lzma-rust2

• Add sans-I/O decoders for LZMA1 and LZIP. Thanks @Black-Frost (#109)

Fixed

• Fixed an out-of-bounds panic when encoding more than ~2 GiB through a single stream (#107)

Added

• Add check_type () getter to XzStream. Thanks @Black-Frost (#106)

Added

• Add sans-I/O decoders for LZMA2 and XZ, which allows to better implement async and also multithreaded abstractions. Thanks @Black-Frost (#105)

Fixed

• Hardened the library against malicious or malformed archives.

Fixed

• Fix invalid XZ output for empty streams
• Fix debug assertion panic on valid flushes

Fixed

• Fixed an issue were custom dictionaries could not be used when using LZMA

Fixed

• We provide our own CRC implementation for lzip and xz, which removes the dependency to the crc crate. No functional change to v0.16.0 or v0.16.1

Fixed

• Make sure to at least use crc version 3.2. No functional updates or changes compared to v0.16.1

Fixed

• Fix compatibility issues with the crc crate by raising MSRV to 1.85. No functional updates or changes compared to v0.15.7

Fixed

• Add missing From<TryReserveError> for the no_std Error

Fixed

• Fix possible panic on out of bound access when reading u32_be correctly

Fixed

• Fix possible panic in the range decoder when trying to read u32_be data
• Fix race condition in multithreaded code in error conditions

Changed

• Fix CRC dependency to exactly 3.3 or else it could break some builds, wrongly selecting a different major version.

Changed

• Pin CRC dependency to not break MSRV. Most likely we will increase the MSRV in the next minor release.

Fixed

• Fix possible underflow in BCJ filters (#72)

Fixed

• Fix fuzzing issues found for the LZMA decoder (#69)

Changed

• Breaking change: LzipReader::new() does not return a Return type anymore, but only itself.

Fixed

• Fix issue when reading some compressed lzma2 data, which could lead to an integer overflow (#64)

Fixed

• Fix issue when compiling with feature xz but without including the feature encoder (#61)
• Fix issue when reading XZ files that allocate too many records (#62)

Fixed

• Fix reading XZ files with multiple streams (#57)
• Fix integer overflow when using LZMA2 dictionary sizes near 4 GiB (#58)

Fixed

• Fix the safe build without the "optimization" feature.

Changed

• Breaking change: Changed CheckType's default from CRC32 to CRC64 @sorairolake (#37)
• Breaking change: All Writer now use the AutoFinisher struct to provide auto finishing writer @sorairolake (#46)

Fixed

• Export missing FilterConfig and FilterType @sorairolake (#50)

Changed

• Breaking change: Rename identifiers to follow Rust API Guidelines @sorairolake (#35)
• Breaking change: All single threaded reader / writer are now implementing UnwindSafe and RefUnwindSafe. Before they saved the last std::io:error and kept returning it for all following red/writes. Now they only return that particular error once.

• Reduced MSRV to 1.82

Fixed

• Fixed documentation @sorairolake
• Improve docs.rs feature compatibility @sorairolake
• Internal cleanup @sorairolake

Changed

• Add writers for lzip and xz that finishes the stream on drop. @sorairolake (#26)

Fixed

• Fixed missing call to .finish () when creating files with BCJ filter with XZ.

Fixed

• Fix broken BCJWriter that couldn't properly finish it's encoding process. Now has a proper finish () function.

Fixed

• Add missing "inner ()" and "inner_mut ()" function to the XZ and LZIP reader and writer.

Changed

• XZReader and XZWriter are now Send.
• Most reader and writer now have "inner ()" and "inner_mut ()" functions.

Updated

• Internally refactoring to reduce code duplication for multi-threading code.

Fixed

• Make sure that LZMA reader errors when going out of bound. This could happen if no EOS was found. This was not a memory a safety issue, but instead a problem in how we propagated a fail state in light of certain optimizations.
• Bound the multithreaded reader as to not use too much memory.

Fixed

• Internally we updated the hash function for the match finders to use a golden ratio based hash instead of the old CRC table based hash. In our test data this was a net win, but it turned out, once tested with bigger datasets, this is a net loss. So we returned back to the CRC table approach (we speak here about a change below 0.01%, but measurable).

Added

• Added single threaded and multithreaded encoder and decoder for the LZIP file format.
• Added multithreaded encoder and decoder for the XZ file format.

Changed

• Renamed LZMA2's "independent work unit" naming from "stream" to "chunk" to not confuse it with XZ streams.
• LZMA2Writer now take a LZMA2Option struct. This enables both the LZMA2Writer and LZMA2WriterMT to encode multiple chunks for multithreaded decoding.
• Changed block size of XZOptions to NonZero type.
• Unified the API of the writers as far as possible.

Fixed

• Fixed unbounded spawning of threads when using the multithreaded version of LZMA2 encoder & decoder.
• Fixed performance regression on linux as reported by @chenxiaolong (#10)
• Fixed compatibility with liblzma when creating XZ files as reported by @chenxiaolong (#14)
• XZWriter properly writes out multiple blocks respecting the block_size.

Added

• Added single threaded encoder and decoder for the XZ file format.
• Ported all pre-filters used by both 7zip and XZ.

Fixed

• Fixed issue with the MT reader discovered by the downstream sevenz-rust2 crate because of incorrect chunk cuts (https://github.com/hasenbanck/sevenz-rust2/issues/44).

Updated

• The multithreading now works more efficient, since new threads are only spawned if really needed.

Added

• Added no_std support by disabling the new std feature that is enabled by default. Custom traits and default implementation for &[u8] and &mut [u8] are provided.

Fixed

• Fixed possible deadlocks in the multithreaded encoder and decoder.

Added

• Added multithreaded compression for LZMA2.
• Added multithreaded decompression for LZMA2.

Updated

• Renamed LZMA2Options to LZMAOptions, since it described the way we encode the LZMA encoder, which is shared between LZMA and LZMA2.

Updated

• Increased the encoding performance. For level 0-3 this crate now is faster than lzma. For 4-9 this crate is on same level with liblzma.

Changed

• Feature "asm" changed to "optimization" and is also enabled by default. Have a look at the "Safety" section of the README.md for more details.

Fixed

• No functional changes.
• Fixed the links to the repository.

Updated

• Increased MSRV to v1.85
• Increased the decoding performance while using only safe Rust. On x86-64 the speed-up was quite large when compared to the v0.2 branch (+50% throughput). Have a look at the "Performance" section of the README.md for more details.
• Added feature flag "asm" which is activated at default which increases the decoding speed when using LZMA2. Have a look at the "Safety" section of the README.md for more details.
• Add EncodeMode and MFType enums to public interface (used for the encoder options).

Removed

• Remove byteorder dependency.
• Remove internal types from public interface.

Updated

• No functional updated.
• Moved into its own repository.

Fixed

• Fix integer overflow when decompressing uncompressed files over u32::MAX
• Allow all byteorder versions with major release 1