puzzle-generator/swedish_generator.js

#!/usr/bin/env node
"use strict";

const fs = require("fs");

const W = 9, H = 8;
const MIN_LEN = 2, MAX_LEN = 8;

const DIRS = {
	"1": [-1, 0], // up
	"2": [0, 1],  // right
	"3": [1, 0],  // down
	"4": [0, -1], // left
};

const IS_DIGIT = (ch) => ch >= "1" && ch <= "4";
const IS_LETTER = (ch) => ch >= "A" && ch <= "Z";
const IS_LETTER_CELL = (ch) => ch === "#" || IS_LETTER(ch);

function usage() {
	console.log(`Usage:
  node swedish_generator.js [--seed N] [--pop N] [--gens N] [--tries N] [--words word-list.txt]

Defaults:
  --seed 1
  --pop 18
  --gens 100
  --tries 50
  --words ./word-list.txt
`);
}

function parseArgs(argv) {
	const out = {seed: 1, pop: 18, gens: 100, tries: 50, wordsPath: "./word-list.txt"};
	for (let i = 2; i < argv.length; i++) {
		const a = argv[i];
		const v = argv[i + 1];
		if (a === "--help" || a === "-h") {
			usage();
			process.exit(0);
		}
		if (a === "--seed") out.seed = parseInt(v, 10), i++;
		else if (a === "--pop") out.pop = parseInt(v, 10), i++;
		else if (a === "--gens") out.gens = parseInt(v, 10), i++;
		else if (a === "--tries") out.tries = parseInt(v, 10), i++;
		else if (a === "--words") out.wordsPath = v, i++;
		else throw new Error(`Unknown arg: ${a}`);
	}
	return out;
}

/** Seeded RNG (xorshift32) */
function makeRng(seed) {
	let x = (seed >>> 0) || 1;
	return {
		nextU32() {
			x ^= x << 13;
			x >>>= 0;
			x ^= x >>> 17;
			x >>>= 0;
			x ^= x << 5;
			x >>>= 0;
			return x >>> 0;
		},
		int(min, max) {
			const r = this.nextU32();
			return min + (r % (max - min + 1));
		},
		float() {
			return this.nextU32() / 0xFFFFFFFF;
		},
	};
}

function clamp(x, a, b) { return Math.max(a, Math.min(b, x)); }

function makeEmptyGrid() {
	return Array.from({length: H}, () => Array.from({length: W}, () => "#"));
}

function deepCopyGrid(g) { return g.map(r => r.slice()); }

function gridToString(g) { return g.map(r => r.join("")).join("\n"); }

function renderHuman(g) {
	return g.map(row => row.map(ch => IS_DIGIT(ch) ? " " : ch).join("")).join("\n");
}

/** --- Words / index --- */
function loadWords(wordsPath) {
	let raw = "";
	try {
		raw = fs.readFileSync(wordsPath, "utf8");
	} catch {
		raw = "EU\nUUR\nAUTO\nBOOM\nHUIS\nKAT\nZEE\nRODE\nDRAAD\nKENNIS\nNETWERK\nPAKTE\n";
	}

	const words = raw
		.split(/\r?\n/g)
		.map(s => s.trim().toUpperCase())
		.filter(s => /^[A-Z]{2,8}$/.test(s));

	// index[len] = { words: string[], pos: Array(len) of [26 arrays of indices] }
	const index = new Map();
	const lenCounts = new Map();

	for (const w of words) {
		const L = w.length;
		lenCounts.set(L, (lenCounts.get(L) || 0) + 1);

		if (!index.has(L)) {
			const pos = Array.from({length: L}, () =>
				Array.from({length: 26}, () => [])
			);
			index.set(L, {words: [], pos});
		}
		const entry = index.get(L);
		const idx = entry.words.length;
		entry.words.push(w);
		for (let i = 0; i < L; i++) {
			entry.pos[i][w.charCodeAt(i) - 65].push(idx);
		}
	}

	return {words, index, lenCounts};
}

function intersectSorted(a, b) {
	const out = [];
	let i = 0, j = 0;
	while (i < a.length && j < b.length) {
		const x = a[i], y = b[j];
		if (x === y) {
			out.push(x);
			i++;
			j++;
		} else if (x < y) i++;
		else j++;
	}
	return out;
}

