eyeling 1.32.0 → 1.33.0

package/lib/eyelang/builtins/registry.mjs

@@ -0,0 +1,54 @@
+// Registry for builtins and their execution metadata.
+// The solver uses the metadata to know when a builtin is deterministic, mode-ready, or should fall back to user clauses.
+import { arithmeticBuiltins } from './arithmetic.mjs';
+import { coreBuiltins } from './core.mjs';
+import { stringBuiltins } from './strings.mjs';
+import { listBuiltins } from './lists.mjs';
+import { aggregationBuiltins } from './aggregation.mjs';
+import { formulaBuiltins } from './formula.mjs';
+import { controlBuiltins } from './control.mjs';
+import { sudokuBuiltins } from './sudoku.mjs';
+import { portfolioBuiltins } from './portfolio.mjs';
+import { searchBuiltins } from './search.mjs';
+import { numberTheoryBuiltins } from './number-theory.mjs';
+import { matrixBuiltins } from './matrix.mjs';
+import { alphameticBuiltins } from './alphametic.mjs';
+import { n3Builtins } from './n3.mjs';
+
+export class BuiltinRegistry {
+  constructor() {
+    this.defs = new Map();
+  }
+  add(name, arity, handler, options = {}) {
+    // ready() describes the argument mode in which the builtin is safe to run;
+    // fallbackWhenNotReady keeps user-defined clauses visible outside that mode.
+    this.defs.set(`${name}/${arity}`, {
+      name,
+      arity,
+      handler,
+      deterministic: options.deterministic ?? false,
+      ready: options.ready ?? null,
+      fallbackWhenNotReady: options.fallbackWhenNotReady ?? false,
+      shouldUse: options.shouldUse ?? null,
+    });
+    return this;
+  }
+  get(name, arity) {
+    return this.defs.get(`${name}/${arity}`) ?? null;
+  }
+}
+
+export function createDefaultRegistry() {
+  const registry = new BuiltinRegistry();
+  for (const mod of [coreBuiltins, arithmeticBuiltins, stringBuiltins, listBuiltins, aggregationBuiltins, formulaBuiltins, controlBuiltins, sudokuBuiltins, portfolioBuiltins, searchBuiltins, numberTheoryBuiltins, matrixBuiltins, alphameticBuiltins, n3Builtins]) {
+    mod.register(registry);
+  }
+  return registry;
+}
+
+let defaultRegistry = null;
+
+export function getDefaultRegistry() {
+  if (defaultRegistry == null) defaultRegistry = createDefaultRegistry();
+  return defaultRegistry;
+}

package/lib/eyelang/builtins/search.mjs

@@ -0,0 +1,579 @@
+// Reusable finite-search builtins for examples that would otherwise spend most
+// of their time in small relational generators.  These predicates are generic
+// entry points (graph, CNF, QMC, range, and n-queens helpers), not compiled
+// replacements for particular example predicate names.
+import { atom, compound, deref, listFromItems, numberTerm, properListItems, unify } from '../term.mjs';
+import { compareLexicalOrNumeric } from './arithmetic.mjs';
+
+export const searchBuiltins = {
+  register(registry) {
+    registry.add('atom_range', 4, atomRange, { fallbackWhenNotReady: true, ready: atomRangeReady });
+    registry.add('atom_ranges', 4, atomRanges, { fallbackWhenNotReady: true, ready: atomRangesReady });
+    registry.add('n_queens', 2, nQueens, { fallbackWhenNotReady: true, ready: firstIntReady });
+    registry.add('weighted_hamiltonian_cycle', 4, weightedHamiltonianCycle, { fallbackWhenNotReady: true, ready: weightedGraphReady });
+    registry.add('weighted_hamiltonian_path', 4, weightedHamiltonianPath, { fallbackWhenNotReady: true, ready: weightedGraphReady });
+    registry.add('hamiltonian_cycle', 3, hamiltonianCycle, { fallbackWhenNotReady: true, ready: graphReady });
+    registry.add('fixed_length_cycle', 4, fixedLengthCycle, { fallbackWhenNotReady: true, ready: fixedCycleReady });
+    registry.add('bounded_path', 5, boundedPath, { fallbackWhenNotReady: true, ready: boundedPathReady });
