lib/Support/JSON.cpp

   1 //=== JSON.cpp - JSON value, parsing and serialization - C++ -----------*-===//
   2 //
   3 //                     The LLVM Compiler Infrastructure
   4 //
   5 // This file is distributed under the University of Illinois Open Source
   6 // License. See LICENSE.TXT for details.
   7 //
   8 //===---------------------------------------------------------------------===//
   9
  10 #include "llvm/Support/JSON.h"
  11 #include "llvm/Support/ConvertUTF.h"
  12 #include "llvm/Support/Format.h"
  13 #include <cctype>
  14
  15 namespace llvm {
  16 namespace json {
  17
  18 Value &Object::operator[](const ObjectKey &K) {
  19   return try_emplace(K, nullptr).first->getSecond();
  20 }
  21 Value &Object::operator[](ObjectKey &&K) {
  22   return try_emplace(std::move(K), nullptr).first->getSecond();
  23 }
  24 Value *Object::get(StringRef K) {
  25   auto I = find(K);
  26   if (I == end())
  27     return nullptr;
  28   return &I->second;
  29 }
  30 const Value *Object::get(StringRef K) const {
  31   auto I = find(K);
  32   if (I == end())
  33     return nullptr;
  34   return &I->second;
  35 }
  36 llvm::Optional<std::nullptr_t> Object::getNull(StringRef K) const {
  37   if (auto *V = get(K))
  38     return V->getAsNull();
  39   return llvm::None;
  40 }
  41 llvm::Optional<bool> Object::getBoolean(StringRef K) const {
  42   if (auto *V = get(K))
  43     return V->getAsBoolean();
  44   return llvm::None;
  45 }
  46 llvm::Optional<double> Object::getNumber(StringRef K) const {
  47   if (auto *V = get(K))
  48     return V->getAsNumber();
  49   return llvm::None;
  50 }
  51 llvm::Optional<int64_t> Object::getInteger(StringRef K) const {
  52   if (auto *V = get(K))
  53     return V->getAsInteger();
  54   return llvm::None;
  55 }
  56 llvm::Optional<llvm::StringRef> Object::getString(StringRef K) const {
  57   if (auto *V = get(K))
  58     return V->getAsString();
  59   return llvm::None;
  60 }
  61 const json::Object *Object::getObject(StringRef K) const {
  62   if (auto *V = get(K))
  63     return V->getAsObject();
  64   return nullptr;
  65 }
  66 json::Object *Object::getObject(StringRef K) {
  67   if (auto *V = get(K))
  68     return V->getAsObject();
  69   return nullptr;
  70 }
  71 const json::Array *Object::getArray(StringRef K) const {
  72   if (auto *V = get(K))
  73     return V->getAsArray();
  74   return nullptr;
  75 }
  76 json::Array *Object::getArray(StringRef K) {
  77   if (auto *V = get(K))
  78     return V->getAsArray();
  79   return nullptr;
  80 }
  81 bool operator==(const Object &LHS, const Object &RHS) {
  82   if (LHS.size() != RHS.size())
  83     return false;
  84   for (const auto &L : LHS) {
  85     auto R = RHS.find(L.first);
  86     if (R == RHS.end() || L.second != R->second)
  87       return false;
  88   }
  89   return true;
  90 }
  91
  92 Array::Array(std::initializer_list<Value> Elements) {
  93   V.reserve(Elements.size());
  94   for (const Value &V : Elements) {
  95     emplace_back(nullptr);
  96     back().moveFrom(std::move(V));
  97   }
  98 }
  99
 100 Value::Value(std::initializer_list<Value> Elements)
 101     : Value(json::Array(Elements)) {}
 102
 103 void Value::copyFrom(const Value &M) {
 104   Type = M.Type;
 105   switch (Type) {
 106   case T_Null:
 107   case T_Boolean:
 108   case T_Double:
 109   case T_Integer:
 110     memcpy(Union.buffer, M.Union.buffer, sizeof(Union.buffer));
 111     break;
 112   case T_StringRef:
 113     create<StringRef>(M.as<StringRef>());
 114     break;
 115   case T_String:
 116     create<std::string>(M.as<std::string>());
 117     break;
 118   case T_Object:
 119     create<json::Object>(M.as<json::Object>());
 120     break;
 121   case T_Array:
 122     create<json::Array>(M.as<json::Array>());
 123     break;
 124   }
