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 | void Value::print(llvm::raw_ostream &OS) const { OS << *this; } |
186 | |
187 | bool operator==(const Value &L, const Value &R) { |
188 | if (L.kind() != R.kind()) |
189 | return false; |
190 | switch (L.kind()) { |
191 | case Value::Null: |
192 | return *L.getAsNull() == *R.getAsNull(); |
193 | case Value::Boolean: |
194 | return *L.getAsBoolean() == *R.getAsBoolean(); |
195 | case Value::Number: |
196 | // Workaround for https://gcc.gnu.org/bugzilla/show_bug.cgi?id=323 |
197 | // The same integer must convert to the same double, per the standard. |
198 | // However we see 64-vs-80-bit precision comparisons with gcc-7 -O3 -m32. |
199 | // So we avoid floating point promotion for exact comparisons. |
200 | if (L.Type == Value::T_Integer || R.Type == Value::T_Integer) |
201 | return L.getAsInteger() == R.getAsInteger(); |
202 | return *L.getAsNumber() == *R.getAsNumber(); |
203 | case Value::String: |
204 | return *L.getAsString() == *R.getAsString(); |
205 | case Value::Array: |
206 | return *L.getAsArray() == *R.getAsArray(); |
207 | case Value::Object: |
208 | return *L.getAsObject() == *R.getAsObject(); |
209 | } |
210 | llvm_unreachable("Unknown value kind" ); |
211 | } |
212 | |
213 | void Path::report(llvm::StringLiteral Msg) { |
214 | // Walk up to the root context, and count the number of segments. |
215 | unsigned Count = 0; |
216 | const Path *P; |
217 | for (P = this; P->Parent != nullptr; P = P->Parent) |
218 | ++Count; |
219 | Path::Root *R = P->Seg.root(); |
220 | // Fill in the error message and copy the path (in reverse order). |
221 | R->ErrorMessage = Msg; |
222 | R->ErrorPath.resize(new_size: Count); |
223 | auto It = R->ErrorPath.begin(); |
224 | for (P = this; P->Parent != nullptr; P = P->Parent) |
225 | *It++ = P->Seg; |
226 | } |
227 | |
228 | Error Path::Root::getError() const { |
229 | std::string S; |
230 | raw_string_ostream OS(S); |
231 | OS << (ErrorMessage.empty() ? "invalid JSON contents" : ErrorMessage); |
232 | if (ErrorPath.empty()) { |
233 | if (!Name.empty()) |
234 | OS << " when parsing " << Name; |
235 | } else { |
236 | OS << " at " << (Name.empty() ? "(root)" : Name); |
237 | for (const Path::Segment &S : llvm::reverse(C: ErrorPath)) { |
238 | if (S.isField()) |
239 | OS << '.' << S.field(); |
240 | else |
241 | OS << '[' << S.index() << ']'; |
242 | } |
243 | } |
244 | return createStringError(EC: llvm::inconvertibleErrorCode(), S); |
245 | } |
246 | |
247 | std::vector<const Object::value_type *> sortedElements(const Object &O) { |
248 | std::vector<const Object::value_type *> Elements; |
249 | for (const auto &E : O) |
250 | Elements.push_back(x: &E); |
251 | llvm::sort(C&: Elements, |
252 | Comp: [](const Object::value_type *L, const Object::value_type *R) { |
253 | return L->first < R->first; |
254 | }); |
255 | return Elements; |
256 | } |
257 | |
258 | // Prints a one-line version of a value that isn't our main focus. |
259 | // We interleave writes to OS and JOS, exploiting the lack of extra buffering. |
260 | // This is OK as we own the implementation. |
261 | static void abbreviate(const Value &V, OStream &JOS) { |
262 | switch (V.kind()) { |
