I like the idea!

Think out loud: advanceExact is a rather frequently used method in past experience of using docvalues. I wonder if we can make it also included in the iterator by exposing a sub interface of DocIdSetIterator? It also helps cluster all operations of docs in the exposed iterator.

Indeed advanceExact is ofter used from within collectors. It proved often faster than calling advance and checking if it returns the target doc ID.

I agree that it is not great the way it is. I was actually thinking of going one step further compared with your suggestion and adding advanceExact directly to DocIdSetIterator, changing its contract to no longer guarantee that docID() returns the target if advanceExact returns false, and adding a default implementation that calls advance.

The benefit of adding advanceExact directly to DocIdSetIterator is that then we no longer need to duplicate (and increase polymorphism) impls such as the one returned by DocIdSetIterator#all for both the case when it supports advanceExact and the case when it doesn't.

The benefit of adding advanceExact directly to DocIdSetIterator is that then we no longer need to duplicate (and increase polymorphism) impls such as the one returned by DocIdSetIterator#all for both the case when it supports advanceExact and the case when it doesn't.

+1 for this approach, thanks for sharing and explanation!

It occurs to me that logic like:

DocValuesIterator iter = DocValuesIterator.of(1, 2, 4);
boolean exist = iter.advanceExact(3);
assert exist == false;
iter.intoBitSet(upto, bitset, offset);

now seems problematic, because the default implementation of intoBitSet starts with a docID() that doesn't actually exist?

+1 again to change the contract, let's make docID() always return a existed doc.

This PR has not had activity in the past 2 weeks, labeling it as stale. If the PR is waiting for review, notify the dev@lucene.apache.org list. Thank you for your contribution!

This PR has not had activity in the past 2 weeks, labeling it as stale. If the PR is waiting for review, notify the dev@lucene.apache.org list. Thank you for your contribution!