 125 }
 126
 127 void Value::moveFrom(const Value &&M) {
 128   Type = M.Type;
 129   switch (Type) {
 130   case T_Null:
 131   case T_Boolean:
 132   case T_Double:
 133   case T_Integer:
 134     memcpy(Union.buffer, M.Union.buffer, sizeof(Union.buffer));
 135     break;
 136   case T_StringRef:
 137     create<StringRef>(M.as<StringRef>());
 138     break;
 139   case T_String:
 140     create<std::string>(std::move(M.as<std::string>()));
 141     M.Type = T_Null;
 142     break;
 143   case T_Object:
 144     create<json::Object>(std::move(M.as<json::Object>()));
 145     M.Type = T_Null;
 146     break;
 147   case T_Array:
 148     create<json::Array>(std::move(M.as<json::Array>()));
 149     M.Type = T_Null;
 150     break;
 151   }
 152 }
 153
 154 void Value::destroy() {
 155   switch (Type) {
 156   case T_Null:
 157   case T_Boolean:
 158   case T_Double:
 159   case T_Integer:
 160     break;
 161   case T_StringRef:
 162     as<StringRef>().~StringRef();
 163     break;
 164   case T_String:
 165     as<std::string>().~basic_string();
 166     break;
 167   case T_Object:
 168     as<json::Object>().~Object();
 169     break;
 170   case T_Array:
 171     as<json::Array>().~Array();
 172     break;
 173   }
 174 }
 175
 176 bool operator==(const Value &L, const Value &R) {
 177   if (L.kind() != R.kind())
 178     return false;
 179   switch (L.kind()) {
 180   case Value::Null:
 181     return *L.getAsNull() == *R.getAsNull();
 182   case Value::Boolean:
 183     return *L.getAsBoolean() == *R.getAsBoolean();
 184   case Value::Number:
 185     return *L.getAsNumber() == *R.getAsNumber();
 186   case Value::String:
 187     return *L.getAsString() == *R.getAsString();
 188   case Value::Array:
 189     return *L.getAsArray() == *R.getAsArray();
 190   case Value::Object:
 191     return *L.getAsObject() == *R.getAsObject();
 192   }
 193   llvm_unreachable("Unknown value kind");
 194 }
 195
 196 namespace {
 197 // Simple recursive-descent JSON parser.
 198 class Parser {
 199 public:
 200   Parser(StringRef JSON)
 201       : Start(JSON.begin()), P(JSON.begin()), End(JSON.end()) {}
 202
 203   bool checkUTF8() {
 204     size_t ErrOffset;
 205     if (isUTF8(StringRef(Start, End - Start), &ErrOffset))
 206       return true;
 207     P = Start + ErrOffset; // For line/column calculation.
 208     return parseError("Invalid UTF-8 sequence");
 209   }
 210
 211   bool parseValue(Value &Out);
 212
 213   bool assertEnd() {
 214     eatWhitespace();
 215     if (P == End)
 216       return true;
 217     return parseError("Text after end of document");
 218   }
 219
 220   Error takeError() {
 221     assert(Err);
 222     return std::move(*Err);
 223   }
 224
 225 private:
 226   void eatWhitespace() {
 227     while (P != End && (*P == ' ' || *P == '\r' || *P == '\n' || *P == '\t'))
 228       ++P;
 229   }
 230
 231   // On invalid syntax, parseX() functions return false and set Err.
 232   bool parseNumber(char First, Value &Out);
 233   bool parseString(std::string &Out);
 234   bool parseUnicode(std::string &Out);
 235   bool parseError(const char *Msg); // always returns false
 236
 237   char next() { return P == End ? 0 : *P++; }
 238   char peek() { return P == End ? 0 : *P; }
 239   static bool isNumber(char C) {
 240     return C == '0' || C == '1' || C == '2' || C == '3' || C == '4' ||
 241            C == '5' || C == '6' || C == '7' || C == '8' || C == '9' ||
 242            C == 'e' || C == 'E' || C == '+' || C == '-' || C == '.';
 243   }
 244
 245   Optional<Error> Err;
 246   const char *Start, *P, *End;
 247 };
 248
 249 bool Parser::parseValue(Value &Out) {
 250   eatWhitespace();
 251   if (P == End)
 252     return parseError("Unexpected EOF");
 253   switch (char C = next()) {
 254   // Bare null/true/false are easy - first char identifies them.
