xref: /llvm-project/llvm/lib/Support/JSON.cpp (revision 2946cd701067404b99c39fb29dc9c74bd7193eb3)
1 //=== JSON.cpp - JSON value, parsing and serialization - C++ -----------*-===//
2 //
3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4 // See https://llvm.org/LICENSE.txt for license information.
5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6 //
7 //===---------------------------------------------------------------------===//
8 
9 #include "llvm/Support/JSON.h"
10 #include "llvm/Support/ConvertUTF.h"
11 #include "llvm/Support/Format.h"
12 #include <cctype>
13 
14 namespace llvm {
15 namespace json {
16 
17 Value &Object::operator[](const ObjectKey &K) {
18   return try_emplace(K, nullptr).first->getSecond();
19 }
20 Value &Object::operator[](ObjectKey &&K) {
21   return try_emplace(std::move(K), nullptr).first->getSecond();
22 }
23 Value *Object::get(StringRef K) {
24   auto I = find(K);
25   if (I == end())
26     return nullptr;
27   return &I->second;
28 }
29 const Value *Object::get(StringRef K) const {
30   auto I = find(K);
31   if (I == end())
32     return nullptr;
33   return &I->second;
34 }
35 llvm::Optional<std::nullptr_t> Object::getNull(StringRef K) const {
36   if (auto *V = get(K))
37     return V->getAsNull();
38   return llvm::None;
39 }
40 llvm::Optional<bool> Object::getBoolean(StringRef K) const {
41   if (auto *V = get(K))
42     return V->getAsBoolean();
43   return llvm::None;
44 }
45 llvm::Optional<double> Object::getNumber(StringRef K) const {
46   if (auto *V = get(K))
47     return V->getAsNumber();
48   return llvm::None;
49 }
50 llvm::Optional<int64_t> Object::getInteger(StringRef K) const {
51   if (auto *V = get(K))
52     return V->getAsInteger();
53   return llvm::None;
54 }
55 llvm::Optional<llvm::StringRef> Object::getString(StringRef K) const {
56   if (auto *V = get(K))
57     return V->getAsString();
58   return llvm::None;
59 }
60 const json::Object *Object::getObject(StringRef K) const {
61   if (auto *V = get(K))
62     return V->getAsObject();
63   return nullptr;
64 }
65 json::Object *Object::getObject(StringRef K) {
66   if (auto *V = get(K))
67     return V->getAsObject();
68   return nullptr;
69 }
70 const json::Array *Object::getArray(StringRef K) const {
71   if (auto *V = get(K))
72     return V->getAsArray();
73   return nullptr;
74 }
75 json::Array *Object::getArray(StringRef K) {
76   if (auto *V = get(K))
77     return V->getAsArray();
78   return nullptr;
79 }
80 bool operator==(const Object &LHS, const Object &RHS) {
81   if (LHS.size() != RHS.size())
82     return false;
83   for (const auto &L : LHS) {
84     auto R = RHS.find(L.first);
85     if (R == RHS.end() || L.second != R->second)
86       return false;
87   }
88   return true;
89 }
90 
91 Array::Array(std::initializer_list<Value> Elements) {
92   V.reserve(Elements.size());
93   for (const Value &V : Elements) {
94     emplace_back(nullptr);
95     back().moveFrom(std::move(V));
96   }
97 }
98 
99 Value::Value(std::initializer_list<Value> Elements)
100     : Value(json::Array(Elements)) {}
101 
102 void Value::copyFrom(const Value &M) {
103   Type = M.Type;
104   switch (Type) {
105   case T_Null:
106   case T_Boolean:
107   case T_Double:
108   case T_Integer:
109     memcpy(Union.buffer, M.Union.buffer, sizeof(Union.buffer));
110     break;
111   case T_StringRef:
112     create<StringRef>(M.as<StringRef>());
113     break;
114   case T_String:
115     create<std::string>(M.as<std::string>());
116     break;
117   case T_Object:
118     create<json::Object>(M.as<json::Object>());
119     break;
120   case T_Array:
121     create<json::Array>(M.as<json::Array>());
122     break;
123   }
124 }
125 
