1 //===--- JSON.h - JSON values, 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 /// \file
10 /// This file supports working with JSON data.
11 ///
12 /// It comprises:
13 ///
14 /// - classes which hold dynamically-typed parsed JSON structures
15 /// These are value types that can be composed, inspected, and modified.
16 /// See json::Value, and the related types json::Object and json::Array.
17 ///
18 /// - functions to parse JSON text into Values, and to serialize Values to text.
19 /// See parse(), operator<<, and format_provider.
20 ///
21 /// - a convention and helpers for mapping between json::Value and user-defined
22 /// types. See fromJSON(), ObjectMapper, and the class comment on Value.
23 ///
24 /// - an output API json::OStream which can emit JSON without materializing
25 /// all structures as json::Value.
26 ///
27 /// Typically, JSON data would be read from an external source, parsed into
28 /// a Value, and then converted into some native data structure before doing
29 /// real work on it. (And vice versa when writing).
30 ///
31 /// Other serialization mechanisms you may consider:
32 ///
33 /// - YAML is also text-based, and more human-readable than JSON. It's a more
34 /// complex format and data model, and YAML parsers aren't ubiquitous.
35 /// YAMLParser.h is a streaming parser suitable for parsing large documents
36 /// (including JSON, as YAML is a superset). It can be awkward to use
37 /// directly. YAML I/O (YAMLTraits.h) provides data mapping that is more
38 /// declarative than the toJSON/fromJSON conventions here.
39 ///
40 /// - LLVM bitstream is a space- and CPU- efficient binary format. Typically it
41 /// encodes LLVM IR ("bitcode"), but it can be a container for other data.
42 /// Low-level reader/writer libraries are in Bitstream/Bitstream*.h
43 ///
44 //===---------------------------------------------------------------------===//
45
46 #ifndef LLVM_SUPPORT_JSON_H
47 #define LLVM_SUPPORT_JSON_H
48
49 #include "llvm/ADT/DenseMap.h"
50 #include "llvm/ADT/SmallVector.h"
51 #include "llvm/ADT/StringRef.h"
52 #include "llvm/Support/Error.h"
53 #include "llvm/Support/FormatVariadic.h"
54 #include "llvm/Support/raw_ostream.h"
55 #include <map>
56
57 namespace llvm {
58 namespace json {
59
60 // === String encodings ===
61 //
62 // JSON strings are character sequences (not byte sequences like std::string).
63 // We need to know the encoding, and for simplicity only support UTF-8.
64 //
65 // - When parsing, invalid UTF-8 is a syntax error like any other
66 //
67 // - When creating Values from strings, callers must ensure they are UTF-8.
68 // with asserts on, invalid UTF-8 will crash the program
69 // with asserts off, we'll substitute the replacement character (U+FFFD)
70 // Callers can use json::isUTF8() and json::fixUTF8() for validation.
71 //
72 // - When retrieving strings from Values (e.g. asString()), the result will
73 // always be valid UTF-8.
74
75 /// Returns true if \p S is valid UTF-8, which is required for use as JSON.
76 /// If it returns false, \p Offset is set to a byte offset near the first error.
77 bool isUTF8(llvm::StringRef S, size_t *ErrOffset = nullptr);
78 /// Replaces invalid UTF-8 sequences in \p S with the replacement character
79 /// (U+FFFD). The returned string is valid UTF-8.
80 /// This is much slower than isUTF8, so test that first.
81 std::string fixUTF8(llvm::StringRef S);
82
83 class Array;
84 class ObjectKey;
85 class Value;
86 template <typename T> Value toJSON(const llvm::Optional<T> &Opt);
87
88 /// An Object is a JSON object, which maps strings to heterogenous JSON values.
89 /// It simulates DenseMap<ObjectKey, Value>. ObjectKey is a maybe-owned string.
