Implement Scrabble move generator (DAWG) with English and Russian rules

A Go library that returns every legal play ranked by score and scores or
validates plays, using the Appel-Jacobson DAWG algorithm over
github.com/iliadenisov/dafsa v1.1.0.

- DAWG move generation (across / down / both), full tournament scoring with a
  per-tile breakdown; public Solver: GenerateMoves (ranked), ScorePlay,
  ValidatePlay.
- Rulesets: English Scrabble, Russian Scrabble, Эрудит (parameterizable Ruleset).
- cmd/builddict (build the DAWG from the dictionaries submodule), cmd/stress
  (self-play benchmark), selfplay engine; brute-force test oracle.
- A GADDAG was implemented, benchmarked and removed (the DAWG was smaller and
  faster for a scoring solver); see RESULTS.md and ALGORITHM.md.
This commit is contained in:
Ilia Denisov
2026-06-01 16:07:32 +02:00
parent f51a1fe2f2
commit 15c7959d96
43 changed files with 3406 additions and 0 deletions
+76
View File
@@ -0,0 +1,76 @@
// Package dict loads the English test dictionary as a DAWG, preferring the serialized
// cache under testdata and falling back to building from the dictionaries submodule.
// Paths are resolved relative to the repository root so it works both from the repo root
// (commands) and from a package directory (tests).
package dict
import (
"os"
"path/filepath"
"github.com/iliadenisov/alphabet"
dawg "github.com/iliadenisov/dafsa"
"scrabble-solver/internal/dictdawg"
"scrabble-solver/internal/wordlist"
)
// MinLen and MaxLen bound playable word lengths (a 15x15 board holds at most 15).
const (
MinLen = 2
MaxLen = 15
)
func exists(p string) bool { _, err := os.Stat(p); return err == nil }
// Root returns the repository root by walking up from the working directory to the
// directory containing go.mod, or "." if none is found.
func Root() string {
dir, err := os.Getwd()
if err != nil {
return "."
}
for {
if exists(filepath.Join(dir, "go.mod")) {
return dir
}
parent := filepath.Dir(dir)
if parent == dir {
return "."
}
dir = parent
}
}
// DAWGCache and WordlistPath locate the English cache file and source word list,
// relative to the repository root.
func DAWGCache() string { return filepath.Join(Root(), "testdata", "sowpods.dawg") }
func WordlistPath() string { return filepath.Join(Root(), "dictionaries", "english", "sowpods.txt") }
// EnglishAvailable reports whether the English dictionary can be loaded (cache or source).
func EnglishAvailable() bool {
return exists(DAWGCache()) || exists(WordlistPath())
}
// EnglishWords returns the encoded English word list (from the submodule source).
func EnglishWords() ([][]byte, error) {
return wordlist.Read(WordlistPath(), alphabet.Latin(), MinLen, MaxLen)
}
// EnglishDAWG returns the English DAWG, loading the cache if present, otherwise building
// it from the word list and caching it (best effort).
func EnglishDAWG() (dawg.Finder, error) {
if exists(DAWGCache()) {
return dictdawg.Load(DAWGCache())
}
words, err := EnglishWords()
if err != nil {
return nil, err
}
f, err := dictdawg.Build(alphabet.Latin(), words)
if err != nil {
return nil, err
}
_ = dictdawg.Save(f, DAWGCache())
return f, nil
}
+30
View File
@@ -0,0 +1,30 @@
// Package dictdawg builds a plain left-to-right DAWG of a dictionary, as used by the
// Appel-Jacobson move generator.
package dictdawg
import (
"github.com/iliadenisov/alphabet"
dawg "github.com/iliadenisov/dafsa"
)
// Build returns a DAWG Finder over words, which must be alphabet-index slices sorted by
// index order and de-duplicated (see wordlist.Encode).
func Build(idx alphabet.Indexer, words [][]byte) (dawg.Finder, error) {
d := dawg.New(idx)
for _, w := range words {
if err := d.AddB(w); err != nil {
return nil, err
}
}
return d.Finish(), nil
}
// Save writes the DAWG to filename. It requires an embedded alphabet (for example
// alphabet.Latin()), so that Load can reconstruct it.
func Save(f dawg.Finder, filename string) error {
_, err := f.Save(filename)
return err
}
// Load reopens a DAWG saved with Save.
