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/ADT/STLExtras.h" |
11 | #include "llvm/ADT/StringExtras.h" |
12 | #include "llvm/Support/ConvertUTF.h" |
13 | #include "llvm/Support/Error.h" |
14 | #include "llvm/Support/Format.h" |
15 | #include "llvm/Support/NativeFormatting.h" |
16 | #include "llvm/Support/raw_ostream.h" |
17 | #include <cctype> |
18 | #include <cerrno> |
19 | #include <optional> |
20 | |
21 | namespace llvm { |
22 | namespace json { |
23 | |
24 | Value &Object::operator[](const ObjectKey &K) { |
25 | return try_emplace(K, Args: nullptr).first->getSecond(); |
26 | } |
27 | Value &Object::operator[](ObjectKey &&K) { |
28 | return try_emplace(K: std::move(K), Args: nullptr).first->getSecond(); |
29 | } |
30 | Value *Object::get(StringRef K) { |
31 | auto I = find(K); |
32 | if (I == end()) |
33 | return nullptr; |
34 | return &I->second; |
35 | } |
36 | const Value *Object::get(StringRef K) const { |
37 | auto I = find(K); |
38 | if (I == end()) |
39 | return nullptr; |
40 | return &I->second; |
41 | } |
42 | std::optional<std::nullptr_t> Object::getNull(StringRef K) const { |
43 | if (auto *V = get(K)) |
44 | return V->getAsNull(); |
45 | return std::nullopt; |
46 | } |
47 | std::optional<bool> Object::getBoolean(StringRef K) const { |
48 | if (auto *V = get(K)) |
49 | return V->getAsBoolean(); |
50 | return std::nullopt; |
51 | } |
52 | std::optional<double> Object::getNumber(StringRef K) const { |
53 | if (auto *V = get(K)) |
54 | return V->getAsNumber(); |
55 | return std::nullopt; |
56 | } |
57 | std::optional<int64_t> Object::getInteger(StringRef K) const { |
58 | if (auto *V = get(K)) |
59 | return V->getAsInteger(); |
60 | return std::nullopt; |
61 | } |
62 | std::optional<llvm::StringRef> Object::getString(StringRef K) const { |
63 | if (auto *V = get(K)) |
64 | return V->getAsString(); |
65 | return std::nullopt; |
66 | } |
67 | const json::Object *Object::getObject(StringRef K) const { |
68 | if (auto *V = get(K)) |
69 | return V->getAsObject(); |
70 | return nullptr; |
71 | } |
72 | json::Object *Object::getObject(StringRef K) { |
73 | if (auto *V = get(K)) |
74 | return V->getAsObject(); |
75 | return nullptr; |
76 | } |
77 | const json::Array *Object::getArray(StringRef K) const { |
78 | if (auto *V = get(K)) |
79 | return V->getAsArray(); |
80 | return nullptr; |
81 | } |
82 | json::Array *Object::getArray(StringRef K) { |
83 | if (auto *V = get(K)) |
84 | return V->getAsArray(); |
85 | return nullptr; |
86 | } |
87 | bool operator==(const Object &LHS, const Object &RHS) { |
88 | if (LHS.size() != RHS.size()) |
89 | return false; |
90 | for (const auto &L : LHS) { |
91 | auto R = RHS.find(K: L.first); |
92 | if (R == RHS.end() || L.second != R->second) |
93 | return false; |
94 | } |
95 | return true; |
96 | } |
97 | |
98 | Array::Array(std::initializer_list<Value> Elements) { |
99 | V.reserve(n: Elements.size()); |
100 | for (const Value &V : Elements) { |
101 | emplace_back(A: nullptr); |
102 | back().moveFrom(M: std::move(V)); |
103 | } |
104 | } |
105 | |
106 | Value::Value(std::initializer_list<Value> Elements) |
107 | : Value(json::Array(Elements)) {} |
108 | |
109 | void Value::copyFrom(const Value &M) { |
110 | Type = M.Type; |
111 | switch (Type) { |
112 | case T_Null: |
113 | case T_Boolean: |
114 | case T_Double: |
115 | case T_Integer: |
116 | case T_UINT64: |
117 | memcpy(dest: &Union, src: &M.Union, n: sizeof(Union)); |
118 | break; |
119 | case T_StringRef: |
120 | create<StringRef>(V&: M.as<StringRef>()); |
121 | break; |
122 | case T_String: |
123 | create<std::string>(V&: M.as<std::string>()); |
124 | break; |
125 | case T_Object: |
126 | create<json::Object>(V&: M.as<json::Object>()); |
127 | break; |
128 | case T_Array: |
129 | create<json::Array>(V&: M.as<json::Array>()); |
130 | break; |
131 | } |
132 | } |
