Making NO_INTERVALS to be used by clients of Lucene

[expani]

Description

A bug was reported in OpenSearch which caused Interval Queries containing sub-query in rules to fail with Sub-iterators of ConjunctionDISI are not on the same document if an no_match query was generated after going through the search analyser.

A similar issue was found in Lucene a few years back and was fixed by this PR

I am proposing to make NO_INTERVALS present in Lucene as public so it can be used by it's clients like OpenSearch rather than creating a clone of the same.

[jainankitk]

Makes sense for IntervalBuilder clients to use the existing IntervalsSource implementation. Also, safe to make public given the variable is static final. Approved!

[romseygeek]

LGTM. Can you add an entry to CHANGES.txt?

[expani]

@romseygeek Should I add only under Lucene 10.0.0 or other versions as well ?

[romseygeek]

We can backport this to 9x so add the entry under the latest 9x version.

[expani]

@jainankitk @romseygeek Requesting an approval.

[romseygeek]

Do you want to add your name here, so that we can assign credit? Github account name is fine if that's your preference.

[expani]

Added

[uschindler]

Hi, I have one addition I stumbled over serveral times. There's no "official" way to get an IntervalSource that matches no documents/intervals (I have this sometimes in a query parser for Solr to match complex patent restrictions). This class is pkg private:
https://github.com/expani/lucene/blob/c1c70de765cc7ec67762b40846c7be1e1375eb1c/lucene/queries/src/java/org/apache/lucene/queries/intervals/NoMatchIntervalsSource.java

To get an instance there's a workaround, but its undocumented: public static IntervalsSource Intervals#atLeast(int minShouldMatch, IntervalsSource... sources). You can misuse it by passing 1 and an empty array: Intervals.atLeast(1)

I wonder how this differs from this one. The implementation there has no iterator, while this one here has.

I'd suggest to merge those implementations and instead of exposing a constant to end user add another static method like Intervals#noMatch("reason").

Unfortunately this is already merged, but IntervalBuilder looks like wrong class for that.

[uschindler]

Should I open a new issue, or should we revert this one and give a better solution?

[romseygeek]

Let's open a new issue and we can discuss a bit more about when you need such a thing and where it should live.

[uschindler]

OK. Just for brevity: How does the one made public here differs from the already existing class (except that one returns null iterator while the other one returns an empty iterator)?

[uschindler]

See #13388

[romseygeek]

From a quick look we can indeed replace everything with NoMatchIntervalIterator. I think there are some other places that we can make use of the null shortcut as well - for example, DisjunctionIntervalsSource.create() can return null if its input collection is empty.