/** returns {indices?: number[], count: number} WITHOUT allocating huge arrays */
function candidateInfoForPattern(entry, pattern /* array char|null */) {
	const lists = [];
	for (let i = 0; i < pattern.length; i++) {
		const ch = pattern[i];
		if (ch && IS_LETTER(ch)) {
			lists.push(entry.pos[i][ch.charCodeAt(0) - 65]);
		}
	}
	if (lists.length === 0) {
		return {indices: null, count: entry.words.length}; // unconstrained
	}
	lists.sort((a, b) => a.length - b.length);
	let cur = lists[0];
	for (let k = 1; k < lists.length; k++) {
		cur = intersectSorted(cur, lists[k]);
		if (cur.length === 0) break;
	}
	return {indices: cur, count: cur.length};
}

/** --- Slots --- */
function extractSlots(grid) {
	const slots = [];
	for (let r = 0; r < H; r++) {
		for (let c = 0; c < W; c++) {
			const d = grid[r][c];
			if (!IS_DIGIT(d)) continue;

			const [dr, dc] = DIRS[d];
			let rr = r + dr, cc = c + dc;
			if (rr < 0 || rr >= H || cc < 0 || cc >= W) continue;
			if (!IS_LETTER_CELL(grid[rr][cc])) continue;

			const cells = [];
			while (rr >= 0 && rr < H && cc >= 0 && cc < W) {
				const ch = grid[rr][cc];
				if (!IS_LETTER_CELL(ch)) break;
				cells.push([rr, cc]);
				rr += dr;
				cc += dc;
				if (cells.length > MAX_LEN) break;
			}

			slots.push({clue: [r, c, d], dir: d, cells, len: cells.length});
		}
	}
	return slots;
}

function hasRoomForClue(grid, r, c, d) {
	const [dr, dc] = DIRS[d];
	let rr = r + dr, cc = c + dc;
	let run = 0;
	while (rr >= 0 && rr < H && cc >= 0 && cc < W && IS_LETTER_CELL(grid[rr][cc]) && run < MAX_LEN) {
		run++;
		rr += dr;
		cc += dc;
	}
	return run >= MIN_LEN;
}

/** --- FAST mask fitness (structural only) --- */
function maskFitness(grid, lenCounts) {
	let penalty = 0;

	// clue density (avoid all digits)
	let clueCount = 0;
	for (let r = 0; r < H; r++) for (let c = 0; c < W; c++) {
		if (IS_DIGIT(grid[r][c])) clueCount++;
	}
	const targetClues = Math.round(W * H * 0.25); // ~18
	penalty += 8 * Math.abs(clueCount - targetClues);

	const slots = extractSlots(grid);
	if (slots.length === 0) return 1e9;

	// coverage counts per letter cell: horiz vs vert
	const covH = Array.from({length: H}, () => Array(W).fill(0));
	const covV = Array.from({length: H}, () => Array(W).fill(0));

	for (const s of slots) {
		const horiz = (s.dir === "2" || s.dir === "4");

		if (s.len < MIN_LEN) penalty += 8000;
		if (s.len > MAX_LEN) penalty += 8000 + (s.len - MAX_LEN) * 500;

		// dictionary availability only (cheap)
		if (s.len >= MIN_LEN && s.len <= MAX_LEN) {
			if (!lenCounts.get(s.len)) penalty += 12000;
		}

		for (const [r, c] of s.cells) {
			if (horiz) covH[r][c] += 1;
			else covV[r][c] += 1;
		}
	}

	// coverage penalties per letter cell
	for (let r = 0; r < H; r++) for (let c = 0; c < W; c++) {
		if (!IS_LETTER_CELL(grid[r][c])) continue;
		const h = covH[r][c], v = covV[r][c];
		if (h === 0 && v === 0) penalty += 1500;
		else if (h > 0 && v > 0) penalty += 0;
		else if (h + v === 1) penalty += 200;
		else penalty += 600;
	}

