ata-validator 1.37.0 → 1.38.0

package/lib/buffer-gate.browser.js

@@ -9,5 +9,10 @@
 
 module.exports = {
   bufferNeedsSlowPath: () => false,
-  installSlowBufferApis() {},
+  // A browser build has no addon and no Buffer, and the core throws before
+  // it gets here. A polyfilled Buffer would reach this: fail loudly rather
+  // than return with the methods still uninstalled.
+  installSlowBufferApis() {
+    throw new Error('The buffer APIs are not available in the browser build. Use validate(), isValidObject() or validateJSON().');
+  },
 };

package/lib/buffer-gate.js

@@ -37,6 +37,30 @@ const SUBSCHEMA_NODES = [
   'allOf', 'anyOf', 'oneOf', 'prefixItems',
 ];
 
+// Where RE2 and ECMA-262 give different answers for the same pattern. RE2's
+// `\s` is the ASCII whitespace alone, so it misses U+00A0, U+2028, U+3000 and
+// U+FEFF, which ECMA-262 counts, and its `.` matches a carriage return and the
+// line and paragraph separators, which ECMA-262's does not. `^\S+$` accepted
+// "a\u2028b" on the native text path while validate() rejected it. Any `\s`
+// or `\S`, and any `.` outside a class that is not escaped, sends the schema to
+// the JS path.
+function re2Diverges(pattern) {
+  let inClass = false;
+  for (let i = 0; i < pattern.length; i++) {
+    const c = pattern[i];
+    if (c === '\\') {
+      const e = pattern[i + 1];
+      if (e === 's' || e === 'S') return true;
+      i++;
+      continue;
+    }
+    if (inClass) { if (c === ']') inClass = false; continue; }
+    if (c === '[') { inClass = true; continue; }
+    if (c === '.') return true;
+  }
+  return false;
+}
+
 function walk(schema, depth) {
   if (schema === true || schema === false) {
     // A boolean root is answered wrong by the walker; nested booleans are
@@ -62,6 +86,7 @@ function walk(schema, depth) {
     // Unicode property escapes: RE2 cannot parse them and the walker then
     // skips the pattern instead of failing.
     if (key === 'pattern' && typeof v === 'string' && /\\[pP]\{/.test(v)) return true;
+    if (key === 'pattern' && typeof v === 'string' && re2Diverges(v)) return true;
     // Tuple forms: prefixItems and the draft-07 array form of items are
     // checked against the wrong positions by the walker.
     if (key === 'prefixItems') return true;

package/lib/js-compiler.js

@@ -135,12 +135,26 @@ function enumCondition(ctx, vals, v) {
 const { codeFor } = require('./error-codes')
 const { ordinalFor } = require('./schema-order')
 const { compileSafe, patternIsSafe } = require('./safe-regex')
+// Whether a pattern goes to the linear-time engine. A pattern the platform
+// RegExp also runs in linear time (nativeIsLinear) takes the RegExp: it answers
+// the same, runs faster, and a compiled module that needs no other pattern
+// does not carry the engine. Patterns the engine cannot represent keep the
+// RegExp, as before.
+// Loaded on the first pattern, so a schema without one does not pay for it.
+let _nativeIsLinear = null
+// The linearity check only parses; it runs first so a pattern bound for the
+// RegExp is never compiled for the engine just to be told it could have been.
+const useSafeEngine = (src) => {
+  if (_nativeIsLinear === null) _nativeIsLinear = require('./regex-linear').nativeIsLinear
+  if (_nativeIsLinear(src)) return false
+  return patternIsSafe(src)
+}
 
 // Closure value for a patternProperties/propertyNames regex: the linear-time
 // safe matcher when the pattern is in the supported subset (and flag the ctx so
 // standalone embeds the engine), else a plain RegExp. Both expose `.test`.
 function safeReClosure (ctx, src) {
-  if (patternIsSafe(src)) { ctx.usesSafeRe = true; return compileSafe(src) }
+  if (useSafeEngine(src)) { ctx.usesSafeRe = true; return compileSafe(src) }
   return new RegExp(src)
 }
 
@@ -523,7 +537,7 @@ function compileToJS(schema, defs, schemaMap) {
   }
   if (schema.pattern) {
     try {
-      const re = patternIsSafe(schema.pattern) ? compileSafe(schema.pattern) : new RegExp(schema.pattern)
+      const re = useSafeEngine(schema.pattern) ? compileSafe(schema.pattern) : new RegExp(schema.pattern)
       checks.push((d) => typeof d !== 'string' || re.test(d))
     } catch {
       return null
@@ -1188,6 +1202,31 @@ function emitGuardedRun (attempt, reset, defSets) {
   )
 }
 
+// Definitions referenced from more than one place whose body is large enough
+// that repeating its error code at every reference costs more than a call.
+function sharedDefNames (root, defs) {
+  const out = new Set()
+  if (!defs || typeof defs !== 'object') return out
+  const counts = new Map()
+  const seen = new WeakSet()
+  const walk = (node) => {
+    if (node === null || typeof node !== 'object' || seen.has(node)) return
+    seen.add(node)
+    if (typeof node.$ref === 'string') {
+      const m = /^#\/(?:\$defs|definitions)\/([^/]+)$/.exec(node.$ref)
+      if (m) counts.set(m[1], (counts.get(m[1]) || 0) + 1)
+    }
+    for (const v of Object.values(node)) walk(v)
+  }
+  walk(root)
+  for (const [name, n] of counts) {
+    const def = defs[name]
+    if (n >= 2 && def && typeof def === 'object' && JSON.stringify(def).length >= SHARED_DEF_MIN_CHARS) out.add(name)
+  }
+  return out
+}
+const SHARED_DEF_MIN_CHARS = 120
+
 function cyclicDefNames (defs) {
   const out = new Set()
   if (!defs || typeof defs !== 'object') return out
@@ -1725,16 +1764,10 @@ function sharedCodegenGate(schema, schemaMap) {
 
