Skip to content

CHANGELOG.md

Latest commit

 

History

History
164 lines (119 loc) · 7.36 KB

CHANGELOG.md

File metadata and controls

164 lines (119 loc) · 7.36 KB

Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[Unreleased]

[0.6.1] - 2026-03-10

Fixed

  • ShardedIndex.search() raises AxisError when merging results from view and active shards with a single-row 2D query of shape (1, ndim) (#22)

0.6.0 - 2026-02-22

Added

  • dirty write counter property on NphdIndex and ShardedIndex for caller-driven flush policies (#16)
  • "For Coding Agents" reference page in documentation with architecture map, decision dispatch tables, constraints catalog, and task recipes

Fixed

  • Accept bytes and bytearray in pad_vectors() per-element loop (#14)
  • Only count existing keys in dirty counter on remove() (#16)
  • Generate API reference markdown from source for llms-full.txt instead of raw mkdocstrings directives

Changed

  • Reduce logging verbosity: default timer level from INFO to DEBUG, reserve INFO for aggregate summaries like shard load/save counts (#15)
  • Bump usearch-iscc minimum version to 2.24.2
  • Move docs overrides directory into docs/overrides for cleaner project root
  • Make shard_size, connectivity, expansion_add, expansion_search explicit parameters on ShardedNphdIndex.__init__ (fixes griffe doc generation warnings)
  • Resolve all ty type checker warnings caused by upstream type stubs and type narrowing gaps
  • Match NphdIndex.search() signature to parent Index.search() (fixes LSP violation)

0.5.0 - 2026-02-20

Changed

  • Replace custom Numba NPHD metric with native usearch-iscc MetricKind.NPHD for 2-3x faster add and search operations
  • Drop numba dependency (reduces install size and complexity)
  • Rewrite pad_vectors/unpad_vectors as plain Python/NumPy (no JIT compilation)

0.4.0 - 2026-02-19

Added

  • remove() method for all sharded index variants — tombstone-based deletion for view shard entries, lazy USearch deletion for active shard entries
  • upsert() method for all sharded index variants — insert-or-update with last-write-wins batch deduplication
  • compact() method for all sharded index variants — rebuilds view shards excluding tombstoned and cross-shard duplicate entries to reclaim space
  • __delitem__ support (del index[key]) as alias for remove()
  • Dedup-aware keys and vectors iterators — active shard authoritative, view shard entries filtered by tombstones and cross-shard overlap
  • Cross-shard search result suppression — tombstoned and stale view shard copies are filtered from search results with adaptive oversampling
  • Persistent tombstone tracking (tombstones.npy) — deletion state survives save/load cycles

Fixed

  • Widen distance tolerance for HNSW approximate search test on macOS

0.3.0 - 2026-02-17

Added

  • read_only mode for all sharded index variants — opens all shards as memory-mapped views, blocking write operations while allowing full read access
  • add_once() method for skip-if-exists semantics on all sharded index variants
  • reset() method for releasing resources without deleting files on disk

Fixed

  • expansion_search setter now propagates to view shards for consistent search quality
  • Batch search() with mixed-length vectors in ShardedNphdIndex
  • save() now raises TypeError if a path argument is passed to ShardedIndex

Changed

  • ShardedIndex128.add() accepts list[bytes] as batch keys

0.2.1 - 2026-02-16

Changed

  • Shard and bloom filter saves use atomic write (temp file + rename) to prevent corruption from interrupted writes

Fixed

  • Fix type error in ShardedNphdIndex.get() causing CI type check failure

0.2.0 - 2026-02-08

Added

  • ShardedIndex128 — sharded index with 128-bit UUID keys (bytes(16) / np.dtype('V16'))
  • ShardedNphdIndex128 — sharded NPHD index with 128-bit UUID keys for variable-length vectors
  • _UuidKeyMixin providing key-handling hooks, validation, and dispatch for all 128-bit subclasses
  • ScalableBloomFilter support for bytes keys (add, contains, add_batch, contains_batch)
  • Strict validation on 128-bit key operations — wrong key type, length, or dtype raises ValueError
  • upsert() support for 128-bit UUID keys (single bytes(16) and batch V16 ndarray)
  • Python 3.10 and 3.11 support

Changed

  • Extract 6 key-handling hook methods on ShardedIndex (_is_single_key, _bloom_key, _bloom_keys, _normalize_batch_keys, _shard_batch_keys, _key_dtype) for subclass customization
  • Remove UUID workaround hooks (_iter_shard_vectors, _get_shard_vector, _register_view_shard, _search_view_shards) — upstream usearch now supports Index.vectors, Indexes.merge(), and Indexes.search() for uuid-keyed indexes
  • UUID sharded indexes now use C++-optimized Indexes multi-shard search instead of Python-side per-shard iteration
  • Make _merge_batch_matches and _apply_radius_filter dtype-safe using np.zeros_like + mask-copy instead of np.where(..., 0) (V16 arrays do not support scalar 0 fill)
  • Rewrite ShardedIndexedKeys.__array__ to shard-aware concatenation preserving correct dtype
  • Use serialized_length instead of stats.allocated_bytes for shard rotation threshold check (exactly matches on-disk file size and is faster)
  • Amortize rotation size check to avoid O(n) serialized_length call on every add()

Fixed

  • Fix size property using wrong shard list — now sums _viewed_indexes which is always maintained
  • Enable serialized_length property test (fixed upstream in usearch-iscc fork)

0.1.0 - 2026-02-06

Initial release of iscc-usearch.

Added

  • NphdIndex class extending USearch Index with variable-length binary vector support
  • Normalized Prefix Hamming Distance (NPHD) custom metric compiled via Numba
  • Length-prefixed padding for storing variable-length vectors in a fixed-dimension index
  • ShardedIndex for scalable vector storage beyond available RAM
  • ShardedNphdIndex combining NPHD metric with transparent sharding
  • Idempotent upsert method for single and batch operations
  • Bloom filter support for fast key membership checks
  • Cross-shard get, contains, count, keys, and vectors operations
  • Auto-detection of ndim/max_dim when opening existing indexes
  • Timer context manager for logging operation durations
  • Comprehensive test suite with 100% coverage requirement
  • Documentation site using Zensical with Diataxis structure
  • CI workflows for tests (Linux, macOS, Windows) and docs deployment
  • Python 3.12, 3.13, and 3.14 support