	// clue clustering (8-connected)
	const seen = Array.from({length: H}, () => Array(W).fill(false));
	const nbrs8 = [[-1, -1], [-1, 0], [-1, 1], [0, -1], [0, 1], [1, -1], [1, 0], [1, 1]];
	for (let r = 0; r < H; r++) for (let c = 0; c < W; c++) {
		if (!IS_DIGIT(grid[r][c]) || seen[r][c]) continue;
		const stack = [[r, c]];
		seen[r][c] = true;
		let size = 0;
		while (stack.length) {
			const [x, y] = stack.pop();
			size++;
			for (const [dr, dc] of nbrs8) {
				const nx = x + dr, ny = y + dc;
				if (nx < 0 || nx >= H || ny < 0 || ny >= W) continue;
				if (seen[nx][ny]) continue;
				if (!IS_DIGIT(grid[nx][ny])) continue;
				seen[nx][ny] = true;
				stack.push([nx, ny]);
			}
		}
		if (size >= 2) penalty += (size - 1) * 120;
	}

	// dead-end-ish letter cell (3+ walls)
	const nbrs4 = [[-1, 0], [1, 0], [0, -1], [0, 1]];
	for (let r = 0; r < H; r++) for (let c = 0; c < W; c++) {
		if (!IS_LETTER_CELL(grid[r][c])) continue;
		let walls = 0;
		for (const [dr, dc] of nbrs4) {
			const rr = r + dr, cc = c + dc;
			if (rr < 0 || rr >= H || cc < 0 || cc >= W) {
				walls++;
				continue;
			}
			if (!IS_LETTER_CELL(grid[rr][cc])) walls++;
		}
		if (walls >= 3) penalty += 400;
	}

	return penalty;
}

/** --- Mask generation (memetic-ish + hillclimb) --- */
function randomMask(rng) {
	const g = makeEmptyGrid();
	const targetClues = Math.round(W * H * 0.25); // ~18
	let placed = 0, guard = 0;

	while (placed < targetClues && guard++ < 4000) {
		const r = rng.int(0, H - 1);
		const c = rng.int(0, W - 1);
		if (IS_DIGIT(g[r][c])) continue;

		const d = String(rng.int(1, 4));
		g[r][c] = d;
		if (!hasRoomForClue(g, r, c, d)) {
			g[r][c] = "#";
			continue;
		}
		placed++;
	}
	return g;
}

function mutate(rng, grid) {
	const g = deepCopyGrid(grid);
	const cx = rng.int(0, H - 1);
	const cy = rng.int(0, W - 1);

	const steps = 4;
	for (let k = 0; k < steps; k++) {
		const rr = clamp(cx + (rng.int(-2, 2) + rng.int(-2, 2)), 0, H - 1);
		const cc = clamp(cy + (rng.int(-2, 2) + rng.int(-2, 2)), 0, W - 1);

		const cur = g[rr][cc];
		if (IS_DIGIT(cur)) {
			g[rr][cc] = "#";
		} else {
			const d = String(rng.int(1, 4));
			g[rr][cc] = d;
			if (!hasRoomForClue(g, rr, cc, d)) g[rr][cc] = "#";
		}
	}
	return g;
}

function crossover(rng, a, b) {
	const out = makeEmptyGrid();
	const cx = (H - 1) / 2;
	const cy = (W - 1) / 2;
	const theta = rng.float() * Math.PI;
	const nx = Math.cos(theta);
	const ny = Math.sin(theta);

	for (let r = 0; r < H; r++) for (let c = 0; c < W; c++) {
		const x = r - cx, y = c - cy;
		const side = x * nx + y * ny;
		out[r][c] = (side >= 0) ? a[r][c] : b[r][c];
	}

	// cleanup invalid clues
	for (let r = 0; r < H; r++) for (let c = 0; c < W; c++) {
		const ch = out[r][c];
		if (IS_DIGIT(ch) && !hasRoomForClue(out, r, c, ch)) out[r][c] = "#";
	}
	return out;
}

function hillclimb(rng, start, lenCounts, limit) {
	let best = deepCopyGrid(start);
	let bestF = maskFitness(best, lenCounts);
	let fails = 0;

	while (fails < limit) {
		const cand = mutate(rng, best);
		const f = maskFitness(cand, lenCounts);
		if (f < bestF) {
			best = cand;
			bestF = f;
			fails = 0;
		} else {
			fails++;
		}
	}
	return best;
}