 // --- Codegen mode: generates a single Function (NOT CSP-safe) ---
 // This matches ajv's approach: one monolithic function, V8 JIT fully inlines it
-// `propertyNames` as the generators read it. They express maxLength,
-// minLength, pattern, const and enum, and the gate declines anything else, so
-// a schema that writes `propertyNames: { $ref: '#/definitions/Key' }` or adds
-// `type: 'string'` went to the interpreted engine. Both are rewritten here, on
-// a copy of the nodes that change: a local reference, through aliases, is
-// replaced by what it names, `type: 'string'` is dropped since every property
-// name is a string, and annotations go. Where what is left is still not all
-// supported keys, the node is left as written and the gate declines as before.
-// Errors keep the path the interpreted engine reports, which runs through the
-// reference without naming it.
+// `propertyNames` as the generators read it (maxLength, minLength, pattern,
+// const, enum): a local $ref, through aliases, becomes what it names, and
+// `type: 'string'` and annotations go, on a copy. Anything else is left as
+// written and the gate declines. Errors keep the interpreted engine's path.
 const PN_SUPPORTED = new Set(['maxLength', 'minLength', 'pattern', 'const', 'enum'])
 const PN_ANNOTATIONS = new Set(['$schema', '$comment', 'title', 'description', 'examples', 'default', 'deprecated', 'readOnly', 'writeOnly'])
 function simplePropertyNames(pn, defs) {
@@ -2476,7 +2509,7 @@ function genCodeNode(schema, v, lines, ctx, knownType) {
       if (!ctx.regExpMap.has(pattern)) {
         const ri = ctx.varCounter++
         ctx.regExpMap.set(pattern, ri)
-        if (patternIsSafe(schema.pattern)) {
+        if (useSafeEngine(schema.pattern)) {
           ctx.helperCode.push(`const _re${ri}=__ataSafeRe(${pattern})`);
           ctx.usesSafeRe = true
         } else {
@@ -4277,9 +4310,14 @@ function genCodeENode(schema, v, pathExpr, lines, ctx, schemaPrefix) {
     if (m && ctx.rootDefs && ctx.rootDefs[m[1]]) {
       const defName = m[1]
       if (!ctx.cyclicDefs) ctx.cyclicDefs = cyclicDefNames(ctx.rootDefs)
-      if (ctx.cyclicDefs.has(defName)) {
-        // Named helper for a def on a cycle; the path comes in as an
-        // argument because every call site has its own.
+      if (!ctx.sharedDefs) ctx.sharedDefs = sharedDefNames(ctx.rootSchema, ctx.rootDefs)
+      if (ctx.cyclicDefs.has(defName) || ctx.sharedDefs.has(defName)) {
+        // Named helper for a def on a cycle, and for one referenced from
+        // several places, which inlined would repeat its whole error code at
+        // each: Uniswap's token-list schema has 16 references to 10
+        // definitions. The path comes in as an argument because every call
+        // site has its own. Only the error path runs these; the verdict
+        // function, which answers valid documents, is unchanged.
         if (!ctx.defFns) ctx.defFns = new Map()
         let fnName = ctx.defFns.get(defName)
         if (!fnName) {
@@ -4459,7 +4497,7 @@ function genCodeENode(schema, v, pathExpr, lines, ctx, schemaPrefix) {
       if (!ctx.regExpMap.has(pattern)) {
         const ri = ctx.varCounter++
         ctx.regExpMap.set(pattern, ri)
-        if (patternIsSafe(schema.pattern)) {
+        if (useSafeEngine(schema.pattern)) {
           ctx.helperCode.push(`const _re${ri}=__ataSafeRe(${pattern})`)
           ctx.usesSafeRe = true
         } else {
@@ -4534,7 +4572,7 @@ function genCodeENode(schema, v, pathExpr, lines, ctx, schemaPrefix) {
       if (!ctx.regExpMap.has(pattern)) {
         const ri = ctx.varCounter++
         ctx.regExpMap.set(pattern, ri)
-        if (patternIsSafe(pat)) {
+        if (useSafeEngine(pat)) {
           ctx.helperCode.push(`const _re${ri}=__ataSafeRe(${pattern})`)
           ctx.usesSafeRe = true
         } else {
@@ -4697,7 +4735,7 @@ function genUnevaluatedFalseE(schema, v, pathExpr, lines, ctx, schemaPrefix, isO
     if (!ctx.regExpMap.has(pattern)) {
       const ri = ctx.varCounter++
       ctx.regExpMap.set(pattern, ri)
-      if (patternIsSafe(pat)) {
+      if (useSafeEngine(pat)) {
         ctx.helperCode.push(`const _re${ri}=__ataSafeRe(${pattern})`)
         ctx.usesSafeRe = true
       } else {
@@ -4728,7 +4766,7 @@ function genPatternPropertiesE(schema, v, pathExpr, lines, ctx, schemaPrefix) {
     if (!ctx.regExpMap.has(pattern)) {
       const ri = ctx.varCounter++
       ctx.regExpMap.set(pattern, ri)
-      if (patternIsSafe(pat)) {
+      if (useSafeEngine(pat)) {
         ctx.helperCode.push(`const _re${ri}=__ataSafeRe(${pattern})`);
         ctx.usesSafeRe = true
       } else {
@@ -4757,7 +4795,7 @@ function genAdditionalSchemaE(schema, v, pathExpr, lines, ctx, schemaPrefix) {
     if (!ctx.regExpMap.has(pattern)) {
       const ri = ctx.varCounter++
       ctx.regExpMap.set(pattern, ri)
-      if (patternIsSafe(pat)) {
+      if (useSafeEngine(pat)) {
         ctx.helperCode.push(`const _re${ri}=__ataSafeRe(${pattern})`)
         ctx.usesSafeRe = true
       } else {
@@ -4809,7 +4847,7 @@ function genPropertyNamesE(schema, v, pathExpr, lines, ctx, schemaPrefix, fail)
     if (!ctx.regExpMap.has(pattern)) {
       const ri = ctx.varCounter++
       ctx.regExpMap.set(pattern, ri)
-      if (patternIsSafe(pn.pattern)) {
+      if (useSafeEngine(pn.pattern)) {
         ctx.helperCode.push(`const _re${ri}=__ataSafeRe(${pattern})`);
         ctx.usesSafeRe = true
       } else {
@@ -5248,7 +5286,7 @@ function genCodeCNode(schema, v, pathExpr, lines, ctx, schemaPrefix) {
       const ri = ctx.varCounter++
       const reVar = `_re${ri}`
       ctx.closureVars.push(reVar)
-      ctx.closureVals.push(patternIsSafe(schema.pattern) ? compileSafe(schema.pattern) : new RegExp(schema.pattern))
+      ctx.closureVals.push(useSafeEngine(schema.pattern) ? compileSafe(schema.pattern) : new RegExp(schema.pattern))
       const c = isStr ? `!${reVar}.test(${v})` : `typeof ${v}==='string'&&!${reVar}.test(${v})`
       lines.push(`if(${c}){${fail('pattern', 'pattern', `{pattern:${JSON.stringify(schema.pattern)}}`, JSON.stringify(`must match pattern "${schema.pattern}"`))}}`)
     }

package/lib/regex-linear.js

@@ -0,0 +1,65 @@
+'use strict'
+
+const { parse } = require('./safe-regex')
+
+// True when the platform RegExp, a backtracking engine, also runs `src` in
+// linear time, so it can take the pattern: every quantifier on one character,
+// class or `.`; at most one unbounded, and then anchored with `^`; few bounded
+// choices; only escapes both engines read alike; nothing outside the BMP, so
+// both see the same UTF-16 units. tests/test_regex_native_linear.js holds it.
+const NATIVE_ESCAPES = new Set(['d', 'D', 'w', 'W', 's', 'S', 'n', 'r', 't', 'f', 'v', '.', '\\', '/', '^', '$', '|', '?', '*', '+', '(', ')', '[', ']', '{', '}', '-', 'x', 'u'])
+function nativeIsLinear (src) {
+  if (typeof src !== 'string' || src.length > 512) return false
+  for (let i = 0; i < src.length; i++) {
+    const c = src.charCodeAt(i)
+    if (c >= 0xd800 && c <= 0xdfff) return false
+    if (src[i] === '\\') {
+      const e = src[i + 1]
+      if (!NATIVE_ESCAPES.has(e)) return false
+      if (e === 'x' && !/^[0-9a-fA-F]{2}$/.test(src.slice(i + 2, i + 4))) return false
+      if (e === 'u' && !/^[0-9a-fA-F]{4}$/.test(src.slice(i + 2, i + 6))) return false
+      i++
+    }
+  }
+  let ast
+  try { ast = parse(src) } catch { return false }
+  let unbounded = 0
+  let bounded = 0
+  let ok = true
+  const atom = (n) => n.t === 'char' || n.t === 'class' || n.t === 'any'
+  const walk = (n) => {
+    if (!ok) return
+    switch (n.t) {
+      case 'star': case 'plus':
+        if (!atom(n.child)) ok = false
+        unbounded++
+        break
+      case 'quest':
+        if (!atom(n.child)) ok = false
+        bounded += 1
+        break
+      case 'repeat':
+        if (!atom(n.child)) ok = false
+        if (n.max === Infinity) unbounded++
+        else bounded += n.max - n.min
+        break
+      case 'group': walk(n.child); break
+      case 'concat': n.parts.forEach(walk); break
+      case 'alt': n.opts.forEach(walk); break
+      default: break
+    }
+  }
+  walk(ast)
+  if (!ok || unbounded > 1 || bounded > 64) return false
+  if (unbounded === 1) {
+    // Anchored at the start on every top-level branch.
+    let top = ast
+    while (top.t === 'group') top = top.child
+    const branches = top.t === 'alt' ? top.opts : [top]
+    const first = (n) => { while (n && n.t === 'group') n = n.child; return n && n.t === 'concat' ? first(n.parts[0]) : n }
+    if (!branches.every((b) => { const f = first(b); return f && f.t === 'bol' })) return false
+  }
+  return true
+}
+
+module.exports = { nativeIsLinear }

package/lib/safe-regex-source.js

@@ -5,4 +5,4 @@
 // into standalone output without a runtime `fs` read. Kept in sync by
 // `tests/test_safe_regex_source_sync.js`.
 
