BlockJoinBulkScorer could check for parent deletions (not children)

[jimczi]

This change ensures that BlockJoinBulkScorer verifies deletions at the parent level.

[jpountz]

Block-joins require deletes to be consistent across parent and child documents. So I wouldn't call it a bug, the implementation should be free to check deletes on child documents or parent documents depending on whichever is more convenient/efficient.

[ChrisHegarty]

Block-joins require deletes to be consistent across parent and child documents.

This seems a little fragile, and place where bugs could hide. I wonder if we should assert this constraint, which would at least catch some of these potential issue usage issues, while not hurt run time performance (too much).

[jpountz]

Indeed. We have a few such assertions already, e.g. to check that docs returned by the child query don't match the parent bitset. Adding a few more would be good if it can be done in a reasonable fashion (e.g. not forcing nextDoc() / advance() to be called just for the sake of validation).

The block-join module also has CheckJoinIndex, which is expected to be used in a more offline fashion and will catch such bugs.

lucene/lucene/join/src/java/org/apache/lucene/search/join/CheckJoinIndex.java

Lines 63 to 84 in a8d8d6b

	for (int child = prevParentDoc + 1; child != parentDoc; child++) {
	final boolean childIsLive = liveDocs.get(child);
	if (parentIsLive != childIsLive) {
	if (parentIsLive) {
	throw new IllegalStateException(
	"Parent doc "
	+ parentDoc
	+ " of segment "
	+ context.reader()
	+ " is live but has a deleted child document "
	+ child);
	} else {
	throw new IllegalStateException(
	"Parent doc "
	+ parentDoc
	+ " of segment "
	+ context.reader()
	+ " is deleted but has a live child document "
	+ child);
	}
	}
	}

[github-actions]

This PR has not had activity in the past 2 weeks, labeling it as stale. If the PR is waiting for review, notify the [email protected] list. Thank you for your contribution!