harshvardhanrp opened a new issue, #16809:
URL: https://github.com/apache/lucene/issues/16809
### Description
> If an index was written with an index sort, and is later opened by an
IndexWriter
> whose config has no index sort, doc-values updates on the sort field are
accepted.
> The same happens after addIndexes(Directory) of a sorted index into an
unsorted writer.
>
> The segments still say they are sorted by that field, but after the update
their
> documents are no longer in that order. Code that trusts the segment sort
then
> returns wrong results with no error:
> - a sorted top-N search returns the wrong top hit (early termination)
> - IndexSortSortedNumericDocValuesRangeQuery misses matching documents
>
>The problem stays after a later merge, even by a writer that has the
correct sort.
> CheckIndex at the slow level reports:
> "segment has indexSort=<long: "sort"> but docID=1998 sorts after
docID=1999"
> The default CheckIndex level does not run this check.
>
> Cause: validateIndexSort() skips everything when config.getIndexSort() is
null,
> and the update guards only check config.getIndexSortFields(), not the sort
> recorded in each segment.
>
> Steps to reproduce:
> 1. Index 2000 docs with NumericDocValuesField("sort", i + 1) and
> setIndexSort(new Sort(new SortField("sort", SortField.Type.LONG))),
then forceMerge(1)
> 2. Open a new IndexWriter with a plain IndexWriterConfig (no index sort)
> 3. updateNumericDocValue(term for the last doc, "sort", 0) -> no exception
> 4. searcher.search(MatchAllDocsQuery.INSTANCE, 1, indexSort) returns
sort=1, expected 0
>
> I have a fix with tests and can open a PR.
### Version and environment details
Lucene main at 18d5c42edb (also checked on current main), JDK 25 (Temurin
25.0.4.1),
Linux (WSL2). Reproduces with assertions on and off.
--
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]
