HyperPrune v2.5.2

Documentation

• Updated the vendored Unity test framework sources to upstream 2.6.3 and removed the local formatting changes
• Added upstream Unity MIT license text at test/unity/LICENSE.txt
• Added a short third-party notices section to README.md pointing to the vendored Unity license

HyperPrune v2.5.1

CI

• Added dedicated verify-5x5 CI job: builds for BOARD_SIZE=5 and runs 1M self-play games with a 200M-entry transposition table to verify perfect play on 5×5

Documentation

• Updated README tagline to reflect verified perfect play on 3×3 to 5×5 (was an unverified claim for all boards up to 8×8)
• Documented bitboard_has_won LUT precondition: player_pieces must have no bits set above MAX_MOVES-1

HyperPrune v2.5.0

Performance

• Replaced bitboard_has_won mask loop with a precomputed lookup table for MAX_MOVES ≤ 16 (i.e. BOARD_SIZE ≤ 4); 512 bytes for 3×3, 64 KiB for 4×4, populated once at startup by init_win_masks()
• Extended lookup table to checkWinner for MAX_MOVES ≤ 16, replacing the bitboard_did_last_move_win mask scan with a single table load (equivalent under the precondition that the board was non-terminal before the move)

Fixed

• Corrected initializeBoard() comment: player_turn is not used by checkWinner(); only move_count is
• Corrected checkWinner() comment: abbreviated return values TIE/CONTINUE changed to actual enum names GAME_TIE/GAME_CONTINUE

HyperPrune v2.4.0

Performance

• Added ENABLE_MOVE_ORDERING compile-time flag to control killer-move and history heuristics; defaults to ON for BOARD_SIZE <= 4 (pruning gains outweigh O(n²) sort cost) and OFF for larger boards where ordering degrades performance

CI

• Reworked test workflow: replaced matrix dimensions for board_size and move_ordering with inner shell loops, reducing job count while covering all board sizes (3–8) × both move-ordering states across every job
• Added Debug build coverage to test (Make), test-cmake, and test-windows jobs
• Added --seed / -S CLI smoke tests to test, test-cmake, and test-windows (3×3 Release runs)
• Added default ENABLE_MOVE_ORDERING selection validation (no explicit override) for BOARD_SIZE=3 and BOARD_SIZE=5 in test, test-cmake, and test-windows, exercising the Makefile and CMake default-derivation logic
• Expanded build-portable to cover all board sizes × both move-ordering states (was 3×3 only)
• Switched test-windows from PowerShell to Bash to fix CMake -D variable expansion

HyperPrune v2.3.4

Performance

• Portable builds (make portable, make install) now use LTO, -fno-semantic-interposition, matching the release build flags (minus -march=native)
• Default CMake Release builds now use LTO, -funroll-loops, and -fno-semantic-interposition; previously these required -DENABLE_NATIVE_OPTIMIZATIONS=ON

Documentation

• Rewrote "How it works" as a step-by-step pipeline walkthrough of a single getAiMove() call, covering board representation, move generation, search entry point, negamax loop, transposition table, terminal conditions, move ordering, and PGO
• Trimmed Highlights section — removed redundant inline explanations already covered in "How it works"
• Added CodeQL, release version, and platforms badges to README
• Added CI/CD bullet to Highlights covering the test matrix and automated release workflow
• Added note after Requirements about 5×5+ boards being impractical for real-time play

HyperPrune v2.3.3

Fixed

• make install now runs make clean after installing, preventing root-owned build artifacts from blocking future make clean calls when installing with sudo
• CI: set cancel-in-progress: false on all workflows so new pushes queue instead of cancelling in-progress runs

Tests

• Renamed nx to x_count in test_restartGame_clears_board_and_resets_turn for consistency with o_count

Documentation

• Clarified that make install and make uninstall require sudo on systems where /usr/local/bin is root-owned
• Corrected PGO description in README from "compiles twice" to the accurate three-step process
• negamax.h: added zobrist_init() and transposition_table_init() to getAiMove() prerequisites

HyperPrune v2.3.2

Added

• -S short flag for --seed (e.g. ttt -S 42 -s 1000)

Documentation

• Added "Prebuilt binaries" section to README with chmod +x instructions for Linux/macOS release downloads
• Added author credit to README and --help output

HyperPrune v2.3.1

Performance

• player_to_index: replaced conditional with player & 1 (ASCII bit-parity)
• getAiMove: cache occupied mask to avoid recomputing x_pieces | o_pieces twice
• Combined throughput improvement: 3×3 release ~7.3 M games/s, PGO ~8.2 M games/s (up from ~7.1 / ~7.7)

HyperPrune v2.3.0

Changed

• Self-play mode now exits with code 1 and prints a diagnostic to stderr if any game is won (perfect play broken); previously continued silently
• Self-play output no longer includes the Outcomes breakdown (x/o/tie counts); all games are expected to be ties
• Self-play output condensed to a single line: N games in X.XXX s (Y.YY M games/s)

Fixed

• --help text now documents that --quiet requires --selfplay
• Corrected inaccurate comments and documentation across the codebase

CI

• CodeQL now runs on all branches and pull requests, not only main
• Simplified self-play verification steps to rely on exit code instead of parsing human-readable output
• Release workflow now runs 1M-game self-play verification before uploading binaries

HyperPrune v2.2.3

Fixed

• Replaced scanf("%d") with fgets+strtol in interactive coordinate input to eliminate undefined behaviour on out-of-range integer input

CI

• Added CodeQL static analysis workflow (security-extended query suite) with weekly schedule
• Bumped github/codeql-action from v3 to v4 for Node.js 24 compatibility