Caideyipi commented on PR #17871: URL: https://github.com/apache/iotdb/pull/17871#issuecomment-6843049505
Independent re-review completed for `1b94fa945b8ae6c7f71db515177888e3aa664767` against master baseline `cca1dbbde6d0dffa38495c92c92a48b2626d2f99`, covering the full 84-file diff and relevant context. Fixed in `063697ab5b89817786a987034b92d718f97e6d18`: - **P1 — Reused rows in isometric sampling:** `RowIterator.next()` reuses one `Row`. Retaining those objects repeated the final row, or read a final null row. Sampling now retains indexes and reads each selected row immediately; the test iterator now follows the API's reuse contract. [UDTFSample.java:182](https://github.com/Caideyipi/iotdb/blob/063697ab5b89817786a987034b92d718f97e6d18/library-udf/src/main/java/org/apache/iotdb/library/dprofile/UDTFSample.java#L182) - **P2 — INT64 precision in triangle sampling:** endpoint/all-points branches converted unchanged values through `double`, e.g. `9007199254740993` became `9007199254740992`. These branches now preserve typed values. [UDTFSample.java:168](https://github.com/Caideyipi/iotdb/blob/063697ab5b89817786a987034b92d718f97e6d18/library-udf/src/main/java/org/apache/iotdb/library/dprofile/UDTFSample.java#L168) - **P2 — Excess DWT layers:** Java's shift-distance masking made `layer=33` reconstruct the wrong signal. Inversion now derives the actual forward levels from the signal and filter lengths. [DWTUtil.java:247](https://github.com/Caideyipi/iotdb/blob/063697ab5b89817786a987034b92d718f97e6d18/library-udf/src/main/java/org/apache/iotdb/library/frequency/util/DWTUtil.java#L247) - **P2 — Spline timestamp precision:** converting absolute timestamps to `double` before subtraction collapsed nearby nanosecond timestamps and caused a non-monotonic-sequence exception. The integer difference is now calculated first. [UDTFSpline.java:86](https://github.com/Caideyipi/iotdb/blob/063697ab5b89817786a987034b92d718f97e6d18/library-udf/src/main/java/org/apache/iotdb/library/dprofile/UDTFSpline.java#L86) - **P2 — Pattern-test false positives:** the result-builder stub fabricated a non-null position even without a write. Tests now use `BinaryColumnBuilder` and assert exactly one output position. [UDAFPatternTest.java:281](https://github.com/Caideyipi/iotdb/blob/063697ab5b89817786a987034b92d718f97e6d18/library-udf/src/test/java/org/apache/iotdb/library/UDAFPatternTest.java#L281) The `UDAFMad` conflict resolution correctly preserves the upstream exception for degenerate sketch results, null/non-finite filtering, and empty-input handling. The `UDAFMadTest` proxy's `isNull` response is effective; both MAD tests pass. No further MAD changes were needed. Validation: - All four behavioral regressions were reproduced before their fixes. - Offline clean reactor test (`library-udf -am`, build-cache reuse disabled): **111 passed, 0 failures, 0 errors, 0 skipped**. All newly added tests were run locally. - Offline clean Chinese-locale reactor `test-compile`: passed. - Spotless apply with `with-integration-tests`, lifecycle Checkstyle, English compilation, en/zh message-key and format-specifier parity, and PR-diff whitespace checks: passed. Final local commit, confirmed source-branch head, and confirmed PR head are all `063697ab5b89817786a987034b92d718f97e6d18`. The update is complete and the PR remains **Draft** because this round found behavioral/test-logic issues; a fresh independent re-review is required. No CI waiting or polling was performed. -- 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]