+    registry.add('cnf_model', 3, cnfModel, { fallbackWhenNotReady: true, ready: cnfReady });
+    registry.add('qm_prime_implicants', 4, qmPrimeImplicants, { deterministic: true, ready: qmReady });
+    registry.add('qm_minimal_cover', 4, qmMinimalCover, { deterministic: true, ready: qmReady });
+  }
+};
+
+function firstIntReady(goal, env) { return intTerm(goal.args[0], env) !== null; }
+function graphReady(goal, env) { return atomKey(deref(goal.args[0], env)) !== null && properListItems(goal.args[1], env) !== null; }
+function weightedGraphReady(goal, env) { return graphReady(goal, env); }
+function fixedCycleReady(goal, env) { return atomKey(deref(goal.args[0], env)) !== null && intTerm(goal.args[1], env) !== null; }
+function boundedPathReady(goal, env) { return atomKey(deref(goal.args[0], env)) !== null && atomKey(deref(goal.args[1], env)) !== null && atomKey(deref(goal.args[2], env)) !== null && intTerm(goal.args[3], env) !== null; }
+function atomRangeReady(goal, env) { return atomKey(deref(goal.args[0], env)) !== null && intTerm(goal.args[1], env) !== null && intTerm(goal.args[2], env) !== null; }
+function atomRangesReady(goal, env) { return atomList(goal.args[0], env) !== null && intTerm(goal.args[1], env) !== null && intTerm(goal.args[2], env) !== null; }
+function cnfReady(goal, env) { return atomList(goal.args[0], env) !== null && clauseList(goal.args[1], env) !== null; }
+function qmReady(goal, env) { return numberList(goal.args[0], env) !== null && numberList(goal.args[1], env) !== null && bitTable(goal.args[2], env) !== null; }
+
+function* atomRange({ goal, env }) {
+  // atom_range(Prefix, Low, High, Atom) is a compact generator for atoms such
+  // as n1, n2, ...; it replaces between/3 + atom_concat/3 in large generated
+  // data sets without committing to any domain-specific predicate name.
+  const prefix = atomKey(deref(goal.args[0], env));
+  const low = intTerm(goal.args[1], env);
+  const high = intTerm(goal.args[2], env);
+  if (prefix == null || low == null || high == null) return;
+  const output = deref(goal.args[3], env);
+  if (output.type !== 'var') {
+    const text = atomKey(output);
+    const n = parsePrefixedInteger(prefix, text);
+    if (n == null || n < low || n > high) return;
+    yield env;
+    return;
+  }
+  for (let i = low; i <= high; i++) {
+    const next = env.clone();
+    next.bind(output.name, atom(`${prefix}${i}`));
+    yield next;
+  }
+}
+
+function parsePrefixedInteger(prefix, text) {
+  if (text == null || !text.startsWith(prefix)) return null;
+  const suffix = text.slice(prefix.length);
+  if (!/^\d+$/.test(suffix)) return null;
+  const n = Number(suffix);
+  return Number.isSafeInteger(n) ? n : null;
+}
+
+function* atomRanges({ goal, env }) {
+  // atom_ranges([p, q], Low, High, Atom) is the multi-prefix companion to
+  // atom_range/4. It is useful when several generated atom families feed the
+  // same relation.
+  const prefixes = atomList(goal.args[0], env);
+  const low = intTerm(goal.args[1], env);
+  const high = intTerm(goal.args[2], env);
+  if (!prefixes || low == null || high == null) return;
+  const output = deref(goal.args[3], env);
+  if (output.type !== 'var') {
+    const text = atomKey(output);
+    for (const prefix of prefixes) {
+      const n = parsePrefixedInteger(prefix, text);
+      if (n != null && n >= low && n <= high) { yield env; return; }
+    }
+    return;
+  }
+  for (const prefix of prefixes) {
+    for (let i = low; i <= high; i++) {
+      const next = env.clone();
+      next.bind(output.name, atom(`${prefix}${i}`));
+      yield next;
+    }
+  }
+}
+
+function* nQueens({ goal, env }) {
+  // n_queens(N, Solution) performs the same finite search as the declarative
+  // select/not_member version, but keeps occupied diagonals in sets.
