TheR1sing3un opened a new pull request, #10005: URL: https://github.com/apache/paimon/pull/10005
### Purpose Multimodal search currently returns projected rows without exposing their relevance scores, and final lookup determines the output order. Add opt-in `with_score(column_name="_score")` and `order_by_score()` for data-evolution tables, including full-text/hybrid queries and local/Ray single and batch vector queries. Scores retain the existing higher-is-better convention; ordering uses descending score and ascending row ID for ties. Queries explicitly selecting only row IDs or scores can skip final lookup when neither a post-filter nor query authorization requires it. Batch queries retain their shared lookup and attach scores separately for each query. Existing queries keep their default behavior. Document score meanings and the scope of the lookup optimization. Related to #8484. ### Tests - Real IVF and raw searches: scores, ordering, metadata-only lookup avoidance, shared batch lookup, duplicate projections, empty results, post-filters, authorization, deletions and pinned snapshots. - Native full-text and hybrid score output; local/Ray parity for single and batch vectors, including refinement. - 167 targeted metadata, batch lookup and multimodal tests passed on Python 3.11 / Arrow 19. Score-output tests also passed with Ray 2.44 / Arrow 18 (43 tests). - Repository-configured Flake8, Apache license checks and Python 3.6 grammar checks 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]