133 | |
134 | void Value::moveFrom(const Value &&M) { |
135 | Type = M.Type; |
136 | switch (Type) { |
137 | case T_Null: |
138 | case T_Boolean: |
139 | case T_Double: |
140 | case T_Integer: |
141 | case T_UINT64: |
142 | memcpy(dest: &Union, src: &M.Union, n: sizeof(Union)); |
143 | break; |
144 | case T_StringRef: |
145 | create<StringRef>(V&: M.as<StringRef>()); |
146 | break; |
147 | case T_String: |
148 | create<std::string>(V: std::move(M.as<std::string>())); |
149 | M.Type = T_Null; |
150 | break; |
151 | case T_Object: |
152 | create<json::Object>(V: std::move(M.as<json::Object>())); |
153 | M.Type = T_Null; |
154 | break; |
155 | case T_Array: |
156 | create<json::Array>(V: std::move(M.as<json::Array>())); |
157 | M.Type = T_Null; |
158 | break; |
159 | } |
160 | } |
161 | |
162 | void Value::destroy() { |
163 | switch (Type) { |
164 | case T_Null: |
165 | case T_Boolean: |
166 | case T_Double: |
167 | case T_Integer: |
168 | case T_UINT64: |
169 | break; |
170 | case T_StringRef: |
171 | as<StringRef>().~StringRef(); |
172 | break; |
173 | case T_String: |
174 | as<std::string>().~basic_string(); |
175 | break; |
176 | case T_Object: |
177 | as<json::Object>().~Object(); |
178 | break; |
179 | case T_Array: |
180 | as<json::Array>().~Array(); |
181 | break; |
182 | } |
183 | } |
184 | |
185 | bool operator==(const Value &L, const Value &R) { |
186 | if (L.kind() != R.kind()) |
187 | return false; |
188 | switch (L.kind()) { |
189 | case Value::Null: |
190 | return *L.getAsNull() == *R.getAsNull(); |
191 | case Value::Boolean: |
192 | return *L.getAsBoolean() == *R.getAsBoolean(); |
193 | case Value::Number: |
194 | // Workaround for https://gcc.gnu.org/bugzilla/show_bug.cgi?id=323 |
195 | // The same integer must convert to the same double, per the standard. |
196 | // However we see 64-vs-80-bit precision comparisons with gcc-7 -O3 -m32. |
197 | // So we avoid floating point promotion for exact comparisons. |
198 | if (L.Type == Value::T_Integer || R.Type == Value::T_Integer) |
199 | return L.getAsInteger() == R.getAsInteger(); |
200 | return *L.getAsNumber() == *R.getAsNumber(); |
201 | case Value::String: |
202 | return *L.getAsString() == *R.getAsString(); |
203 | case Value::Array: |
204 | return *L.getAsArray() == *R.getAsArray(); |
205 | case Value::Object: |
206 | return *L.getAsObject() == *R.getAsObject(); |
207 | } |
208 | llvm_unreachable("Unknown value kind" ); |
209 | } |
210 | |
211 | void Path::report(llvm::StringLiteral Msg) { |
212 | // Walk up to the root context, and count the number of segments. |
213 | unsigned Count = 0; |
214 | const Path *P; |
215 | for (P = this; P->Parent != nullptr; P = P->Parent) |
216 | ++Count; |
217 | Path::Root *R = P->Seg.root(); |
218 | // Fill in the error message and copy the path (in reverse order). |
219 | R->ErrorMessage = Msg; |
220 | R->ErrorPath.resize(new_size: Count); |
221 | auto It = R->ErrorPath.begin(); |
222 | for (P = this; P->Parent != nullptr; P = P->Parent) |
223 | *It++ = P->Seg; |
224 | } |
225 | |
226 | Error Path::Root::getError() const { |
227 | std::string S; |
228 | raw_string_ostream OS(S); |
229 | OS << (ErrorMessage.empty() ? "invalid JSON contents" : ErrorMessage); |
230 | if (ErrorPath.empty()) { |
231 | if (!Name.empty()) |
232 | OS << " when parsing " << Name; |
233 | } else { |
234 | OS << " at " << (Name.empty() ? "(root)" : Name); |
235 | for (const Path::Segment &S : llvm::reverse(C: ErrorPath)) { |
236 | if (S.isField()) |
237 | OS << '.' << S.field(); |
238 | else |
239 | OS << '[' << S.index() << ']'; |
240 | } |
241 | } |
242 | return createStringError(EC: llvm::inconvertibleErrorCode(), S); |
243 | } |
244 | |
245 | std::vector<const Object::value_type *> sortedElements(const Object &O) { |
246 | std::vector<const Object::value_type *> Elements; |
247 | for (const auto &E : O) |