90 class Object {
91 using Storage = DenseMap<ObjectKey, Value, llvm::DenseMapInfo<StringRef>>;
92 Storage M;
93
94 public:
95 using key_type = ObjectKey;
96 using mapped_type = Value;
97 using value_type = Storage::value_type;
98 using iterator = Storage::iterator;
99 using const_iterator = Storage::const_iterator;
100
101 Object() = default;
102 // KV is a trivial key-value struct for list-initialization.
103 // (using std::pair forces extra copies).
104 struct KV;
105 explicit Object(std::initializer_list<KV> Properties);
106
begin()107 iterator begin() { return M.begin(); }
begin()108 const_iterator begin() const { return M.begin(); }
end()109 iterator end() { return M.end(); }
end()110 const_iterator end() const { return M.end(); }
111
empty()112 bool empty() const { return M.empty(); }
size()113 size_t size() const { return M.size(); }
114
clear()115 void clear() { M.clear(); }
116 std::pair<iterator, bool> insert(KV E);
117 template <typename... Ts>
try_emplace(const ObjectKey & K,Ts &&...Args)118 std::pair<iterator, bool> try_emplace(const ObjectKey &K, Ts &&... Args) {
119 return M.try_emplace(K, std::forward<Ts>(Args)...);
120 }
121 template <typename... Ts>
try_emplace(ObjectKey && K,Ts &&...Args)122 std::pair<iterator, bool> try_emplace(ObjectKey &&K, Ts &&... Args) {
123 return M.try_emplace(std::move(K), std::forward<Ts>(Args)...);
124 }
125 bool erase(StringRef K);
erase(iterator I)126 void erase(iterator I) { M.erase(I); }
127
find(StringRef K)128 iterator find(StringRef K) { return M.find_as(K); }
find(StringRef K)129 const_iterator find(StringRef K) const { return M.find_as(K); }
130 // operator[] acts as if Value was default-constructible as null.
131 Value &operator[](const ObjectKey &K);
132 Value &operator[](ObjectKey &&K);
133 // Look up a property, returning nullptr if it doesn't exist.
134 Value *get(StringRef K);
135 const Value *get(StringRef K) const;
136 // Typed accessors return None/nullptr if
137 // - the property doesn't exist
138 // - or it has the wrong type
139 llvm::Optional<std::nullptr_t> getNull(StringRef K) const;
140 llvm::Optional<bool> getBoolean(StringRef K) const;
141 llvm::Optional<double> getNumber(StringRef K) const;
142 llvm::Optional<int64_t> getInteger(StringRef K) const;
143 llvm::Optional<llvm::StringRef> getString(StringRef K) const;
144 const json::Object *getObject(StringRef K) const;
145 json::Object *getObject(StringRef K);
146 const json::Array *getArray(StringRef K) const;
147 json::Array *getArray(StringRef K);
148 };
149 bool operator==(const Object &LHS, const Object &RHS);
150 inline bool operator!=(const Object &LHS, const Object &RHS) {
151 return !(LHS == RHS);
152 }
153
154 /// An Array is a JSON array, which contains heterogeneous JSON values.
155 /// It simulates std::vector<Value>.
156 class Array {
157 std::vector<Value> V;
158
159 public:
160 using value_type = Value;
161 using iterator = std::vector<Value>::iterator;
162 using const_iterator = std::vector<Value>::const_iterator;
163
164 Array() = default;
165 explicit Array(std::initializer_list<Value> Elements);
Array(const Collection & C)166 template <typename Collection> explicit Array(const Collection &C) {
167 for (const auto &V : C)
168 emplace_back(V);
169 }
170
171 Value &operator[](size_t I);
172 const Value &operator[](size_t I) const;
173 Value &front();
174 const Value &front() const;
175 Value &back();
176 const Value &back() const;
177 Value *data();
178 const Value *data() const;
179
180 iterator begin();
181 const_iterator begin() const;
182 iterator end();
183 const_iterator end() const;
184
185 bool empty() const;
186 size_t size() const;
187 void reserve(size_t S);
188
189 void clear();
190 void push_back(const Value &E);
191 void push_back(Value &&E);
192 template <typename... Args> void emplace_back(Args &&...A);
193 void pop_back();
194 // FIXME: insert() takes const_iterator since C++11, old libstdc++ disagrees.
