Apache Lucene open-source search software
Sync CHANGES for 10.5.2
Michael Sokolov2bb5ad256e0a
diff --git a/lucene/CHANGES.txt b/lucene/CHANGES.txtindex 151a044cb13..7d5a01d6125 100644--- a/lucene/CHANGES.txt+++ b/lucene/CHANGES.txt@@ -691,7 +691,7 @@ OtherAPI Changes----------------------+* GITHUB#16654: Public constructors of [Byte|Float]VectorSimilarityQuery are now private (binary-incompatible change).Lucene 10.5.0 (GITHUB#15784) switched these queries from explicit to adaptive HNSW graph traversal but kept theconstructor signatures, so the two positional float arguments silently changed meaning from (traversalSimilarity,@@ -709,6 +709,17 @@ API ChangesBug Fixes---------------------+* GITHUB#16553: Fix stored-fields corruption when PerField#finish throws: the stored-fields frame+ was left open, causing "Wrote N docs, finish called with numDocs=M" at flush. (Michael Marshall)++* GITHUB#16552: Reconnect HNSW nodes whose out-degree decays across successive+ merge-reuse generations, preventing progressive recall loss under low delete+ rates. (Ethan Baker)++* GITHUB#16527: Hunspell: check for cancellation while building the suggestible entry cache. (Ilia Permiashkin)++* GITHUB#16577: Fix incorrect position increment generated by FixedShingleFilter. (Kishor Kharbas)+* GITHUB#16665: Fix Lucene104HnswScalarQuantizedVectorsFormat merges with an asymmetric encoding(1 and 2 bit quantization today) and COSINE similarity: the merge scorer's query side wasquantized from un-normalized vectors while the index side was normalized, so the merge trips an@@ -764,6 +775,9 @@ API Changes* GITHUB#16043: Add field parameter to FlatVectorsReader.getFlatVectorScorer to match the otherscorer methods in that class. (Simon Cooper)+* GITHUB#16042: Deprecate TopFieldCollectorManager#getCollectors. The result of the search operation is returned+ as part of a TopFieldDocs instance, there is no need to check the state of each collector after search. (Prudhvi Godithi)+* GITHUB#15990: Add experimental api to IndexWriter for columnar indexing. (Tim Brooks)* GITHUB#16062: Add FlatFieldVectorsWriter.asKnnVectorValues to return directly a KnnVectorValues "view" over@@ -877,6 +891,10 @@ ImprovementsOptimizations---------------------+* GITHUB#16254: Enable the vectorized findNextGEQ path on aarch64, where it was disabled on cores+ with fewer than 8 int lanes (Graviton2 and Graviton4). The vector compare beats the scalar+ fallback by +75% on Graviton4 and +35% on Graviton2 (AdvanceBenchmark). (Shitij Bhargava)+* GITHUB#16111: Optimize FieldExistsQuery.count() when all docs have the field. (Prithvi S)* GITHUB#16222: MultiTermQuery constant-score wrapper now defers term collection to ScorerSupplier#get()@@ -888,6 +906,11 @@ Optimizations* GITHUB#16176: Restore WANDScorer for TOP_SCORES + minShouldMatch > 1. (Tianxiao Wei)+* GITHUB#16050, GITHUB#16177: Doc-values range queries are now always two-phase iterators, with+ SIMD-accelerated bulk range evaluation for dense numeric doc values exposed through+ TwoPhaseIterator#intoBitSet (overridable for SIMD block evaluation) instead of a separate plain+ DocIdSetIterator, speeding up conjunctions of range filters. (Sagar Upadhyaya, Jim Ferenczi)+* GITHUB#16153: Use TernaryLongHeap in UpdateGraphsUtils for faster HNSW graph merging. (Prithvi S)* GITHUB#15861: Optimise PhraseScorer by short circuiting non competitive documents in TOP_SCORES mode. (Prithvi S)