126 void Value::moveFrom(const Value &&M) {
127   Type = M.Type;
128   switch (Type) {
129   case T_Null:
130   case T_Boolean:
131   case T_Double:
132   case T_Integer:
133     memcpy(Union.buffer, M.Union.buffer, sizeof(Union.buffer));
134     break;
135   case T_StringRef:
136     create<StringRef>(M.as<StringRef>());
137     break;
138   case T_String:
139     create<std::string>(std::move(M.as<std::string>()));
140     M.Type = T_Null;
141     break;
142   case T_Object:
143     create<json::Object>(std::move(M.as<json::Object>()));
144     M.Type = T_Null;
145     break;
146   case T_Array:
147     create<json::Array>(std::move(M.as<json::Array>()));
148     M.Type = T_Null;
149     break;
150   }
151 }
152 
153 void Value::destroy() {
154   switch (Type) {
155   case T_Null:
156   case T_Boolean:
157   case T_Double:
158   case T_Integer:
159     break;
160   case T_StringRef:
161     as<StringRef>().~StringRef();
162     break;
163   case T_String:
164     as<std::string>().~basic_string();
165     break;
166   case T_Object:
167     as<json::Object>().~Object();
168     break;
169   case T_Array:
170     as<json::Array>().~Array();
171     break;
172   }
173 }
174 
175 bool operator==(const Value &L, const Value &R) {
176   if (L.kind() != R.kind())
177     return false;
178   switch (L.kind()) {
179   case Value::Null:
180     return *L.getAsNull() == *R.getAsNull();
181   case Value::Boolean:
182     return *L.getAsBoolean() == *R.getAsBoolean();
183   case Value::Number:
184     return *L.getAsNumber() == *R.getAsNumber();
185   case Value::String:
186     return *L.getAsString() == *R.getAsString();
187   case Value::Array:
188     return *L.getAsArray() == *R.getAsArray();
189   case Value::Object:
190     return *L.getAsObject() == *R.getAsObject();
191   }
192   llvm_unreachable("Unknown value kind");
193 }
194 
195 namespace {
196 // Simple recursive-descent JSON parser.
197 class Parser {
198 public:
199   Parser(StringRef JSON)
200       : Start(JSON.begin()), P(JSON.begin()), End(JSON.end()) {}
201 
202   bool checkUTF8() {
203     size_t ErrOffset;
204     if (isUTF8(StringRef(Start, End - Start), &ErrOffset))
205       return true;
206     P = Start + ErrOffset; // For line/column calculation.
207     return parseError("Invalid UTF-8 sequence");
208   }
209 
210   bool parseValue(Value &Out);
211 
212   bool assertEnd() {
213     eatWhitespace();
214     if (P == End)
215       return true;
216     return parseError("Text after end of document");
217   }
218 
219   Error takeError() {
220     assert(Err);
221     return std::move(*Err);
222   }
223 
224 private:
225   void eatWhitespace() {
226     while (P != End && (*P == ' ' || *P == '\r' || *P == '\n' || *P == '\t'))
227       ++P;
228   }
229 
230   // On invalid syntax, parseX() functions return false and set Err.
231   bool parseNumber(char First, Value &Out);
232   bool parseString(std::string &Out);
233   bool parseUnicode(std::string &Out);
234   bool parseError(const char *Msg); // always returns false
235 
236   char next() { return P == End ? 0 : *P++; }
237   char peek() { return P == End ? 0 : *P; }
238   static bool isNumber(char C) {
239     return C == '0' || C == '1' || C == '2' || C == '3' || C == '4' ||
240            C == '5' || C == '6' || C == '7' || C == '8' || C == '9' ||
241            C == 'e' || C == 'E' || C == '+' || C == '-' || C == '.';
242   }
243 
244   Optional<Error> Err;
245   const char *Start, *P, *End;
246 };
247 
248 bool Parser::parseValue(Value &Out) {
249   eatWhitespace();
250   if (P == End)
251     return parseError("Unexpected EOF");
252   switch (char C = next()) {
253   // Bare null/true/false are easy - first char identifies them.