 255   case 'n':
 256     Out = nullptr;
 257     return (next() == 'u' && next() == 'l' && next() == 'l') ||
 258            parseError("Invalid JSON value (null?)");
 259   case 't':
 260     Out = true;
 261     return (next() == 'r' && next() == 'u' && next() == 'e') ||
 262            parseError("Invalid JSON value (true?)");
 263   case 'f':
 264     Out = false;
 265     return (next() == 'a' && next() == 'l' && next() == 's' && next() == 'e') ||
 266            parseError("Invalid JSON value (false?)");
 267   case '"': {
 268     std::string S;
 269     if (parseString(S)) {
 270       Out = std::move(S);
 271       return true;
 272     }
 273     return false;
 274   }
 275   case '[': {
 276     Out = Array{};
 277     Array &A = *Out.getAsArray();
 278     eatWhitespace();
 279     if (peek() == ']') {
 280       ++P;
 281       return true;
 282     }
 283     for (;;) {
 284       A.emplace_back(nullptr);
 285       if (!parseValue(A.back()))
 286         return false;
 287       eatWhitespace();
 288       switch (next()) {
 289       case ',':
 290         eatWhitespace();
 291         continue;
 292       case ']':
 293         return true;
 294       default:
 295         return parseError("Expected , or ] after array element");
 296       }
 297     }
 298   }
 299   case '{': {
 300     Out = Object{};
 301     Object &O = *Out.getAsObject();
 302     eatWhitespace();
 303     if (peek() == '}') {
 304       ++P;
 305       return true;
 306     }
 307     for (;;) {
 308       if (next() != '"')
 309         return parseError("Expected object key");
 310       std::string K;
 311       if (!parseString(K))
 312         return false;
 313       eatWhitespace();
 314       if (next() != ':')
 315         return parseError("Expected : after object key");
 316       eatWhitespace();
 317       if (!parseValue(O[std::move(K)]))
 318         return false;
 319       eatWhitespace();
 320       switch (next()) {
 321       case ',':
 322         eatWhitespace();
 323         continue;
 324       case '}':
 325         return true;
 326       default:
 327         return parseError("Expected , or } after object property");
 328       }
 329     }
 330   }
 331   default:
 332     if (isNumber(C))
 333       return parseNumber(C, Out);
 334     return parseError("Invalid JSON value");
 335   }
 336 }
 337
 338 bool Parser::parseNumber(char First, Value &Out) {
 339   // Read the number into a string. (Must be null-terminated for strto*).
 340   SmallString<24> S;
 341   S.push_back(First);
 342   while (isNumber(peek()))
 343     S.push_back(next());
 344   char *End;
 345   // Try first to parse as integer, and if so preserve full 64 bits.
 346   // strtoll returns long long >= 64 bits, so check it's in range too.
 347   auto I = std::strtoll(S.c_str(), &End, 10);
 348   if (End == S.end() && I >= std::numeric_limits<int64_t>::min() &&
 349       I <= std::numeric_limits<int64_t>::max()) {
 350     Out = int64_t(I);
 351     return true;
 352   }
 353   // If it's not an integer
 354   Out = std::strtod(S.c_str(), &End);
 355   return End == S.end() || parseError("Invalid JSON value (number?)");
 356 }
 357
 358 bool Parser::parseString(std::string &Out) {
 359   // leading quote was already consumed.
 360   for (char C = next(); C != '"'; C = next()) {
 361     if (LLVM_UNLIKELY(P == End))
 362       return parseError("Unterminated string");
 363     if (LLVM_UNLIKELY((C & 0x1f) == C))
 364       return parseError("Control character in string");
 365     if (LLVM_LIKELY(C != '\\')) {
 366       Out.push_back(C);
 367       continue;
 368     }
 369     // Handle escape sequence.