263 | case Value::Array: |
264 | JOS.rawValue(Contents: V.getAsArray()->empty() ? "[]" : "[ ... ]" ); |
265 | break; |
266 | case Value::Object: |
267 | JOS.rawValue(Contents: V.getAsObject()->empty() ? "{}" : "{ ... }" ); |
268 | break; |
269 | case Value::String: { |
270 | llvm::StringRef S = *V.getAsString(); |
271 | if (S.size() < 40) { |
272 | JOS.value(V); |
273 | } else { |
274 | std::string Truncated = fixUTF8(S: S.take_front(N: 37)); |
275 | Truncated.append(s: "..." ); |
276 | JOS.value(V: Truncated); |
277 | } |
278 | break; |
279 | } |
280 | default: |
281 | JOS.value(V); |
282 | } |
283 | } |
284 | |
285 | // Prints a semi-expanded version of a value that is our main focus. |
286 | // Array/Object entries are printed, but not recursively as they may be huge. |
287 | static void abbreviateChildren(const Value &V, OStream &JOS) { |
288 | switch (V.kind()) { |
289 | case Value::Array: |
290 | JOS.array(Contents: [&] { |
291 | for (const auto &I : *V.getAsArray()) |
292 | abbreviate(V: I, JOS); |
293 | }); |
294 | break; |
295 | case Value::Object: |
296 | JOS.object(Contents: [&] { |
297 | for (const auto *KV : sortedElements(O: *V.getAsObject())) { |
298 | JOS.attributeBegin(Key: KV->first); |
299 | abbreviate(V: KV->second, JOS); |
300 | JOS.attributeEnd(); |
301 | } |
302 | }); |
303 | break; |
304 | default: |
305 | JOS.value(V); |
306 | } |
307 | } |
308 | |
309 | void Path::Root::printErrorContext(const Value &R, raw_ostream &OS) const { |
310 | OStream JOS(OS, /*IndentSize=*/2); |
311 | // PrintValue recurses down the path, printing the ancestors of our target. |
312 | // Siblings of nodes along the path are printed with abbreviate(), and the |
313 | // target itself is printed with the somewhat richer abbreviateChildren(). |
314 | // 'Recurse' is the lambda itself, to allow recursive calls. |
315 | auto PrintValue = [&](const Value &V, ArrayRef<Segment> Path, auto &Recurse) { |
316 | // Print the target node itself, with the error as a comment. |
317 | // Also used if we can't follow our path, e.g. it names a field that |
318 | // *should* exist but doesn't. |
319 | auto HighlightCurrent = [&] { |
320 | std::string = "error: " ; |
321 | Comment.append(s: ErrorMessage.data(), n: ErrorMessage.size()); |
322 | JOS.comment(Comment); |
323 | abbreviateChildren(V, JOS); |
324 | }; |
325 | if (Path.empty()) // We reached our target. |
326 | return HighlightCurrent(); |
327 | const Segment &S = Path.back(); // Path is in reverse order. |
328 | if (S.isField()) { |
329 | // Current node is an object, path names a field. |
330 | llvm::StringRef FieldName = S.field(); |
331 | const Object *O = V.getAsObject(); |
332 | if (!O || !O->get(K: FieldName)) |
333 | return HighlightCurrent(); |
334 | JOS.object(Contents: [&] { |
335 | for (const auto *KV : sortedElements(O: *O)) { |
336 | JOS.attributeBegin(Key: KV->first); |
337 | if (FieldName == StringRef(KV->first)) |
338 | Recurse(KV->second, Path.drop_back(), Recurse); |
339 | else |
340 | abbreviate(V: KV->second, JOS); |
341 | JOS.attributeEnd(); |
342 | } |
343 | }); |
344 | } else { |
345 | // Current node is an array, path names an element. |
346 | const Array *A = V.getAsArray(); |
