lzo_neo 0.2.2

Pure Dart implementation of the LZO1X lossless compression algorithm, with liblzo2-compatible streams and zero native dependencies.

Changelog

0.2.2

• Upgrade to lit 0.2.1 and use its new UnsafeBytes external memory for lzop block scratch buffers (explicitly freed in inally).
• Benchmarked UnsafeBytes for the hot LZO1X-1 dictionary and LZO1X-999 workspace; external memory was 10-20% slower there, so those tables stay in GC typed arrays. The web arena's grow-invalidates-views behavior is also avoided in the codec.
• Lzo1x.dispose and Lzo999Workspace.dispose eagerly drop work memory; reusing a disposed codec throws StateError.
• lzop continues to use BytesReader/BytesMut/Bytes internally.

0.2.1

• Upgrade to lit 0.2.0.
• lzop internals now use blit: BytesReader for header/block parsing, BytesMut for output assembly, and zero-copy Bytes views for payloads.
• Lzop.compressBytes/decompressBytes no longer copy the result.
• Core imports use package:blit/blit_buffer.dart, keeping the web build free of the file-I/O layer.

0.2.0

Performance-focused redesign, with no stream-format changes.

• LZO1X-1 match finder rewritten around 64-bit unaligned loads:
  • 8-byte match extension with mask-based first-difference detection
  • ByteData.getUint64 on the VM, 32-bit fallback on the web
  • 65 536-entry hash table (256 KiB) to remove most collisions
  • inline VM fast path for the match loop
• Decoder copy paths:
  • short non-overlapping matches copied with 8-byte loads
  • trailing state literals copied with explicit stores
  • inlined literal/match helpers
• New Lzo1xSpeed (fastest, balanced, bestRatio) to tune the speed/ratio trade-off; all modes emit standard LZO1X.
• Measured on a 17.5 MiB mixed corpus against liblzo2 2.10 -O2:
  • decompression: ~875-915 MB/s, 80-83% of native
  • LZO1X-1 compression: ~570-585 MB/s mixed (65-67%), ~6.0-6.8 GB/s on 8 MiB text (~100%)
• Lzo1xLevel.fast1 now uses a 256 KiB work table (was 64 KiB). Existing streams are unaffected.
• Tests: 888 pass; full-matrix differential remains 1936/1936.

0.1.2

• Production hardening release.
• CI: all-family fixtures no longer depend on README.md, so documentation edits cannot break the reproducibility gate.
• Performance: LZO1X-1 uses unaligned 32-bit match scanning and 8-byte match extension (~20-30% faster than 0.1.1); family hash tables clear in bulk.
• API: LzoCodec.compressedSizeBound for manual output buffers.
• Added an all-family fuzz suite and codec-API tests.
• Benchmark chart switches to a linear scale; the unused Windows benchmark job and its CMake build were removed.

0.1.1

• Complete liblzo2 2.10 algorithm matrix in both directions: LZO1, LZO1A, LZO1B, LZO1C, LZO1F, LZO1X, LZO1Y, LZO1Z and LZO2A, including all compression levels and LZO1X-999 quality 1..9.
• LZO1X-999 encoder with a reusable 448 KiB workspace, plus preset-dictionary compression/decompression for LZO1X, LZO1Y and LZO1Z.
• lzo1x_optimize and true in-place decompression (Lzo1x.optimizeInto, Lzo1x.decompressInPlace).
• LZO-RLE bitstream version 1 (Linux kernel / zram), encoder and decoder.
• lzop container reader/writer with Adler-32/CRC-32 checksums and stored blocks (package:lzo_neo/lzop.dart).
• Unified LzoAlgorithm / lzoCodec factory.
• Full-matrix vectors and differential verification against liblzo2 2.10.

0.1.0

• Initial release.
• LZO1X decompressor: safe, fully validating, no allocation, liblzo2-compatible error codes.
• LZO1X compressor: lzo1x_1, lzo1x_1_11, lzo1x_1_12, lzo1x_1_15.
• Zero-copy Bytes API (package:blit) and raw Uint8List entry points.
• Overlapping source/destination support for compression.