v0.7.4

Added

• (zstd) accept libzstd's negative "fast" levels
• (zstd) configurable window limit, bounded history, Frame_Content_Size

Fixed

• (xz,bzip2) let finish drain decoded bytes before reporting truncation
• (xz) decode Delta and BCJ filter chains before LZMA2
• (xz) decode concatenated Streams and Stream Padding
• (xz) verify CRC64 and SHA-256 Block Checks
• (bzip2) decoder stops at the stream end and accepts what libbzip2 does
• (bzip2) make encoder output byte-identical to libbzip2, add flush
• (brotli) report StreamEnd from decode and hand back trailing bytes
• (brotli) accept a complex prefix code with a single code-length code
• (brotli) decode a whole stream handed over in one call in linear time
• (brotli) stop quadratic memmoves when one encode call holds many blocks
• (gzip) report StreamEnd when non-gzip bytes follow a member
• (brotli) build without std (replace std-only f32::log2)
• (lzma) make flush encode buffered input and emit settled bytes
• (lzma2) make flush emit pending input as complete chunks
• (xz) make flush emit pending input as complete LZMA2 chunks
• (zstd) make flush emit pending input
• (quantum) keep an odd final input byte when padding at EOF
• (amiga_lzx) accept an odd-sized uncompressed block before another block
• (lzx) close the frame when a match ends exactly on its boundary
• (quantum) allow >16-bit runs in read_many_bits' peek
• (lzx) zero-pad past end of input so the last symbols decode
• (lzx) accept odd-sized uncompressed blocks; split words at frame ends
• (lzx) skip a full pad word before a word-aligned uncompressed block
• (lzx) realign the bitstream to 16 bits at each 32 KiB frame boundary
• (lzx) measure E8 curpos at the 0xE8 byte and translate whole frames
• (bzip2) space out speculative block re-parses to stop quadratic decode
• (bzip2) decode concatenated streams and ignore trailing garbage
• (huffman) bound declared length by payload bits before allocating
• (lzma,lzma2) keep the match finder working past 4 GiB of input
• (lzma) stop the .lzma encoder retaining all compressed output
• (bcj2) reject out_len larger than the input streams before allocating
• (bcj) implement the real xz/7-Zip RISC-V filter
• (bcj) convert ARM64 ADRP like xz (+/-512 MiB only, sign-extended)
• (bcj) make the x86 filter match xz's prev-mask handling
• (deflate,deflate64,zlib) drain pending output through finish
• (lzss) let finish drain a literal or match paused on a full buffer
• (lznt1) use the MS-XCA offset/length split, not its mirror image
• (xpress_huffman) reject streams longer than the u32 length header
• (xpress_huffman) let a match run past the 64 KiB block end
• (lzs) cap match distance at 2047 so it fits the 11-bit offset field
• (arc_squeeze) finish no longer fails when output fills exactly
• (io,tokio_io) make EncoderWriter flush sync-flush the encoder
• (io) keep writer output staged when the inner write fails
• (tokio_io) no WriteZero on zero-input writes; deliver pending decoder output
• (io) don't replay buffered input when the inner reader errors
• (lha) stream lh4-lh7 decoder output instead of materialising it
• (lha) send an unused position table in single-symbol form
• (lha) use the reference lh6/lh7 dictionary sizes and position alphabets
• (rar5) document that reset clears registered file boundaries
• (rar5) drop a filter that starts inside already-emitted output
• (rar5) yield decoded output instead of buffering a whole block
• (rar5) restore the declared unpack size on reset
• (rar2) cap the output preallocation taken from the unpack size
• (rar2) push the repeat-last-match distance into the old-offset ring
• (lzfse) do not consume bytes past the end-of-stream marker
• (xz) do not consume bytes past the Stream Footer
• (lzfse) cap bvx2 commands and literals per block
• (lzfse) decode real bvx2 streams
• (zstd) recognise Skippable_Frame magics
• (traits) default discard_output stops at and reports the stream end
• (embed) decode to the end of the stream into an exact-size buffer
• (deflate,deflate64) keep the preset dictionary across reset
• (docs) deprecations ship in 0.7.4; drop redundant embed::flate link targets
• (hpack,qpack) cap decoded header size to stop reference amplification
• (lha) split static-Huffman blocks at 65535 tokens; cap lh4 distances
• (bzip2, vec) drain pending decoded output after all input is consumed
• (zstd) implicit last Huffman weight when explicit weights sum to a power of two

Other

• derive the alloc-only build's feature list from all
• (lzma,lha) fix the .lzma dictionary note and a private doc link
• (cli) gate the block-buffer drain test on the codecs it drives
• add an MSRV job that checks on the rust-version toolchain
• raise MSRV to Rust 1.95