347 | if (!A || S.index() >= A->size()) |
348 | return HighlightCurrent(); |
349 | JOS.array(Contents: [&] { |
350 | unsigned Current = 0; |
351 | for (const auto &V : *A) { |
352 | if (Current++ == S.index()) |
353 | Recurse(V, Path.drop_back(), Recurse); |
354 | else |
355 | abbreviate(V, JOS); |
356 | } |
357 | }); |
358 | } |
359 | }; |
360 | PrintValue(R, ErrorPath, PrintValue); |
361 | } |
362 | |
363 | namespace { |
364 | // Simple recursive-descent JSON parser. |
365 | class Parser { |
366 | public: |
367 | Parser(StringRef JSON) |
368 | : Start(JSON.begin()), P(JSON.begin()), End(JSON.end()) {} |
369 | |
370 | bool checkUTF8() { |
371 | size_t ErrOffset; |
372 | if (isUTF8(S: StringRef(Start, End - Start), ErrOffset: &ErrOffset)) |
373 | return true; |
374 | P = Start + ErrOffset; // For line/column calculation. |
375 | return parseError(Msg: "Invalid UTF-8 sequence" ); |
376 | } |
377 | |
378 | bool parseValue(Value &Out); |
379 | |
380 | bool assertEnd() { |
381 | eatWhitespace(); |
382 | if (P == End) |
383 | return true; |
384 | return parseError(Msg: "Text after end of document" ); |
385 | } |
386 | |
387 | Error takeError() { |
388 | assert(Err); |
389 | return std::move(*Err); |
390 | } |
391 | |
392 | private: |
393 | void eatWhitespace() { |
394 | while (P != End && (*P == ' ' || *P == '\r' || *P == '\n' || *P == '\t')) |
395 | ++P; |
396 | } |
397 | |
398 | // On invalid syntax, parseX() functions return false and set Err. |
399 | bool parseNumber(char First, Value &Out); |
400 | bool parseString(std::string &Out); |
401 | bool parseUnicode(std::string &Out); |
402 | bool parseError(const char *Msg); // always returns false |
403 | |
404 | char next() { return P == End ? 0 : *P++; } |
405 | char peek() { return P == End ? 0 : *P; } |
406 | static bool isNumber(char C) { |
407 | return C == '0' || C == '1' || C == '2' || C == '3' || C == '4' || |
408 | C == '5' || C == '6' || C == '7' || C == '8' || C == '9' || |
409 | C == 'e' || C == 'E' || C == '+' || C == '-' || C == '.'; |
410 | } |
411 | |
412 | std::optional<Error> Err; |
413 | const char *Start, *P, *End; |
414 | }; |
415 | } // namespace |
416 | |
417 | bool Parser::parseValue(Value &Out) { |
418 | eatWhitespace(); |
419 | if (P == End) |
420 | return parseError(Msg: "Unexpected EOF" ); |
421 | switch (char C = next()) { |
422 | // Bare null/true/false are easy - first char identifies them. |
423 | case 'n': |
424 | Out = nullptr; |
425 | return (next() == 'u' && next() == 'l' && next() == 'l') || |
426 | parseError(Msg: "Invalid JSON value (null?)" ); |
427 | case 't': |
428 | Out = true; |
429 | return (next() == 'r' && next() == 'u' && next() == 'e') || |
430 | parseError(Msg: "Invalid JSON value (true?)" ); |
431 | case 'f': |
432 | Out = false; |
433 | return (next() == 'a' && next() == 'l' && next() == 's' && next() == 'e') || |
434 | parseError(Msg: "Invalid JSON value (false?)" ); |
435 | case '"': { |
436 | std::string S; |
437 | if (parseString(Out&: S)) { |
438 | Out = std::move(S); |
439 | return true; |
440 | } |
441 | return false; |
442 | } |
443 | case '[': { |
444 | Out = Array{}; |
445 | Array &A = *Out.getAsArray(); |
446 | eatWhitespace(); |
447 | if (peek() == ']') { |