-module.exports = "'use strict'\n\n// Linear-time regex engine for JSON Schema `pattern`, used in place of JS RegExp\n// so an adversarial input cannot trigger catastrophic backtracking (ReDoS).\n//\n// It is a Pike VM: the pattern compiles to a small instruction program, and the\n// VM simulates all NFA threads in lockstep over the input, deduping by program\n// counter. Runtime is O(input * program), with no backtracking.\n//\n// Supported (the RE2 subset, which is what ata's native path also accepts):\n// literals, ., character classes, \\d \\w \\s \\D \\W \\S, anchors ^ $, quantifiers\n// * + ? {n} {n,} {n,m} (greedy or lazy, same language for a boolean test),\n// groups ( ) (?: ), alternation |. Backreferences and lookaround are not\n// supported by linear engines; compileSafe throws on them so the caller can\n// decide (ata's codegen rejects such schemas rather than risk a hang).\n\nconst WS = [[9, 13], [32, 32], [160, 160]]\nconst DIGIT = [[48, 57]]\nconst WORD = [[48, 57], [65, 90], [97, 122], [95, 95]]\n\nfunction parse (src) {\n  let i = 0\n  const len = src.length\n  const peek = () => src[i]\n  const eof = () => i >= len\n\n  function parseAlt () {\n    const opts = [parseConcat()]\n    while (!eof() && peek() === '|') { i++; opts.push(parseConcat()) }\n    return opts.length === 1 ? opts[0] : { t: 'alt', opts }\n  }\n\n  function parseConcat () {\n    const parts = []\n    while (!eof() && peek() !== '|' && peek() !== ')') parts.push(parseRepeat())\n    if (parts.length === 0) return { t: 'empty' }\n    return parts.length === 1 ? parts[0] : { t: 'concat', parts }\n  }\n\n  function parseRepeat () {\n    let node = parseAtom()\n    while (!eof()) {\n      const ch = peek()\n      if (ch === '*') { i++; node = { t: 'star', child: node } }\n      else if (ch === '+') { i++; node = { t: 'plus', child: node } }\n      else if (ch === '?') { i++; node = { t: 'quest', child: node } }\n      else if (ch === '{') {\n        const saved = i\n        const q = tryQuantifier()\n        if (!q) { i = saved; break }\n        node = { t: 'repeat', child: node, min: q.min, max: q.max }\n      } else break\n      // a trailing ? makes the quantifier lazy; same language for a boolean test\n      if (!eof() && peek() === '?') i++\n    }\n    return node\n  }\n\n  function tryQuantifier () {\n    // assumes current char is '{'\n    i++\n    let min = ''\n    while (!eof() && /[0-9]/.test(peek())) { min += peek(); i++ }\n    if (min === '') return null\n    let max\n    if (peek() === '}') { i++; return { min: +min, max: +min } }\n    if (peek() === ',') {\n      i++\n      let m = ''\n      while (!eof() && /[0-9]/.test(peek())) { m += peek(); i++ }\n      if (peek() !== '}') return null\n      i++\n      max = m === '' ? Infinity : +m\n      return { min: +min, max }\n    }\n    return null\n  }\n\n  function parseAtom () {\n    const ch = peek()\n    if (ch === '(') {\n      i++\n      if (src[i] === '?') {\n        if (src[i + 1] === ':') { i += 2 }\n        else throw new Error('unsupported group (lookaround/named) in pattern')\n      }\n      const child = parseAlt()\n      if (peek() !== ')') throw new Error('unbalanced ( in pattern')\n      i++\n      return { t: 'group', child }\n    }\n    if (ch === '[') return parseClass()\n    if (ch === '.') { i++; return { t: 'any' } }\n    if (ch === '^') { i++; return { t: 'bol' } }\n    if (ch === '$') { i++; return { t: 'eol' } }\n    if (ch === '\\\\') return parseEscape(false)\n    if (ch === ')' || ch === '|') return { t: 'empty' }\n    i++\n    return { t: 'char', c: ch.charCodeAt(0) }\n  }\n\n  function parseClass () {\n    i++ // [\n    let neg = false\n    if (peek() === '^') { neg = true; i++ }\n    const ranges = []\n    while (!eof() && peek() !== ']') {\n      let lo\n      if (peek() === '\\\\') {\n        const esc = parseEscape(true)\n        if (esc.t === 'classpart') { for (const r of esc.ranges) ranges.push(r); continue }\n        lo = esc.c\n      } else { lo = peek().charCodeAt(0); i++ }\n      if (peek() === '-' && src[i + 1] !== ']' && i + 1 < len) {\n        i++ // -\n        let hi\n        if (peek() === '\\\\') { const e = parseEscape(true); hi = e.c } else { hi = peek().charCodeAt(0); i++ }\n        ranges.push([lo, hi])\n      } else {\n        ranges.push([lo, lo])\n      }\n    }\n    if (peek() !== ']') throw new Error('unbalanced [ in pattern')\n    i++\n    return { t: 'class', neg, ranges }\n  }\n\n  function parseEscape (inClass) {\n    i++ // backslash\n    if (eof()) throw new Error('trailing backslash in pattern')\n    const ch = peek(); i++\n    switch (ch) {\n      case 'd': return inClass ? { t: 'classpart', ranges: DIGIT } : { t: 'class', neg: false, ranges: DIGIT }\n      case 'w': return inClass ? { t: 'classpart', ranges: WORD } : { t: 'class', neg: false, ranges: WORD }\n      case 's': return inClass ? { t: 'classpart', ranges: WS } : { t: 'class', neg: false, ranges: WS }\n      case 'D': if (inClass) throw new Error('\\\\D inside a class is not supported'); return { t: 'class', neg: true, ranges: DIGIT }\n      case 'W': if (inClass) throw new Error('\\\\W inside a class is not supported'); return { t: 'class', neg: true, ranges: WORD }\n      case 'S': if (inClass) throw new Error('\\\\S inside a class is not supported'); return { t: 'class', neg: true, ranges: WS }\n      case 'n': return { t: 'char', c: 10 }\n      case 'r': return { t: 'char', c: 13 }\n      case 't': return { t: 'char', c: 9 }\n      case 'f': return { t: 'char', c: 12 }\n      case 'v': return { t: 'char', c: 11 }\n      case '0': return { t: 'char', c: 0 }\n      case 'x': { const h = src.slice(i, i + 2); i += 2; return { t: 'char', c: parseInt(h, 16) } }\n      case 'u': { const h = src.slice(i, i + 4); i += 4; return { t: 'char', c: parseInt(h, 16) } }\n      case 'b': if (inClass) return { t: 'char', c: 8 }; throw new Error('\\\\b word boundary is not supported')\n      default:\n        if (/[1-9]/.test(ch)) throw new Error('backreferences are not supported in pattern')\n        return { t: 'char', c: ch.charCodeAt(0) }\n    }\n  }\n\n  const ast = parseAlt()\n  if (!eof()) throw new Error('unexpected \"' + peek() + '\" in pattern')\n  return ast\n}\n\nfunction compileProg (ast) {\n  const prog = []\n  const emit = (op, extra) => { const idx = prog.length; prog.push(Object.assign({ op }, extra)); return idx }\n\n  function rec (n) {\n    switch (n.t) {\n      case 'empty': break\n      case 'char': emit('char', { c: n.c }); break\n      case 'any': emit('any'); break\n      case 'class': emit('class', { neg: n.neg, ranges: n.ranges }); break\n      case 'bol': emit('bol'); break\n      case 'eol': emit('eol'); break\n      case 'group': rec(n.child); break\n      case 'concat': for (const p of n.parts) rec(p); break\n      case 'alt': {\n        const jmps = []\n        for (let k = 0; k < n.opts.length; k++) {\n          if (k < n.opts.length - 1) {\n            const sp = emit('split', { x: 0, y: 0 })\n            prog[sp].x = prog.length\n            rec(n.opts[k])\n            jmps.push(emit('jmp', { x: 0 }))\n            prog[sp].y = prog.length\n          } else {\n            rec(n.opts[k])\n          }\n        }\n        for (const j of jmps) prog[j].x = prog.length\n        break\n      }\n      case 'star': {\n        const sp = emit('split', { x: 0, y: 0 })\n        prog[sp].x = prog.length\n        rec(n.child)\n        emit('jmp', { x: sp })\n        prog[sp].y = prog.length\n        break\n      }\n      case 'plus': {\n        const start = prog.length\n        rec(n.child)\n        const sp = emit('split', { x: start, y: 0 })\n        prog[sp].y = prog.length\n        break\n      }\n      case 'quest': {\n        const sp = emit('split', { x: 0, y: 0 })\n        prog[sp].x = prog.length\n        rec(n.child)\n        prog[sp].y = prog.length\n        break\n      }\n      case 'repeat': {\n        for (let k = 0; k < n.min; k++) rec(n.child)\n        if (n.max === Infinity) {\n          if (n.min === 0) rec({ t: 'star', child: n.child })\n          else rec({ t: 'star', child: n.child })\n        } else {\n          for (let k = 0; k < n.max - n.min; k++) rec({ t: 'quest', child: n.child })\n        }\n        break\n      }\n    }\n  }\n\n  rec(ast)\n  emit('match')\n  return prog\n}\n\n// Numeric opcodes for the runner. The program is compiled once into flat\n// typed arrays so the inner loop does no property lookups or string compares.\nconst OP_CHAR = 0\nconst OP_ANY = 1\nconst OP_CLASS = 2\nconst OP_SPLIT = 3\nconst OP_JMP = 4\nconst OP_BOL = 5\nconst OP_EOL = 6\nconst OP_MATCH = 7\n\nfunction classMatcher (instr) {\n  // ASCII is answered from a bitmap; anything above 0x7f walks the ranges.