254   case 'n':
255     Out = nullptr;
256     return (next() == 'u' && next() == 'l' && next() == 'l') ||
257            parseError("Invalid JSON value (null?)");
258   case 't':
259     Out = true;
260     return (next() == 'r' && next() == 'u' && next() == 'e') ||
261            parseError("Invalid JSON value (true?)");
262   case 'f':
263     Out = false;
264     return (next() == 'a' && next() == 'l' && next() == 's' && next() == 'e') ||
265            parseError("Invalid JSON value (false?)");
266   case '"': {
267     std::string S;
268     if (parseString(S)) {
269       Out = std::move(S);
270       return true;
271     }
272     return false;
273   }
274   case '[': {
275     Out = Array{};
276     Array &A = *Out.getAsArray();
277     eatWhitespace();
278     if (peek() == ']') {
279       ++P;
280       return true;
281     }
282     for (;;) {
283       A.emplace_back(nullptr);
284       if (!parseValue(A.back()))
285         return false;
286       eatWhitespace();
287       switch (next()) {
288       case ',':
289         eatWhitespace();
290         continue;
291       case ']':
292         return true;
293       default:
294         return parseError("Expected , or ] after array element");
295       }
296     }
297   }
298   case '{': {
299     Out = Object{};
300     Object &O = *Out.getAsObject();
301     eatWhitespace();
302     if (peek() == '}') {
303       ++P;
304       return true;
305     }
306     for (;;) {
307       if (next() != '"')
308         return parseError("Expected object key");
309       std::string K;
310       if (!parseString(K))
311         return false;
312       eatWhitespace();
313       if (next() != ':')
314         return parseError("Expected : after object key");
315       eatWhitespace();
316       if (!parseValue(O[std::move(K)]))
317         return false;
318       eatWhitespace();
319       switch (next()) {
320       case ',':
321         eatWhitespace();
322         continue;
323       case '}':
324         return true;
325       default:
326         return parseError("Expected , or } after object property");
327       }
328     }
329   }
330   default:
331     if (isNumber(C))
332       return parseNumber(C, Out);
333     return parseError("Invalid JSON value");
334   }
335 }
336 
337 bool Parser::parseNumber(char First, Value &Out) {
338   // Read the number into a string. (Must be null-terminated for strto*).
339   SmallString<24> S;
340   S.push_back(First);
341   while (isNumber(peek()))
342     S.push_back(next());
343   char *End;
344   // Try first to parse as integer, and if so preserve full 64 bits.
345   // strtoll returns long long >= 64 bits, so check it's in range too.
346   auto I = std::strtoll(S.c_str(), &End, 10);
347   if (End == S.end() && I >= std::numeric_limits<int64_t>::min() &&
348       I <= std::numeric_limits<int64_t>::max()) {
349     Out = int64_t(I);
350     return true;
351   }
352   // If it's not an integer
353   Out = std::strtod(S.c_str(), &End);
354   return End == S.end() || parseError("Invalid JSON value (number?)");
355 }
356 
357 bool Parser::parseString(std::string &Out) {
358   // leading quote was already consumed.
359   for (char C = next(); C != '"'; C = next()) {
360     if (LLVM_UNLIKELY(P == End))
361       return parseError("Unterminated string");
362     if (LLVM_UNLIKELY((C & 0x1f) == C))
363       return parseError("Control character in string");
364     if (LLVM_LIKELY(C != '\\')) {
365       Out.push_back(C);
366       continue;
367     }
368     // Handle escape sequence.