195 iterator insert(iterator P, const Value &E);
196 iterator insert(iterator P, Value &&E);
197 template <typename It> iterator insert(iterator P, It A, It Z);
198 template <typename... Args> iterator emplace(const_iterator P, Args &&...A);
199
200 friend bool operator==(const Array &L, const Array &R);
201 };
202 inline bool operator!=(const Array &L, const Array &R) { return !(L == R); }
203
204 /// A Value is an JSON value of unknown type.
205 /// They can be copied, but should generally be moved.
206 ///
207 /// === Composing values ===
208 ///
209 /// You can implicitly construct Values from:
210 /// - strings: std::string, SmallString, formatv, StringRef, char*
211 /// (char*, and StringRef are references, not copies!)
212 /// - numbers
213 /// - booleans
214 /// - null: nullptr
215 /// - arrays: {"foo", 42.0, false}
216 /// - serializable things: types with toJSON(const T&)->Value, found by ADL
217 ///
218 /// They can also be constructed from object/array helpers:
219 /// - json::Object is a type like map<ObjectKey, Value>
220 /// - json::Array is a type like vector<Value>
221 /// These can be list-initialized, or used to build up collections in a loop.
222 /// json::ary(Collection) converts all items in a collection to Values.
223 ///
224 /// === Inspecting values ===
225 ///
226 /// Each Value is one of the JSON kinds:
227 /// null (nullptr_t)
228 /// boolean (bool)
229 /// number (double or int64)
230 /// string (StringRef)
231 /// array (json::Array)
232 /// object (json::Object)
233 ///
234 /// The kind can be queried directly, or implicitly via the typed accessors:
235 /// if (Optional<StringRef> S = E.getAsString()
236 /// assert(E.kind() == Value::String);
237 ///
238 /// Array and Object also have typed indexing accessors for easy traversal:
239 /// Expected<Value> E = parse(R"( {"options": {"font": "sans-serif"}} )");
240 /// if (Object* O = E->getAsObject())
241 /// if (Object* Opts = O->getObject("options"))
242 /// if (Optional<StringRef> Font = Opts->getString("font"))
243 /// assert(Opts->at("font").kind() == Value::String);
244 ///
245 /// === Converting JSON values to C++ types ===
246 ///
247 /// The convention is to have a deserializer function findable via ADL:
248 /// fromJSON(const json::Value&, T&)->bool
249 /// Deserializers are provided for:
250 /// - bool
251 /// - int and int64_t
252 /// - double
253 /// - std::string
254 /// - vector<T>, where T is deserializable
255 /// - map<string, T>, where T is deserializable
256 /// - Optional<T>, where T is deserializable
257 /// ObjectMapper can help writing fromJSON() functions for object types.
258 ///
259 /// For conversion in the other direction, the serializer function is:
260 /// toJSON(const T&) -> json::Value
261 /// If this exists, then it also allows constructing Value from T, and can
262 /// be used to serialize vector<T>, map<string, T>, and Optional<T>.
263 ///
264 /// === Serialization ===
265 ///
266 /// Values can be serialized to JSON:
267 /// 1) raw_ostream << Value // Basic formatting.
268 /// 2) raw_ostream << formatv("{0}", Value) // Basic formatting.
269 /// 3) raw_ostream << formatv("{0:2}", Value) // Pretty-print with indent 2.
270 ///
271 /// And parsed:
272 /// Expected<Value> E = json::parse("[1, 2, null]");
273 /// assert(E && E->kind() == Value::Array);
274 class Value {
275 public:
276 enum Kind {
277 Null,
278 Boolean,
279 /// Number values can store both int64s and doubles at full precision,
280 /// depending on what they were constructed/parsed from.