function similarity(a, b) {
	let same = 0;
	for (let r = 0; r < H; r++) for (let c = 0; c < W; c++) {
		if (a[r][c] === b[r][c]) same++;
	}
	return same / (W * H);
}

function generateMask(rng, lenCounts, popSize, gens) {
	console.log(`generateMask init pop: ${popSize}`);
	let pop = [];
	for (let i = 0; i < popSize; i++) {
		const g = randomMask(rng);
		pop.push(hillclimb(rng, g, lenCounts, 180)); // faster init
	}

	for (let gen = 0; gen < gens; gen++) {
		const children = [];
		const pairs = Math.max(popSize, Math.floor(popSize * 1.5));
		for (let k = 0; k < pairs; k++) {
			const p1 = pop[rng.int(0, pop.length - 1)];
			const p2 = pop[rng.int(0, pop.length - 1)];
			const child = crossover(rng, p1, p2);
			children.push(hillclimb(rng, child, lenCounts, 70)); // light repair
		}

		pop = pop.concat(children);
		pop.sort((x, y) => maskFitness(x, lenCounts) - maskFitness(y, lenCounts));

		// similarity cull
		const next = [];
		for (const cand of pop) {
			if (next.length >= popSize) break;
			let ok = true;
			for (const kept of next) {
				if (similarity(cand, kept) > 0.92) {
					ok = false;
					break;
				}
			}
			if (ok) next.push(cand);
		}
		pop = next;

		if ((gen % 10) === 0) {
			const bestF = maskFitness(pop[0], lenCounts);
			console.log(`  gen ${gen}/${gens} bestFitness=${bestF}`);
		}
	}

	pop.sort((x, y) => maskFitness(x, lenCounts) - maskFitness(y, lenCounts));
	return pop[0];
}

/** --- Fill (CSP) with NO huge candidate arrays --- */
function fillMask(rng, mask, dictIndex, opts = {}) {
	const grid = deepCopyGrid(mask);
	const slots = extractSlots(grid).filter(s => s.len >= MIN_LEN && s.len <= MAX_LEN);

	const used = new Set();
	const assigned = new Map();

	// progress options
	const logEveryMs = opts.logEveryMs ?? 250;
	const timeLimitMs = opts.timeLimitMs ?? 0; // 0 = no limit

	// crossing weight precompute
	const cellCount = Array.from({length: H}, () => Array(W).fill(0));
	for (const s of slots) for (const [r, c] of s.cells) cellCount[r][c]++;

	function slotKey(s) { return `${s.clue[0]},${s.clue[1]}:${s.clue[2]}`; }

	function patternForSlot(s) {
		return s.cells.map(([r, c]) => {
			const ch = grid[r][c];
			return IS_LETTER(ch) ? ch : null;
		});
	}

	function slotScore(s) {
		let cross = 0;
		for (const [r, c] of s.cells) cross += (cellCount[r][c] - 1);
		return cross * 10 + s.len;
	}

	function placeWord(s, w) {
		const undo = [];
		for (let i = 0; i < s.cells.length; i++) {
			const [r, c] = s.cells[i];
			const prev = grid[r][c];
			const ch = w[i];
			if (prev === "#") {
				undo.push([r, c, prev]);
				grid[r][c] = ch;
			} else if (prev !== ch) {
				return null;
			}
		}
		return undo;
	}

	function undoPlace(undo) { for (const [r, c, prev] of undo) grid[r][c] = prev; }

	// ---- progress bar ----
	const t0 = Date.now();
	let lastLog = t0;
	let nodes = 0;
	let backtracks = 0;
	let lastMRV = 0;

	function renderProgress(final = false) {
		const now = Date.now();
		if (!final && (now - lastLog) < logEveryMs) return;
		lastLog = now;

		const done = assigned.size;
		const total = slots.length;
		const pct = total ? Math.floor((done / total) * 100) : 100;
		const barLen = 22;
		const filled = Math.min(barLen, Math.floor((pct / 100) * barLen));
		const bar = `[${"#".repeat(filled)}${"-".repeat(barLen - filled)}]`;

		const elapsed = ((now - t0) / 1000).toFixed(1);
		const msg =
			`${bar} ${done}/${total} slots | nodes=${nodes} | backtracks=${backtracks} | mrv=${lastMRV} | ${elapsed}s`;

