brk/crates/quickmatch/docs/CHANGELOG.md

Changelog

v0.4.0 - 2026-03-22

Breaking Changes

Rust & JavaScript

• Word indexing now stores all prefixes of each word and compound, increasing memory usage proportionally to total word character count — previously only full words and full compounds were indexed (Rust source, JS source)

New Features

Rust & JavaScript

• Added prefix indexing: every prefix of each word is now stored in the word index (e.g., "dominance" registers "d", "do", "dom", ..., "dominance"), so short queries like "dom" directly match items containing "dominance" without relying on trigram fallback (Rust source, JS source)
• Added prefix indexing for compound words: adjacent word pairs index all prefixes starting from the second word's first character (e.g., "hash" + "rate" indexes "hashr", "hashra", ..., "hashrate"), enabling partial compound matching
• Changed prefix_score from exact word matching to prefix matching (starts_with in Rust, length comparison in JS), so query word "pric" is recognized as a prefix of item word "price" and scored accordingly (Rust source, JS source)
• Added union fallback: when word intersection yields no common items and trigram matching produces no results, the algorithm falls back to union of all known word sets, returning partial matches instead of empty results (Rust source, JS source)
• Changed search flow to try trigram matching first when unknown words exist; if trigram results are found they're returned immediately, otherwise falls back to ranking known-word candidates (intersection or union)

Examples

• Added examples/compare.rs with 80+ test queries covering exact matches, compound words, typos, short prefixes, acronyms, reverse word order, and unknown terms against a metrics dataset (source)

Internal Changes

Rust

• Extracted ranking logic into a standalone fn rank() method that takes an iterator of (*const str, usize) candidates and produces bucket-sorted results (source)
• Added intersect_sets() method that returns Option<FxHashSet> (returning None instead of empty set to distinguish "no known words" from "empty intersection")
• Added union_sets() method that merges multiple FxHashSet references into a single set
• Changed query word deduplication from a separate FxHashSet to inline seen.insert() filtering during collection, then dropping the set immediately
• Used lifetime elision (QuickMatch<'_>) in Send and Sync trait implementations
• Removed doc comments from prefix_score free function

JavaScript

• Replaced flat sorting by (prefixScore, trigramScore, length) with a 3-bucket ranking approach: items are bucketed by prefix score (0, 1, 2), each bucket is independently sorted by (score, length), and results are filled from the highest bucket first until the limit is reached (source)
• Added union() helper function that merges multiple index arrays into a deduplicated result using a Set
• Changed intersect() to return null instead of an empty array when intersection is empty, distinguishing "no results" from "no input"
• Added duplicate prevention in addToIndex(): checks arr[arr.length - 1] !== value before pushing, preventing duplicate entries from prefix indexing
• Removed standalone indexTrigrams() function; trigram indexing is now inlined in the constructor alongside prefix indexing
• Condensed JSDoc class-level documentation to include behavioral descriptions (e.g., "Supports exact words, prefixes, joined words, and typo tolerance")

View changes

---

v0.3.2 - 2026-03-21

Breaking Changes

Rust & JavaScript

• Default separators changed from ['_', '-', ' '] to ['_', '-', ' ', ':', '/'], meaning colon-separated and path-like strings are now split into words by default (Rust source, JS source)

New Features

Rust & JavaScript

• Added compound word indexing: adjacent words in items are indexed as concatenated pairs (e.g., "hash" + "rate" → "hashrate"), so queries like hashrate now match items like hash_rate without relying on trigram fallback (Rust source, JS source)
• Added prefix-based result ranking via a new prefix_score function that scores items as exact match (2), prefix match (1), or no match (0) — results are now sorted by prefix score first, then trigram score, then length, producing significantly more relevant ordering for autocomplete (Rust source, JS source)
• Added with_min_score(n) (Rust) / withMinScore(n) (JS) configuration option to set the minimum trigram score required for fuzzy matches — higher values require more trigram overlap, reducing noise in results (default: 2) (Rust source, JS source)

Documentation

• Added docs/README.md with comprehensive usage examples for both Rust and JavaScript, a "How it works" section explaining the 3-stage matching pipeline (word → compound → trigram), a configuration options table, and performance benchmarks (~26μs/query Rust, ~29μs/query JS) (source)

Examples

• Added examples/colon_test.rs and examples/colon_test.mjs demonstrating separator behavior with colon-containing items and queries under both default and custom separator configurations (Rust source, JS source)

CI

• Added .github/workflows/outdated.yml workflow that runs daily (and on manual trigger) to detect outdated Rust dependencies using cargo-outdated (source)