448 | ++P; |
449 | return true; |
450 | } |
451 | for (;;) { |
452 | A.emplace_back(A: nullptr); |
453 | if (!parseValue(Out&: A.back())) |
454 | return false; |
455 | eatWhitespace(); |
456 | switch (next()) { |
457 | case ',': |
458 | eatWhitespace(); |
459 | continue; |
460 | case ']': |
461 | return true; |
462 | default: |
463 | return parseError(Msg: "Expected , or ] after array element" ); |
464 | } |
465 | } |
466 | } |
467 | case '{': { |
468 | Out = Object{}; |
469 | Object &O = *Out.getAsObject(); |
470 | eatWhitespace(); |
471 | if (peek() == '}') { |
472 | ++P; |
473 | return true; |
474 | } |
475 | for (;;) { |
476 | if (next() != '"') |
477 | return parseError(Msg: "Expected object key" ); |
478 | std::string K; |
479 | if (!parseString(Out&: K)) |
480 | return false; |
481 | eatWhitespace(); |
482 | if (next() != ':') |
483 | return parseError(Msg: "Expected : after object key" ); |
484 | eatWhitespace(); |
485 | if (!parseValue(Out&: O[std::move(K)])) |
486 | return false; |
487 | eatWhitespace(); |
488 | switch (next()) { |
489 | case ',': |
490 | eatWhitespace(); |
491 | continue; |
492 | case '}': |
493 | return true; |
494 | default: |
495 | return parseError(Msg: "Expected , or } after object property" ); |
496 | } |
497 | } |
498 | } |
499 | default: |
500 | if (isNumber(C)) |
501 | return parseNumber(First: C, Out); |
502 | return parseError(Msg: "Invalid JSON value" ); |
503 | } |
504 | } |
505 | |
506 | bool Parser::parseNumber(char First, Value &Out) { |
507 | // Read the number into a string. (Must be null-terminated for strto*). |
508 | SmallString<24> S; |
509 | S.push_back(Elt: First); |
510 | while (isNumber(C: peek())) |
511 | S.push_back(Elt: next()); |
512 | char *End; |
513 | // Try first to parse as integer, and if so preserve full 64 bits. |
514 | // We check for errno for out of bounds errors and for End == S.end() |
515 | // to make sure that the numeric string is not malformed. |
516 | errno = 0; |
517 | int64_t I = std::strtoll(nptr: S.c_str(), endptr: &End, base: 10); |
518 | if (End == S.end() && errno != ERANGE) { |
519 | Out = int64_t(I); |
520 | return true; |
521 | } |
522 | // strtroull has a special handling for negative numbers, but in this |
523 | // case we don't want to do that because negative numbers were already |
524 | // handled in the previous block. |
525 | if (First != '-') { |
526 | errno = 0; |
527 | uint64_t UI = std::strtoull(nptr: S.c_str(), endptr: &End, base: 10); |
528 | if (End == S.end() && errno != ERANGE) { |
529 | Out = UI; |
530 | return true; |
531 | } |
532 | } |
533 | // If it's not an integer |
534 | Out = std::strtod(nptr: S.c_str(), endptr: &End); |
535 | return End == S.end() || parseError(Msg: "Invalid JSON value (number?)" ); |
536 | } |
537 | |
538 | bool Parser::parseString(std::string &Out) { |
539 | // leading quote was already consumed. |
540 | for (char C = next(); C != '"'; C = next()) { |
541 | if (LLVM_UNLIKELY(P == End)) |
542 | return parseError(Msg: "Unterminated string" ); |
543 | if (LLVM_UNLIKELY((C & 0x1f) == C)) |
544 | return parseError(Msg: "Control character in string" ); |
545 | if (LLVM_LIKELY(C != '\\')) { |
546 | Out.push_back(c: C); |
