Adrien Grand

There is ByteArrayDataInput too.

This looks very similar to the implementation of `Weight#count` on `PointRangeQuery` and should only perform marginally faster? It's uncreal to me whether this PR buys us much.

One thing I'm not totally happy with is that the change assumes that the field is either not indexed or indexed with `LongPoint` while this query would previousy also work...

I was wondering about descending sorts too! Do we actually need to make this configurable on BKD trees, I would rather not add this option and make the binary search...

I'm (maybe naively) assuming that we could work around this case at the inner node level by skipping inner nodes whose max value is equal to the min value if...

I had to add a loop, which makes the patch bigger than it should be due to indentation changes. It's best to review the patch with `?w=1`: https://github.com/apache/lucene/pull/319/files?w=1.

I agree with you that this change is making the code harder to reason about. I'll look into ways to make the logic easier.

Agreed. Maybe another way of fixing it would consist of preserving more significant bits in the scaled values. I'll look into that as well.

That change makes sense to me. FWIW my recollection from profiling DocIdSetBuilder is that the deduplication logic is cheap and most of the time is spent in `LSBRadixSorter#reorder` so it's...

Sorry @zacharymorn I have not forgotten about it but the change is very large and it's hard to find enough adjacent time to review it. I'll do my best to...