+1 bumping this up! The list of cool features in Lucene 11 has grown over the past year (not exhaustive, please add more here!):
* Native vector computations (https://github.com/apache/lucene/pull/15508) * More performant query cache (https://github.com/apache/lucene/pull/15558) * FP16 vector encoding (https://github.com/apache/lucene/pull/16383) ... Panama FFM was also finalized in Java 22 (https://openjdk.org/jeps/454), I wonder if that makes memory segment related changes easier to maintain too? I'm also happy to help with the release process, unless someone else wants to do it! - Kaival On 2025/07/09 12:52:21 Uwe Schindler wrote: > Hi, > > as you might have followed by talk at BerlinBuzzwords > (https://2025.berlinbuzzwords.de/session/state-of-native-access-in-apache-lucene/, > > https://www.youtube.com/watch?v=4RxkahnUBvg), I'd like to make the > proposal to release Lucene 11 with Java 25 in November or December this > year. > > The development model of Lucene has changed a bit the last years like > the main branch is now always on latest Java version and we only > backport features which are easy to backport to stable branch (no > general refactorings or features that require new Java versions). The > effect of this (see discussion in the talk above) is that we may need to > align the release of new major versions of Lucene a bit with the LTS > schedule of Java. Java releases a LTS relaese every 2 years, so we could > do the same and release a Lucene major version also every 2 years based > on main branch once it has all updates for the previous LTS release of > Java. This does not mean we release directly after the LTS, but it > should be done between the LTS release of Java and before we upgrade > main branch again. E.g., we could fork the next stable branch after the > LTS release and update of build system. > > I think a new major release every 2 years is a good plan and also fits > what Adrien and Luca told us on Berlinbuzzwords > (https://2025.berlinbuzzwords.de/session/shipping-lucene-10-0-25-years-in-the-making/, > > https://www.youtube.com/watch?v=GRhzgCEL_ac). So lets start with this. > OK, it is a bit shorter than 1 year, but we were lazy with > > New features for Lucene 11: > > * rewrite of IOContext and hopefully better madvise soon (it is ongoing) > * FAISS vector format in sandbox module (this was backported, but > reverted as I don't think we should backport this due to the Java 21 > problems with preview APIs and incompatible changes in the FFM Api: > https://github.com/apache/lucene/pull/14843) > * Binary partitioning merge policy > * expressions module no longer depending on ASM > * priorityqueue hopefully gets faster once its final > * ongoing panama vectorization improvements (I would not backport > those), so they are new features > * new fast and speedy build system > * cleanup of all AccessController/SecurityManager/SecurityException > (to be done...., will open PR soon) > > If you have any cool feature that I forgot please add it here! > > In general due to the changes in the build system, backports are getting > harder from week to week. Sometimes even one-line backports get hard > (change from Groovy to Java code,...), so I'd like to see Lucene 10 as > some intermediate version and 11 should come soon. > > We should possibly solve all issues with backwards compatibility that > are long ongoing discussions before release: > > * allow to read older indexes or not? I would say we should only > prevent anybody from reading indexes older than Lucene 8 or like that. > > Uwe > > -- > Uwe Schindler > Achterdiek 19, D-28357 Bremen > https://www.thetaphi.de > eMail:[email protected] > --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
