Skip to content

Apache Lucene open-source search software

Fix VectorScorer#bulk eagerly advancing its iterator (#16377)

Michael Marshall9d09aa71eefe
diff --git a/lucene/CHANGES.txt b/lucene/CHANGES.txt
index 979f742b308..e38cbe1a216 100644
--- a/lucene/CHANGES.txt
+++ b/lucene/CHANGES.txt
@@ -224,6 +224,8 @@ Bug Fixes
* GITHUB#16303: Add suppressed exception details when lock acquisition fails in NativeFSLockFactory. (Bharathi-Kanna)
+* Make VectorScorer#bulk default method lazily position its iterator on first use. (Michael Marshall)
+
Changes in Runtime Behavior
---------------------
* GITHUB#14187: The query cache is now disabled by default. (Adrien Grand)
diff --git a/lucene/core/src/java/org/apache/lucene/search/VectorScorer.java b/lucene/core/src/java/org/apache/lucene/search/VectorScorer.java
index f6a2c1f6d6b..1969427ddaf 100644
--- a/lucene/core/src/java/org/apache/lucene/search/VectorScorer.java
+++ b/lucene/core/src/java/org/apache/lucene/search/VectorScorer.java
@@ -50,8 +50,8 @@ public interface VectorScorer {
* docs. The iterator of this instance of VectorScorer should be used and iterated in conjunction
* with the provided matchingDocs iterator to score only the documents that are present in both
* iterators. If the provided matchingDocs iterator is null, then all documents should be scored.
- * Additionally, if the iterators are unpositioned (docID() == -1), this method should position
- * them to the first document.
+ * Additionally, if the iterators are unpositioned (docID() == -1), the first call to {@link
+ * Bulk#nextDocsAndScores} should position them to the first document.
*
* @param matchingDocs Optional filter to iterate over the documents to score
* @return a {@link Bulk} scorer
@@ -63,11 +63,11 @@ public interface VectorScorer {
matchingDocs == null
? iterator()
: ConjunctionUtils.createConjunction(List.of(matchingDocs, iterator()), List.of());
- if (iterator.docID() == -1) {
- iterator.nextDoc();
- }
return (upTo, liveDocs, buffer) -> {
assert upTo > 0;
+ if (iterator.docID() == -1) {
+ iterator.nextDoc();
+ }
buffer.growNoCopy(DEFAULT_BULK_BATCH_SIZE);
int size = 0;
float maxScore = Float.NEGATIVE_INFINITY;
diff --git a/lucene/core/src/test/org/apache/lucene/search/TestVectorScorer.java b/lucene/core/src/test/org/apache/lucene/search/TestVectorScorer.java
index 5432879a134..de728f4db4d 100644
--- a/lucene/core/src/test/org/apache/lucene/search/TestVectorScorer.java
+++ b/lucene/core/src/test/org/apache/lucene/search/TestVectorScorer.java
@@ -32,6 +32,8 @@ import org.apache.lucene.index.VectorEncoding;
import org.apache.lucene.store.Directory;
import org.apache.lucene.tests.index.RandomIndexWriter;
import org.apache.lucene.tests.util.LuceneTestCase;
+import org.apache.lucene.util.BitSetIterator;
+import org.apache.lucene.util.FixedBitSet;
public class TestVectorScorer extends LuceneTestCase {
@@ -64,6 +66,36 @@ public class TestVectorScorer extends LuceneTestCase {
}
}
+ public void testBulkDoesNotEagerlyAdvanceIterator() throws IOException {
+ FixedBitSet bits = new FixedBitSet(10);
+ bits.set(2);
+ bits.set(5);
+ bits.set(7);
+ BitSetIterator sharedIterator = new BitSetIterator(bits, bits.cardinality());
+ VectorScorer scorer =
+ new VectorScorer() {
+ @Override
+ public float score() {
+ return 1f;
+ }
+
+ @Override
+ public DocIdSetIterator iterator() {
+ return sharedIterator;
+ }
+ };
+
+ assertEquals(-1, scorer.iterator().docID());
+ scorer.bulk(null); // constructing a Bulk must not have any observable side effect
+ assertEquals(-1, scorer.iterator().docID());
+
+ FixedBitSet matchBits = new FixedBitSet(10);
+ matchBits.set(2);
+ matchBits.set(7);
+ scorer.bulk(new BitSetIterator(matchBits, matchBits.cardinality()));
+ assertEquals(-1, scorer.iterator().docID());
+ }
+
/** Creates a new directory and adds documents with the given vectors as kNN vector fields */
private Directory getIndexStore(String field, VectorEncoding encoding, float[]... contents)
throws IOException {