281 Number,
282 String,
283 Array,
284 Object,
285 };
286
287 // It would be nice to have Value() be null. But that would make {} null too.
Value(const Value & M)288 Value(const Value &M) { copyFrom(M); }
Value(Value && M)289 Value(Value &&M) { moveFrom(std::move(M)); }
290 Value(std::initializer_list<Value> Elements);
Value(json::Array && Elements)291 Value(json::Array &&Elements) : Type(T_Array) {
292 create<json::Array>(std::move(Elements));
293 }
294 template <typename Elt>
Value(const std::vector<Elt> & C)295 Value(const std::vector<Elt> &C) : Value(json::Array(C)) {}
Value(json::Object && Properties)296 Value(json::Object &&Properties) : Type(T_Object) {
297 create<json::Object>(std::move(Properties));
298 }
299 template <typename Elt>
Value(const std::map<std::string,Elt> & C)300 Value(const std::map<std::string, Elt> &C) : Value(json::Object(C)) {}
301 // Strings: types with value semantics. Must be valid UTF-8.
Value(std::string V)302 Value(std::string V) : Type(T_String) {
303 if (LLVM_UNLIKELY(!isUTF8(V))) {
304 assert(false && "Invalid UTF-8 in value used as JSON");
305 V = fixUTF8(std::move(V));
306 }
307 create<std::string>(std::move(V));
308 }
Value(const llvm::SmallVectorImpl<char> & V)309 Value(const llvm::SmallVectorImpl<char> &V)
310 : Value(std::string(V.begin(), V.end())) {}
Value(const llvm::formatv_object_base & V)311 Value(const llvm::formatv_object_base &V) : Value(V.str()) {}
312 // Strings: types with reference semantics. Must be valid UTF-8.
Value(StringRef V)313 Value(StringRef V) : Type(T_StringRef) {
314 create<llvm::StringRef>(V);
315 if (LLVM_UNLIKELY(!isUTF8(V))) {
316 assert(false && "Invalid UTF-8 in value used as JSON");
317 *this = Value(fixUTF8(V));
318 }
319 }
Value(const char * V)320 Value(const char *V) : Value(StringRef(V)) {}
Value(std::nullptr_t)321 Value(std::nullptr_t) : Type(T_Null) {}
322 // Boolean (disallow implicit conversions).
323 // (The last template parameter is a dummy to keep templates distinct.)
324 template <
325 typename T,
326 typename = typename std::enable_if<std::is_same<T, bool>::value>::type,
327 bool = false>
Value(T B)328 Value(T B) : Type(T_Boolean) {
329 create<bool>(B);
330 }
331 // Integers (except boolean). Must be non-narrowing convertible to int64_t.
332 template <
333 typename T,
334 typename = typename std::enable_if<std::is_integral<T>::value>::type,
335 typename = typename std::enable_if<!std::is_same<T, bool>::value>::type>
Value(T I)336 Value(T I) : Type(T_Integer) {
337 create<int64_t>(int64_t{I});
338 }
339 // Floating point. Must be non-narrowing convertible to double.
340 template <typename T,
341 typename =
342 typename std::enable_if<std::is_floating_point<T>::value>::type,
343 double * = nullptr>
Value(T D)344 Value(T D) : Type(T_Double) {
345 create<double>(double{D});
346 }
347 // Serializable types: with a toJSON(const T&)->Value function, found by ADL.