547 | continue; |
548 | } |
549 | // Handle escape sequence. |
550 | switch (C = next()) { |
551 | case '"': |
552 | case '\\': |
553 | case '/': |
554 | Out.push_back(c: C); |
555 | break; |
556 | case 'b': |
557 | Out.push_back(c: '\b'); |
558 | break; |
559 | case 'f': |
560 | Out.push_back(c: '\f'); |
561 | break; |
562 | case 'n': |
563 | Out.push_back(c: '\n'); |
564 | break; |
565 | case 'r': |
566 | Out.push_back(c: '\r'); |
567 | break; |
568 | case 't': |
569 | Out.push_back(c: '\t'); |
570 | break; |
571 | case 'u': |
572 | if (!parseUnicode(Out)) |
573 | return false; |
574 | break; |
575 | default: |
576 | return parseError(Msg: "Invalid escape sequence" ); |
577 | } |
578 | } |
579 | return true; |
580 | } |
581 | |
582 | static void encodeUtf8(uint32_t Rune, std::string &Out) { |
583 | if (Rune < 0x80) { |
584 | Out.push_back(c: Rune & 0x7F); |
585 | } else if (Rune < 0x800) { |
586 | uint8_t FirstByte = 0xC0 | ((Rune & 0x7C0) >> 6); |
587 | uint8_t SecondByte = 0x80 | (Rune & 0x3F); |
588 | Out.push_back(c: FirstByte); |
589 | Out.push_back(c: SecondByte); |
590 | } else if (Rune < 0x10000) { |
591 | uint8_t FirstByte = 0xE0 | ((Rune & 0xF000) >> 12); |
592 | uint8_t SecondByte = 0x80 | ((Rune & 0xFC0) >> 6); |
593 | uint8_t ThirdByte = 0x80 | (Rune & 0x3F); |
594 | Out.push_back(c: FirstByte); |
595 | Out.push_back(c: SecondByte); |
596 | Out.push_back(c: ThirdByte); |
597 | } else if (Rune < 0x110000) { |
598 | uint8_t FirstByte = 0xF0 | ((Rune & 0x1F0000) >> 18); |
599 | uint8_t SecondByte = 0x80 | ((Rune & 0x3F000) >> 12); |
600 | uint8_t ThirdByte = 0x80 | ((Rune & 0xFC0) >> 6); |
601 | uint8_t FourthByte = 0x80 | (Rune & 0x3F); |
602 | Out.push_back(c: FirstByte); |
603 | Out.push_back(c: SecondByte); |
604 | Out.push_back(c: ThirdByte); |
605 | Out.push_back(c: FourthByte); |
606 | } else { |
607 | llvm_unreachable("Invalid codepoint" ); |
608 | } |
609 | } |
610 | |
611 | // Parse a UTF-16 \uNNNN escape sequence. "\u" has already been consumed. |
612 | // May parse several sequential escapes to ensure proper surrogate handling. |
613 | // We do not use ConvertUTF.h, it can't accept and replace unpaired surrogates. |
614 | // These are invalid Unicode but valid JSON (RFC 8259, section 8.2). |
615 | bool Parser::parseUnicode(std::string &Out) { |
616 | // Invalid UTF is not a JSON error (RFC 8529§8.2). It gets replaced by U+FFFD. |
617 | auto Invalid = [&] { Out.append(/* UTF-8 */ l: {'\xef', '\xbf', '\xbd'}); }; |
618 | // Decodes 4 hex digits from the stream into Out, returns false on error. |
619 | auto Parse4Hex = [this](uint16_t &Out) -> bool { |
620 | Out = 0; |
621 | char Bytes[] = {next(), next(), next(), next()}; |
622 | for (unsigned char C : Bytes) { |
623 | if (!std::isxdigit(C)) |
624 | return parseError(Msg: "Invalid \\u escape sequence" ); |
625 | Out <<= 4; |
626 | Out |= (C > '9') ? (C & ~0x20) - 'A' + 10 : (C - '0'); |
627 | } |
628 | return true; |
629 | }; |
630 | uint16_t First; // UTF-16 code unit from the first \u escape. |
631 | if (!Parse4Hex(First)) |
632 | return false; |
633 | |
634 | // We loop to allow proper surrogate-pair error handling. |
635 | while (true) { |
636 | // Case 1: the UTF-16 code unit is already a codepoint in the BMP. |