		process.stdout.write("\r" + msg.padEnd(120));
		if (final) process.stdout.write("\n");
	}

	function chooseMRV() {
		let best = null;
		let bestInfo = null;

		for (const s of slots) {
			const k = slotKey(s);
			if (assigned.has(k)) continue;

			const entry = dictIndex.get(s.len);
			if (!entry) return {slot: null, info: null};

			const pat = patternForSlot(s);
			const info = candidateInfoForPattern(entry, pat);

			if (info.count === 0) return {slot: null, info: null};

			if (
				!best ||
				info.count < bestInfo.count ||
				(info.count === bestInfo.count && slotScore(s) > slotScore(best))
			) {
				best = s;
				bestInfo = info;
				if (info.count <= 1) break;
			}
		}

		if (!best) return {slot: null, info: {done: true}};
		return {slot: best, info: bestInfo};
	}

	const MAX_TRIES_PER_SLOT = 500;

	function backtrack() {
		nodes++;

		if (timeLimitMs && (Date.now() - t0) > timeLimitMs) return false;

		const pick = chooseMRV();
		if (!pick.slot && pick.info && pick.info.done) return true;
		if (!pick.slot) {
			backtracks++;
			return false;
		}

		lastMRV = pick.info.count;
		renderProgress(false);

		const s = pick.slot;
		const k = slotKey(s);
		const entry = dictIndex.get(s.len);
		const pat = patternForSlot(s);

		const tryWord = (w) => {
			if (!w) return false;
			if (used.has(w)) return false;

			for (let i = 0; i < pat.length; i++) {
				if (pat[i] && pat[i] !== w[i]) return false;
			}

			const undo = placeWord(s, w);
			if (!undo) return false;

			used.add(w);
			assigned.set(k, w);

			if (backtrack()) return true;

			assigned.delete(k);
			used.delete(w);
			undoPlace(undo);
			return false;
		};

		// constrained: iterate indices (bounded)
		if (pick.info.indices && pick.info.indices.length) {
			const idxs = pick.info.indices;
			const L = idxs.length;
			const tries = Math.min(MAX_TRIES_PER_SLOT, L);

			// safe stepping even for L=1
			const start = (L === 1) ? 0 : rng.int(0, L - 1);
			const step = (L <= 1) ? 1 : rng.int(1, L - 1);

			for (let t = 0; t < tries; t++) {
				const idx = idxs[(start + t * step) % L];
				const w = entry.words[idx];
				if (tryWord(w)) return true;
			}
			backtracks++;
			return false;
		}

		// unconstrained: sample without building arrays
		const N = entry.words.length;
		if (N === 0) {
			backtracks++;
			return false;
		}

		const tries = Math.min(MAX_TRIES_PER_SLOT, N);
		const start = (N === 1) ? 0 : rng.int(0, N - 1);
		const step = (N <= 1) ? 1 : rng.int(1, N - 1);

		for (let t = 0; t < tries; t++) {
			const idx = (start + t * step) % N;
			const w = entry.words[idx];
			if (tryWord(w)) return true;
		}

		backtracks++;
		return false;
	}

	renderProgress(false);
	const ok = backtrack();
	renderProgress(true);

	const clueMap = {};
	for (const [k, v] of assigned.entries()) clueMap[k] = v;
	return {ok, grid, clueMap, stats: {nodes, backtracks, seconds: (Date.now() - t0) / 1000}};
}

/** --- Top-level: try mask+fill until success --- */
function generatePuzzle(opts) {
	const rng = makeRng(opts.seed);
	console.time("LOAD_WORDS");
	const dict = loadWords(opts.wordsPath);
	console.timeEnd("LOAD_WORDS");

	for (let attempt = 1; attempt <= opts.tries; attempt++) {
		console.log(`\nAttempt ${attempt}/${opts.tries}`);
		console.time("MASK");
		const mask = generateMask(rng, dict.lenCounts, opts.pop, opts.gens);
		console.timeEnd("MASK");

		console.time("FILL");
		const filled = fillMask(rng, mask, dict.index, {logEveryMs: 200, timeLimitMs: 30000});
		console.timeEnd("FILL");

		if (filled.ok) return {mask, filled};
	}
	return null;
}

module.exports = {parseArgs, generatePuzzle, gridToString};