\n  const bits = new Uint8Array(128)\n  const r = instr.ranges\n  for (let k = 0; k < r.length; k++) {\n    const hi = Math.min(r[k][1], 127)\n    for (let c = r[k][0]; c <= hi; c++) bits[c] = 1\n  }\n  return { bits, ranges: r, neg: instr.neg }\n}\n\nfunction matchClass (cls, c) {\n  let inside\n  if (c < 128) {\n    inside = cls.bits[c] === 1\n  } else {\n    inside = false\n    const r = cls.ranges\n    for (let k = 0; k < r.length; k++) { if (c >= r[k][0] && c <= r[k][1]) { inside = true; break } }\n  }\n  return cls.neg ? !inside : inside\n}\n\nfunction makeRunner (prog) {\n  const n = prog.length\n  const ops = new Uint8Array(n)\n  const xs = new Int32Array(n)\n  const ys = new Int32Array(n)\n  const cs = new Int32Array(n)\n  const classes = new Array(n)\n  for (let i = 0; i < n; i++) {\n    const I = prog[i]\n    switch (I.op) {\n      case 'char': ops[i] = OP_CHAR; cs[i] = I.c; break\n      case 'any': ops[i] = OP_ANY; break\n      case 'class': ops[i] = OP_CLASS; classes[i] = classMatcher(I); break\n      case 'split': ops[i] = OP_SPLIT; xs[i] = I.x; ys[i] = I.y; break\n      case 'jmp': ops[i] = OP_JMP; xs[i] = I.x; break\n      case 'bol': ops[i] = OP_BOL; break\n      case 'eol': ops[i] = OP_EOL; break\n      case 'match': ops[i] = OP_MATCH; break\n    }\n  }\n\n  const lastGen = new Int32Array(n).fill(-1)\n  let gen = 0\n  // Each unvisited instruction is popped once and pushes at most two, so the\n  // stack never holds more than 2n + 1 entries.\n  const stack = new Int32Array(2 * n + 2)\n  // Thread lists hold at most one entry per instruction per step.\n  let clist = new Int32Array(n)\n  let nlist = new Int32Array(n)\n  let clen = 0\n  let nlen = 0\n\n  // Follows epsilon edges from `pc` and records every consuming instruction\n  // (or match) reached in `list`. `lastGen` dedupes per step.\n  function addThread (list, len0, pc, pos, len) {\n    // Most transitions land directly on a consuming instruction; skip the\n    // stack walk for those.\n    if (ops[pc] <= OP_CLASS || ops[pc] === OP_MATCH) {\n      if (lastGen[pc] === gen) return len0\n      lastGen[pc] = gen\n      list[len0] = pc\n      return len0 + 1\n    }\n    let sp = 0\n    stack[sp++] = pc\n    let count = len0\n    while (sp > 0) {\n      const p = stack[--sp]\n      if (lastGen[p] === gen) continue\n      lastGen[p] = gen\n      switch (ops[p]) {\n        case OP_JMP: stack[sp++] = xs[p]; break\n        case OP_SPLIT: stack[sp++] = ys[p]; stack[sp++] = xs[p]; break\n        case OP_BOL: if (pos === 0) stack[sp++] = p + 1; break\n        case OP_EOL: if (pos === len) stack[sp++] = p + 1; break\n        default: list[count++] = p\n      }\n    }\n    return count\n  }\n\n  // A pattern is anchored when starting it anywhere but position 0 yields no\n  // thread, which is the case for `^...` and its alternations. The probe sits\n  // at position 1 of a length-1 string so that only `^` can fail. For anchored\n  // patterns the per-position restart below is skipped, and an empty thread\n  // list means the match has already failed.\n  gen++\n  const anchored = addThread(nlist, 0, 0, 1, 1) === 0\n\n  function testNFA (s) {\n    const len = s.length\n    gen++\n    clen = addThread(clist, 0, 0, 0, len)\n    for (let pos = 0; pos <= len; pos++) {\n      const c = pos < len ? s.charCodeAt(pos) : -1\n      gen++\n      nlen = 0\n      for (let k = 0; k < clen; k++) {\n        const pc = clist[k]\n        switch (ops[pc]) {\n          case OP_MATCH: return true\n          case OP_CHAR: if (c === cs[pc]) nlen = addThread(nlist, nlen, pc + 1, pos + 1, len); break\n          case OP_ANY: if (c !== -1 && c !== 10) nlen = addThread(nlist, nlen, pc + 1, pos + 1, len); break\n          case OP_CLASS: if (c !== -1 && matchClass(classes[pc], c)) nlen = addThread(nlist, nlen, pc + 1, pos + 1, len); break\n        }\n      }\n      if (pos < len) {\n        if (!anchored) nlen = addThread(nlist, nlen, 0, pos + 1, len)\n        else if (nlen === 0) return false\n      }\n      const tmp = clist; clist = nlist; nlist = tmp\n      clen = nlen\n    }\n    return false\n  }\n\n  // Lazy DFA on top of the NFA. A DFA state is the set of consuming\n  // instructions live at a position; transitions are computed on first use\n  // and cached per ASCII character. `^` and `$` depend on position, so the\n  // closure is taken with flags for \"at start\" and \"at end\", which gives two\n  // start states and two transition tables per state. The state count is\n  // capped; past the cap the matcher falls back to the NFA walk above, so the\n  // time bound stays linear either way.\n  const MAX_STATES = 256\n  const states = []\n  const stateIds = new Map()\n  let overflow = false\n\n  function closure (list, count, pc, atStart, atEnd) {\n    // Same walk as addThread, with the position replaced by the two flags.\n    let sp = 0\n    stack[sp++] = pc\n    while (sp > 0) {\n      const p = stack[--sp]\n      if (lastGen[p] === gen) continue\n      lastGen[p] = gen\n      switch (ops[p]) {\n        case OP_JMP: stack[sp++] = xs[p]; break\n        case OP_SPLIT: stack[sp++] = ys[p]; stack[sp++] = xs[p]; break\n        case OP_BOL: if (atStart) stack[sp++] = p + 1; break\n        case OP_EOL: if (atEnd) stack[sp++] = p + 1; break\n        default: list[count++] = p\n      }\n    }\n    return count\n  }\n\n  function internState (list, count) {\n    const pcs = Array.from(list.subarray(0, count)).sort((a, b) => a - b)\n    const key = pcs.join(',')\n    let id = stateIds.get(key)\n    if (id !== undefined) return id\n    if (states.length >= MAX_STATES) { overflow = true; return -1 }\n    id = states.length\n    let isMatch = false\n    for (let k = 0; k < pcs.length; k++) if (ops[pcs[k]] === OP_MATCH) { isMatch = true; break }\n    states.push({ pcs: Int32Array.from(pcs), isMatch, next: new Int32Array(128).fill(-2), nextEnd: new Int32Array(128).fill(-2) })\n    stateIds.set(key, id)\n    return id\n  }\n\n  function step (state, c, atEnd) {\n    gen++\n    let count = 0\n    const pcs = state.pcs\n    for (let k = 0; k < pcs.length; k++) {\n      const pc = pcs[k]\n      switch (ops[pc]) {\n        case OP_CHAR: if (c === cs[pc]) count = closure(nlist, count, pc + 1, false, atEnd); break\n        case OP_ANY: if (c !== 10) count = closure(nlist, count, pc + 1, false, atEnd); break\n        case OP_CLASS: if (matchClass(classes[pc], c)) count = closure(nlist, count, pc + 1, false, atEnd); break\n      }\n    }\n    if (!anchored) count = closure(nlist, count, 0, false, atEnd)\n    return internState(nlist, count)\n  }\n\n  let startEmpty = -2\n  let startNonEmpty = -2\n\n  function startState (atEnd) {\n    gen++\n    const count = closure(nlist, 0, 0, true, atEnd)\n    return internState(nlist, count)\n  }\n\n  function testDFA (s) {\n    const len = s.length\n    let id\n    if (len === 0) {\n      if (startEmpty === -2) startEmpty = startState(true)\n      id = startEmpty\n    } else {\n      if (startNonEmpty === -2) startNonEmpty = startState(false)\n      id = startNonEmpty\n    }\n    if (id < 0) return testNFA(s)\n    let state = states[id]\n    for (let pos = 0; pos < len; pos++) {\n      if (state.isMatch) return true\n      const c = s.charCodeAt(pos)\n      const atEnd = pos + 1 === len\n      let nid\n      if (c < 128) {\n        const table = atEnd ? state.nextEnd : state.next\n        nid = table[c]\n        if (nid === -2) { nid = step(state, c, atEnd); table[c] = nid }\n      } else {\n        nid = step(state, c, atEnd)\n      }\n      if (nid < 0) return testNFA(s)\n      state = states[nid]\n      if (anchored && state.pcs.length === 0) return false\n    }\n    return state.isMatch\n  }\n\n  return function test (s) {\n    return overflow ? testNFA(s) : testDFA(s)\n  }\n}\n\nfunction compileSafe (pattern) {\n  const prog = compileProg(parse(pattern))\n  const runner = makeRunner(prog)\n  // `__ataSafe` brands the result so the standalone serializer can tell a safe\n  // matcher apart from a RegExp and emit `__ataSafeRe(source)` instead.\n  return { test: runner, source: pattern, __ataSafe: true }\n}\n\n// True when the linear engine can represent `src`. Used by the codegen to decide\n// between the safe matcher and a JS RegExp fallback for patterns outside the\n// supported (RE2) subset (backreferences, lookaround, etc.).\nfunction patternIsSafe (src) {\n  try { compileSafe(src); return true } catch { return false }\n}\n\nmodule.exports = { compileSafe, patternIsSafe }\n";