369     switch (C = next()) {
370     case '"':
371     case '\\':
372     case '/':
373       Out.push_back(C);
374       break;
375     case 'b':
376       Out.push_back('\b');
377       break;
378     case 'f':
379       Out.push_back('\f');
380       break;
381     case 'n':
382       Out.push_back('\n');
383       break;
384     case 'r':
385       Out.push_back('\r');
386       break;
387     case 't':
388       Out.push_back('\t');
389       break;
390     case 'u':
391       if (!parseUnicode(Out))
392         return false;
393       break;
394     default:
395       return parseError("Invalid escape sequence");
396     }
397   }
398   return true;
399 }
400 
401 static void encodeUtf8(uint32_t Rune, std::string &Out) {
402   if (Rune < 0x80) {
403     Out.push_back(Rune & 0x7F);
404   } else if (Rune < 0x800) {
405     uint8_t FirstByte = 0xC0 | ((Rune & 0x7C0) >> 6);
406     uint8_t SecondByte = 0x80 | (Rune & 0x3F);
407     Out.push_back(FirstByte);
408     Out.push_back(SecondByte);
409   } else if (Rune < 0x10000) {
410     uint8_t FirstByte = 0xE0 | ((Rune & 0xF000) >> 12);
411     uint8_t SecondByte = 0x80 | ((Rune & 0xFC0) >> 6);
412     uint8_t ThirdByte = 0x80 | (Rune & 0x3F);
413     Out.push_back(FirstByte);
414     Out.push_back(SecondByte);
415     Out.push_back(ThirdByte);
416   } else if (Rune < 0x110000) {
417     uint8_t FirstByte = 0xF0 | ((Rune & 0x1F0000) >> 18);
418     uint8_t SecondByte = 0x80 | ((Rune & 0x3F000) >> 12);
419     uint8_t ThirdByte = 0x80 | ((Rune & 0xFC0) >> 6);
420     uint8_t FourthByte = 0x80 | (Rune & 0x3F);
421     Out.push_back(FirstByte);
422     Out.push_back(SecondByte);
423     Out.push_back(ThirdByte);
424     Out.push_back(FourthByte);
425   } else {
426     llvm_unreachable("Invalid codepoint");
427   }
428 }
429 
430 // Parse a UTF-16 \uNNNN escape sequence. "\u" has already been consumed.
431 // May parse several sequential escapes to ensure proper surrogate handling.
432 // We do not use ConvertUTF.h, it can't accept and replace unpaired surrogates.
433 // These are invalid Unicode but valid JSON (RFC 8259, section 8.2).
434 bool Parser::parseUnicode(std::string &Out) {
435   // Invalid UTF is not a JSON error (RFC 8529§8.2). It gets replaced by U+FFFD.
436   auto Invalid = [&] { Out.append(/* UTF-8 */ {'\xef', '\xbf', '\xbd'}); };
437   // Decodes 4 hex digits from the stream into Out, returns false on error.
438   auto Parse4Hex = [this](uint16_t &Out) -> bool {
439     Out = 0;
440     char Bytes[] = {next(), next(), next(), next()};
441     for (unsigned char C : Bytes) {
442       if (!std::isxdigit(C))
443         return parseError("Invalid \\u escape sequence");
444       Out <<= 4;
445       Out |= (C > '9') ? (C & ~0x20) - 'A' + 10 : (C - '0');
446     }
447     return true;
448   };
449   uint16_t First; // UTF-16 code unit from the first \u escape.
450   if (!Parse4Hex(First))
451     return false;
452 
453   // We loop to allow proper surrogate-pair error handling.
454   while (true) {
455     // Case 1: the UTF-16 code unit is already a codepoint in the BMP.
456     if (LLVM_LIKELY(First < 0xD800 || First >= 0xE000)) {
457       encodeUtf8(First, Out);
458       return true;
459     }
460 
461     // Case 2: it's an (unpaired) trailing surrogate.
