Fix/engine correctness and performance

[ujandey]

Summary

• Fix 1 (LMR): Single null-window search with conditional full-depth re-search only when reduced or score < beta; eliminates the double full-depth re-search bug.
• Fix 2 (fast eval): New _evaluate_material_fast() — O(64) material+PST+bishop-pair only — for quiescence stand_pat; avoids slow king-safety/pawn-map eval on every capture node.
• Fix 3 (adaptive timeout): _timeout_mask tightened from 2047→255 for time controls < 100 ms so short moves don't overshoot their budget.
• Fix 4 (TT mate scores): _score_to_tt / _score_from_tt encode/decode mate scores relative to the storing ply so retrieved values remain correct at other plies.
• Fix 5 (two-pass legal moves): generate_all_legal_moves / get_legal_moves now set board.turn to match the requested side via try/finally, removing the stale-turn bug that caused wrong move counts when called out of turn; two-pass refresh replaces the unbounded recursive retry.
• Fix 6 (in-place sort): order_moves uses list.sort() instead of sorted(), eliminating a full list copy per negamax node.
• Fix 7 (perft try/finally): push/pop in perft wrapped in try/finally so board state is always restored even on exception.
• Fix 8 (_extract_pv): Removed ValueError from the suppressed-exception tuple so real bugs surface instead of silently truncating the PV.
• Fix 9 (en passant undo): Restructured board.undo_move to set the destination square to "." before restoring the captured pawn, closing a window where piece_positions temporarily held a phantom piece.

Test changes

• test_perft.py: corrected kiwipete FEN to standard CPW (no c5 pawn, D2=2039 D3=97862); replaced broken promotions FEN with classic CPW Position 5 (n1n5/PPPk4/..., D1=24 D2=496 D3=9483); added depth-4 for startpos and endgame_ep.
• test_state_integrity.py (new): 5 tests verifying full board-state restore after push/pop, timeout, stop flag, and UCI stop, plus generate_all_legal_moves side-independence.
• test_notation_pgn_uci.py: relaxed go-background timing assertion 50 ms→500 ms for Windows thread-spawn overhead.

Test plan

• pytest -q — all 97 tests pass (perft depth-4 included)
• Kiwipete D1=48, D2=2039, D3=97862 verified against CPW reference
• CPW Position 5 D1=24, D2=496, D3=9483 verified against CPW reference
• All 5 state-integrity tests pass

Summary by CodeRabbit

Release Notes

• Bug Fixes

  • Fixed en passant move undoing to correctly restore game state and captured pieces.

• Performance & Stability

  • Improved move generation accuracy and search robustness to enhance overall engine quality and game strength.
  • Optimized time management and evaluation speed for better performance under tight time controls.
  • Strengthened error recovery and state handling to ensure greater engine stability during searches.

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.

[coderabbitai]

Caution

Review failed

Pull request was closed or merged during review

📝 Walkthrough

Walkthrough

This PR enhances the chess engine's robustness and performance through three interconnected improvements: corrected en passant undo logic in the board state manager, significant refactoring of the move generator to add time management, TT integration with mate scoring, fast material evaluation, and defensive state guards, and expanded test coverage validating state integrity across search and board operations.

Changes

Board State and Search Engine Improvements

Layer / File(s)	Summary
Board State Fix engine/board.py	undo_move now correctly handles en passant captures by restoring the captured piece to its original square (not the end square) and clearing the end square, ensuring proper state restoration.
Core Move Generation and Search engine/move_generator.py	Major refactoring: get_legal_moves now verifies pseudo-legal moves via make_move/undo_move; generate_all_legal_moves refreshes zobrist hash on stale state; perft and perft_divide wrapped in try/finally guards; order_moves sorts in place; new _score_to_tt and _score_from_tt encode/decode mate distances for transposition table storage; _evaluate_material_fast replaces full evaluation in quiescence stand-pat; negamax updated to use TT encoding/decoding and adds late-move reductions; timeout polling made configurable via _timeout_mask.
Exception Handling engine/move_generator.py	_extract_pv exception handling widened to catch IndexError, KeyError, and TypeError for robustness.
Test Coverage—Timing Adjustment tests/test_notation_pgn_uci.py	Tolerance in test_go_runs_in_background_and_stop_finishes_bestmove increased from 0.05s to 0.5s to account for slower execution.
Test Coverage—Perft Data tests/test_perft.py	Test expectations expanded: startpos and endgame_ep now include depth-4 node counts; kiwipete updated with corrected FEN and depth 2–3 expectations; new promotions test case added with depths 1–3.
Test Coverage—State Integrity tests/test_state_integrity.py	New comprehensive module introducing snapshot(board) helper and SearchStateIntegrityTests class verifying board state restoration after timeouts, stop signals, and fuzzing; UciStopStateIntegrityTests verifying position consistency after UCI stop sequences.

Estimated code review effort

🎯 4 (Complex) | ⏱️ ~60 minutes

Poem

🐰 With en passant fixed and search so bright,
Our moves now legal, states set right—
Fast evals dash through quiescence's night,
TT scores dance in encoded light,
While fuzzy tests guard logic's might! 🌟

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 inconclusive)

Check name	Status	Explanation	Resolution
Title check	❓ Inconclusive	The title "Fix/engine correctness and performance" is overly broad and vague, using generic descriptors that don't clearly convey the specific nature of the changes despite the PR involving multiple targeted fixes.	Revise the title to be more specific, such as "Fix en passant undo and improve move generation correctness" or "Improve engine correctness with faster evaluation and TT handling" to better reflect the primary changes.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch fix/engine-correctness-and-performance

Tip

💬 Introducing Slack Agent: The best way for teams to turn conversations into code.

Slack Agent is built on CodeRabbit's deep understanding of your code, so your team can collaborate across the entire SDLC without losing context.

• Generate code and open pull requests
• Plan features and break down work
• Investigate incidents and troubleshoot customer tickets together
• Automate recurring tasks and respond to alerts with triggers
• Summarize progress and report instantly

Built for teams:

• Shared memory across your entire org—no repeating context
• Per-thread sandboxes to safely plan and execute work
• Governance built-in—scoped access, auditability, and budget controls

One agent for your entire SDLC. Right inside Slack.

👉 Get started

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}