+module.exports = "'use strict'\n\n// Linear-time regex engine for JSON Schema `pattern`, used in place of JS RegExp\n// so an adversarial input cannot trigger catastrophic backtracking (ReDoS).\n//\n// It is a Pike VM: the pattern compiles to a small instruction program, and the\n// VM simulates all NFA threads in lockstep over the input, deduping by program\n// counter. Runtime is O(input * program), with no backtracking.\n//\n// Supported (the RE2 subset, which is what ata's native path also accepts):\n// literals, ., character classes, \\d \\w \\s \\D \\W \\S, anchors ^ $, quantifiers\n// * + ? {n} {n,} {n,m} (greedy or lazy, same language for a boolean test),\n// groups ( ) (?: ), alternation |. Backreferences and lookaround are not\n// supported by linear engines; compileSafe throws on them so the caller can\n// decide (ata's codegen rejects such schemas rather than risk a hang).\n\n// ECMA-262 WhiteSpace and LineTerminator, what `\\s` matches: tab, line feed,\n// vertical tab, form feed, carriage return, space, no-break space, the\n// Unicode space separators, the line and paragraph separators, and the byte\n// order mark. The set used to stop at U+00A0, so `^\\S+$` accepted a string\n// holding U+2028 or U+3000.\nconst WS = [[9, 13], [32, 32], [160, 160], [0x1680, 0x1680], [0x2000, 0x200a], [0x2028, 0x2029], [0x202f, 0x202f], [0x205f, 0x205f], [0x3000, 0x3000], [0xfeff, 0xfeff]]\n// `.` matches anything but a line terminator: LF, CR and U+2028/U+2029. It\n// excluded only LF, so `^.$` accepted \"\\r\".\nconst isLineTerminator = (c) => c === 10 || c === 13 || c === 0x2028 || c === 0x2029\nconst DIGIT = [[48, 57]]\nconst WORD = [[48, 57], [65, 90], [97, 122], [95, 95]]\n\nfunction parse (src) {\n  let i = 0\n  const len = src.length\n  const peek = () => src[i]\n  const eof = () => i >= len\n\n  function parseAlt () {\n    const opts = [parseConcat()]\n    while (!eof() && peek() === '|') { i++; opts.push(parseConcat()) }\n    return opts.length === 1 ? opts[0] : { t: 'alt', opts }\n  }\n\n  function parseConcat () {\n    const parts = []\n    while (!eof() && peek() !== '|' && peek() !== ')') parts.push(parseRepeat())\n    if (parts.length === 0) return { t: 'empty' }\n    return parts.length === 1 ? parts[0] : { t: 'concat', parts }\n  }\n\n  function parseRepeat () {\n    let node = parseAtom()\n    while (!eof()) {\n      const ch = peek()\n      if (ch === '*') { i++; node = { t: 'star', child: node } }\n      else if (ch === '+') { i++; node = { t: 'plus', child: node } }\n      else if (ch === '?') { i++; node = { t: 'quest', child: node } }\n      else if (ch === '{') {\n        const saved = i\n        const q = tryQuantifier()\n        if (!q) { i = saved; break }\n        node = { t: 'repeat', child: node, min: q.min, max: q.max }\n      } else break\n      // a trailing ? makes the quantifier lazy; same language for a boolean test\n      if (!eof() && peek() === '?') i++\n    }\n    return node\n  }\n\n  function tryQuantifier () {\n    // assumes current char is '{'\n    i++\n    let min = ''\n    while (!eof() && /[0-9]/.test(peek())) { min += peek(); i++ }\n    if (min === '') return null\n    let max\n    if (peek() === '}') { i++; return { min: +min, max: +min } }\n    if (peek() === ',') {\n      i++\n      let m = ''\n      while (!eof() && /[0-9]/.test(peek())) { m += peek(); i++ }\n      if (peek() !== '}') return null\n      i++\n      max = m === '' ? Infinity : +m\n      return { min: +min, max }\n    }\n    return null\n  }\n\n  function parseAtom () {\n    const ch = peek()\n    if (ch === '(') {\n      i++\n      if (src[i] === '?') {\n        if (src[i + 1] === ':') { i += 2 }\n        else throw new Error('unsupported group (lookaround/named) in pattern')\n      }\n      const child = parseAlt()\n      if (peek() !== ')') throw new Error('unbalanced ( in pattern')\n      i++\n      return { t: 'group', child }\n    }\n    if (ch === '[') return parseClass()\n    if (ch === '.') { i++; return { t: 'any' } }\n    if (ch === '^') { i++; return { t: 'bol' } }\n    if (ch === '$') { i++; return { t: 'eol' } }\n    if (ch === '\\\\') return parseEscape(false)\n    if (ch === ')' || ch === '|') return { t: 'empty' }\n    i++\n    return { t: 'char', c: ch.charCodeAt(0) }\n  }\n\n  function parseClass () {\n    i++ // [\n    let neg = false\n    if (peek() === '^') { neg = true; i++ }\n    const ranges = []\n    while (!eof() && peek() !== ']') {\n      let lo\n      if (peek() === '\\\\') {\n        const esc = parseEscape(true)\n        if (esc.t === 'classpart') { for (const r of esc.ranges) ranges.push(r); continue }\n        lo = esc.c\n      } else { lo = peek().charCodeAt(0); i++ }\n      if (peek() === '-' && src[i + 1] !== ']' && i + 1 < len) {\n        i++ // -\n        let hi\n        if (peek() === '\\\\') { const e = parseEscape(true); hi = e.c } else { hi = peek().charCodeAt(0); i++ }\n        ranges.push([lo, hi])\n      } else {\n        ranges.push([lo, lo])\n      }\n    }\n    if (peek() !== ']') throw new Error('unbalanced [ in pattern')\n    i++\n    return { t: 'class', neg, ranges }\n  }\n\n  function parseEscape (inClass) {\n    i++ // backslash\n    if (eof()) throw new Error('trailing backslash in pattern')\n    const ch = peek(); i++\n    switch (ch) {\n      case 'd': return inClass ? { t: 'classpart', ranges: DIGIT } : { t: 'class', neg: false, ranges: DIGIT }\n      case 'w': return inClass ? { t: 'classpart', ranges: WORD } : { t: 'class', neg: false, ranges: WORD }\n      case 's': return inClass ? { t: 'classpart', ranges: WS } : { t: 'class', neg: false, ranges: WS }\n      case 'D': if (inClass) throw new Error('\\\\D inside a class is not supported'); return { t: 'class', neg: true, ranges: DIGIT }\n      case 'W': if (inClass) throw new Error('\\\\W inside a class is not supported'); return { t: 'class', neg: true, ranges: WORD }\n      case 'S': if (inClass) throw new Error('\\\\S inside a class is not supported'); return { t: 'class', neg: true, ranges: WS }\n      case 'n': return { t: 'char', c: 10 }\n      case 'r': return { t: 'char', c: 13 }\n      case 't': return { t: 'char', c: 9 }\n      case 'f': return { t: 'char', c: 12 }\n      case 'v': return { t: 'char', c: 11 }\n      case '0': return { t: 'char', c: 0 }\n      case 'x': { const h = src.slice(i, i + 2); i += 2; return { t: 'char', c: parseInt(h, 16) } }\n      case 'u': { const h = src.slice(i, i + 4); i += 4; return { t: 'char', c: parseInt(h, 16) } }\n      case 'b': if (inClass) return { t: 'char', c: 8 }; throw new Error('\\\\b word boundary is not supported')\n      default:\n        if (/[1-9]/.test(ch)) throw new Error('backreferences are not supported in pattern')\n        return { t: 'char', c: ch.charCodeAt(0) }\n    }\n  }\n\n  const ast = parseAlt()\n  if (!eof()) throw new Error('unexpected \"' + peek() + '\" in pattern')\n  return ast\n}\n\nfunction compileProg (ast) {\n  const prog = []\n  const emit = (op, extra) => { const idx = prog.length; prog.push(Object.assign({ op }, extra)); return idx }\n\n  function rec (n) {\n    switch (n.t) {\n      case 'empty': break\n      case 'char': emit('char', { c: n.c }); break\n      case 'any': emit('any'); break\n      case 'class': emit('class', { neg: n.neg, ranges: n.ranges }); break\n      case 'bol': emit('bol'); break\n      case 'eol': emit('eol'); break\n      case 'group': rec(n.child); break\n      case 'concat': for (const p of n.parts) rec(p); break\n      case 'alt': {\n        const jmps = []\n        for (let k = 0; k < n.opts.length; k++) {\n          if (k < n.opts.length - 1) {\n            const sp = emit('split', { x: 0, y: 0 })\n            prog[sp].x = prog.length\n            rec(n.opts[k])\n            jmps.push(emit('jmp', { x: 0 }))\n            prog[sp].y = prog.length\n          } else {\n            rec(n.opts[k])\n          }\n        }\n        for (const j of jmps) prog[j].x = prog.length\n        break\n      }\n      case 'star': {\n        const sp = emit('split', { x: 0, y: 0 })\n        prog[sp].x = prog.length\n        rec(n.child)\n        emit('jmp', { x: sp })\n        prog[sp].y = prog.length\n        break\n      }\n      case 'plus': {\n        const start = prog.length\n        rec(n.child)\n        const sp = emit('split', { x: start, y: 0 })\n        prog[sp].y = prog.length\n        break\n      }\n      case 'quest': {\n        const sp = emit('split', { x: 0, y: 0 })\n        prog[sp].x = prog.length\n        rec(n.child)\n        prog[sp].y = prog.length\n        break\n      }\n      case 'repeat': {\n        for (let k = 0; k < n.min; k++) rec(n.child)\n        if (n.max === Infinity) {\n          if (n.min === 0) rec({ t: 'star', child: n.child })\n          else rec({ t: 'star', child: n.child })\n        } else {\n          for (let k = 0; k < n.max - n.min; k++) rec({ t: 'quest', child: n.child })\n        }\n        break\n      }\n    }\n  }\n\n  rec(ast)\n  emit('match')\n  return prog\n}\n\n// Numeric opcodes for the runner. The program is compiled once into flat\n// typed arrays so the inner loop does no property lookups or string compares.