Caideyipi commented on PR #17871:
URL: https://github.com/apache/iotdb/pull/17871#issuecomment-6846746464

   ### Independent re-review: fixes completed, Draft retained
   
   Completed one review round of the full diff and relevant context, including 
MAD merge/isNull, sampling Row reuse and INT64 precision, DWT levels, 
Histogram/MinMax extremes, Spline timestamps/coordinates/endpoints, and test 
fixtures.
   
   - Reviewed head: `a8e5d9e1f70c240a74c67a7f9135cc5bbc3fb25b`
   - Comparison base: `cca1dbbde6d0dffa38495c92c92a48b2626d2f99`
   - Final local commit: `c1b81b14a3041e5cdb8cce797ee5cd3e9c835d1f`
   - Confirmed source branch and PR head: 
`c1b81b14a3041e5cdb8cce797ee5cd3e9c835d1f`
   
   Six behavior issues and one test-fixture issue were found and fixed:
   
   1. [Spline output 
coordinates](https://github.com/Caideyipi/iotdb/blob/c1b81b14a3041e5cdb8cce797ee5cd3e9c835d1f/library-udf/src/main/java/org/apache/iotdb/library/dprofile/UDTFSpline.java#L120):
 for `t=y=0..4`, `points=4`, rounded timestamps `[0,1,3,4]` previously carried 
values `[0,1.3333,2.6667,4]`. Interpolation now evaluates at the emitted 
timestamps.
   2. [Spline terminal coordinate 
collisions](https://github.com/Caideyipi/iotdb/blob/c1b81b14a3041e5cdb8cce797ee5cd3e9c835d1f/library-udf/src/main/java/org/apache/iotdb/library/dprofile/UDTFSpline.java#L108):
 timestamps near `Long.MAX_VALUE` can share a double coordinate across the full 
long range. The final input value is now retained instead of being discarded.
   3. [AR small-amplitude 
inputs](https://github.com/Caideyipi/iotdb/blob/c1b81b14a3041e5cdb8cce797ee5cd3e9c835d1f/library-udf/src/main/java/org/apache/iotdb/library/dlearn/UDTFAR.java#L143):
 the absolute epsilon cutoff suppressed valid coefficients, e.g. 
`[1e-8,2e-8,3e-8]` should yield `6/7`. The guard now rejects only zero or 
non-finite denominators.
   4. [DTW pattern 
validation](https://github.com/Caideyipi/iotdb/blob/c1b81b14a3041e5cdb8cce797ee5cd3e9c835d1f/library-udf/src/main/java/org/apache/iotdb/library/match/UDAFDTWMatch.java#L213):
 comma-only patterns could pass validation because `split` discarded empty 
elements. Empty elements, including trailing commas, are now rejected.
   5. [Histogram 
boundaries](https://github.com/Caideyipi/iotdb/blob/c1b81b14a3041e5cdb8cce797ee5cd3e9c835d1f/library-udf/src/main/java/org/apache/iotdb/library/dprofile/UDTFHistogram.java#L88):
 with `[0,100]` and 100 buckets, input `29` could fall into bucket `28`. 
Division by bucket width now preserves this exact boundary, with the existing 
extreme-range scaling and a fallback when width underflows.
   6. [Moving-average 
recovery](https://github.com/Caideyipi/iotdb/blob/c1b81b14a3041e5cdb8cce797ee5cd3e9c835d1f/library-udf/src/main/java/org/apache/iotdb/library/dprofile/UDTFMvAvg.java#L76):
 an overflowing sum permanently poisoned later finite windows. A non-finite sum 
is now rebuilt from the current window; non-finite results are skipped and 
subsequent finite windows recover.
   7. [RowWindow 
fixture](https://github.com/Caideyipi/iotdb/blob/c1b81b14a3041e5cdb8cce797ee5cd3e9c835d1f/library-udf/src/test/java/org/apache/iotdb/library/UDFWindowAndQueueTest.java#L3013):
 `getRow()` returned independent objects despite the intended reuse coverage. 
It now exposes a shared mutable Row and actual data types, and sampling tests 
explicitly assert object reuse.
   
   Validation: the six behavior regressions reproduced their failures before 
the production fixes. Spotless, Checkstyle, and `git diff --check` passed. The 
English clean source reactor passed all 123 tests (0 failures/errors/skips); 
the Chinese clean source reactor passed test compilation. Both reactors 
included the required IoTDB source dependencies and disabled local build-cache 
reuse. A local offline fallback was used because remote Maven plugin downloads 
were blocked. All newly added tests were run locally. This was focused 
library-udf verification, not a full-project fresh-dependency build.
   
   Separate baseline observations remain unchanged: 
`ExactOrderStatistics.getMedian` can overflow when adding the two middle 
extreme values, affecting median/MAD; `UDAFDTWMatch.combineState` replays only 
the source's final buffer without merging its historical match results. Both 
were present in the comparison base and were identified by static inspection, 
without executable reproduction in this round. They warrant separate follow-up.
   
   The source branch had not changed before the push. The push and final 
source/PR head confirmation are complete; the PR remains **Draft** because this 
round found behavior/test-logic issues. A fresh session must review the full 
final diff before Ready. No remote 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]

Reply via email to