+  const n = intTerm(goal.args[0], env);
+  if (n == null || n < 0 || n > 14) return;
+  const cols = Array.from({ length: n }, (_, i) => i + 1);
+  const chosen = [];
+  const down = new Set();
+  const up = new Set();
+  function* place(row, remaining) {
+    if (remaining.length === 0) { yield chosen.slice(); return; }
+    for (let i = 0; i < remaining.length; i++) {
+      const q = remaining[i];
+      const d = row + q;
+      const u = row - q;
+      if (down.has(d) || up.has(u)) continue;
+      chosen.push(q); down.add(d); up.add(u);
+      const rest = remaining.slice(0, i).concat(remaining.slice(i + 1));
+      yield* place(row + 1, rest);
+      up.delete(u); down.delete(d); chosen.pop();
+    }
+  }
+  for (const solution of place(1, cols)) {
+    const next = env.clone();
+    if (unify(goal.args[1], listFromItems(solution.map(numberTerm)), next)) yield next;
+  }
+}
+
+function* weightedHamiltonianCycle({ solver, goal, env }) {
+  // weighted_hamiltonian_cycle(EdgePred, Cities, Cycle, Cost) treats
+  // EdgePred/3 facts as undirected weighted edges and enumerates
+  // symmetry-broken cycles beginning at the first city.
+  const predicate = atomKey(deref(goal.args[0], env));
+  const cities = properListItems(goal.args[1], env);
+  if (!predicate || !cities || cities.length < 2) return;
+  const weights = edgeWeightMap(solver.program, predicate);
+  if (!weights) return;
+  const start = cities[0];
+  const rest = cities.slice(1);
+  for (const order of permutations(rest)) {
+    if (compareLexicalOrNumeric(atomKey(order[0]), atomKey(order[order.length - 1])) >= 0) continue;
+    const cost = weightedPathCost([start, ...order, start], weights);
+    if (cost == null) continue;
+    const next = env.clone();
+    if (unify(goal.args[2], listFromItems([start, ...order, start]), next) && unify(goal.args[3], numberTerm(cost), next)) yield next;
+  }
+}
+
+function* weightedHamiltonianPath({ solver, goal, env }) {
+  // weighted_hamiltonian_path(EdgePred, Cities, Path, Cost) enumerates paths
+  // from the first city through every remaining city exactly once.
+  const predicate = atomKey(deref(goal.args[0], env));
+  const cities = properListItems(goal.args[1], env);
+  if (!predicate || !cities || cities.length < 1) return;
+  const weights = edgeWeightMap(solver.program, predicate);
+  if (!weights) return;
+  const start = cities[0];
+  const rest = cities.slice(1);
+  for (const order of permutations(rest)) {
+    const path = [start, ...order];
+    const cost = weightedPathCost(path, weights);
+    if (cost == null) continue;
+    const next = env.clone();
+    if (unify(goal.args[2], listFromItems(path), next) && unify(goal.args[3], numberTerm(cost), next)) yield next;
+  }
+}
+
+function weightedPathCost(path, weights) {
+  let cost = 0;
+  for (let i = 1; i < path.length; i++) {
+    const w = weights.get(`${atomKey(path[i - 1])}\x1f${atomKey(path[i])}`);
+    if (w == null) return null;
+    cost += w;
+  }
+  return cost;
+}
+
+function edgeWeightMap(program, predicate) {
+  const map = new Map();
+  let count = 0;
+  for (const clause of program.clauses) {
+    if (clause.body.length !== 0) continue;
+    const h = clause.head;
+    if (h.type !== 'compound' || h.name !== predicate || h.arity !== 3) continue;
+    const a = atomKey(h.args[0]), b = atomKey(h.args[1]);
+    const w = intTerm(h.args[2], null);
+    if (a == null || b == null || w == null) continue;
+    map.set(`${a}\x1f${b}`, w);
+    map.set(`${b}\x1f${a}`, w);
+    count++;
+  }
+  return count ? map : null;
+}
+
+function* hamiltonianCycle({ solver, goal, env }) {
+  // hamiltonian_cycle(EdgePred, Vertices, Cycle) treats EdgePred/2 facts as an
+  // undirected graph and enumerates cycles starting at Vertices[0].
