Skip to content

CHANGELOG.md

Latest commit

 

History

History
89 lines (68 loc) · 6.35 KB

CHANGELOG.md

File metadata and controls

89 lines (68 loc) · 6.35 KB
Raw

Changelog

All notable user-facing changes to CompressKit are tracked here.

The project follows Keep a Changelog style categories and uses semantic versioning for releases.

Unreleased

Changed

  • BREAKING: Project refactored to C++17-only. Go and Rust implementations removed.
  • BREAKING: Build system migrated from raw g++ Makefile to CMake.
  • Removed OpenSpec, Cursor, and Claude skill meta-tooling directories.
  • Removed cross-language conformance matrix and streaming API contract tests.
  • Buffer layer rewritten to use in-memory transforms instead of temporary files.
  • Huffman encoding now uses uint64_t code words instead of std::string for 8x density.
  • Huffman decoding now uses 8-bit lookup table per internal node (~8x faster than bit-by-bit tree walk).
  • Range Coder now uses shared frequency_table utilities (was duplicating scale/cumulative/header logic).
  • scale_frequencies rewritten from O(N×M) decrement loop to O(N log N) proportional reduction.
  • build_cumulative now signals all-zero tables via empty result instead of silent fallback.
  • Centralized constants (SYMBOL_LIMIT, EOF_SYMBOL, magic bytes, size limits) in constants.hpp.
  • VitePress documentation pruned of Go/Rust content; bilingual (en/zh) structure retained.
  • Issue templates pruned of Go/Rust/OpenSpec/cross-language references; Language scope reduced to C++17 / Python scripts / Docs (feature template adds CI).

Added

  • CMakeLists.txt with static library target, 4 algorithm executables, and CTest integration.
  • algorithms/shared/cpp/include/compresskit/constants.hpp for shared constants.
  • Shared count_frequencies, scale_frequencies, build_cumulative utilities in frequency_table.hpp.
  • Architecture Decision Records 0001, 0003, 0004, 0005 under docs/adr/ (validation metadata module shape, range coder corpus cap policy, RLE buffered streaming stance, range C++ bench mode migration path).
  • tests/metadata.py validation metadata module (C++17-only: LANGUAGE_ORDER = ("cpp",)).

Changed (merge integration)

  • Merged remote master (PR #9 architecture-deepening) into the C++17-only tree.
  • Dropped ADR 0002 (cross-language semantic error alignment) and docs/architecture/contract-inventory.md: both depend on the removed Go/Rust implementations.
  • CONTEXT.md "参考资料" section rebuilt: keeps ADR 0001/0003/0004/0005 references, drops the 0002 link and the deleted openspec/ spec links.
  • docs/en/architecture/index.md CLI section kept on the C++17-only single-binary contract (./build/<algo>_cpp); dropped multi-language --lang and <algo>_<lang> invocations.

Changed (Clean Code: extract shared utilities)

  • Extracted in-memory little-endian serialization helpers (write_u32_le, write_magic, write_frequency_header, read_frequency_header) to new compresskit/serialization.hpp. Eliminates duplicated push_u32 lambdas and read_frequencies functions across huffman/arithmetic/range (~90 lines removed).
  • Extracted BitWriter and BitReader to new compresskit/bit_io.hpp. Eliminates duplicated BitWriter class across huffman/arithmetic (~28 lines removed).
  • RLE encode now uses shared write_magic and write_u32_le instead of inline copies.
  • Removed unused name parameter from compresskit::cli::run (and all 4 algorithm call sites).
  • Renamed kInitialEncodeOverhead (Google-style) to INITIAL_ENCODE_OVERHEAD (matches codebase UPPER_CASE convention for local constants).

Changed (Clean Code: replace magic numbers with named constants)

  • Added shared binary-format constants to constants.hpp: MAGIC_SIZE, U32_SIZE, BITS_PER_BYTE, BYTE_VALUES, RLE_PAIR_SIZE, STREAM_READ_BUFFER_SIZE, INITIAL_ENCODE_OVERHEAD, INITIAL_DECODE_OVERHEAD.
  • Replaced bare 4 (magic size) with MAGIC_SIZE across huffman/arithmetic/range/rle decode paths and serialization.hpp.
  • Replaced bare 4 (uint32 LE size) with U32_SIZE in serialization.hpp, frequency_table.cpp, and rle inline count decode.
  • Replaced bare 8 / 7 (bits per byte) with BITS_PER_BYTE in bit_io.hpp, huffman decode table, and buffer_api.cpp encode-limit calculation.
  • Replaced bare 256 (byte value count) with BYTE_VALUES in huffman 8-bit decode table.
  • Replaced bare 5 (RLE count+value pair size) with RLE_PAIR_SIZE in rle decode.
  • Replaced bare 32 * 1024 (stream read buffer) with STREAM_READ_BUFFER_SIZE in frequency_table.cpp.
  • Promoted INITIAL_ENCODE_OVERHEAD from buffer_api.cpp anonymous namespace to constants.hpp; replaced bare 2048 reserve overhead in huffman/range encode.
  • Added INITIAL_DECODE_OVERHEAD constant; replaced bare 1024 decode buffer overhead in buffer_api.cpp.
  • Replaced 0xFFFFFFFFu with UINT32_MAX in range coder encoder/decoder state.
  • Named STATE_BYTES = 4 (range coder 32-bit state width) in range main.
  • Named MAX_TREE_NODES = 2 * SYMBOL_LIMIT (huffman worst-case node count) in huffman main.
  • Named EXPECTED_ARGC = 4 (program + mode + input + output) in cli_launcher.cpp.
  • Refactored write_u32_le / write_magic / read_frequency_header and rle count decode from unrolled byte shifts to U32_SIZE/MAGIC_SIZE loops.

Changed (Clean Code: finish magic-number sweep)

  • Converted stream-based write_u32_le / read_u32_le in frequency_table.cpp from unrolled byte shifts to U32_SIZE loops (matches serialization.hpp style).
  • Replaced bare <<8 / >>24 in range coder byte-renormalisation with << BITS_PER_BYTE / >> TOP_BYTE_SHIFT; added local TOP_BYTE_SHIFT = (STATE_BYTES - 1) * BITS_PER_BYTE constant.

1.0.0 - 2026-01-07

Added

  • Huffman Coding, Arithmetic Coding, Range Coder, and Run-Length Encoding implementations.
  • C++17, Go, and Rust command-line tools for all four algorithms.
  • Unified CLI shape: <binary> <encode|decode> <input> <output>.
  • Cross-language file compatibility goals for educational verification.
  • Test data generation scripts and benchmark scripts.
  • VitePress documentation site with English and Chinese content.
  • MIT license, contribution guide, code of conduct, security policy, issue templates, and pull request template.

Security

  • Documented maximum input size of 4 GiB.
  • Documented maximum decoded output size of 1 GiB for decompression-bomb protection.