462     if (LLVM_UNLIKELY(First >= 0xDC00)) {
463       Invalid();
464       return true;
465     }
466 
467     // Case 3: it's a leading surrogate. We expect a trailing one next.
468     // Case 3a: there's no trailing \u escape. Don't advance in the stream.
469     if (LLVM_UNLIKELY(P + 2 > End || *P != '\\' || *(P + 1) != 'u')) {
470       Invalid(); // Leading surrogate was unpaired.
471       return true;
472     }
473     P += 2;
474     uint16_t Second;
475     if (!Parse4Hex(Second))
476       return false;
477     // Case 3b: there was another \u escape, but it wasn't a trailing surrogate.
478     if (LLVM_UNLIKELY(Second < 0xDC00 || Second >= 0xE000)) {
479       Invalid();      // Leading surrogate was unpaired.
480       First = Second; // Second escape still needs to be processed.
481       continue;
482     }
483     // Case 3c: a valid surrogate pair encoding an astral codepoint.
484     encodeUtf8(0x10000 | ((First - 0xD800) << 10) | (Second - 0xDC00), Out);
485     return true;
486   }
487 }
488 
489 bool Parser::parseError(const char *Msg) {
490   int Line = 1;
491   const char *StartOfLine = Start;
492   for (const char *X = Start; X < P; ++X) {
493     if (*X == 0x0A) {
494       ++Line;
495       StartOfLine = X + 1;
496     }
497   }
498   Err.emplace(
499       llvm::make_unique<ParseError>(Msg, Line, P - StartOfLine, P - Start));
500   return false;
501 }
502 } // namespace
503 
504 Expected<Value> parse(StringRef JSON) {
505   Parser P(JSON);
506   Value E = nullptr;
507   if (P.checkUTF8())
508     if (P.parseValue(E))
509       if (P.assertEnd())
510         return std::move(E);
511   return P.takeError();
512 }
513 char ParseError::ID = 0;
514 
515 static std::vector<const Object::value_type *> sortedElements(const Object &O) {
516   std::vector<const Object::value_type *> Elements;
517   for (const auto &E : O)
518     Elements.push_back(&E);
519   llvm::sort(Elements,
520              [](const Object::value_type *L, const Object::value_type *R) {
521                return L->first < R->first;
522              });
523   return Elements;
524 }
525 
526 bool isUTF8(llvm::StringRef S, size_t *ErrOffset) {
527   // Fast-path for ASCII, which is valid UTF-8.
528   if (LLVM_LIKELY(isASCII(S)))
529     return true;
530 
531   const UTF8 *Data = reinterpret_cast<const UTF8 *>(S.data()), *Rest = Data;
532   if (LLVM_LIKELY(isLegalUTF8String(&Rest, Data + S.size())))
533     return true;
534 
535   if (ErrOffset)
536     *ErrOffset = Rest - Data;
537   return false;
538 }
539 
540 std::string fixUTF8(llvm::StringRef S) {
541   // This isn't particularly efficient, but is only for error-recovery.
542   std::vector<UTF32> Codepoints(S.size()); // 1 codepoint per byte suffices.
543   const UTF8 *In8 = reinterpret_cast<const UTF8 *>(S.data());
544   UTF32 *Out32 = Codepoints.data();
545   ConvertUTF8toUTF32(&In8, In8 + S.size(), &Out32, Out32 + Codepoints.size(),
546                      lenientConversion);
547   Codepoints.resize(Out32 - Codepoints.data());
548   std::string Res(4 * Codepoints.size(), 0); // 4 bytes per codepoint suffice
549   const UTF32 *In32 = Codepoints.data();
550   UTF8 *Out8 = reinterpret_cast<UTF8 *>(&Res[0]);
551   ConvertUTF32toUTF8(&In32, In32 + Codepoints.size(), &Out8, Out8 + Res.size(),
552                      strictConversion);
553   Res.resize(reinterpret_cast<char *>(Out8) - Res.data());
554   return Res;
555 }
556 
557 } // namespace json
558 } // namespace llvm
559 
560 static void quote(llvm::raw_ostream &OS, llvm::StringRef S) {
561   OS << '\"';
562   for (unsigned char C : S) {
563     if (C == 0x22 || C == 0x5C)
564       OS << '\\';
565     if (C >= 0x20) {
566       OS << C;
567       continue;
568     }
569     OS << '\\';
570     switch (C) {
571     // A few characters are common enough to make short escapes worthwhile.