348 template <typename T,
349 typename = typename std::enable_if<std::is_same<
350 Value, decltype(toJSON(*(const T *)nullptr))>::value>,
351 Value * = nullptr>
Value(const T & V)352 Value(const T &V) : Value(toJSON(V)) {}
353
354 Value &operator=(const Value &M) {
355 destroy();
356 copyFrom(M);
357 return *this;
358 }
359 Value &operator=(Value &&M) {
360 destroy();
361 moveFrom(std::move(M));
362 return *this;
363 }
~Value()364 ~Value() { destroy(); }
365
kind()366 Kind kind() const {
367 switch (Type) {
368 case T_Null:
369 return Null;
370 case T_Boolean:
371 return Boolean;
372 case T_Double:
373 case T_Integer:
374 return Number;
375 case T_String:
376 case T_StringRef:
377 return String;
378 case T_Object:
379 return Object;
380 case T_Array:
381 return Array;
382 }
383 llvm_unreachable("Unknown kind");
384 }
385
386 // Typed accessors return None/nullptr if the Value is not of this type.
getAsNull()387 llvm::Optional<std::nullptr_t> getAsNull() const {
388 if (LLVM_LIKELY(Type == T_Null))
389 return nullptr;
390 return llvm::None;
391 }
getAsBoolean()392 llvm::Optional<bool> getAsBoolean() const {
393 if (LLVM_LIKELY(Type == T_Boolean))
394 return as<bool>();
395 return llvm::None;
396 }
getAsNumber()397 llvm::Optional<double> getAsNumber() const {
398 if (LLVM_LIKELY(Type == T_Double))
399 return as<double>();
400 if (LLVM_LIKELY(Type == T_Integer))
401 return as<int64_t>();
402 return llvm::None;
403 }
404 // Succeeds if the Value is a Number, and exactly representable as int64_t.
getAsInteger()405 llvm::Optional<int64_t> getAsInteger() const {
406 if (LLVM_LIKELY(Type == T_Integer))
407 return as<int64_t>();
408 if (LLVM_LIKELY(Type == T_Double)) {
409 double D = as<double>();
410 if (LLVM_LIKELY(std::modf(D, &D) == 0.0 &&
411 D >= double(std::numeric_limits<int64_t>::min()) &&
412 D <= double(std::numeric_limits<int64_t>::max())))
413 return D;
414 }
415 return llvm::None;
416 }
getAsString()417 llvm::Optional<llvm::StringRef> getAsString() const {
418 if (Type == T_String)
419 return llvm::StringRef(as<std::string>());
420 if (LLVM_LIKELY(Type == T_StringRef))
421 return as<llvm::StringRef>();
422 return llvm::None;
423 }
getAsObject()424 const json::Object *getAsObject() const {
425 return LLVM_LIKELY(Type == T_Object) ? &as<json::Object>() : nullptr;
426 }
getAsObject()427 json::Object *getAsObject() {
428 return LLVM_LIKELY(Type == T_Object) ? &as<json::Object>() : nullptr;
429 }
getAsArray()430 const json::Array *getAsArray() const {
431 return LLVM_LIKELY(Type == T_Array) ? &as<json::Array>() : nullptr;
432 }
getAsArray()433 json::Array *getAsArray() {
434 return LLVM_LIKELY(Type == T_Array) ? &as<json::Array>() : nullptr;
435 }
436
437 private:
438 void destroy();
439 void copyFrom(const Value &M);
440 // We allow moving from *const* Values, by marking all members as mutable!
441 // This hack is needed to support initializer-list syntax efficiently.
442 // (std::initializer_list<T> is a container of const T).
443 void moveFrom(const Value &&M);
444 friend class Array;
445 friend class Object;
446
create(U &&...V)447 template <typename T, typename... U> void create(U &&... V) {
448 new (reinterpret_cast<T *>(Union.buffer)) T(std::forward<U>(V)...);
449 }
as()450 template <typename T> T &as() const {
451 // Using this two-step static_cast via void * instead of reinterpret_cast
452 // silences a -Wstrict-aliasing false positive from GCC6 and earlier.
453 void *Storage = static_cast<void *>(Union.buffer);
454 return *static_cast<T *>(Storage);
455 }
456
457 friend class OStream;
458
459 enum ValueType : char {
460 T_Null,
461 T_Boolean,
462 T_Double,
463 T_Integer,
464 T_StringRef,
465 T_String,
466 T_Object,
467 T_Array,
468 };
469 // All members mutable, see moveFrom().
