github-actions[bot] commented on code in PR #69646:
URL: https://github.com/apache/doris/pull/69646#discussion_r4384585804
##########
be/src/format_v2/table_reader.h:
##########
@@ -461,6 +461,8 @@ class TableReader {
RETURN_IF_ERROR(_data_reader.column_mapper->create_scan_request(
_table_filters, _projected_columns, file_request.get(),
_runtime_state));
file_request->row_ids = _row_ids;
+ file_request->allow_row_reordering =
+ std::ranges::all_of(_conjuncts, _is_safe_to_pre_execute);
Review Comment:
[P2] Fence implicit projection conversions before reordering
This flag checks only `_conjuncts`, while `TableReader::finalize_chunk` also
executes file-to-table projections that FE cannot see. A reduced triggering
plan is:
```text
TopN(k ASC, ts ASC LIMIT 1)
Iceberg FileScan(k INT, ts TIMESTAMP_NS)
retained Parquet schema: k INT, ts DATE
```
In one split/scanner, put `(k=100, ts=1000-01-01)` in physical RG0, followed
by several full batches of `(k=0, ts=2024-01-02)`. `rebuild_projection` in
`column_mapper.cpp` injects an `EvolutionCast` for `ts`; its
DATE-to-TIMESTAMP_NS path forces strict conversion even when SQL strict casts
are disabled. The existing
`IcebergV2ReaderTest.DatePromotionToNanosecondsChecksOverflow` establishes this
overflow behavior. Both FE order keys are safe slots, and the localized TopN
predicate references only unchanged `k`, so neither safety gate sees that
conversion.
Physical order fails on RG0 before a bound exists. The new ordering can read
valid low-key groups first, publish bound `k=0`, and skip RG0 at the footer
recheck before its conversion executes. The struct-accessor projection
safeguard does not inspect an independent top-level `ts` output. This is
distinct from the existing explicit FE-expression threads.
Include implicit projected materialization in the table-layer safety proof,
preserve it on refresh, and add this multi-row-group error 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]
