Skip to content

v15.0.0-beta.20

Pre-release
Pre-release

Choose a tag to compare

@lance-community lance-community released this 13 Nov 15:36
· 70 commits to main since this release

What's Changed

Breaking Changes 🛠

  • feat(index)!: change ngram, rtree, and minhash lsh scalar indexes to row-address domain by @westonpace in #9858
  • perf(encoding): push down MiniBlock range decode and O(1) bitpack take by @Xuanwo in #10909
  • fix(namespace): preserve timestamp and duration precision by @Xuanwo in #11059

Critical Fixes ‼️

  • fix(dataset): honor target stores in shallow clones by @app/lance-gatefixer in #10942

New Features 🎉

Bug Fixes 🐛

  • fix(dataset): reject branch names that exceed storage limits by @app/lance-gatefixer in #10734
  • fix: persist nested view arrays in Arrow Variant columns by @beinan in #10739
  • fix(namespace): publish S3 markers with conditional PUT by @app/lance-gatefixer in #10764
  • fix(dataset): reject incompatible UDF checkpoints by @jackylee-ch in #10771
  • fix(python): train ANN test fixture on all rows by @app/lance-gatefixer in #10785
  • fix(index): don't cache the tagged FRI listing before vector details are inferred by @zehiko in #10791
  • fix(index): reject incompatible SQ models during segment merge by @ddupg in #10797
  • fix(encoding): decode empty variable packed structs by @BubbleCal in #10830
  • fix(index): avoid false consolidation conflicts after refreshing a dataset by @ddupg in #10888
  • fix(java): release row address lookup lock before I/O by @Xuanwo in #10931
  • fix(file): restore Blob v2 descriptor children in the raw file reader by @ump45nose in #10946
  • fix(mem-wal): reject dictionary overflow before accepting writes by @app/lance-gatefixer in #11001
  • fix(fts): consolidate small partitions during incremental updates by @BubbleCal in #11011
  • fix: reject a merge or update whose right side has only the join key by @LuciferYang in #11030
  • fix: reject a replacement data file whose column types differ from the dataset by @LuciferYang in #11031
  • fix: apply chained field metadata updates for one field in order by @LuciferYang in #11033
  • fix(index): classify invalid search parameters as input errors by @jackye1995 in #11034
  • fix(test): serialize allocation group registration by @Xuanwo in #11044

Documentation 📚

Performance Improvements 🚀

  • perf(vector): support adaptive nprobes in shared-scan batch IVF search by @sezruby in #10443
  • perf(index): compile IVF search/build scheduling once, not per S×Q by @wjones127 in #10727
  • perf(index): overlap RTree page reads in prewarm and search_bbox by @LuciferYang in #10748
  • perf(index): stop LabelList has-all-labels search once the intersection is empty by @LuciferYang in #10749
  • perf(index): size BloomFilter zones without serializing and skip the copy on remapped load by @LuciferYang in #10750
  • perf(scan): resolve SQL filter column types against the scanner's schema by @zehiko in #10867
  • perf(encoding): optimize FSST symbol-table reuse and buffer handling by @Xuanwo in #10907
  • perf(encoding): accelerate dictionary interning, sampling, and decode gather by @Xuanwo in #10911
  • perf(encoding): compute block stats lazily and accelerate repdef validity paths by @Xuanwo in #10958
  • perf(dataset): read manifests concurrently in Dataset::versions by @jackylee-ch in #10966
  • perf(encoding): add AVX-512/AVX2 FastLanes bitpacking dispatch and trim chunk allocations by @Xuanwo in #10967
  • perf(encoding): accelerate ALP, RLE, BinaryMiniBlock, AxisDelta, and FullZip paths by @Xuanwo in #10972
  • perf(fts): skip filtered partitions before loading postings by @BubbleCal in #10988
  • perf: binary-search the deletions when resolving fragment take offsets by @LuciferYang in #11028
  • perf(fts): share the jieba dictionary between tokenizer clones by @ethqunzhong in #11056
  • perf(index): overlap shared RaBitQ prewarm planes by @BubbleCal in #11058

Full Changelog: v15.0.0-beta.19...v15.0.0-beta.20