+  const predicate = atomKey(deref(goal.args[0], env));
+  const vertices = properListItems(goal.args[1], env);
+  if (!predicate || !vertices || vertices.length < 2) return;
+  const graph = undirectedGraph(solver.program, predicate);
+  if (!graph) return;
+  const start = atomKey(vertices[0]);
+  if (start == null) return;
+  const rest = vertices.slice(1).map(atomKey);
+  if (rest.some((v) => v == null)) return;
+  const path = [start];
+  function* dfs(current, remaining) {
+    if (remaining.length === 0) {
+      if (!graph.has(`${current}\x1f${start}`)) return;
+      const next = env.clone();
+      if (unify(goal.args[2], listFromItems([...path, start].map(atom)), next)) yield next;
+      return;
+    }
+    for (let i = 0; i < remaining.length; i++) {
+      const v = remaining[i];
+      if (!graph.has(`${current}\x1f${v}`)) continue;
+      path.push(v);
+      yield* dfs(v, remaining.slice(0, i).concat(remaining.slice(i + 1)));
+      path.pop();
+    }
+  }
+  yield* dfs(start, rest);
+}
+
+function undirectedGraph(program, predicate) {
+  const edges = new Set();
+  let count = 0;
+  for (const clause of program.clauses) {
+    if (clause.body.length !== 0) continue;
+    const h = clause.head;
+    if (h.type !== 'compound' || h.name !== predicate || h.arity !== 2) continue;
+    const a = atomKey(h.args[0]), b = atomKey(h.args[1]);
+    if (a == null || b == null) continue;
+    edges.add(`${a}\x1f${b}`);
+    edges.add(`${b}\x1f${a}`);
+    count++;
+  }
+  return count ? edges : null;
+}
+
+function* fixedLengthCycle({ solver, goal, env }) {
+  // fixed_length_cycle(EdgePred, Length, Relation, Cycle) is useful for RDF-like
+  // edge(Source, Relation, Target) data. It enumerates closed walks of exactly
+  // Length steps and returns the relation label and node list.
+  const predicate = atomKey(deref(goal.args[0], env));
+  const length = intTerm(goal.args[1], env);
+  if (!predicate || length == null || length < 0) return;
+  const graph = labelledGraph(solver.program, predicate);
+  if (!graph) return;
+
+  const requestedRelation = atomKey(deref(goal.args[2], env));
+  const requestedCycle = fixedLengthCycleRequestedPath(goal, env, length);
+  if (requestedRelation && requestedCycle) {
+    const byStart = graph.byRelation.get(requestedRelation);
+    if (byStart && fixedLengthCyclePathExists(byStart, requestedCycle)) {
+      const next = env.clone();
+      if (unify(goal.args[2], atom(requestedRelation), next) && unify(goal.args[3], listFromItems(requestedCycle.map(atom)), next)) yield next;
+    }
+    return;
+  }
+
+  const relations = requestedRelation
+    ? [[requestedRelation, graph.byRelation.get(requestedRelation)]].filter(([, byStart]) => byStart)
+    : graph.byRelation.entries();
+  for (const [relation, byStart] of relations) {
+    for (const start of byStart.keys()) {
+      const path = [atom(start)];
+      yield* fixedLengthCycleDfs(goal, env, byStart, relation, start, start, length, path);
+    }
+  }
+}
+
+function fixedLengthCycleRequestedPath(goal, env, length) {
+  const items = properListItems(goal.args[3], env);
+  if (!items || items.length !== length + 1) return null;
+  const nodes = items.map((item) => atomKey(deref(item, env)));
+  if (nodes.some((node) => node == null)) return null;
+  if (nodes[0] !== nodes[nodes.length - 1]) return null;
+  return nodes;
+}
+
+function fixedLengthCyclePathExists(byStart, nodes) {
+  for (let i = 0; i < nodes.length - 1; i++) {
+    if (!(byStart.get(nodes[i]) ?? []).includes(nodes[i + 1])) return false;
+  }
+  return true;
+}
+
+function* fixedLengthCycleDfs(goal, env, byStart, relation, start, current, remaining, path) {
+  if (remaining === 0) {
+    if (current !== start) return;
+    const next = env.clone();
+    if (unify(goal.args[2], atom(relation), next) && unify(goal.args[3], listFromItems(path), next)) yield next;
+    return;
+  }
+  const nexts = byStart.get(current) ?? [];
+  for (const dst of nexts) {
+    path.push(atom(dst));
+    yield* fixedLengthCycleDfs(goal, env, byStart, relation, start, dst, remaining - 1, path);
+    path.pop();
+  }
+}
+
+function labelledGraph(program, predicate) {
+  let count = 0;
+  const byRelation = new Map();
+  for (const clause of program.clauses) {
+    if (clause.body.length !== 0) continue;
+    const h = clause.head;
+    if (h.type !== 'compound' || h.name !== predicate || h.arity !== 3) continue;
+    const src = atomKey(h.args[0]), rel = atomKey(h.args[1]), dst = atomKey(h.args[2]);
+    if (src == null || rel == null || dst == null) continue;
+    let byStart = byRelation.get(rel);
+    if (!byStart) byRelation.set(rel, byStart = new Map());
+    let arr = byStart.get(src);
+    if (!arr) byStart.set(src, arr = []);
+    arr.push(dst); count++;
+  }
+  return count ? { byRelation } : null;
+}
+
+
+function* boundedPath({ solver, goal, env }) {
+  // bounded_path(EdgePred, Source, Target, MaxEdges, Path) enumerates simple
+  // directed paths with at most MaxEdges edges. EdgePred is read from EdgePred/2
+  // facts in source order so declarative examples retain stable answer order.