572     case '\t':
573       OS << 't';
574       break;
575     case '\n':
576       OS << 'n';
577       break;
578     case '\r':
579       OS << 'r';
580       break;
581     default:
582       OS << 'u';
583       llvm::write_hex(OS, C, llvm::HexPrintStyle::Lower, 4);
584       break;
585     }
586   }
587   OS << '\"';
588 }
589 
590 enum IndenterAction {
591   Indent,
592   Outdent,
593   Newline,
594   Space,
595 };
596 
597 // Prints JSON. The indenter can be used to control formatting.
598 template <typename Indenter>
599 void llvm::json::Value::print(raw_ostream &OS, const Indenter &I) const {
600   switch (Type) {
601   case T_Null:
602     OS << "null";
603     break;
604   case T_Boolean:
605     OS << (as<bool>() ? "true" : "false");
606     break;
607   case T_Double:
608     OS << format("%.*g", std::numeric_limits<double>::max_digits10,
609                  as<double>());
610     break;
611   case T_Integer:
612     OS << as<int64_t>();
613     break;
614   case T_StringRef:
615     quote(OS, as<StringRef>());
616     break;
617   case T_String:
618     quote(OS, as<std::string>());
619     break;
620   case T_Object: {
621     bool Comma = false;
622     OS << '{';
623     I(Indent);
624     for (const auto *P : sortedElements(as<json::Object>())) {
625       if (Comma)
626         OS << ',';
627       Comma = true;
628       I(Newline);
629       quote(OS, P->first);
630       OS << ':';
631       I(Space);
632       P->second.print(OS, I);
633     }
634     I(Outdent);
635     if (Comma)
636       I(Newline);
637     OS << '}';
638     break;
639   }
640   case T_Array: {
641     bool Comma = false;
642     OS << '[';
643     I(Indent);
644     for (const auto &E : as<json::Array>()) {
645       if (Comma)
646         OS << ',';
647       Comma = true;
648       I(Newline);
649       E.print(OS, I);
650     }
651     I(Outdent);
652     if (Comma)
653       I(Newline);
654     OS << ']';
655     break;
656   }
657   }
658 }
659 
660 void llvm::format_provider<llvm::json::Value>::format(
661     const llvm::json::Value &E, raw_ostream &OS, StringRef Options) {
662   if (Options.empty()) {
663     OS << E;
664     return;
665   }
666   unsigned IndentAmount = 0;
667   if (Options.getAsInteger(/*Radix=*/10, IndentAmount))
668     llvm_unreachable("json::Value format options should be an integer");
669   unsigned IndentLevel = 0;
670   E.print(OS, [&](IndenterAction A) {
671     switch (A) {
672     case Newline:
673       OS << '\n';
674       OS.indent(IndentLevel);
675       break;
676     case Space:
677       OS << ' ';
678       break;
679     case Indent:
680       IndentLevel += IndentAmount;
681       break;
682     case Outdent:
683       IndentLevel -= IndentAmount;
684       break;
685     };
686   });
687 }
688 
689 llvm::raw_ostream &llvm::json::operator<<(raw_ostream &OS, const Value &E) {
690   E.print(OS, [](IndenterAction A) { /*ignore*/ });
691   return OS;
692 }
693