248 | Elements.push_back(x: &E); |
249 | llvm::sort(C&: Elements, |
250 | Comp: [](const Object::value_type *L, const Object::value_type *R) { |
251 | return L->first < R->first; |
252 | }); |
253 | return Elements; |
254 | } |
255 | |
256 | // Prints a one-line version of a value that isn't our main focus. |
257 | // We interleave writes to OS and JOS, exploiting the lack of extra buffering. |
258 | // This is OK as we own the implementation. |
259 | static void abbreviate(const Value &V, OStream &JOS) { |
260 | switch (V.kind()) { |
261 | case Value::Array: |
262 | JOS.rawValue(Contents: V.getAsArray()->empty() ? "[]" : "[ ... ]" ); |
263 | break; |
264 | case Value::Object: |
265 | JOS.rawValue(Contents: V.getAsObject()->empty() ? "{}" : "{ ... }" ); |
266 | break; |
267 | case Value::String: { |
268 | llvm::StringRef S = *V.getAsString(); |
269 | if (S.size() < 40) { |
270 | JOS.value(V); |
271 | } else { |
272 | std::string Truncated = fixUTF8(S: S.take_front(N: 37)); |
273 | Truncated.append(s: "..." ); |
274 | JOS.value(V: Truncated); |
275 | } |
276 | break; |
277 | } |
278 | default: |
279 | JOS.value(V); |
280 | } |
281 | } |
282 | |
283 | // Prints a semi-expanded version of a value that is our main focus. |
284 | // Array/Object entries are printed, but not recursively as they may be huge. |
285 | static void abbreviateChildren(const Value &V, OStream &JOS) { |
286 | switch (V.kind()) { |
287 | case Value::Array: |
288 | JOS.array(Contents: [&] { |
289 | for (const auto &I : *V.getAsArray()) |
290 | abbreviate(V: I, JOS); |
291 | }); |
292 | break; |
293 | case Value::Object: |
294 | JOS.object(Contents: [&] { |
295 | for (const auto *KV : sortedElements(O: *V.getAsObject())) { |
296 | JOS.attributeBegin(Key: KV->first); |
297 | abbreviate(V: KV->second, JOS); |
298 | JOS.attributeEnd(); |
299 | } |
300 | }); |
301 | break; |
302 | default: |
303 | JOS.value(V); |
304 | } |
305 | } |
306 | |
307 | void Path::Root::printErrorContext(const Value &R, raw_ostream &OS) const { |
308 | OStream JOS(OS, /*IndentSize=*/2); |
309 | // PrintValue recurses down the path, printing the ancestors of our target. |
310 | // Siblings of nodes along the path are printed with abbreviate(), and the |
311 | // target itself is printed with the somewhat richer abbreviateChildren(). |
312 | // 'Recurse' is the lambda itself, to allow recursive calls. |
313 | auto PrintValue = [&](const Value &V, ArrayRef<Segment> Path, auto &Recurse) { |
314 | // Print the target node itself, with the error as a comment. |
315 | // Also used if we can't follow our path, e.g. it names a field that |
316 | // *should* exist but doesn't. |
317 | auto HighlightCurrent = [&] { |
318 | std::string = "error: " ; |
319 | Comment.append(s: ErrorMessage.data(), n: ErrorMessage.size()); |
320 | JOS.comment(Comment); |
321 | abbreviateChildren(V, JOS); |
322 | }; |
323 | if (Path.empty()) // We reached our target. |
324 | return HighlightCurrent(); |
325 | const Segment &S = Path.back(); // Path is in reverse order. |
326 | if (S.isField()) { |
327 | // Current node is an object, path names a field. |
328 | llvm::StringRef FieldName = S.field(); |
329 | const Object *O = V.getAsObject(); |
330 | if (!O || !O->get(K: FieldName)) |
331 | return HighlightCurrent(); |
332 | JOS.object(Contents: [&] { |
333 | for (const auto *KV : sortedElements(O: *O)) { |
334 | JOS.attributeBegin(Key: KV->first); |
335 | if (FieldName == StringRef(KV->first)) |
336 | Recurse(KV->second, Path.drop_back(), Recurse); |
337 | else |
338 | abbreviate(V: KV->second, JOS); |
339 | JOS.attributeEnd(); |
340 | } |
341 | }); |
342 | } else { |
343 | // Current node is an array, path names an element. |
344 | const Array *A = V.getAsArray(); |
345 | if (!A || S.index() >= A->size()) |
346 | return HighlightCurrent(); |