637 | if (LLVM_LIKELY(First < 0xD800 || First >= 0xE000)) { |
638 | encodeUtf8(Rune: First, Out); |
639 | return true; |
640 | } |
641 | |
642 | // Case 2: it's an (unpaired) trailing surrogate. |
643 | if (LLVM_UNLIKELY(First >= 0xDC00)) { |
644 | Invalid(); |
645 | return true; |
646 | } |
647 | |
648 | // Case 3: it's a leading surrogate. We expect a trailing one next. |
649 | // Case 3a: there's no trailing \u escape. Don't advance in the stream. |
650 | if (LLVM_UNLIKELY(P + 2 > End || *P != '\\' || *(P + 1) != 'u')) { |
651 | Invalid(); // Leading surrogate was unpaired. |
652 | return true; |
653 | } |
654 | P += 2; |
655 | uint16_t Second; |
656 | if (!Parse4Hex(Second)) |
657 | return false; |
658 | // Case 3b: there was another \u escape, but it wasn't a trailing surrogate. |
659 | if (LLVM_UNLIKELY(Second < 0xDC00 || Second >= 0xE000)) { |
660 | Invalid(); // Leading surrogate was unpaired. |
661 | First = Second; // Second escape still needs to be processed. |
662 | continue; |
663 | } |
664 | // Case 3c: a valid surrogate pair encoding an astral codepoint. |
665 | encodeUtf8(Rune: 0x10000 | ((First - 0xD800) << 10) | (Second - 0xDC00), Out); |
666 | return true; |
667 | } |
668 | } |
669 | |
670 | bool Parser::parseError(const char *Msg) { |
671 | int Line = 1; |
672 | const char *StartOfLine = Start; |
673 | for (const char *X = Start; X < P; ++X) { |
674 | if (*X == 0x0A) { |
675 | ++Line; |
676 | StartOfLine = X + 1; |
677 | } |
678 | } |
679 | Err.emplace( |
680 | args: std::make_unique<ParseError>(args&: Msg, args&: Line, args: P - StartOfLine, args: P - Start)); |
681 | return false; |
682 | } |
683 | |
684 | Expected<Value> parse(StringRef JSON) { |
685 | Parser P(JSON); |
686 | Value E = nullptr; |
687 | if (P.checkUTF8()) |
688 | if (P.parseValue(Out&: E)) |
689 | if (P.assertEnd()) |
690 | return std::move(E); |
691 | return P.takeError(); |
692 | } |
693 | char ParseError::ID = 0; |
694 | |
695 | bool isUTF8(llvm::StringRef S, size_t *ErrOffset) { |
696 | // Fast-path for ASCII, which is valid UTF-8. |
697 | if (LLVM_LIKELY(isASCII(S))) |
698 | return true; |
699 | |
700 | const UTF8 *Data = reinterpret_cast<const UTF8 *>(S.data()), *Rest = Data; |
701 | if (LLVM_LIKELY(isLegalUTF8String(&Rest, Data + S.size()))) |
702 | return true; |
703 | |
704 | if (ErrOffset) |
705 | *ErrOffset = Rest - Data; |
706 | return false; |
707 | } |
708 | |
709 | std::string fixUTF8(llvm::StringRef S) { |
710 | // This isn't particularly efficient, but is only for error-recovery. |
711 | std::vector<UTF32> Codepoints(S.size()); // 1 codepoint per byte suffices. |
712 | const UTF8 *In8 = reinterpret_cast<const UTF8 *>(S.data()); |
713 | UTF32 *Out32 = Codepoints.data(); |
714 | ConvertUTF8toUTF32(sourceStart: &In8, sourceEnd: In8 + S.size(), targetStart: &Out32, targetEnd: Out32 + Codepoints.size(), |
715 | flags: lenientConversion); |
716 | Codepoints.resize(new_size: Out32 - Codepoints.data()); |
717 | std::string Res(4 * Codepoints.size(), 0); // 4 bytes per codepoint suffice |
718 | const UTF32 *In32 = Codepoints.data(); |
719 | UTF8 *Out8 = reinterpret_cast<UTF8 *>(&Res[0]); |