+  const predicate = atomKey(deref(goal.args[0], env));
+  const source = atomKey(deref(goal.args[1], env));
+  const target = atomKey(deref(goal.args[2], env));
+  const maxEdges = intTerm(goal.args[3], env);
+  if (!predicate || source == null || target == null || maxEdges == null || maxEdges < 0) return;
+  const graph = directedAdjacency(solver.program, predicate);
+  if (!graph) return;
+  const path = [source];
+  const visited = new Set([source]);
+  function* dfs(current, remaining) {
+    if (current === target) {
+      const next = env.clone();
+      if (unify(goal.args[4], listFromItems(path.map(atom)), next)) yield next;
+      return;
+    }
+    if (remaining <= 0) return;
+    for (const dst of graph.get(current) ?? []) {
+      if (visited.has(dst)) continue;
+      visited.add(dst);
+      path.push(dst);
+      yield* dfs(dst, remaining - 1);
+      path.pop();
+      visited.delete(dst);
+    }
+  }
+  yield* dfs(source, maxEdges);
+}
+
+function directedAdjacency(program, predicate) {
+  const map = new Map();
+  let count = 0;
+  for (const clause of program.clauses) {
+    if (clause.body.length !== 0) continue;
+    const h = clause.head;
+    if (h.type !== 'compound' || h.name !== predicate || h.arity !== 2) continue;
+    const a = atomKey(h.args[0]), b = atomKey(h.args[1]);
+    if (a == null || b == null) continue;
+    let arr = map.get(a);
+    if (!arr) map.set(a, arr = []);
+    arr.push(b);
+    count++;
+  }
+  return count ? map : null;
+}
+
+function* cnfModel({ goal, env }) {
+  // cnf_model(Variables, Clauses, Assignment) enumerates finite truth
+  // assignments in false-before-true order.  It deliberately preserves the
+  // proof multiplicity of the declarative clause_true/2 + cnf_true/2 program:
+  // a clause with two true literals has two derivations.