\nconst OP_CHAR = 0\nconst OP_ANY = 1\nconst OP_CLASS = 2\nconst OP_SPLIT = 3\nconst OP_JMP = 4\nconst OP_BOL = 5\nconst OP_EOL = 6\nconst OP_MATCH = 7\n\nfunction classMatcher (instr) {\n  // ASCII is answered from a bitmap; anything above 0x7f walks the ranges.\n  const bits = new Uint8Array(128)\n  const r = instr.ranges\n  for (let k = 0; k < r.length; k++) {\n    const hi = Math.min(r[k][1], 127)\n    for (let c = r[k][0]; c <= hi; c++) bits[c] = 1\n  }\n  return { bits, ranges: r, neg: instr.neg }\n}\n\nfunction matchClass (cls, c) {\n  let inside\n  if (c < 128) {\n    inside = cls.bits[c] === 1\n  } else {\n    inside = false\n    const r = cls.ranges\n    for (let k = 0; k < r.length; k++) { if (c >= r[k][0] && c <= r[k][1]) { inside = true; break } }\n  }\n  return cls.neg ? !inside : inside\n}\n\nfunction makeRunner (prog) {\n  const n = prog.length\n  const ops = new Uint8Array(n)\n  const xs = new Int32Array(n)\n  const ys = new Int32Array(n)\n  const cs = new Int32Array(n)\n  const classes = new Array(n)\n  for (let i = 0; i < n; i++) {\n    const I = prog[i]\n    switch (I.op) {\n      case 'char': ops[i] = OP_CHAR; cs[i] = I.c; break\n      case 'any': ops[i] = OP_ANY; break\n      case 'class': ops[i] = OP_CLASS; classes[i] = classMatcher(I); break\n      case 'split': ops[i] = OP_SPLIT; xs[i] = I.x; ys[i] = I.y; break\n      case 'jmp': ops[i] = OP_JMP; xs[i] = I.x; break\n      case 'bol': ops[i] = OP_BOL; break\n      case 'eol': ops[i] = OP_EOL; break\n      case 'match': ops[i] = OP_MATCH; break\n    }\n  }\n\n  const lastGen = new Int32Array(n).fill(-1)\n  let gen = 0\n  // Each unvisited instruction is popped once and pushes at most two, so the\n  // stack never holds more than 2n + 1 entries.\n  const stack = new Int32Array(2 * n + 2)\n  // Thread lists hold at most one entry per instruction per step.\n  let clist = new Int32Array(n)\n  let nlist = new Int32Array(n)\n  let clen = 0\n  let nlen = 0\n\n  // Follows epsilon edges from `pc` and records every consuming instruction\n  // (or match) reached in `list`. `lastGen` dedupes per step.\n  function addThread (list, len0, pc, pos, len) {\n    // Most transitions land directly on a consuming instruction; skip the\n    // stack walk for those.\n    if (ops[pc] <= OP_CLASS || ops[pc] === OP_MATCH) {\n      if (lastGen[pc] === gen) return len0\n      lastGen[pc] = gen\n      list[len0] = pc\n      return len0 + 1\n    }\n    let sp = 0\n    stack[sp++] = pc\n    let count = len0\n    while (sp > 0) {\n      const p = stack[--sp]\n      if (lastGen[p] === gen) continue\n      lastGen[p] = gen\n      switch (ops[p]) {\n        case OP_JMP: stack[sp++] = xs[p]; break\n        case OP_SPLIT: stack[sp++] = ys[p]; stack[sp++] = xs[p]; break\n        case OP_BOL: if (pos === 0) stack[sp++] = p + 1; break\n        case OP_EOL: if (pos === len) stack[sp++] = p + 1; break\n        default: list[count++] = p\n      }\n    }\n    return count\n  }\n\n  // A pattern is anchored when starting it anywhere but position 0 yields no\n  // thread, which is the case for `^...` and its alternations. The probe sits\n  // at position 1 of a length-1 string so that only `^` can fail. For anchored\n  // patterns the per-position restart below is skipped, and an empty thread\n  // list means the match has already failed.\n  gen++\n  const anchored = addThread(nlist, 0, 0, 1, 1) === 0\n\n  function testNFA (s) {\n    const len = s.length\n    gen++\n    clen = addThread(clist, 0, 0, 0, len)\n    for (let pos = 0; pos <= len; pos++) {\n      const c = pos < len ? s.charCodeAt(pos) : -1\n      gen++\n      nlen = 0\n      for (let k = 0; k < clen; k++) {\n        const pc = clist[k]\n        switch (ops[pc]) {\n          case OP_MATCH: return true\n          case OP_CHAR: if (c === cs[pc]) nlen = addThread(nlist, nlen, pc + 1, pos + 1, len); break\n          case OP_ANY: if (c !== -1 && !isLineTerminator(c)) nlen = addThread(nlist, nlen, pc + 1, pos + 1, len); break\n          case OP_CLASS: if (c !== -1 && matchClass(classes[pc], c)) nlen = addThread(nlist, nlen, pc + 1, pos + 1, len); break\n        }\n      }\n      if (pos < len) {\n        if (!anchored) nlen = addThread(nlist, nlen, 0, pos + 1, len)\n        else if (nlen === 0) return false\n      }\n      const tmp = clist; clist = nlist; nlist = tmp\n      clen = nlen\n    }\n    return false\n  }\n\n  // Lazy DFA on top of the NFA. A DFA state is the set of consuming\n  // instructions live at a position; transitions are computed on first use\n  // and cached per ASCII character. `^` and `$` depend on position, so the\n  // closure is taken with flags for \"at start\" and \"at end\", which gives two\n  // start states and two transition tables per state. The state count is\n  // capped; past the cap the matcher falls back to the NFA walk above, so the\n  // time bound stays linear either way.\n  const MAX_STATES = 256\n  const states = []\n  const stateIds = new Map()\n  let overflow = false\n\n  function closure (list, count, pc, atStart, atEnd) {\n    // Same walk as addThread, with the position replaced by the two flags.\n    let sp = 0\n    stack[sp++] = pc\n    while (sp > 0) {\n      const p = stack[--sp]\n      if (lastGen[p] === gen) continue\n      lastGen[p] = gen\n      switch (ops[p]) {\n        case OP_JMP: stack[sp++] = xs[p]; break\n        case OP_SPLIT: stack[sp++] = ys[p]; stack[sp++] = xs[p]; break\n        case OP_BOL: if (atStart) stack[sp++] = p + 1; break\n        case OP_EOL: if (atEnd) stack[sp++] = p + 1; break\n        default: list[count++] = p\n      }\n    }\n    return count\n  }\n\n  function internState (list, count) {\n    const pcs = Array.from(list.subarray(0, count)).sort((a, b) => a - b)\n    const key = pcs.join(',')\n    let id = stateIds.get(key)\n    if (id !== undefined) return id\n    if (states.length >= MAX_STATES) { overflow = true; return -1 }\n    id = states.length\n    let isMatch = false\n    for (let k = 0; k < pcs.length; k++) if (ops[pcs[k]] === OP_MATCH) { isMatch = true; break }\n    states.push({ pcs: Int32Array.from(pcs), isMatch, next: new Int32Array(128).fill(-2), nextEnd: new Int32Array(128).fill(-2) })\n    stateIds.set(key, id)\n    return id\n  }\n\n  function step (state, c, atEnd) {\n    gen++\n    let count = 0\n    const pcs = state.pcs\n    for (let k = 0; k < pcs.length; k++) {\n      const pc = pcs[k]\n      switch (ops[pc]) {\n        case OP_CHAR: if (c === cs[pc]) count = closure(nlist, count, pc + 1, false, atEnd); break\n        case OP_ANY: if (!isLineTerminator(c)) count = closure(nlist, count, pc + 1, false, atEnd); break\n        case OP_CLASS: if (matchClass(classes[pc], c)) count = closure(nlist, count, pc + 1, false, atEnd); break\n      }\n    }\n    if (!anchored) count = closure(nlist, count, 0, false, atEnd)\n    return internState(nlist, count)\n  }\n\n  let startEmpty = -2\n  let startNonEmpty = -2\n\n  function startState (atEnd) {\n    gen++\n    const count = closure(nlist, 0, 0, true, atEnd)\n    return internState(nlist, count)\n  }\n\n  function testDFA (s) {\n    const len = s.length\n    let id\n    if (len === 0) {\n      if (startEmpty === -2) startEmpty = startState(true)\n      id = startEmpty\n    } else {\n      if (startNonEmpty === -2) startNonEmpty = startState(false)\n      id = startNonEmpty\n    }\n    if (id < 0) return testNFA(s)\n    let state = states[id]\n    for (let pos = 0; pos < len; pos++) {\n      if (state.isMatch) return true\n      const c = s.charCodeAt(pos)\n      const atEnd = pos + 1 === len\n      let nid\n      if (c < 128) {\n        const table = atEnd ? state.nextEnd : state.next\n        nid = table[c]\n        if (nid === -2) { nid = step(state, c, atEnd); table[c] = nid }\n      } else {\n        nid = step(state, c, atEnd)\n      }\n      if (nid < 0) return testNFA(s)\n      state = states[nid]\n      if (anchored && state.pcs.length === 0) return false\n    }\n    return state.isMatch\n  }\n\n  return function test (s) {\n    return overflow ? testNFA(s) : testDFA(s)\n  }\n}\n\nfunction compileSafe (pattern) {\n  const prog = compileProg(parse(pattern))\n  const runner = makeRunner(prog)\n  // `__ataSafe` brands the result so the standalone serializer can tell a safe\n  // matcher apart from a RegExp and emit `__ataSafeRe(source)` instead.\n  return { test: runner, source: pattern, __ataSafe: true }\n}\n\n// True when the linear engine can represent `src`. Used by the codegen to decide\n// between the safe matcher and a JS RegExp fallback for patterns outside the\n// supported (RE2) subset (backreferences, lookaround, etc.).\nfunction patternIsSafe (src) {\n  try { compileSafe(src); return true } catch { return false }\n}\n\nmodule.exports = { compileSafe, patternIsSafe, parse }\n";