470 mutable ValueType Type;
471 mutable llvm::AlignedCharArrayUnion<bool, double, int64_t, llvm::StringRef,
472 std::string, json::Array, json::Object>
473 Union;
474 friend bool operator==(const Value &, const Value &);
475 };
476
477 bool operator==(const Value &, const Value &);
478 inline bool operator!=(const Value &L, const Value &R) { return !(L == R); }
479
480 // Array Methods
481 inline Value &Array::operator[](size_t I) { return V[I]; }
482 inline const Value &Array::operator[](size_t I) const { return V[I]; }
front()483 inline Value &Array::front() { return V.front(); }
front()484 inline const Value &Array::front() const { return V.front(); }
back()485 inline Value &Array::back() { return V.back(); }
back()486 inline const Value &Array::back() const { return V.back(); }
data()487 inline Value *Array::data() { return V.data(); }
data()488 inline const Value *Array::data() const { return V.data(); }
489
begin()490 inline typename Array::iterator Array::begin() { return V.begin(); }
begin()491 inline typename Array::const_iterator Array::begin() const { return V.begin(); }
end()492 inline typename Array::iterator Array::end() { return V.end(); }
end()493 inline typename Array::const_iterator Array::end() const { return V.end(); }
494
empty()495 inline bool Array::empty() const { return V.empty(); }
size()496 inline size_t Array::size() const { return V.size(); }
reserve(size_t S)497 inline void Array::reserve(size_t S) { V.reserve(S); }
498
clear()499 inline void Array::clear() { V.clear(); }
push_back(const Value & E)500 inline void Array::push_back(const Value &E) { V.push_back(E); }
push_back(Value && E)501 inline void Array::push_back(Value &&E) { V.push_back(std::move(E)); }
emplace_back(Args &&...A)502 template <typename... Args> inline void Array::emplace_back(Args &&...A) {
503 V.emplace_back(std::forward<Args>(A)...);
504 }
pop_back()505 inline void Array::pop_back() { V.pop_back(); }
insert(iterator P,const Value & E)506 inline typename Array::iterator Array::insert(iterator P, const Value &E) {
507 return V.insert(P, E);
508 }
insert(iterator P,Value && E)509 inline typename Array::iterator Array::insert(iterator P, Value &&E) {
510 return V.insert(P, std::move(E));
511 }
512 template <typename It>
insert(iterator P,It A,It Z)513 inline typename Array::iterator Array::insert(iterator P, It A, It Z) {
514 return V.insert(P, A, Z);
515 }
516 template <typename... Args>
emplace(const_iterator P,Args &&...A)517 inline typename Array::iterator Array::emplace(const_iterator P, Args &&...A) {
518 return V.emplace(P, std::forward<Args>(A)...);
519 }
520 inline bool operator==(const Array &L, const Array &R) { return L.V == R.V; }
521
522 /// ObjectKey is a used to capture keys in Object. Like Value but:
523 /// - only strings are allowed
524 /// - it's optimized for the string literal case (Owned == nullptr)
525 /// Like Value, strings must be UTF-8. See isUTF8 documentation for details.