+  const variables = atomList(goal.args[0], env);
+  const clauses = clauseList(goal.args[1], env);
+  if (!variables || !clauses) return;
+  for (const assignment of enumerateAssignments(variables)) {
+    const derivations = cnfDerivationCount(clauses, assignment);
+    if (derivations === 0) continue;
+    const terms = variables.map((v) => compound('value', [atom(v), atom(assignment.get(v) ? 'true' : 'false')]));
+    for (let i = 0; i < derivations; i++) {
+      const next = env.clone();
+      if (unify(goal.args[2], listFromItems(terms), next)) yield next;
+    }
+  }
+}
+
+function* enumerateAssignments(vars, index = 0, assignment = new Map()) {
+  if (index >= vars.length) { yield new Map(assignment); return; }
+  const v = vars[index];
+  assignment.set(v, false); yield* enumerateAssignments(vars, index + 1, assignment);
+  assignment.set(v, true); yield* enumerateAssignments(vars, index + 1, assignment);
+  assignment.delete(v);
+}
+
+function cnfDerivationCount(clauses, assignment) {
+  let count = 1;
+  for (const clause of clauses) {
+    let trueLiterals = 0;
+    for (const lit of clause) if (assignment.get(lit.var) === lit.positive) trueLiterals++;
+    if (trueLiterals === 0) return 0;
+    count *= trueLiterals;
+  }
+  return count;
+}
+
+function* qmPrimeImplicants({ goal, env }) {
+  const data = qmDataFromArgs(goal, env);
+  if (!data) return;
+  const primes = computePrimeImplicants(data).map(patternTerm);
+  const next = env.clone();
+  if (unify(goal.args[3], listFromItems(primes), next)) yield next;
+}
+
+function* qmMinimalCover({ goal, env }) {
+  const data = qmDataFromArgs(goal, env);
+  if (!data) return;
+  const primes = computePrimeImplicants(data);
+  const cover = computeMinimalCover(primes, data.minterms);
+  if (!cover) return;
+  const next = env.clone();
+  if (unify(goal.args[3], listFromItems(cover.map(patternTerm)), next)) yield next;
+}
+
+function qmDataFromArgs(goal, env) {
+  const minterms = numberList(goal.args[0], env);
+  const dontCares = numberList(goal.args[1], env);
+  const bits = bitTable(goal.args[2], env);
+  if (!minterms || !dontCares || !bits?.size) return null;
+  return { minterms, dontCares, bits };
+}
+
+function computePrimeImplicants(data) {
+  const initial = [...data.minterms, ...data.dontCares].map((n) => data.bits.get(n)).filter(Boolean);
+  const once = [];
+  for (const a of initial) for (const b of initial) { const c = combinePatterns(a, b); if (c) once.push(c); }
+  const twice = [];
+  for (const a of once) for (const b of once) { const c = combinePatterns(a, b); if (c) twice.push(c); }
+  const raw = [];
+  for (const p of initial) if (!initial.some((q) => combinePatterns(p, q))) raw.push(p);
+  for (const p of once) if (!once.some((q) => combinePatterns(p, q))) raw.push(p);
+  raw.push(...twice);
+  return uniqueSortedPatterns(raw);
+}
+
+function combinePatterns(a, b) {
+  if (a.length !== b.length) return null;
+  let diffs = 0;
+  const out = [];
+  for (let i = 0; i < a.length; i++) {
+    if (a[i] === b[i]) out.push(a[i]);
+    else {
+      diffs++;
+      if (diffs > 1) return null;
+      out.push('x');
+    }
+  }
+  return diffs === 1 ? out : null;
+}
+
+function uniqueSortedPatterns(patterns) {
+  const map = new Map();
+  for (const p of patterns) map.set(patternKey(p), p);
+  return [...map.values()].sort(comparePattern);
+}
+
+function comparePattern(a, b) {
+  for (let i = 0; i < a.length; i++) {
+    const ra = a[i] === 'x' ? 2 : 1;
+    const rb = b[i] === 'x' ? 2 : 1;
+    if (ra !== rb) return ra - rb;
+    if (a[i] !== b[i]) return a[i] < b[i] ? -1 : 1;
+  }
+  return 0;
+}
+
+function computeMinimalCover(primes, minterms) {
+  const sortedMinterms = [...new Set(minterms)].sort((a, b) => a - b);
+  const candidates = [];
+  for (let i = 0; i < primes.length; i++) {
+    for (let j = i + 1; j < primes.length; j++) {
+      const cover = uniqueSortedPatterns([primes[i], primes[j]]);
+      if (sortedMinterms.every((m) => cover.some((p) => patternCoversInt(p, m)))) candidates.push(cover);
+    }
+  }
+  candidates.sort((a, b) => comparePatternList(a, b));
+  return candidates[0] ?? null;
+}
+
+function patternCoversInt(pattern, n) {
+  const bits = n.toString(2).padStart(pattern.length, '0').split('').map(Number);
+  return pattern.every((v, i) => v === 'x' || v === bits[i]);
+}
+
+function comparePatternList(a, b) {