 370     switch (C = next()) {
 371     case '"':
 372     case '\\':
 373     case '/':
 374       Out.push_back(C);
 375       break;
 376     case 'b':
 377       Out.push_back('\b');
 378       break;
 379     case 'f':
 380       Out.push_back('\f');
 381       break;
 382     case 'n':
 383       Out.push_back('\n');
 384       break;
 385     case 'r':
 386       Out.push_back('\r');
 387       break;
 388     case 't':
 389       Out.push_back('\t');
 390       break;
 391     case 'u':
 392       if (!parseUnicode(Out))
 393         return false;
 394       break;
 395     default:
 396       return parseError("Invalid escape sequence");
 397     }
 398   }
 399   return true;
 400 }
 401
 402 static void encodeUtf8(uint32_t Rune, std::string &Out) {
 403   if (Rune < 0x80) {
 404     Out.push_back(Rune & 0x7F);
 405   } else if (Rune < 0x800) {
 406     uint8_t FirstByte = 0xC0 | ((Rune & 0x7C0) >> 6);
 407     uint8_t SecondByte = 0x80 | (Rune & 0x3F);
 408     Out.push_back(FirstByte);
 409     Out.push_back(SecondByte);
 410   } else if (Rune < 0x10000) {
 411     uint8_t FirstByte = 0xE0 | ((Rune & 0xF000) >> 12);
 412     uint8_t SecondByte = 0x80 | ((Rune & 0xFC0) >> 6);
 413     uint8_t ThirdByte = 0x80 | (Rune & 0x3F);
 414     Out.push_back(FirstByte);
 415     Out.push_back(SecondByte);
 416     Out.push_back(ThirdByte);
 417   } else if (Rune < 0x110000) {
 418     uint8_t FirstByte = 0xF0 | ((Rune & 0x1F0000) >> 18);
 419     uint8_t SecondByte = 0x80 | ((Rune & 0x3F000) >> 12);
 420     uint8_t ThirdByte = 0x80 | ((Rune & 0xFC0) >> 6);
 421     uint8_t FourthByte = 0x80 | (Rune & 0x3F);
 422     Out.push_back(FirstByte);
 423     Out.push_back(SecondByte);
 424     Out.push_back(ThirdByte);
 425     Out.push_back(FourthByte);
 426   } else {
 427     llvm_unreachable("Invalid codepoint");
 428   }
 429 }
 430
 431 // Parse a UTF-16 \uNNNN escape sequence. "\u" has already been consumed.
 432 // May parse several sequential escapes to ensure proper surrogate handling.
 433 // We do not use ConvertUTF.h, it can't accept and replace unpaired surrogates.
 434 // These are invalid Unicode but valid JSON (RFC 8259, section 8.2).
 435 bool Parser::parseUnicode(std::string &Out) {
 436   // Invalid UTF is not a JSON error (RFC 8529§8.2). It gets replaced by U+FFFD.
 437   auto Invalid = [&] { Out.append(/* UTF-8 */ {'\xef', '\xbf', '\xbd'}); };
 438   // Decodes 4 hex digits from the stream into Out, returns false on error.
 439   auto Parse4Hex = [this](uint16_t &Out) -> bool {
 440     Out = 0;
 441     char Bytes[] = {next(), next(), next(), next()};
 442     for (unsigned char C : Bytes) {
 443       if (!std::isxdigit(C))
 444         return parseError("Invalid \\u escape sequence");
 445       Out <<= 4;
 446       Out |= (C > '9') ? (C & ~0x20) - 'A' + 10 : (C - '0');
 447     }
 448     return true;
 449   };
 450   uint16_t First; // UTF-16 code unit from the first \u escape.
 451   if (!Parse4Hex(First))
 452     return false;
 453
 454   // We loop to allow proper surrogate-pair error handling.
 455   while (true) {
 456     // Case 1: the UTF-16 code unit is already a codepoint in the BMP.
 457     if (LLVM_LIKELY(First < 0xD800 || First >= 0xE000)) {
 458       encodeUtf8(First, Out);
 459       return true;
 460     }
 461
 462     // Case 2: it's an (unpaired) trailing surrogate.
