dsmiley commented on PR #12306: URL: https://github.com/apache/lucene/pull/12306#issuecomment-1553317548
> A user configures this system property, indexes a 10Billion dimension vector. We can block that with a maximum of 2048, the tested and thus supported maximum limit. I'm presently looking at incorporating this. > Given how simple this change is, is there harm in putting it in our back pocket as an alternative if no other option pans out closer to the 9.7 release? I don't see a justifiable reason to rush this right now. At the very moment, we're just discussing (no merge) because there are vetos. I want to continue this discussion now. If the vetos were lifted then sure; could be delayed but wouldn't want to jeopardize it missing the release. > I don't think anyone is saying we don't make it configurable. I haven't _quite_ heard anyone say that either, but I haven't heard you or Rob suggest an alternative way to make it configurable. I've heard suggestions of writing new codecs or subclassing codecs but that's not configurability; it's _customizability_ at best. And IMO it's too much to ask of most users. > I'm proposing we do it via Lucene95HnswVectorsFormat as per https://github.com/apache/lucene/issues/12309 so we can provide better bwc guarantees instead of a global system property that's harder to provide bwc. I just saw #12309 which is strictly _moving_ the limit, not about making it configurable. I like that it's focused on that, and I welcome that change. This PR here (#12306) is about the configurable characteristic. How is a global system property hard to provide backwards compatibility for? -- 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: issues-unsubscr...@lucene.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org --------------------------------------------------------------------- To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org
