Pupil

Live link

Overview

Pupil is a chess engine written in Go. The goal of writing this program was to create an engine capable of beating me in a 10 minute game.

Perft tests pass for the initial board state and for kiwipete up to depths 8 and 6 respectively.

Some of the major optimizations include: bitboard piece representation, magic bitboards for move generation, quiescence search, null move pruning, killer move heuristic, MVV-LVA move ordering, object pooling, and a transposition table keyed via zobrist hashes.

Although nowhere near as complete, the code is heavily inspired by the open-source Stockfish project, whose source code I dipped into heavily.

Running the code

GOOS=js GOARCH=wasm go build -o docs/pupil.wasm     # fresh build of wasm
go run .                                            # start local server

Then visit http://localhost:8080.

---

Description

Feature	Implementation
Board representation	Bitboard
Square mapping	Little-endian Rank-file Mapping
Bitboard Serialization	Forward-scanning
Move encoding	16 bit From-to based
Move Generation	Magic Bitboard approach
Search	Alpha-beta with quiescence search, null move pruning, transposition tables, killer moves, and MVV-LVA move ordering.
Evaluation	Material, piece-square tables, pawn structure (doubled/isolated/passed pawns), bishop pair, rook on open files, king safety (pawn shield), and endgame-specific king tables.

---

History

This is my third attempt at a chess engine.

The predecessors:

1. Rhess - a command line game written in Ruby using a mailbox 8x8 board representation. Styled with 100% unicode and playable via keyboard.

2. Gogochess - a chess-move generator written in Go with a custom HashMap based move-list. Passing perft tests up to depth 5 and capable of solving Mate-in-Twos in less than a minute. Configured with HTTP API for browser integration and easy testing.

---

Todo

• Test alpha-beta against negamax.
• Update transposition table for non-perft search
• Cache best move
• Update alpha-beta to search best move first
• Limit cache entry size
• Implement LRU/better cache clearing (simple array)
• Flesh out frontend for better UX
• Experiment compiling to WebAssembly
• Deploy with WebAssembly
• Quiescence search
• Killer moves
• MVV-LVA move ordering
• Move list pooling
• Null move pruning
• Improved evaluation (pawn structure, king safety, endgame)
• Lint

Nice to haves:

• Iterative deepening with time limit
• Concurrent alpha-beta