 463     if (LLVM_UNLIKELY(First >= 0xDC00)) {
 464       Invalid();
 465       return true;
 466     }
 467
 468     // Case 3: it's a leading surrogate. We expect a trailing one next.
 469     // Case 3a: there's no trailing \u escape. Don't advance in the stream.
 470     if (LLVM_UNLIKELY(P + 2 > End || *P != '\\' || *(P + 1) != 'u')) {
 471       Invalid(); // Leading surrogate was unpaired.
 472       return true;
 473     }
 474     P += 2;
 475     uint16_t Second;
 476     if (!Parse4Hex(Second))
 477       return false;
 478     // Case 3b: there was another \u escape, but it wasn't a trailing surrogate.
 479     if (LLVM_UNLIKELY(Second < 0xDC00 || Second >= 0xE000)) {
 480       Invalid();      // Leading surrogate was unpaired.
 481       First = Second; // Second escape still needs to be processed.
 482       continue;
 483     }
 484     // Case 3c: a valid surrogate pair encoding an astral codepoint.
 485     encodeUtf8(0x10000 | ((First - 0xD800) << 10) | (Second - 0xDC00), Out);
 486     return true;
 487   }
 488 }
 489
 490 bool Parser::parseError(const char *Msg) {
 491   int Line = 1;
 492   const char *StartOfLine = Start;
 493   for (const char *X = Start; X < P; ++X) {
 494     if (*X == 0x0A) {
 495       ++Line;
 496       StartOfLine = X + 1;
 497     }
 498   }
 499   Err.emplace(
 500       llvm::make_unique<ParseError>(Msg, Line, P - StartOfLine, P - Start));
 501   return false;
 502 }
 503 } // namespace
 504
 505 Expected<Value> parse(StringRef JSON) {
 506   Parser P(JSON);
 507   Value E = nullptr;
 508   if (P.checkUTF8())
 509     if (P.parseValue(E))
 510       if (P.assertEnd())
 511         return std::move(E);
 512   return P.takeError();
 513 }
 514 char ParseError::ID = 0;
 515
 516 static std::vector<const Object::value_type *> sortedElements(const Object &O) {
 517   std::vector<const Object::value_type *> Elements;
 518   for (const auto &E : O)
 519     Elements.push_back(&E);
 520   llvm::sort(Elements,
 521              [](const Object::value_type *L, const Object::value_type *R) {
 522                return L->first < R->first;
 523              });
 524   return Elements;
 525 }
 526
 527 bool isUTF8(llvm::StringRef S, size_t *ErrOffset) {
 528   // Fast-path for ASCII, which is valid UTF-8.
 529   if (LLVM_LIKELY(isASCII(S)))
 530     return true;
 531
 532   const UTF8 *Data = reinterpret_cast<const UTF8 *>(S.data()), *Rest = Data;
 533   if (LLVM_LIKELY(isLegalUTF8String(&Rest, Data + S.size())))
 534     return true;
 535
 536   if (ErrOffset)
 537     *ErrOffset = Rest - Data;
 538   return false;
 539 }
 540
 541 std::string fixUTF8(llvm::StringRef S) {
 542   // This isn't particularly efficient, but is only for error-recovery.
 543   std::vector<UTF32> Codepoints(S.size()); // 1 codepoint per byte suffices.
 544   const UTF8 *In8 = reinterpret_cast<const UTF8 *>(S.data());
 545   UTF32 *Out32 = Codepoints.data();
 546   ConvertUTF8toUTF32(&In8, In8 + S.size(), &Out32, Out32 + Codepoints.size(),
 547                      lenientConversion);
 548   Codepoints.resize(Out32 - Codepoints.data());
 549   std::string Res(4 * Codepoints.size(), 0); // 4 bytes per codepoint suffice
 550   const UTF32 *In32 = Codepoints.data();
 551   UTF8 *Out8 = reinterpret_cast<UTF8 *>(&Res[0]);
 552   ConvertUTF32toUTF8(&In32, In32 + Codepoints.size(), &Out8, Out8 + Res.size(),
 553                      strictConversion);
 554   Res.resize(reinterpret_cast<char *>(Out8) - Res.data());
 555   return Res;
 556 }
 557
 558 } // namespace json
 559 } // namespace llvm
 560
 561 static void quote(llvm::raw_ostream &OS, llvm::StringRef S) {
 562   OS << '\"';
 563   for (unsigned char C : S) {
 564     if (C == 0x22 || C == 0x5C)
 565       OS << '\\';
 566     if (C >= 0x20) {
 567       OS << C;
 568       continue;
 569     }
 570     OS << '\\';
 571     switch (C) {
 572     // A few characters are common enough to make short escapes worthwhile.