347 | JOS.array(Contents: [&] { |
348 | unsigned Current = 0; |
349 | for (const auto &V : *A) { |
350 | if (Current++ == S.index()) |
351 | Recurse(V, Path.drop_back(), Recurse); |
352 | else |
353 | abbreviate(V, JOS); |
354 | } |
355 | }); |
356 | } |
357 | }; |
358 | PrintValue(R, ErrorPath, PrintValue); |
359 | } |
360 | |
361 | namespace { |
362 | // Simple recursive-descent JSON parser. |
363 | class Parser { |
364 | public: |
365 | Parser(StringRef JSON) |
366 | : Start(JSON.begin()), P(JSON.begin()), End(JSON.end()) {} |
367 | |
368 | bool checkUTF8() { |
369 | size_t ErrOffset; |
370 | if (isUTF8(S: StringRef(Start, End - Start), ErrOffset: &ErrOffset)) |
371 | return true; |
372 | P = Start + ErrOffset; // For line/column calculation. |
373 | return parseError(Msg: "Invalid UTF-8 sequence" ); |
374 | } |
375 | |
376 | bool parseValue(Value &Out); |
377 | |
378 | bool assertEnd() { |
379 | eatWhitespace(); |
380 | if (P == End) |
381 | return true; |
382 | return parseError(Msg: "Text after end of document" ); |
383 | } |
384 | |
385 | Error takeError() { |
386 | assert(Err); |
387 | return std::move(*Err); |
388 | } |
389 | |
390 | private: |
391 | void eatWhitespace() { |
392 | while (P != End && (*P == ' ' || *P == '\r' || *P == '\n' || *P == '\t')) |
393 | ++P; |
394 | } |
395 | |
396 | // On invalid syntax, parseX() functions return false and set Err. |
397 | bool parseNumber(char First, Value &Out); |
398 | bool parseString(std::string &Out); |
399 | bool parseUnicode(std::string &Out); |
400 | bool parseError(const char *Msg); // always returns false |
401 | |
402 | char next() { return P == End ? 0 : *P++; } |
403 | char peek() { return P == End ? 0 : *P; } |
404 | static bool isNumber(char C) { |
405 | return C == '0' || C == '1' || C == '2' || C == '3' || C == '4' || |
406 | C == '5' || C == '6' || C == '7' || C == '8' || C == '9' || |
407 | C == 'e' || C == 'E' || C == '+' || C == '-' || C == '.'; |
408 | } |
409 | |
410 | std::optional<Error> Err; |
411 | const char *Start, *P, *End; |
412 | }; |
413 | } // namespace |
414 | |
415 | bool Parser::parseValue(Value &Out) { |
416 | eatWhitespace(); |
417 | if (P == End) |
418 | return parseError(Msg: "Unexpected EOF" ); |
419 | switch (char C = next()) { |
420 | // Bare null/true/false are easy - first char identifies them. |
421 | case 'n': |
422 | Out = nullptr; |
423 | return (next() == 'u' && next() == 'l' && next() == 'l') || |
424 | parseError(Msg: "Invalid JSON value (null?)" ); |
425 | case 't': |
426 | Out = true; |
427 | return (next() == 'r' && next() == 'u' && next() == 'e') || |
428 | parseError(Msg: "Invalid JSON value (true?)" ); |
429 | case 'f': |
430 | Out = false; |
431 | return (next() == 'a' && next() == 'l' && next() == 's' && next() == 'e') || |
432 | parseError(Msg: "Invalid JSON value (false?)" ); |
433 | case '"': { |
434 | std::string S; |
435 | if (parseString(Out&: S)) { |
436 | Out = std::move(S); |
437 | return true; |
438 | } |
439 | return false; |
440 | } |
441 | case '[': { |
442 | Out = Array{}; |
443 | Array &A = *Out.getAsArray(); |
444 | eatWhitespace(); |
445 | if (peek() == ']') { |
446 | ++P; |
447 | return true; |
448 | } |
449 | for (;;) { |
450 | A.emplace_back(A: nullptr); |
451 | if (!parseValue(Out&: A.back())) |
452 | return false; |
453 | eatWhitespace(); |
454 | switch (next()) { |
455 | case ',': |
456 | eatWhitespace(); |
457 | continue; |
458 | case ']': |
459 | return true; |
460 | default: |
461 | return parseError(Msg: "Expected , or ] after array element" ); |
462 | } |
463 | } |
464 | } |
465 | case '{': { |
466 | Out = Object{}; |
467 | Object &O = *Out.getAsObject(); |
468 | eatWhitespace(); |
469 | if (peek() == '}') { |
470 | ++P; |
471 | return true; |
472 | } |
473 | for (;;) { |
474 | if (next() != '"') |
475 | return parseError(Msg: "Expected object key" ); |