func Load(filename string) (dawg.Finder, error) { return dawg.Load(filename) }
+44
View File
@@ -0,0 +1,44 @@
package dictdawg_test
import (
"path/filepath"
"testing"
"github.com/iliadenisov/alphabet"
"scrabble-solver/internal/dictdawg"
"scrabble-solver/internal/wordlist"
)
func TestBuildAndQuery(t *testing.T) {
words := wordlist.Encode([]string{"care", "cares", "cat"}, alphabet.Latin(), 2, 15)
f, err := dictdawg.Build(alphabet.Latin(), words)
if err != nil {
t.Fatal(err)
}
if f.NumAdded() != 3 {
t.Fatalf("NumAdded = %d, want 3", f.NumAdded())
}
if i := f.IndexOfB([]byte{2, 0, 17, 4}); i != 0 { // care
t.Errorf("IndexOf(care) = %d, want 0", i)
}
if i := f.IndexOfB([]byte{2, 0, 19}); i != 2 { // cat
t.Errorf("IndexOf(cat) = %d, want 2", i)
}
if i := f.IndexOfB([]byte{2, 0, 17}); i != -1 { // car (absent)
t.Errorf("IndexOf(car) = %d, want -1", i)
}
path := filepath.Join(t.TempDir(), "d.dawg")
if err := dictdawg.Save(f, path); err != nil {
t.Fatal(err)
}
g, err := dictdawg.Load(path)
if err != nil {
t.Fatal(err)
}
defer g.Close()
if i := g.IndexOfB([]byte{2, 0, 17, 4, 18}); i != 1 { // cares
t.Errorf("loaded IndexOf(cares) = %d, want 1", i)
}
}
+43
View File
@@ -0,0 +1,43 @@
// Package encoding defines the compact byte conventions shared by the board, rack,
// move output and (for letters) the dictionary graph.
//
// One uniform "symbol byte" is used everywhere:
//
// bits 0..5 the alphabet letter index plus one (1..63); 0 means "empty / no tile"
// bit 6 reserved (unused)
// bit 7 Blank — the tile is a blank standing for that letter; it scores 0
//
// The +1 offset lets 0 mean an empty board square. The same byte represents a board
// cell, a placed tile and a rack tile; the graph stores raw letter indexes (without the
// +1).
package encoding
const (
// Blank flags a tile as a blank standing for its letter; a blank scores 0.
Blank byte = 0x80
// Empty is the value of an unoccupied board square.
Empty byte = 0x00
letterBits byte = 0x3f // low 6 bits: letter index + 1
)
// Cell builds the byte for a tile of the given alphabet letter index. When blank is
// true the tile is marked as a blank (it scores 0).
func Cell(letter byte, blank bool) byte {
c := (letter + 1) & letterBits
if blank {
c |= Blank
}
return c
}
// IsEmpty reports whether a board cell is unoccupied.
func IsEmpty(cell byte) bool { return cell&letterBits == 0 }
// Letter returns the alphabet letter index of a non-empty cell or tile byte. The
// result is meaningless for an empty cell.
func Letter(cell byte) byte { return (cell & letterBits) - 1 }
// IsBlank reports whether a cell or tile byte is a blank (scores 0).
func IsBlank(cell byte) bool { return cell&Blank != 0 }
+39
View File
@@ -0,0 +1,39 @@
package encoding
import "testing"
func TestCellRoundTrip(t *testing.T) {
for letter := range byte(26) {
c := Cell(letter, false)
if IsEmpty(c) {
t.Errorf("Cell(%d,false) reports empty", letter)
}
if IsBlank(c) {
t.Errorf("Cell(%d,false) reports blank", letter)
}
if got := Letter(c); got != letter {
t.Errorf("Letter(Cell(%d,false)) = %d", letter, got)
}
b := Cell(letter, true)
if !IsBlank(b) {
t.Errorf("Cell(%d,true) not blank", letter)
}
if got := Letter(b); got != letter {
t.Errorf("Letter(Cell(%d,true)) = %d, want %d", letter, got, letter)
}
}
}
func TestEmpty(t *testing.T) {
if !IsEmpty(Empty) {
t.Error("IsEmpty(Empty) = false")
}
if IsEmpty(Cell(0, false)) {
t.Error("IsEmpty(Cell('a')) = true")
}
// 'a' (index 0) must not collide with empty.
if Cell(0, false) == Empty {
t.Error("Cell('a') collides with Empty")
}
}
+21
View File
@@ -0,0 +1,21 @@
// Package graph provides thin, reusable helpers over a dafsa Cursor that the move
// generator builds on. It keeps the rest of the solver from depending on dafsa
// traversal details directly.
package graph
import dawg "github.com/iliadenisov/dafsa"
// Spell follows the given alphabet indices from the cursor's root. It returns the
// state reached, whether that state is accepting, and whether the whole path exists.
