Adrien Grand

What are the two cases that you have in mind? I don't think that having a collector with a cache makes sense since it has an assumption that leaves are...

The idea makes sense to me, it's interesting to gain additional visibility into what merging decisions the merge policy actually makes (and specifically whether it thinks that some merging is...

> Give up segments. To expand on Robert's point: segments are so core to Lucene that removing segments from Lucene is going to be a full rewrite anyway. It sounds...

See also this PR for reference: https://github.com/apache/lucene/pull/13586, where @benwtrent explores better parallelizing I/O when fetching vectors for the visited nodes of the HNSW graph. It stalled because it made things...

Lucene has some minimal support for asynchronous I/O via `IndexInput#prefetch`, which indicates that the given byte range is going to be needed in the near future. It's used to allow...

I'm happy to see this API being used as it was only added in the last minor release. The change that you are suggested makes sense to me. I'd like...

I think it'd be fine to document this limitation when sorting by field.

Why are the recall values so bad with parent-join queries (whether merging is enabled or not)? Is there a bug?

This is certainly interesting and possibly applicable to Lucene as indexing involves a lot of sorting, but also looks complicated to integrate. Contributions are welcome. :)

> because of the way our vector codecs have different codecs for every possible combination of options (Byte vs Float, quantized vs not) this will be a lot of code...