package/lib/safe-regex.js

@@ -14,7 +14,15 @@
 // supported by linear engines; compileSafe throws on them so the caller can
 // decide (ata's codegen rejects such schemas rather than risk a hang).
 
-const WS = [[9, 13], [32, 32], [160, 160]]
+// ECMA-262 WhiteSpace and LineTerminator, what `\s` matches: tab, line feed,
+// vertical tab, form feed, carriage return, space, no-break space, the
+// Unicode space separators, the line and paragraph separators, and the byte
+// order mark. The set used to stop at U+00A0, so `^\S+$` accepted a string
+// holding U+2028 or U+3000.
+const WS = [[9, 13], [32, 32], [160, 160], [0x1680, 0x1680], [0x2000, 0x200a], [0x2028, 0x2029], [0x202f, 0x202f], [0x205f, 0x205f], [0x3000, 0x3000], [0xfeff, 0xfeff]]
+// `.` matches anything but a line terminator: LF, CR and U+2028/U+2029. It
+// excluded only LF, so `^.$` accepted "\r".
+const isLineTerminator = (c) => c === 10 || c === 13 || c === 0x2028 || c === 0x2029
 const DIGIT = [[48, 57]]
 const WORD = [[48, 57], [65, 90], [97, 122], [95, 95]]
 
@@ -342,7 +350,7 @@ function makeRunner (prog) {
         switch (ops[pc]) {
           case OP_MATCH: return true
           case OP_CHAR: if (c === cs[pc]) nlen = addThread(nlist, nlen, pc + 1, pos + 1, len); break
-          case OP_ANY: if (c !== -1 && c !== 10) nlen = addThread(nlist, nlen, pc + 1, pos + 1, len); break
+          case OP_ANY: if (c !== -1 && !isLineTerminator(c)) nlen = addThread(nlist, nlen, pc + 1, pos + 1, len); break
           case OP_CLASS: if (c !== -1 && matchClass(classes[pc], c)) nlen = addThread(nlist, nlen, pc + 1, pos + 1, len); break
         }
       }
@@ -409,7 +417,7 @@ function makeRunner (prog) {
       const pc = pcs[k]
       switch (ops[pc]) {
         case OP_CHAR: if (c === cs[pc]) count = closure(nlist, count, pc + 1, false, atEnd); break
-        case OP_ANY: if (c !== 10) count = closure(nlist, count, pc + 1, false, atEnd); break
+        case OP_ANY: if (!isLineTerminator(c)) count = closure(nlist, count, pc + 1, false, atEnd); break
         case OP_CLASS: if (matchClass(classes[pc], c)) count = closure(nlist, count, pc + 1, false, atEnd); break
       }
     }
@@ -477,4 +485,4 @@ function patternIsSafe (src) {
   try { compileSafe(src); return true } catch { return false }
 }
 
-module.exports = { compileSafe, patternIsSafe }
+module.exports = { compileSafe, patternIsSafe, parse }

package/lib/validator-core.js

@@ -1293,8 +1293,14 @@ class Validator {
   _installBufferApis(name) {
     this._ensureCompiled();
     const native = getNative();
-    if (!native) throw new Error(`Native addon required for ${name}(). Use validate(), isValidObject() or validateJSON(), which do not need it.`);
+    // Without the addon the buffer APIs answer through validate(), as they do
+    // for the shapes the gate routes: the same answers, slower. They threw
+    // before, which made the addon a requirement for these five methods where
+    // everywhere else it is an accelerator. Without Buffer, in a browser, there
+    // is nothing to decode with and the error stays.
+    if (!native && typeof Buffer === 'undefined') throw new Error(`Native addon required for ${name}(). Use validate(), isValidObject() or validateJSON(), which do not need it.`);
     const { bufferNeedsSlowPath, installSlowBufferApis } = require('./buffer-gate.js');
+    if (!native) return installSlowBufferApis(this);
     if (bufferNeedsSlowPath(this._schemaObj, this._schemaMap, this._keywords)) return installSlowBufferApis(this);
     this._ensureNative();
     const slot = this._fastSlot;

package/lib/version.js

@@ -7,4 +7,4 @@
 //
 // Kept in lockstep with package.json by `tests/test_version_sync.js`.
 
-module.exports = '1.37.0';
+module.exports = '1.38.0';