476 | std::string K; |
477 | if (!parseString(Out&: K)) |
478 | return false; |
479 | eatWhitespace(); |
480 | if (next() != ':') |
481 | return parseError(Msg: "Expected : after object key" ); |
482 | eatWhitespace(); |
483 | if (!parseValue(Out&: O[std::move(K)])) |
484 | return false; |
485 | eatWhitespace(); |
486 | switch (next()) { |
487 | case ',': |
488 | eatWhitespace(); |
489 | continue; |
490 | case '}': |
491 | return true; |
492 | default: |
493 | return parseError(Msg: "Expected , or } after object property" ); |
494 | } |
495 | } |
496 | } |
497 | default: |
498 | if (isNumber(C)) |
499 | return parseNumber(First: C, Out); |
500 | return parseError(Msg: "Invalid JSON value" ); |
501 | } |
502 | } |
503 | |
504 | bool Parser::parseNumber(char First, Value &Out) { |
505 | // Read the number into a string. (Must be null-terminated for strto*). |
506 | SmallString<24> S; |
507 | S.push_back(Elt: First); |
508 | while (isNumber(C: peek())) |
509 | S.push_back(Elt: next()); |
510 | char *End; |
511 | // Try first to parse as integer, and if so preserve full 64 bits. |
512 | // We check for errno for out of bounds errors and for End == S.end() |
513 | // to make sure that the numeric string is not malformed. |
514 | errno = 0; |
515 | int64_t I = std::strtoll(nptr: S.c_str(), endptr: &End, base: 10); |
516 | if (End == S.end() && errno != ERANGE) { |
517 | Out = int64_t(I); |
518 | return true; |
519 | } |
520 | // strtroull has a special handling for negative numbers, but in this |
521 | // case we don't want to do that because negative numbers were already |
522 | // handled in the previous block. |
523 | if (First != '-') { |
524 | errno = 0; |
525 | uint64_t UI = std::strtoull(nptr: S.c_str(), endptr: &End, base: 10); |
526 | if (End == S.end() && errno != ERANGE) { |
527 | Out = UI; |
528 | return true; |
529 | } |
530 | } |
531 | // If it's not an integer |
532 | Out = std::strtod(nptr: S.c_str(), endptr: &End); |
533 | return End == S.end() || parseError(Msg: "Invalid JSON value (number?)" ); |
534 | } |
535 | |
536 | bool Parser::parseString(std::string &Out) { |
537 | // leading quote was already consumed. |
538 | for (char C = next(); C != '"'; C = next()) { |
539 | if (LLVM_UNLIKELY(P == End)) |
540 | return parseError(Msg: "Unterminated string" ); |
541 | if (LLVM_UNLIKELY((C & 0x1f) == C)) |
542 | return parseError(Msg: "Control character in string" ); |
543 | if (LLVM_LIKELY(C != '\\')) { |
544 | Out.push_back(c: C); |
545 | continue; |
546 | } |
547 | // Handle escape sequence. |
548 | switch (C = next()) { |
549 | case '"': |
550 | case '\\': |
551 | case '/': |
552 | Out.push_back(c: C); |
553 | break; |
554 | case 'b': |
555 | Out.push_back(c: '\b'); |
556 | break; |
557 | case 'f': |
558 | Out.push_back(c: '\f'); |
559 | break; |
560 | case 'n': |
561 | Out.push_back(c: '\n'); |
562 | break; |
563 | case 'r': |
564 | Out.push_back(c: '\r'); |
565 | break; |
566 | case 't': |
567 | Out.push_back(c: '\t'); |
568 | break; |
569 | case 'u': |
570 | if (!parseUnicode(Out)) |
571 | return false; |
572 | break; |
573 | default: |
574 | return parseError(Msg: "Invalid escape sequence" ); |
575 | } |
576 | } |
577 | return true; |
578 | } |
579 | |
580 | static void encodeUtf8(uint32_t Rune, std::string &Out) { |
581 | if (Rune < 0x80) { |
582 | Out.push_back(c: Rune & 0x7F); |
583 | } else if (Rune < 0x800) { |
584 | uint8_t FirstByte = 0xC0 | ((Rune & 0x7C0) >> 6); |
585 | uint8_t SecondByte = 0x80 | (Rune & 0x3F); |
586 | Out.push_back(c: FirstByte); |
587 | Out.push_back(c: SecondByte); |
588 | } else if (Rune < 0x10000) { |
589 | uint8_t FirstByte = 0xE0 | ((Rune & 0xF000) >> 12); |
590 | uint8_t SecondByte = 0x80 | ((Rune & 0xFC0) >> 6); |
591 | uint8_t ThirdByte = 0x80 | (Rune & 0x3F); |
592 | Out.push_back(c: FirstByte); |