720 | ConvertUTF32toUTF8(sourceStart: &In32, sourceEnd: In32 + Codepoints.size(), targetStart: &Out8, targetEnd: Out8 + Res.size(), |
721 | flags: strictConversion); |
722 | Res.resize(n: reinterpret_cast<char *>(Out8) - Res.data()); |
723 | return Res; |
724 | } |
725 | |
726 | static void quote(llvm::raw_ostream &OS, llvm::StringRef S) { |
727 | OS << '\"'; |
728 | for (unsigned char C : S) { |
729 | if (C == 0x22 || C == 0x5C) |
730 | OS << '\\'; |
731 | if (C >= 0x20) { |
732 | OS << C; |
733 | continue; |
734 | } |
735 | OS << '\\'; |
736 | switch (C) { |
737 | // A few characters are common enough to make short escapes worthwhile. |
738 | case '\t': |
739 | OS << 't'; |
740 | break; |
741 | case '\n': |
742 | OS << 'n'; |
743 | break; |
744 | case '\r': |
745 | OS << 'r'; |
746 | break; |
747 | default: |
748 | OS << 'u'; |
749 | llvm::write_hex(S&: OS, N: C, Style: llvm::HexPrintStyle::Lower, Width: 4); |
750 | break; |
751 | } |
752 | } |
753 | OS << '\"'; |
754 | } |
755 | |
756 | void llvm::json::OStream::value(const Value &V) { |
757 | switch (V.kind()) { |
758 | case Value::Null: |
759 | valueBegin(); |
760 | OS << "null" ; |
761 | return; |
762 | case Value::Boolean: |
763 | valueBegin(); |
764 | OS << (*V.getAsBoolean() ? "true" : "false" ); |
765 | return; |
766 | case Value::Number: |
767 | valueBegin(); |
768 | if (V.Type == Value::T_Integer) |
769 | OS << *V.getAsInteger(); |
770 | else if (V.Type == Value::T_UINT64) |
771 | OS << *V.getAsUINT64(); |
772 | else |
773 | OS << format(Fmt: "%.*g" , Vals: std::numeric_limits<double>::max_digits10, |
774 | Vals: *V.getAsNumber()); |
775 | return; |
776 | case Value::String: |
777 | valueBegin(); |
778 | quote(OS, S: *V.getAsString()); |
779 | return; |
780 | case Value::Array: |
781 | return array(Contents: [&] { |
782 | for (const Value &E : *V.getAsArray()) |
783 | value(V: E); |
784 | }); |
785 | case Value::Object: |
786 | return object(Contents: [&] { |
787 | for (const Object::value_type *E : sortedElements(O: *V.getAsObject())) |
788 | attribute(Key: E->first, Contents: E->second); |
789 | }); |
790 | } |
791 | } |
792 | |
793 | void llvm::json::OStream::valueBegin() { |
794 | assert(Stack.back().Ctx != Object && "Only attributes allowed here" ); |
795 | if (Stack.back().HasValue) { |
796 | assert(Stack.back().Ctx != Singleton && "Only one value allowed here" ); |
797 | OS << ','; |
798 | } |
799 | if (Stack.back().Ctx == Array) |
800 | newline(); |
801 | flushComment(); |
802 | Stack.back().HasValue = true; |
803 | } |
804 | |
805 | void OStream::(llvm::StringRef ) { |
806 | assert(PendingComment.empty() && "Only one comment per value!" ); |
807 | PendingComment = Comment; |
808 | } |
809 | |
810 | void OStream::() { |
811 | if (PendingComment.empty()) |
812 | return; |
813 | OS << (IndentSize ? "/* " : "/*" ); |
814 | // Be sure not to accidentally emit "*/". Transform to "* /". |
815 | while (!PendingComment.empty()) { |
816 | auto Pos = PendingComment.find(Str: "*/" ); |
817 | if (Pos == StringRef::npos) { |
818 | OS << PendingComment; |
819 | PendingComment = "" ; |
820 | } else { |
821 | OS << PendingComment.take_front(N: Pos) << "* /" ; |
822 | PendingComment = PendingComment.drop_front(N: Pos + 2); |
823 | } |
