JingsongLi commented on PR #10135: URL: https://github.com/apache/paimon/pull/10135#issuecomment-5805784716
This fixes a correctness issue with direct user impact: projected reads and `COUNT(*)` can otherwise retain rows deleted through a sequence group. I traced `adjustReadType` through `create(readType)`: adding the delete group’s protected field to the required set causes its full comparator ordering fields to be read, while the requested projection remains the visible schema. The new tests cover empty, reordered, and other-group projections plus composite ordering. Local verification on the isolated patch: all 47 `PartialUpdateMergeFunctionTest` cases passed on JDK 8. I also built the Spark 3.5 test module, but its suite aborted before executing tests because this sandbox blocks Spark’s local socket bind (`java.net.SocketException: Operation not permitted`). The current PR checks show failures in Spark 3 and Flink 1 Connectors/CDC, so please resolve or clearly establish those failures as unrelated before merge; I cannot call this production-ready while CI is red. The fix itself has strong end-to-end value. -- 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]
