Upgrade to Python 3.12+ and optimize hot-path performance#47
Merged
Upgrade to Python 3.12+ and optimize hot-path performance#47
Conversation
Add a node counter to AlphaBeta and a bench mode that searches 48 positions from Stockfish's bench suite, reporting per-position and total nodes, time, and NPS. Node count is deterministic and serves as the primary signal for detecting search behavior changes. Includes a CI workflow that runs on PRs and posts results as a comment.
Bump minimum Python version to 3.12 for interpreter speedups (PEP 709
comprehension inlining, faster f-strings). Fix CI bug where test matrix
python-version was hardcoded instead of using the matrix variable.
Optimize the search hot path (~42% NPS improvement at depth 3):
- Replace board.fen() with board._transposition_key() in caches
- Precompute float("inf"), float("-inf"), Move.null() as module constants
- Use positional arguments in recursive negamax/quiescence calls
- Use board.piece_map() instead of iterating 64 squares
- Convert piece value dicts to tuple indexing
- Replace per-eval dict accumulators with plain integer variables
- Short-circuit syzygy tablebase check when no tablebase loaded
- Add __slots__ to Config dataclass
- Remove copy() calls on immutable integers
- Replace typing imports with built-in generics
Greptile SummaryThis PR upgrades the minimum Python version to 3.12 and implements targeted hot-path optimizations that achieve a ~42% performance improvement (17,430 → 24,765 NPS at depth 3). Key changes:
The changes are well-targeted performance optimizations that maintain correctness while significantly improving search speed. Confidence Score: 4/5
|
| Filename | Overview |
|---|---|
| .github/workflows/ci.yml | Fixed CI bug where test job hardcoded python-version instead of using matrix variable, updated matrix to test 3.12-3.13 |
| moonfish/config.py | Added slots=True to dataclass for faster attribute access and migrated to PEP 604 union syntax |
| moonfish/psqt.py | Converted piece-value dicts to tuples for integer indexing, optimized board_evaluation to use piece_map() instead of iterating 64 squares, replaced board.fen() with _transposition_key() in cache |
| moonfish/engines/alpha_beta.py | Major hot-path optimizations: precomputed constants (INF/NEG_INF/NULL_MOVE), replaced board.fen() with _transposition_key(), removed copy() calls, switched to positional args, added Syzygy short-circuit |
Flowchart
flowchart TD
A[negamax entry] --> B{Check cache with<br/>_transposition_key}
B -->|Hit| C[Return cached result]
B -->|Miss| D{depth <= 0?}
D -->|Yes| E[quiescence_search]
E --> F[eval_board]
F --> G{tablebase loaded?}
G -->|Yes| H[count_pieces<br/>short-circuit]
G -->|No| I[board_evaluation]
I --> J[piece_map iteration<br/>~32 squares]
J --> K[Tuple-indexed<br/>piece tables]
K --> L[Return eval]
D -->|No| M{null_move pruning?}
M -->|Yes| N[Push NULL_MOVE<br/>precomputed constant]
M -->|No| O[organize_moves]
O --> P[Loop moves]
P --> Q[Recursive negamax<br/>positional args]
Q --> R[alpha-beta cutoff]
R --> S[Cache & return]
Last reviewed commit: 035811f
![greptile-apps[bot]](https://avatars.githubusercontent.com/in/867647?s=60&v=4){kind=small}
🔬 Stockfish Benchmark Resultsvs Stockfish Skill Level 3
Non-checkmate endings:
vs Stockfish Skill Level 4
Non-checkmate endings:
vs Stockfish Skill Level 5
Non-checkmate endings:
Configuration
|
⚡ NPS Benchmark Results
Per-position breakdown |
run_bench() was hardcoded to depth=5, ignoring config.negamax_depth from the CLI. Now passes the user-specified depth through.
🔬 Stockfish Benchmark Resultsvs Stockfish Skill Level 3
Non-checkmate endings:
vs Stockfish Skill Level 4
Non-checkmate endings:
vs Stockfish Skill Level 5
Non-checkmate endings:
Configuration
|
⚡ NPS Benchmark Results
Per-position breakdown |
- Fix import sorting in engine files (alphabetical order) - Fix mypy errors in psqt.py by replacing None sentinel with empty list in PESTO tables so element type is consistently list[int] - Fix CI install: use `uv pip install -e .` directly instead of `make install` which creates a venv that conflicts with UV_SYSTEM_PYTHON=1 on macOS (packages installed to framework Python while tests run setup-python Python) - Remove bash -l login shell from test steps to avoid PATH issues
Remove blank lines between third-party and local imports (usort treats chess and moonfish as the same category). Reformat PESTO tuple definitions per black line length.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
python-version: '3.10'instead of using${{ matrix.python-version }}Performance
Depth 3 benchmark: ~17,430 NPS → ~24,765 NPS (~42% improvement)
Key optimizations by impact:
board._transposition_key()overboard.fen()piece_map()+ integer localsINF/NEG_INF/NULL_MOVEfloat()string parsing andMove.null()allocation per callcount_pieces()entirely when no tablebase loaded@dataclass(slots=True)on Configself.config.Xattribute accessNode count changes from ~543K to ~20.9M at depth 3 due to
_transposition_key()capturing different state than FEN (different cache hit patterns). NPS is the metric that matters.Test plan
python -m unittest tests/test.py— parallel tests excluded due to sandbox)moonfish --mode bench --depth 3runs successfully with correct output${{ matrix.python-version }}for test job