593 | Out.push_back(c: SecondByte); |
594 | Out.push_back(c: ThirdByte); |
595 | } else if (Rune < 0x110000) { |
596 | uint8_t FirstByte = 0xF0 | ((Rune & 0x1F0000) >> 18); |
597 | uint8_t SecondByte = 0x80 | ((Rune & 0x3F000) >> 12); |
598 | uint8_t ThirdByte = 0x80 | ((Rune & 0xFC0) >> 6); |
599 | uint8_t FourthByte = 0x80 | (Rune & 0x3F); |
600 | Out.push_back(c: FirstByte); |
601 | Out.push_back(c: SecondByte); |
602 | Out.push_back(c: ThirdByte); |
603 | Out.push_back(c: FourthByte); |
604 | } else { |
605 | llvm_unreachable("Invalid codepoint" ); |
606 | } |
607 | } |
608 | |
609 | // Parse a UTF-16 \uNNNN escape sequence. "\u" has already been consumed. |
610 | // May parse several sequential escapes to ensure proper surrogate handling. |
611 | // We do not use ConvertUTF.h, it can't accept and replace unpaired surrogates. |
612 | // These are invalid Unicode but valid JSON (RFC 8259, section 8.2). |
613 | bool Parser::parseUnicode(std::string &Out) { |
614 | // Invalid UTF is not a JSON error (RFC 8529§8.2). It gets replaced by U+FFFD. |
615 | auto Invalid = [&] { Out.append(/* UTF-8 */ l: {'\xef', '\xbf', '\xbd'}); }; |
616 | // Decodes 4 hex digits from the stream into Out, returns false on error. |
617 | auto Parse4Hex = [this](uint16_t &Out) -> bool { |
618 | Out = 0; |
619 | char Bytes[] = {next(), next(), next(), next()}; |
620 | for (unsigned char C : Bytes) { |
621 | if (!std::isxdigit(C)) |
622 | return parseError(Msg: "Invalid \\u escape sequence" ); |
623 | Out <<= 4; |
624 | Out |= (C > '9') ? (C & ~0x20) - 'A' + 10 : (C - '0'); |
625 | } |
626 | return true; |
627 | }; |
628 | uint16_t First; // UTF-16 code unit from the first \u escape. |
629 | if (!Parse4Hex(First)) |
630 | return false; |
631 | |
632 | // We loop to allow proper surrogate-pair error handling. |
633 | while (true) { |
634 | // Case 1: the UTF-16 code unit is already a codepoint in the BMP. |
635 | if (LLVM_LIKELY(First < 0xD800 || First >= 0xE000)) { |
636 | encodeUtf8(Rune: First, Out); |
637 | return true; |
638 | } |
639 | |
640 | // Case 2: it's an (unpaired) trailing surrogate. |
641 | if (LLVM_UNLIKELY(First >= 0xDC00)) { |
642 | Invalid(); |
643 | return true; |
644 | } |
645 | |
646 | // Case 3: it's a leading surrogate. We expect a trailing one next. |
647 | // Case 3a: there's no trailing \u escape. Don't advance in the stream. |
648 | if (LLVM_UNLIKELY(P + 2 > End || *P != '\\' || *(P + 1) != 'u')) { |
649 | Invalid(); // Leading surrogate was unpaired. |
650 | return true; |
651 | } |
652 | P += 2; |
653 | uint16_t Second; |
654 | if (!Parse4Hex(Second)) |
655 | return false; |
656 | // Case 3b: there was another \u escape, but it wasn't a trailing surrogate. |
657 | if (LLVM_UNLIKELY(Second < 0xDC00 || Second >= 0xE000)) { |
658 | Invalid(); // Leading surrogate was unpaired. |
659 | First = Second; // Second escape still needs to be processed. |
660 | continue; |
661 | } |
662 | // Case 3c: a valid surrogate pair encoding an astral codepoint. |
663 | encodeUtf8(Rune: 0x10000 | ((First - 0xD800) << 10) | (Second - 0xDC00), Out); |
664 | return true; |
665 | } |
666 | } |
667 | |
668 | bool Parser::parseError(const char *Msg) { |
669 | int Line = 1; |
670 | const char *StartOfLine = Start; |
671 | for (const char *X = Start; X < P; ++X) { |
672 | if (*X == 0x0A) { |
673 | ++Line; |
674 | StartOfLine = X + 1; |
675 | } |
676 | } |
677 | Err.emplace( |
678 | args: std::make_unique<ParseError>(args&: Msg, args&: Line, args: P - StartOfLine, args: P - Start)); |
679 | return false; |
680 | } |
681 | |
682 | Expected<Value> parse(StringRef JSON) { |
683 | Parser P(JSON); |
684 | Value E = nullptr; |
685 | if (P.checkUTF8()) |
686 | if (P.parseValue(Out&: E)) |
687 | if (P.assertEnd()) |
688 | return std::move(E); |
689 | return P.takeError(); |
690 | } |
691 | char ParseError::ID = 0; |