 573     case '\t':
 574       OS << 't';
 575       break;
 576     case '\n':
 577       OS << 'n';
 578       break;
 579     case '\r':
 580       OS << 'r';
 581       break;
 582     default:
 583       OS << 'u';
 584       llvm::write_hex(OS, C, llvm::HexPrintStyle::Lower, 4);
 585       break;
 586     }
 587   }
 588   OS << '\"';
 589 }
 590
 591 enum IndenterAction {
 592   Indent,
 593   Outdent,
 594   Newline,
 595   Space,
 596 };
 597
 598 // Prints JSON. The indenter can be used to control formatting.
 599 template <typename Indenter>
 600 void llvm::json::Value::print(raw_ostream &OS, const Indenter &I) const {
 601   switch (Type) {
 602   case T_Null:
 603     OS << "null";
 604     break;
 605   case T_Boolean:
 606     OS << (as<bool>() ? "true" : "false");
 607     break;
 608   case T_Double:
 609     OS << format("%.*g", std::numeric_limits<double>::max_digits10,
 610                  as<double>());
 611     break;
 612   case T_Integer:
 613     OS << as<int64_t>();
 614     break;
 615   case T_StringRef:
 616     quote(OS, as<StringRef>());
 617     break;
 618   case T_String:
 619     quote(OS, as<std::string>());
 620     break;
 621   case T_Object: {
 622     bool Comma = false;
 623     OS << '{';
 624     I(Indent);
 625     for (const auto *P : sortedElements(as<json::Object>())) {
 626       if (Comma)
 627         OS << ',';
 628       Comma = true;
 629       I(Newline);
 630       quote(OS, P->first);
 631       OS << ':';
 632       I(Space);
 633       P->second.print(OS, I);
 634     }
 635     I(Outdent);
 636     if (Comma)
 637       I(Newline);
 638     OS << '}';
 639     break;
 640   }
 641   case T_Array: {
 642     bool Comma = false;
 643     OS << '[';
 644     I(Indent);
 645     for (const auto &E : as<json::Array>()) {
 646       if (Comma)
 647         OS << ',';
 648       Comma = true;
 649       I(Newline);
 650       E.print(OS, I);
 651     }
 652     I(Outdent);
 653     if (Comma)
 654       I(Newline);
 655     OS << ']';
 656     break;
 657   }
 658   }
 659 }
 660
 661 void llvm::format_provider<llvm::json::Value>::format(
 662     const llvm::json::Value &E, raw_ostream &OS, StringRef Options) {
 663   if (Options.empty()) {
 664     OS << E;
 665     return;
 666   }
 667   unsigned IndentAmount = 0;
 668   if (Options.getAsInteger(/*Radix=*/10, IndentAmount))
 669     llvm_unreachable("json::Value format options should be an integer");
 670   unsigned IndentLevel = 0;
 671   E.print(OS, [&](IndenterAction A) {
 672     switch (A) {
 673     case Newline:
 674       OS << '\n';
 675       OS.indent(IndentLevel);
 676       break;
 677     case Space:
 678       OS << ' ';
 679       break;
 680     case Indent:
 681       IndentLevel += IndentAmount;
 682       break;
 683     case Outdent:
 684       IndentLevel -= IndentAmount;
 685       break;
 686     };
 687   });
 688 }
 689
 690 llvm::raw_ostream &llvm::json::operator<<(raw_ostream &OS, const Value &E) {
 691   E.print(OS, [](IndenterAction A) { /*ignore*/ });
 692   return OS;
 693 }