Gabriel39 opened a new pull request, #67979:
URL: https://github.com/apache/doris/pull/67979
### What problem does this PR solve?
File-level MIN/MAX pushdown accepts numeric casts that can turn non-null
source values into NULL. For a BIGINT column containing `-2147483649`, `0`, and
`2147483648`, `MIN(CAST(value AS INT))` and `MAX(CAST(value AS INT))` can both
return NULL: the scan keeps only the original endpoints and discards the valid
interior value.
Reject file aggregate pushdown when the resolved numeric cast can introduce
NULL, using the existing cast nullability rules independently of the source
column's nullability. Safe widening casts continue to use pushdown. This covers
aggregate arguments both directly and through projection aliases.
### Release note
Fix incorrect MIN/MAX results over external files when a numeric cast can
overflow or introduce NULL.
### Check List (For Author)
- Test:
- [x] Unit Test: `PhysicalStorageLayerAggregateTest` and
`AggregateStrategiesTest` (16 passed, 3 existing skips). New unsafe-cast test
fails before the fix and passes after it.
- [x] FE Checkstyle.
- Paimon end-to-end regression was not run. An unrelated existing
`IvmNormalizeMTMVJoinTest` constructor mismatch was temporarily excluded from
local test compilation and restored afterward.
- Behavior changed:
- [x] Yes. Unsafe casts evaluate file rows before aggregation; safe
widening casts retain pushdown.
- Does this need documentation?
- [x] 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]
