Is it possible to also include * GITHUB#16400: Restore HNSW graph reuse during merges under per-field vectors formats, including the default codec. Since 10.4.0 the merger tested the per-field wrapper instead of the unwrapped reader, so every merge rebuilt the graph from scratch. (Jeho Jeong)
this was a pretty bad regression I think, although indexing performance only. It is lacking a CHANGES entry on main, sorry (because the bug did not occur on main)! I can address that separately On Wed, Aug 5, 2026 at 4:46 AM Ignacio Vera <[email protected]> wrote: > > +1 > > I spoke with Alan offline and I will be the release manager. I am planning to > build the release candidate tomorrow if there are no concerns. > > Thanks, > > Ignacio > > On Wed, Aug 5, 2026 at 9:31 AM Alan Woodward <[email protected]> wrote: >> >> Hi all, >> >> We’ve identified a nasty bug in lucene 10.5.0, where boolean queries using >> one of the doc-values based range queries as an exclusion clause can return >> incorrect results (see https://github.com/apache/lucene/pull/16450). I >> think this warrants a bugfix release, and I’m happy to act as release >> manager (unless anybody else would like to volunteer :). >> >> We have a number of bug-fixes in the 10.6 changes section, but none of them >> look to me as if they are major enough to need back porting. There are a >> couple of performance bugs which could go either way but I’d lean towards >> keeping the release as small as possible to get this fix out quickly. >> >> - Alan --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