692 | |
693 | bool isUTF8(llvm::StringRef S, size_t *ErrOffset) { |
694 | // Fast-path for ASCII, which is valid UTF-8. |
695 | if (LLVM_LIKELY(isASCII(S))) |
696 | return true; |
697 | |
698 | const UTF8 *Data = reinterpret_cast<const UTF8 *>(S.data()), *Rest = Data; |
699 | if (LLVM_LIKELY(isLegalUTF8String(&Rest, Data + S.size()))) |
700 | return true; |
701 | |
702 | if (ErrOffset) |
703 | *ErrOffset = Rest - Data; |
704 | return false; |
705 | } |
706 | |
707 | std::string fixUTF8(llvm::StringRef S) { |
708 | // This isn't particularly efficient, but is only for error-recovery. |
709 | std::vector<UTF32> Codepoints(S.size()); // 1 codepoint per byte suffices. |
710 | const UTF8 *In8 = reinterpret_cast<const UTF8 *>(S.data()); |
711 | UTF32 *Out32 = Codepoints.data(); |
712 | ConvertUTF8toUTF32(sourceStart: &In8, sourceEnd: In8 + S.size(), targetStart: &Out32, targetEnd: Out32 + Codepoints.size(), |
713 | flags: lenientConversion); |
714 | Codepoints.resize(new_size: Out32 - Codepoints.data()); |
715 | std::string Res(4 * Codepoints.size(), 0); // 4 bytes per codepoint suffice |
716 | const UTF32 *In32 = Codepoints.data(); |
717 | UTF8 *Out8 = reinterpret_cast<UTF8 *>(&Res[0]); |
718 | ConvertUTF32toUTF8(sourceStart: &In32, sourceEnd: In32 + Codepoints.size(), targetStart: &Out8, targetEnd: Out8 + Res.size(), |
719 | flags: strictConversion); |
720 | Res.resize(n: reinterpret_cast<char *>(Out8) - Res.data()); |
721 | return Res; |
722 | } |
723 | |
724 | static void quote(llvm::raw_ostream &OS, llvm::StringRef S) { |
725 | OS << '\"'; |
726 | for (unsigned char C : S) { |
727 | if (C == 0x22 || C == 0x5C) |
728 | OS << '\\'; |
729 | if (C >= 0x20) { |
730 | OS << C; |
731 | continue; |
732 | } |
733 | OS << '\\'; |
734 | switch (C) { |
735 | // A few characters are common enough to make short escapes worthwhile. |
736 | case '\t': |
737 | OS << 't'; |
738 | break; |
739 | case '\n': |
740 | OS << 'n'; |
741 | break; |
742 | case '\r': |
743 | OS << 'r'; |
744 | break; |
745 | default: |
746 | OS << 'u'; |
747 | llvm::write_hex(S&: OS, N: C, Style: llvm::HexPrintStyle::Lower, Width: 4); |
748 | break; |
749 | } |
750 | } |
751 | OS << '\"'; |
752 | } |
753 | |
754 | void llvm::json::OStream::value(const Value &V) { |
755 | switch (V.kind()) { |
756 | case Value::Null: |
757 | valueBegin(); |
758 | OS << "null" ; |
759 | return; |
760 | case Value::Boolean: |
761 | valueBegin(); |
762 | OS << (*V.getAsBoolean() ? "true" : "false" ); |
763 | return; |
764 | case Value::Number: |
765 | valueBegin(); |
766 | if (V.Type == Value::T_Integer) |
767 | OS << *V.getAsInteger(); |
768 | else if (V.Type == Value::T_UINT64) |
769 | OS << *V.getAsUINT64(); |
770 | else |
771 | OS << format(Fmt: "%.*g" , Vals: std::numeric_limits<double>::max_digits10, |
772 | Vals: *V.getAsNumber()); |
773 | return; |
774 | case Value::String: |
775 | valueBegin(); |
776 | quote(OS, S: *V.getAsString()); |
777 | return; |
778 | case Value::Array: |
779 | return array(Contents: [&] { |
780 | for (const Value &E : *V.getAsArray()) |
781 | value(V: E); |
782 | }); |
783 | case Value::Object: |
784 | return object(Contents: [&] { |
785 | for (const Object::value_type *E : sortedElements(O: *V.getAsObject())) |
786 | attribute(Key: E->first, Contents: E->second); |
787 | }); |
788 | } |
789 | } |
790 | |
791 | void llvm::json::OStream::valueBegin() { |
792 | assert(Stack.back().Ctx != Object && "Only attributes allowed here" ); |
793 | if (Stack.back().HasValue) { |
794 | assert(Stack.back().Ctx != Singleton && "Only one value allowed here" ); |
795 | OS << ','; |
796 | } |
797 | if (Stack.back().Ctx == Array) |
798 | newline(); |
799 | flushComment(); |
800 | Stack.back().HasValue = true; |
801 | } |
802 | |
803 | void OStream::(llvm::StringRef ) { |
