torex_local_store 0.1.6

Ultra-high-performance local storage engine for Flutter powered by Rust. LSM-tree based, zero-copy reads, memory-mapped files, crash recovery, and reactive streams.

Changelog

All notable changes to TorexLocalStore will be documented here.

Format follows Keep a Changelog.

---

0.1.6

Fixed macOS crash on startup — Rust FFI symbols not linked in macOS plugin.

Infrastructure

• Fixed dlsym(RTLD_DEFAULT, frb_get_rust_content_hash): symbol not found on macOS
• macOS TorexLocalStorePlugin.register() now calls frb_get_rust_content_hash() via @_silgen_name, matching iOS approach, to force-link Rust static library symbols
• Cleaned up macOS podspec source_files pattern and OTHER_LDFLAGS

---

0.1.5

Fixed Android read-only filesystem error — Torex runtime now initializes using Dart's path_provider instead of Rust's dirs crate.

Infrastructure

• Fixed Android runtime initialization on devices with read-only default paths
• Torex runtime now uses getApplicationDocumentsDirectory from path_provider for correct writable path on all platforms
• Removed unused Android Flutter embedding dependency and Kotlin config

---

0.1.4

Internal improvements and optimizations.

---

0.1.3

Internal improvements and bug fixes.

---

0.1.2

Fixed pub.dev score issues — improved static analysis, corrected repository URLs, included FRB-generated bindings in package.

Infrastructure

• Included FRB-generated Dart bindings (lib/src/rust/) in package to fix static analysis on pub.dev
• Fixed homepage, repository, issue tracker and documentation URLs to point to correct GitHub repository
• Resolved all Dart analysis warnings and lints

---

0.1.1

Internal improvements and bug fixes.

Rust

• Various internal engine improvements and optimizations.

---

0.1.0

Initial release — ultra-high-performance embedded storage for Flutter.

Storage Engine (Rust)

• LSM-tree core: Memtable → WAL → immutable Segments
• Memory-mapped I/O (mmap) for zero-copy segment reads
• Per-segment Bloom filters — O(1) miss detection at 28 ns
• Append-only Write-Ahead Log (WAL) with 256 KB BufWriter
• Background WAL flush worker (every 200 ms)
• Background compaction worker (every 5 s, adaptive threshold)
• LZ4 compression for segment data blocks (>256 B)
• CRC32 integrity verification per WAL entry and segment
• Lock-free read path — readers never block writers
• Crash recovery via WAL replay on startup

Dart API

• Torex.box(name) — zero-config, auto-initialising collections
• Full CRUD: put, get, delete, exists, getOrDefault, update
• JSON helpers: putJson, getJson, getJsonOrDefault, updateJson
• Typed objects: putObject<T>, getObject<T> with TorexCodec<T>
• Built-in codecs: TorexJsonCodec, TorexStringCodec, TorexListCodec
• Batch operations (single WAL fsync): batchPut, batchDelete, batchPutJson, batchPutObjects
• Bulk read: getAll, getAllJson
• Scan API: scan, scanStrings, scanPrefix, scanRange
• Fluent query builder: .prefix(), .startKey(), .endKey(), .limit(), .offset(), .reverse()
• Query terminals: .find(), .findJson(), .findObjects<T>(), .findKeys()
• Reactive streams: watch(), watchPrefix()
• Collection management: flush(), clear(), count(), keys(), stats()

Configuration

• TorexConfig.new(path) — safe defaults (sync_writes: true, 4 MB memtable)
• TorexConfig.high_throughput(path) — 64 MB memtable, async writes, auto CPU workers
• TorexConfig.low_memory(path) — 1 MB memtable, minimal footprint
• TorexConfig.ultra(path) — 128 MB memtable, no fsync, maximum throughput

Platform Support

• iOS (arm64 device · arm64 + x86_64 simulator)
• Android (arm64-v8a · armeabi-v7a · x86_64)
• macOS (Apple Silicon + Intel universal binary)
• Linux (x86_64)
• Windows (x86_64)

Benchmark Results (Apple M-series)

Operation	Latency
get — memtable hit	34 ns
get — segment (mmap)	294 ns
get — miss (Bloom)	28 ns
put 64 B	~1 µs
batch_put 100 entries	17 µs (5.8 M/s)
delete	595 ns
Mixed put + get	6.3 M ops/s