+  const len = Math.min(a.length, b.length);
+  for (let i = 0; i < len; i++) { const c = comparePattern(a[i], b[i]); if (c) return c; }
+  return a.length - b.length;
+}
+
+function patternTerm(pattern) {
+  return listFromItems(pattern.map((v) => v === 'x' ? atom('x') : numberTerm(v)));
+}
+
+function patternKey(pattern) { return pattern.join(''); }
+
+function* permutations(items, start = 0) {
+  if (start >= items.length) { yield items.slice(); return; }
+  for (let i = start; i < items.length; i++) {
+    [items[start], items[i]] = [items[i], items[start]];
+    yield* permutations(items, start + 1);
+    [items[start], items[i]] = [items[i], items[start]];
+  }
+}
+
+function atomKey(term) {
+  if (!term) return null;
+  return (term.type === 'atom' || term.type === 'string' || term.type === 'number') ? term.name : null;
+}
+
+function intTerm(term, env) {
+  const t = env ? deref(term, env) : term;
+  if (!t || t.type !== 'number' || !/^-?\d+$/.test(t.name)) return null;
+  const n = Number(t.name);
+  return Number.isSafeInteger(n) ? n : null;
+}
+
+function numberList(term, env) {
+  const items = properListItems(term, env);
+  if (!items) return null;
+  const out = [];
+  for (const item of items) {
+    const n = intTerm(item, env);
+    if (n == null) return null;
+    out.push(n);
+  }
+  return out;
+}
+
+function atomList(term, env) {
+  const items = properListItems(term, env);
+  if (!items) return null;
+  const out = [];
+  for (const item of items) {
+    const v = atomKey(deref(item, env));
+    if (v == null) return null;
+    out.push(v);
+  }
+  return out;
+}
+
+function bitTable(term, env) {
+  const items = properListItems(term, env);
+  if (!items) return null;
+  const map = new Map();
+  for (const item of items) {
+    const entry = deref(item, env);
+    if (entry.type !== 'compound' || entry.name !== 'bit' || entry.arity !== 2) return null;
+    const n = intTerm(entry.args[0], env);
+    const bits = numberList(entry.args[1], env);
+    if (n == null || !bits) return null;
+    map.set(n, bits);
+  }
+  return map;
+}
+
+function clauseList(term, env) {
+  const clauses = properListItems(term, env);
+  if (!clauses) return null;
+  const out = [];
+  for (const clause of clauses) {
+    const lits = properListItems(clause, env);
+    if (!lits) return null;
+    const row = [];
+    for (const litTerm of lits) {
+      const lit = deref(litTerm, env);
+      if (lit.type !== 'compound' || lit.arity !== 1 || (lit.name !== 'pos' && lit.name !== 'neg')) return null;
+      const v = atomKey(deref(lit.args[0], env));
+      if (v == null) return null;
+      row.push({ var: v, positive: lit.name === 'pos' });
+    }
+    out.push(row);
+  }
+  return out;
+}

package/lib/eyelang/builtins/strings.mjs

@@ -0,0 +1,41 @@
+// String and atom conversion builtins.
+// They mostly project from already-ground terms to avoid guessing string domains.
+import { atom, lexicalValue, stringTerm, unify } from '../term.mjs';
+
+export const stringBuiltins = {
+  register(registry) {
+    for (const name of ['atom_concat', 'str_concat']) registry.add(name, 3, concat(name), { deterministic: true });
+    for (const name of ['contains', 'not_contains', 'matches', 'not_matches']) registry.add(name, 2, contains(name), { deterministic: true });
+  }
+};
+
+
+function concat(name) {
+  return function* ({ goal, env }) {
+    const left = lexicalValue(goal.args[0], env);
+    const right = lexicalValue(goal.args[1], env);
+    if (left == null || right == null) return;
+    const result = name === 'str_concat' ? stringTerm(left + right) : atom(left + right);
+    const next = env.clone();
+    if (unify(goal.args[2], result, next)) yield next;
+  };
+}
+
+function contains(name) {
+  return function* ({ goal, env }) {
+    const haystack = lexicalValue(goal.args[0], env);
+    const needle = lexicalValue(goal.args[1], env);
+    if (haystack == null || needle == null) return;
+    const has = haystack.includes(needle);
+    const matches = simpleAlternationMatch(haystack, needle);
+    const pass = (name === 'contains' && has) ||
+      (name === 'not_contains' && !has) ||
+      (name === 'matches' && matches) ||
+      (name === 'not_matches' && !matches);
+    if (pass) yield env;
+  };
+}
+
+function simpleAlternationMatch(haystack, pattern) {
+  return pattern.split('|').some((part) => part === '' || haystack.includes(part));
+}