// When ok is false the path ran into a missing edge; n and final are meaningless.
func Spell(c *dawg.Cursor, indices []byte) (n dawg.Node, final, ok bool) {
n = c.Root()
final = c.Final(n)
for _, ix := range indices {
n, final, ok = c.Next(n, ix)
if !ok {
return n, false, false
}
}
return n, final, true
}
+43
View File
@@ -0,0 +1,43 @@
package graph_test
import (
"testing"
"github.com/iliadenisov/alphabet"
dawg "github.com/iliadenisov/dafsa"
"scrabble-solver/internal/graph"
)
// TestSpellSmoke also exercises the go.mod replace => ../dafsa wiring and the new
// dafsa traversal API end-to-end from the solver module.
func TestSpellSmoke(t *testing.T) {
d := dawg.New(alphabet.Latin())
for _, w := range []string{"cat", "cats", "dog"} {
if err := d.Add(w); err != nil {
t.Fatalf("Add(%q): %v", w, err)
}
}
c, err := dawg.NewCursor(d.Finish())
if err != nil {
t.Fatal(err)
}
enc := func(s string) []byte {
b, err := alphabet.Latin().Encode(s)
if err != nil {
t.Fatalf("Encode(%q): %v", s, err)
}
return b
}
if _, final, ok := graph.Spell(c, enc("cat")); !ok || !final {
t.Errorf("Spell(cat): ok=%v final=%v, want both true", ok, final)
}
if _, final, ok := graph.Spell(c, enc("ca")); !ok || final {
t.Errorf("Spell(ca): ok=%v final=%v, want ok=true final=false", ok, final)
}
if _, _, ok := graph.Spell(c, enc("xyz")); ok {
t.Errorf("Spell(xyz): ok=true, want false")
}
}
+77
View File
@@ -0,0 +1,77 @@
// Package wordlist reads dictionaries and encodes them into alphabet-index words,
// ready to add to a DAWG.
package wordlist
import (
"bufio"
"bytes"
"os"
"sort"
"strings"
"github.com/iliadenisov/alphabet"
)
// Encode turns words into alphabet-index slices, keeping only those whose length is in
// [minLen, maxLen] and whose characters all belong to idx's alphabet (case-folded).
// The result is sorted by index order and de-duplicated, as a DAWG builder requires.
func Encode(words []string, idx alphabet.Indexer, minLen, maxLen int) [][]byte {
out := make([][]byte, 0, len(words))
for _, w := range words {
w = strings.TrimSpace(w)
if w == "" {
continue
}
b, err := idx.Encode(strings.ToLower(w))
if err != nil {
continue
}
if len(b) < minLen || len(b) > maxLen {
continue
}
out = append(out, b)
}
sort.Slice(out, func(i, j int) bool { return bytes.Compare(out[i], out[j]) < 0 })
return Dedupe(out)
}
// Read is Encode applied to the lines (one word per line) of the file at path.
func Read(path string, idx alphabet.Indexer, minLen, maxLen int) ([][]byte, error) {
f, err := os.Open(path)
if err != nil {
return nil, err
}
defer f.Close()
var words []string
sc := bufio.NewScanner(f)
sc.Buffer(make([]byte, 1<<20), 1<<20)
for sc.Scan() {
words = append(words, sc.Text())
}
if err := sc.Err(); err != nil {
return nil, err
}
return Encode(words, idx, minLen, maxLen), nil
}
// FoldYo replaces Ё/ё with Е/е. The Russian "Эрудит" variant has no Ё tile and treats
// Е and Ё as the same letter, so apply this when preparing an Эрудит dictionary (it is a
// dictionary-preparation step, not an engine behaviour).
func FoldYo(s string) string {
return strings.NewReplacer("ё", "е", "Ё", "Е").Replace(s)
}
// Dedupe removes adjacent duplicates from a sorted slice of index words in place.
func Dedupe(s [][]byte) [][]byte {
if len(s) == 0 {
return s
}
out := s[:1]
for i := 1; i < len(s); i++ {
if !bytes.Equal(s[i], s[i-1]) {
out = append(out, s[i])
}
}
return out
}
+37
View File
@@ -0,0 +1,37 @@
package wordlist
import (
"testing"
"github.com/iliadenisov/alphabet"
)
func TestFoldYo(t *testing.T) {
if got := FoldYo("ёлка"); got != "елка" {
t.Errorf("FoldYo(ёлка) = %q, want елка", got)
}
if got := FoldYo("Ёжик"); got != "Ежик" {
t.Errorf("FoldYo(Ёжик) = %q, want Ежик", got)
}
}
func TestEncodeFilterSortDedupe(t *testing.T) {
got := Encode([]string{
"cat", "CATS", "ab", "b", "abcdefghi", "cat", " do ", "qu1rk",
}, alphabet.Latin(), 2, 8)
want := [][]byte{
{0, 1}, // ab
{2, 0, 19}, // cat
{2, 0, 19, 18}, // cats (from CATS, case-folded)
{3, 14}, // do (trimmed)
}
if len(got) != len(want) {
t.Fatalf("got %d words %v, want %d", len(got), got, len(want))
}
for i := range want {
if string(got[i]) != string(want[i]) {
t.Errorf("word %d = %v, want %v", i, got[i], want[i])
}
}
}