Bug Fixes

Rust

• Fixed max_word_len calculation that was incorrectly using the full item length instead of individual word length, which could cause valid long words to be rejected from queries (source)
• Fixed query normalization ordering so that emptiness and length checks now happen after trimming whitespace and lowercasing, correctly rejecting whitespace-only or non-ASCII-only queries (source)

Internal Changes

Rust

• Changed query_words collection from FxHashSet to Vec to preserve word ordering, which is required for the new prefix scoring algorithm (source)
• Removed doc comments from matches() and matches_with() public methods

JavaScript

• Replaced string-based separator checking (separators.includes()) with a Uint8Array(128) lookup table for O(1) separator detection instead of O(n) string scanning (source)
• Replaced Map<number, number> score tracking with a pre-allocated Uint32Array(items.length) and a dirty index list, eliminating Map allocation overhead during each search (source)
• Merged sortedByLength() and rankedResults() into a single _rank() method that handles both exact-match and fuzzy-match result paths
• Renamed internal functions for brevity: normalizeQuery → normalize, parseWords → splitWords, addTrigramsToIndex → indexTrigrams, intersectAll → intersect, binarySearch → bsearch, pickTrigramPosition → trigramPosition, scoreByTrigrams → _scoreTrigrams
• Condensed JSDoc annotations to single-line @param format

Workspace

• Reorganized project files: moved LICENSE to docs/LICENSE.md, set Cargo.toml readme field to docs/README.md
• Added *metrics* pattern to .gitignore
• Added examples to jsconfig.json exclude list

View changes

---

v0.3.1 - 2026-01-22

New Features

JavaScript

• Added comprehensive JSDoc type annotations for QuickMatchConfig class properties (separators, limit, trigramBudget) and all public methods (withLimit(), withTrigramBudget(), withSeparators(), matches(), matchesWith()) (source)
• Added @private annotations for internal methods (scoreByTrigrams(), sortedByLength(), rankedResults()) to indicate non-public API
• Added explicit type annotations to Map types: Map<string, number[]> for wordIndex and trigramIndex
• Added jsconfig.json with TypeScript-style type checking enabled (checkJs: true, strict: true) for improved IDE support and static analysis (source)

Internal Changes

JavaScript

• Renamed trigramCount variable to hitCount for consistency with Rust implementation
• Normalized string quotes to double quotes throughout the codebase
• Reformatted multi-line function parameters for readability

View changes

---

v0.3.0 - 2026-01-22

New Features

JavaScript

• Added complete JavaScript port of the library with identical API to Rust, published as quickmatch-js npm package (source)
• Implemented QuickMatch class with word indexing and trigram-based fuzzy matching
• Implemented QuickMatchConfig class with builder methods (withLimit(), withTrigramBudget(), withSeparators())
• Implemented matches() and matchesWith() methods returning results sorted by relevance
• Uses ES modules ("type": "module") for modern JavaScript compatibility

Documentation

• Added JavaScript installation instructions to README (npm install quickmatch-js)

Internal Changes

Rust

• Renamed trigram_count to hit_count to clarify it tracks successful trigram matches rather than total attempts (source)
• Changed trigram budget tracking from increment-based (trigram_count += 1) to decrement-based (budget -= 1) for clearer remaining budget visibility
• Simplified trigram processing loop by removing processed_trigrams flag and using direct budget check (if budget == 0) for early termination

View changes

---

v0.2.0 - 2026-01-14

New Features

Rust

• Initial release of quickmatch - a fuzzy string matching library for autocomplete and search-as-you-type interfaces
• Added QuickMatch struct implementing hybrid matching with word-level indexing and trigram-based fuzzy fallback (source)
• Added QuickMatchConfig for customizing match behavior with builder pattern methods (source):
  • with_limit(n) - Set maximum results (default: 100)
  • with_trigram_budget(n) - Set fuzzy matching depth (default: 6, range: 0-20)
  • with_separators(chars) - Set word separator characters (default: _, -, )
• Added new() constructor for default configuration and new_with() for custom configuration
• Added matches() and matches_with() methods returning results ranked by match quality and length
• Implemented thread-safe design with Send and Sync traits for concurrent usage
• Added zero-copy string storage using pointer-based indexing for memory efficiency
• Implemented smart trigram selection algorithm that samples trigrams at strategic positions (first, last, middle, then alternating) for efficient fuzzy matching
• Added interactive autocomplete example demonstrating product search functionality (source)

View changes