824 | } |
825 | OS << (IndentSize ? " */" : "*/" ); |
826 | // Comments are on their own line unless attached to an attribute value. |
827 | if (Stack.size() > 1 && Stack.back().Ctx == Singleton) { |
828 | if (IndentSize) |
829 | OS << ' '; |
830 | } else { |
831 | newline(); |
832 | } |
833 | } |
834 | |
835 | void llvm::json::OStream::newline() { |
836 | if (IndentSize) { |
837 | OS.write(C: '\n'); |
838 | OS.indent(NumSpaces: Indent); |
839 | } |
840 | } |
841 | |
842 | void llvm::json::OStream::arrayBegin() { |
843 | valueBegin(); |
844 | Stack.emplace_back(); |
845 | Stack.back().Ctx = Array; |
846 | Indent += IndentSize; |
847 | OS << '['; |
848 | } |
849 | |
850 | void llvm::json::OStream::arrayEnd() { |
851 | assert(Stack.back().Ctx == Array); |
852 | Indent -= IndentSize; |
853 | if (Stack.back().HasValue) |
854 | newline(); |
855 | OS << ']'; |
856 | assert(PendingComment.empty()); |
857 | Stack.pop_back(); |
858 | assert(!Stack.empty()); |
859 | } |
860 | |
861 | void llvm::json::OStream::objectBegin() { |
862 | valueBegin(); |
863 | Stack.emplace_back(); |
864 | Stack.back().Ctx = Object; |
865 | Indent += IndentSize; |
866 | OS << '{'; |
867 | } |
868 | |
869 | void llvm::json::OStream::objectEnd() { |
870 | assert(Stack.back().Ctx == Object); |
871 | Indent -= IndentSize; |
872 | if (Stack.back().HasValue) |
873 | newline(); |
874 | OS << '}'; |
875 | assert(PendingComment.empty()); |
876 | Stack.pop_back(); |
877 | assert(!Stack.empty()); |
878 | } |
879 | |
880 | void llvm::json::OStream::attributeBegin(llvm::StringRef Key) { |
881 | assert(Stack.back().Ctx == Object); |
882 | if (Stack.back().HasValue) |
883 | OS << ','; |
884 | newline(); |
885 | flushComment(); |
886 | Stack.back().HasValue = true; |
887 | Stack.emplace_back(); |
888 | Stack.back().Ctx = Singleton; |
889 | if (LLVM_LIKELY(isUTF8(Key))) { |
890 | quote(OS, S: Key); |
891 | } else { |
892 | assert(false && "Invalid UTF-8 in attribute key" ); |
893 | quote(OS, S: fixUTF8(S: Key)); |
894 | } |
895 | OS.write(C: ':'); |
896 | if (IndentSize) |
897 | OS.write(C: ' '); |
898 | } |
899 | |
900 | void llvm::json::OStream::attributeEnd() { |
901 | assert(Stack.back().Ctx == Singleton); |
902 | assert(Stack.back().HasValue && "Attribute must have a value" ); |
903 | assert(PendingComment.empty()); |
904 | Stack.pop_back(); |
905 | assert(Stack.back().Ctx == Object); |
906 | } |
907 | |
908 | raw_ostream &llvm::json::OStream::rawValueBegin() { |
909 | valueBegin(); |
910 | Stack.emplace_back(); |
911 | Stack.back().Ctx = RawValue; |
912 | return OS; |
913 | } |
914 | |
915 | void llvm::json::OStream::rawValueEnd() { |
916 | assert(Stack.back().Ctx == RawValue); |
917 | Stack.pop_back(); |
918 | } |
919 | |
920 | } // namespace json |
921 | } // namespace llvm |
922 | |
923 | void llvm::format_provider<llvm::json::Value>::format( |
924 | const llvm::json::Value &E, raw_ostream &OS, StringRef Options) { |
925 | unsigned IndentAmount = 0; |
926 | if (!Options.empty() && Options.getAsInteger(/*Radix=*/10, Result&: IndentAmount)) |
927 | llvm_unreachable("json::Value format options should be an integer" ); |
928 | json::OStream(OS, IndentAmount).value(V: E); |
929 | } |
930 | |
931 | |