xumingming commented on PR #58144: URL: https://github.com/apache/spark/pull/58144#issuecomment-6558828827
Thanks for the thorough review, @cloud-fan! All 8 findings are addressed at head 00d32934229: 1. Alias peel: the rename lookup is now an `AttributeMap` keyed on attribute equality without the qualifier, so rank filters written through subquery or DataFrame aliases keep their group limit. 2. Fallback-path pin: the pin is now declined when the bounded plan costs more exchanges than the unbounded legacy pass — the same single exchange comparison the grouped path uses — so the opt-in plan no longer pays an extra exchange on rejected chains. 3. Wide-output riders: chains now emit riders narrow output before wide, then longest ordering first, keeping a wide rider's growing payload out of later windows' partition buffers. 4. Conf doc: the sort-sharing direction is corrected and the rank-filter caveat now covers the no-shuffle-saved top-k loss, in both the conf doc and the rule comments. 5. Test comments: the three stale comments are corrected, plus a full name-comment-body alignment pass over both new suites (two over-claiming test names softened, a fabricated hypothetical rewritten, duplicate-fixture tests cross-referenced). 6. Probe-bound test: the threshold-off pin test now fails without the conf-gated merge check it claims to cover. 7. `pinThroughFilter` guard: the drop-link test now fails without the reorderable guard. 8. The pin rule is now decided by one exchange comparison on every path and described once, consistently, in the conf doc, the TransposeWindow overview, and the `reorderPinnedChain` comment. Both suites stay green: `TransposeWindowSuite` 103/103 and `TransposeWindowQuerySuite` 31/31, run on the final fixtures. -- 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]