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "ata-validator",
-  "version": "1.37.0",
-  "description": "JSON Schema validation that compiles for speed and still runs where code generation is blocked. Compiled and interpreted engines answer identically at 100% of the official suite. TypeScript inference, Standard Schema V1, and a build step that emits dependency-free modules.",
+  "version": "1.38.0",
+  "description": "JSON Schema validator that compiles to plain JavaScript, still runs where code generation is blocked, and can be compiled away at build time. Passes the full official test suite.",
   "main": "index.js",
   "module": "index.mjs",
   "types": "index.d.ts",
@@ -59,7 +59,7 @@
     "release:check": "node scripts/regen-safe-regex-source.js && node tests/test_pack_purity.js && node scripts/check-doc-coverage.js && node tests/test_error_codes_lock.js && node tests/test_safe_regex_source_sync.js && node tests/test_version_sync.js",
     "build": "cmake-js build --target ata",
     "rebuild": "cmake-js rebuild --target ata",
-    "test": "node test.js && node tests/test_removed_aot_methods.js && node tests/test_no_native.js && node tests/test_no_eval.js && node tests/test_property_dependencies.js && node tests/test_v1_dialect.js && node tests/test_buffer_path_parity.js && node tests/test_text_path_errors.js && node tests/test_compiled_parity.js && node tests/test_buffer_gate.js && node tests/test_scanner_differential.js && node tests/test_validate_json_scanner_cost.js && node tests/test_buffer_reject_cost.js && node tests/test_nan_verdict.js && node tests/test_remove_additional_nested.js && node tests/test_exclusive_bounds.js && node tests/test_engine_differential.js && node tests/test_error_shape_differential.js && node tests/test_output_format.js && node tests/test_standard_json_schema.js && node tests/test_retry_message.js && node tests/test_describe_schema.js && node tests/test_aot_parse.js && node tests/test_draft7_semantics.js && node tests/test_metaschema_ref.js && node tests/test_pure_js_unsupported.js && node tests/test_native_load_order.js && node tests/test_pack_purity.js && node tests/test_make_native_package.js && node tests/test_browser_nofs.js && node tests/test_browser_imports_guard.js && node tests/test_esm_exports.js && node tests/test_version_sync.js && node tests/test_native_loaded.js && node tests/test_native_lazy.js && node tests/test_cold_start_modules.js && node tests/test_safe_regex_source_sync.js && node tests/test_t_builder.js && node tests/test_async_refine.js && node tests/test_safe_regex.js && node tests/test_safe_regex_integration.js && node tests/test_aot_build.js && node tests/test_aot_differential.js && node tests/test_aot_cli_build.js && node tests/test_aot_cli_smoke.js && node tests/test_bundle_standalone.js && node tests/test_standalone_anyof.js && node tests/test_standalone_formats.js && node tests/test_aot_format_mode.js && node tests/test_aot_additional_props_errors.js && node tests/test_id_anchor_refs.js && node tests/test_engine_routing.js && node tests/test_engine_option.js && node tests/test_compile_cache_order.js && node tests/test_engine_diagnostic.js && node tests/test_format_engine_parity.js && node tests/test_error_accumulator_collision.js && node tests/test_uri_helper_parity.js && node tests/test_uri_fast_path.js && node tests/test_extend_verdict.js && node tests/test_ipv4_regex.js && node tests/test_ipv6_fast_path.js && node tests/test_nested_recursion_decline.js && node tests/test_generator_error_content.js && node tests/test_inherited_key_presence.js && node tests/test_proto_accessor.js && node tests/test_hybrid_tier.js && node tests/test_first_rejection_builds_one.js && node tests/test_container_local.js && node tests/test_one_shot_validate.js && node tests/test_fixed_pattern_inline.js && node tests/test_plan_source.js && node tests/test_fused_remove_additional.js && node tests/test_runtime_parse.js && node tests/test_formats_single_pass.js && node tests/test_email_format_cost.js && node tests/test_format_single_error.js && node tests/test_defs_pointer_alias.js && node tests/test_cross_doc_root_ref.js && node tests/test_codegen_entrypoint_agreement.js && node tests/test_ref_annotation_siblings.js && node tests/test_gate_wrapped_schemas.js && node tests/test_pattern_message_escape.js && node tests/test_nested_defaults_engines.js && node tests/test_nested_coercion_parity.js && node tests/test_preprocess_cache.js && node tests/test_large_schema_codegen.js && node tests/test_base_uri.js && node tests/test_hoisted_def_recursion.js && node tests/test_alias_definitions.js && node tests/test_property_names_ref.js && node tests/test_hybrid_agreement.js && node tests/test_codegen_edge_shapes.js && node tests/test_deferred_def_scope.js && node tests/test_node_scoped_ctx_flags.js && node tests/test_fast_slot_exhaustion.js && node tests/test_ajv_errors.js && node tests/test_custom_keywords.js && node tests/test_strict_schema.js && node tests/test_aot_external_checks.js && node tests/test_ajv_parity.js && node tests/test_user_format_error_path.js && node tests/test_unevaluated_error_path.js && node tests/test_unevaluated_error_shape.js && node tests/test_pattern_properties_errors.js && node tests/test_additional_properties_scaling.js && node tests/test_additional_props_combined_cost.js && node tests/test_no_input_mutation.js && node tests/test_typed_validator_runner.js && node tests/test_define_schema.js && node tests/test_error_codes_lock.js && node tests/test_error_code_lookup.js && node tests/test_error_order.js && node tests/test_error_order_ordinal.js && node tests/test_ref_error_order.js && node tests/test_rejection_shape.js && node tests/test_lazy_normalization.js && node tests/test_value_equality.js && node tests/test_vocabulary.js && node tests/test_schema_scan.js && node tests/test_lazy_errors.js && node tests/test_single_pass_errors.js && node tests/test_lazy_json_errors.js && node tests/test_lazy_instance.js && node tests/test_cyclic_input.js && node tests/test_native_error_codes.js && node tests/test_verdict_preprocess.js && node tests/test_lite_parity.js && node tests/test_lite_bundle_size.js && node tests/test_defaults_own_keys.js && node tests/test_plan_compiler.js && node tests/test_nullable.js && node tests/test_validate_and_parse.js && node tests/test_validate_data.js && node tests/test_enrich_error.js && node tests/test_enrich_received.js && node tests/test_rich_errors_optout.js && node tests/test_error_messages.js && node tests/test_source_positions.js && node tests/fuzz_positions.js && node tests/test_malformed_json_termination.js && node tests/test_data_positions.js && node tests/test_targeted_positions.js && node tests/test_position_map_cost.js && node tests/test_targeted_positions_cost.js && node tests/test_aot_staleness.js && node tests/test_file_positions.js && node tests/test_aot_positions_cost.js && node tests/test_schema_hash.js && node tests/test_render_shared.js && node tests/test_renderers.js && node tests/test_additive_fields.js && node tests/test_diagnostic_source.js && node tests/test_diagnose.js && node tests/test_correlate.js && node tests/test_diagnostics_score.js && node tests/test_runtime_error_dx.js && node tests/test_aot_error_dx.js && node tests/test_abort_early.js && node tests/test_branch_collapse.js && node tests/test_suggestions.js && node tests/test_cli_validate.js && node tests/test_cli_version.js && node benchmark/bench_aot_size.mjs",
+    "test": "node test.js && node tests/test_removed_aot_methods.js && node tests/test_no_native.js && node tests/test_no_eval.js && node tests/test_property_dependencies.js && node tests/test_v1_dialect.js && node tests/test_buffer_path_parity.js && node tests/test_text_path_errors.js && node tests/test_compiled_parity.js && node tests/test_buffer_gate.js && node tests/test_scanner_differential.js && node tests/test_validate_json_scanner_cost.js && node tests/test_buffer_reject_cost.js && node tests/test_nan_verdict.js && node tests/test_remove_additional_nested.js && node tests/test_exclusive_bounds.js && node tests/test_engine_differential.js && node tests/test_error_shape_differential.js && node tests/test_output_format.js && node tests/test_standard_json_schema.js && node tests/test_retry_message.js && node tests/test_describe_schema.js && node tests/test_aot_parse.js && node tests/test_draft7_semantics.js && node tests/test_metaschema_ref.js && node tests/test_pure_js_unsupported.js && node tests/test_native_load_order.js && node tests/test_pack_purity.js && node tests/test_make_native_package.js && node tests/test_browser_nofs.js && node tests/test_browser_imports_guard.js && node tests/test_esm_exports.js && node tests/test_version_sync.js && node tests/test_native_loaded.js && node tests/test_native_lazy.js && node tests/test_buffer_apis_without_native.js && node tests/test_cold_start_modules.js && node tests/test_safe_regex_source_sync.js && node tests/test_t_builder.js && node tests/test_async_refine.js && node tests/test_safe_regex.js && node tests/test_safe_regex_integration.js && node tests/test_regex_ecma_whitespace.js && node tests/test_regex_native_linear.js && node tests/test_aot_build.js && node tests/test_aot_differential.js && node tests/test_aot_cli_build.js && node tests/test_aot_cli_smoke.js && node tests/test_bundle_standalone.js && node tests/test_standalone_anyof.js && node tests/test_standalone_formats.js && node tests/test_aot_format_mode.js && node tests/test_aot_additional_props_errors.js && node tests/test_id_anchor_refs.js && node tests/test_engine_routing.js && node tests/test_engine_option.js && node tests/test_compile_cache_order.js && node tests/test_engine_diagnostic.js && node tests/test_format_engine_parity.js && node tests/test_error_accumulator_collision.js && node tests/test_uri_helper_parity.js && node tests/test_uri_fast_path.js && node tests/test_extend_verdict.js && node tests/test_ipv4_regex.js && node tests/test_ipv6_fast_path.js && node tests/test_nested_recursion_decline.js && node tests/test_generator_error_content.js && node tests/test_inherited_key_presence.js && node tests/test_proto_accessor.js && node tests/test_hybrid_tier.js && node tests/test_first_rejection_builds_one.js && node tests/test_container_local.js && node tests/test_one_shot_validate.js && node tests/test_fixed_pattern_inline.js && node tests/test_plan_source.js && node tests/test_fused_remove_additional.js && node tests/test_runtime_parse.js && node tests/test_formats_single_pass.js && node tests/test_email_format_cost.js && node tests/test_format_single_error.js && node tests/test_defs_pointer_alias.js && node tests/test_cross_doc_root_ref.js && node tests/test_codegen_entrypoint_agreement.js && node tests/test_ref_annotation_siblings.js && node tests/test_gate_wrapped_schemas.js && node tests/test_pattern_message_escape.js && node tests/test_nested_defaults_engines.js && node tests/test_nested_coercion_parity.js && node tests/test_preprocess_cache.js && node tests/test_large_schema_codegen.js && node tests/test_base_uri.js && node tests/test_hoisted_def_recursion.js && node tests/test_alias_definitions.js && node tests/test_property_names_ref.js && node tests/test_hybrid_agreement.js && node tests/test_codegen_edge_shapes.js && node tests/test_deferred_def_scope.js && node tests/test_node_scoped_ctx_flags.js && node tests/test_fast_slot_exhaustion.js && node tests/test_ajv_errors.js && node tests/test_custom_keywords.js && node tests/test_strict_schema.js && node tests/test_aot_external_checks.js && node tests/test_ajv_parity.js && node tests/test_user_format_error_path.js && node tests/test_unevaluated_error_path.js && node tests/test_unevaluated_error_shape.js && node tests/test_pattern_properties_errors.js && node tests/test_additional_properties_scaling.js && node tests/test_additional_props_combined_cost.js && node tests/test_no_input_mutation.js && node tests/test_typed_validator_runner.js && node tests/test_define_schema.js && node tests/test_error_codes_lock.js && node tests/test_error_code_lookup.js && node tests/test_error_order.js && node tests/test_error_order_ordinal.js && node tests/test_ref_error_order.js && node tests/test_rejection_shape.js && node tests/test_lazy_normalization.js && node tests/test_value_equality.js && node tests/test_vocabulary.js && node tests/test_schema_scan.js && node tests/test_lazy_errors.js && node tests/test_single_pass_errors.js && node tests/test_lazy_json_errors.js && node tests/test_lazy_instance.js && node tests/test_cyclic_input.js && node tests/test_native_error_codes.js && node tests/test_verdict_preprocess.js && node tests/test_lite_parity.js && node tests/test_lite_bundle_size.js && node tests/test_defaults_own_keys.js && node tests/test_plan_compiler.js && node tests/test_nullable.js && node tests/test_validate_and_parse.js && node tests/test_validate_data.js && node tests/test_enrich_error.js && node tests/test_enrich_received.js && node tests/test_rich_errors_optout.js && node tests/test_error_messages.js && node tests/test_source_positions.js && node tests/fuzz_positions.js && node tests/test_malformed_json_termination.js && node tests/test_data_positions.js && node tests/test_targeted_positions.js && node tests/test_position_map_cost.js && node tests/test_targeted_positions_cost.js && node tests/test_aot_staleness.js && node tests/test_file_positions.js && node tests/test_aot_positions_cost.js && node tests/test_schema_hash.js && node tests/test_render_shared.js && node tests/test_renderers.js && node tests/test_additive_fields.js && node tests/test_diagnostic_source.js && node tests/test_diagnose.js && node tests/test_correlate.js && node tests/test_diagnostics_score.js && node tests/test_runtime_error_dx.js && node tests/test_aot_error_dx.js && node tests/test_abort_early.js && node tests/test_branch_collapse.js && node tests/test_suggestions.js && node tests/test_cli_validate.js && node tests/test_cli_version.js && node benchmark/bench_aot_size.mjs",
     "bench:size": "node benchmark/bench_aot_size.mjs",
     "test:suite": "node tests/run_suite.js && node tests/run_suite.js draft7 && node tests/run_suite.js v1",
     "test:compat": "node tests/test_compat.js",
@@ -135,13 +135,13 @@
     "LICENSE"
   ],
   "optionalDependencies": {
-    "@ata-validator/native-darwin-arm64": "1.37.0",
-    "@ata-validator/native-darwin-x64": "1.37.0",
-    "@ata-validator/native-linux-arm64-gnu": "1.37.0",
-    "@ata-validator/native-linux-arm64-musl": "1.37.0",
-    "@ata-validator/native-linux-x64-gnu": "1.37.0",
-    "@ata-validator/native-linux-x64-musl": "1.37.0",
-    "@ata-validator/native-win32-x64": "1.37.0"
+    "@ata-validator/native-darwin-arm64": "1.38.0",
+    "@ata-validator/native-darwin-x64": "1.38.0",
+    "@ata-validator/native-linux-arm64-gnu": "1.38.0",
+    "@ata-validator/native-linux-arm64-musl": "1.38.0",
+    "@ata-validator/native-linux-x64-gnu": "1.38.0",
+    "@ata-validator/native-linux-x64-musl": "1.38.0",
+    "@ata-validator/native-win32-x64": "1.38.0"
   },
   "peerDependencies": {
     "yaml": "^2.0.0"