TheR1sing3un opened a new pull request, #9934: URL: https://github.com/apache/paimon/pull/9934
### Purpose Keep batch raw vector scoring vectorized when a score block contains null vectors. Previously, one null vector sent every valid row in the block through scalar scoring for every query. Compact valid rows within the existing bounded block and restore their original score positions, leaving null rows skipped and invalid-vector fallback behavior unchanged. The change is limited to batch scoring and does not alter distance arithmetic or top-k selection. In a local kernel benchmark with 4,096 rows, 128 dimensions, 16 queries, and one null per 1,024 rows, median L2 scoring decreased from 298 ms to 23 ms. This is a warm local kernel measurement, not an end-to-end performance guarantee. ### Tests - 123 tests passed across vector scoring, batch raw scans, batch streaming refinement, metric consistency, and vector filtering. - Added coverage for list, large-list, fixed-size-list, sliced and chunked Arrow arrays; exact per-query scores and row positions; all-null blocks; invalid dimensions/elements; non-finite fallback; and invalid physical children under null parents. - Flake8, Python 3.6 syntax parsing, and `git diff --check` passed. -- 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]
