distinct from the existing `assert_true` thread
because `ToBitmapWithCheck` does not implement `NoneMovableFunction` and
therefore passes the newly added guard. Please require a general
safe-on-additional-rows/total-expression property (or mark and audit all
status-error functions) and add this unmatched-row regression.
--
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]