804 | assert(PendingComment.empty() && "Only one comment per value!" ); |
805 | PendingComment = Comment; |
806 | } |
807 | |
808 | void OStream::() { |
809 | if (PendingComment.empty()) |
810 | return; |
811 | OS << (IndentSize ? "/* " : "/*" ); |
812 | // Be sure not to accidentally emit "*/". Transform to "* /". |
813 | while (!PendingComment.empty()) { |
814 | auto Pos = PendingComment.find(Str: "*/" ); |
815 | if (Pos == StringRef::npos) { |
816 | OS << PendingComment; |
817 | PendingComment = "" ; |
818 | } else { |
819 | OS << PendingComment.take_front(N: Pos) << "* /" ; |
820 | PendingComment = PendingComment.drop_front(N: Pos + 2); |
821 | } |
822 | } |
823 | OS << (IndentSize ? " */" : "*/" ); |
824 | // Comments are on their own line unless attached to an attribute value. |
825 | if (Stack.size() > 1 && Stack.back().Ctx == Singleton) { |
826 | if (IndentSize) |
827 | OS << ' '; |
828 | } else { |
829 | newline(); |
830 | } |
831 | } |
832 | |
833 | void llvm::json::OStream::newline() { |
834 | if (IndentSize) { |
835 | OS.write(C: '\n'); |
836 | OS.indent(NumSpaces: Indent); |
837 | } |
838 | } |
839 | |
840 | void llvm::json::OStream::arrayBegin() { |
841 | valueBegin(); |
842 | Stack.emplace_back(); |
843 | Stack.back().Ctx = Array; |
844 | Indent += IndentSize; |
845 | OS << '['; |
846 | } |
847 | |
848 | void llvm::json::OStream::arrayEnd() { |
849 | assert(Stack.back().Ctx == Array); |
850 | Indent -= IndentSize; |
851 | if (Stack.back().HasValue) |
852 | newline(); |
853 | OS << ']'; |
854 | assert(PendingComment.empty()); |
855 | Stack.pop_back(); |
856 | assert(!Stack.empty()); |
857 | } |
858 | |
859 | void llvm::json::OStream::objectBegin() { |
860 | valueBegin(); |
861 | Stack.emplace_back(); |
862 | Stack.back().Ctx = Object; |
863 | Indent += IndentSize; |
864 | OS << '{'; |
865 | } |
866 | |
867 | void llvm::json::OStream::objectEnd() { |
868 | assert(Stack.back().Ctx == Object); |
869 | Indent -= IndentSize; |
870 | if (Stack.back().HasValue) |
871 | newline(); |
872 | OS << '}'; |
873 | assert(PendingComment.empty()); |
874 | Stack.pop_back(); |
875 | assert(!Stack.empty()); |
876 | } |
877 | |
878 | void llvm::json::OStream::attributeBegin(llvm::StringRef Key) { |
879 | assert(Stack.back().Ctx == Object); |
880 | if (Stack.back().HasValue) |
881 | OS << ','; |
882 | newline(); |
883 | flushComment(); |
884 | Stack.back().HasValue = true; |
885 | Stack.emplace_back(); |
886 | Stack.back().Ctx = Singleton; |
887 | if (LLVM_LIKELY(isUTF8(Key))) { |
888 | quote(OS, S: Key); |
889 | } else { |
890 | assert(false && "Invalid UTF-8 in attribute key" ); |
891 | quote(OS, S: fixUTF8(S: Key)); |
892 | } |
893 | OS.write(C: ':'); |
894 | if (IndentSize) |
895 | OS.write(C: ' '); |
896 | } |
897 | |
898 | void llvm::json::OStream::attributeEnd() { |
899 | assert(Stack.back().Ctx == Singleton); |
900 | assert(Stack.back().HasValue && "Attribute must have a value" ); |
901 | assert(PendingComment.empty()); |
902 | Stack.pop_back(); |
903 | assert(Stack.back().Ctx == Object); |
904 | } |
905 | |
906 | raw_ostream &llvm::json::OStream::rawValueBegin() { |
907 | valueBegin(); |
908 | Stack.emplace_back(); |
909 | Stack.back().Ctx = RawValue; |
910 | return OS; |
911 | } |
912 | |
913 | void llvm::json::OStream::rawValueEnd() { |
914 | assert(Stack.back().Ctx == RawValue); |
915 | Stack.pop_back(); |
916 | } |
917 | |
918 | } // namespace json |
919 | } // namespace llvm |
920 | |
921 | void llvm::format_provider<llvm::json::Value>::format( |
922 | const llvm::json::Value &E, raw_ostream &OS, StringRef Options) { |
923 | unsigned IndentAmount = 0; |
924 | if (!Options.empty() && Options.getAsInteger(/*Radix=*/10, Result&: IndentAmount)) |
925 | llvm_unreachable("json::Value format options should be an integer" ); |
926 | json::OStream(OS, IndentAmount).value(V: E); |
927 | } |
928 | |
929 | |