526 class ObjectKey {
527 public:
ObjectKey(const char * S)528 ObjectKey(const char *S) : ObjectKey(StringRef(S)) {}
ObjectKey(std::string S)529 ObjectKey(std::string S) : Owned(new std::string(std::move(S))) {
530 if (LLVM_UNLIKELY(!isUTF8(*Owned))) {
531 assert(false && "Invalid UTF-8 in value used as JSON");
532 *Owned = fixUTF8(std::move(*Owned));
533 }
534 Data = *Owned;
535 }
ObjectKey(llvm::StringRef S)536 ObjectKey(llvm::StringRef S) : Data(S) {
537 if (LLVM_UNLIKELY(!isUTF8(Data))) {
538 assert(false && "Invalid UTF-8 in value used as JSON");
539 *this = ObjectKey(fixUTF8(S));
540 }
541 }
ObjectKey(const llvm::SmallVectorImpl<char> & V)542 ObjectKey(const llvm::SmallVectorImpl<char> &V)
543 : ObjectKey(std::string(V.begin(), V.end())) {}
ObjectKey(const llvm::formatv_object_base & V)544 ObjectKey(const llvm::formatv_object_base &V) : ObjectKey(V.str()) {}
545
ObjectKey(const ObjectKey & C)546 ObjectKey(const ObjectKey &C) { *this = C; }
ObjectKey(ObjectKey && C)547 ObjectKey(ObjectKey &&C) : ObjectKey(static_cast<const ObjectKey &&>(C)) {}
548 ObjectKey &operator=(const ObjectKey &C) {
549 if (C.Owned) {
550 Owned.reset(new std::string(*C.Owned));
551 Data = *Owned;
552 } else {
553 Data = C.Data;
554 }
555 return *this;
556 }
557 ObjectKey &operator=(ObjectKey &&) = default;
558
StringRef()559 operator llvm::StringRef() const { return Data; }
str()560 std::string str() const { return Data.str(); }
561
562 private:
563 // FIXME: this is unneccesarily large (3 pointers). Pointer + length + owned
564 // could be 2 pointers at most.
565 std::unique_ptr<std::string> Owned;
566 llvm::StringRef Data;
567 };
568
569 inline bool operator==(const ObjectKey &L, const ObjectKey &R) {
570 return llvm::StringRef(L) == llvm::StringRef(R);
571 }
572 inline bool operator!=(const ObjectKey &L, const ObjectKey &R) {
573 return !(L == R);
574 }
575 inline bool operator<(const ObjectKey &L, const ObjectKey &R) {
576 return StringRef(L) < StringRef(R);
577 }
578
579 struct Object::KV {
580 ObjectKey K;
581 Value V;
582 };
583
Object(std::initializer_list<KV> Properties)584 inline Object::Object(std::initializer_list<KV> Properties) {
585 for (const auto &P : Properties) {
586 auto R = try_emplace(P.K, nullptr);
587 if (R.second)
588 R.first->getSecond().moveFrom(std::move(P.V));
589 }
590 }
insert(KV E)591 inline std::pair<Object::iterator, bool> Object::insert(KV E) {
592 return try_emplace(std::move(E.K), std::move(E.V));
593 }
erase(StringRef K)594 inline bool Object::erase(StringRef K) {
595 return M.erase(ObjectKey(K));
596 }
597
598 // Standard deserializers are provided for primitive types.
599 // See comments on Value.
fromJSON(const Value & E,std::string & Out)600 inline bool fromJSON(const Value &E, std::string &Out) {
601 if (auto S = E.getAsString()) {
602 Out = *S;
603 return true;
604 }
605 return false;
606 }
fromJSON(const Value & E,int & Out)607 inline bool fromJSON(const Value &E, int &Out) {
608 if (auto S = E.getAsInteger()) {
609 Out = *S;
610 return true;
611 }
612 return false;
613 }
fromJSON(const Value & E,int64_t & Out)614 inline bool fromJSON(const Value &E, int64_t &Out) {
615 if (auto S = E.getAsInteger()) {
616 Out = *S;
617 return true;
618 }
619 return false;
620 }
fromJSON(const Value & E,double & Out)621 inline bool fromJSON(const Value &E, double &Out) {
622 if (auto S = E.getAsNumber()) {
623 Out = *S;
624 return true;
625 }
626 return false;
627 }
fromJSON(const Value & E,bool & Out)628 inline bool fromJSON(const Value &E, bool &Out) {
629 if (auto S = E.getAsBoolean()) {
630 Out = *S;
631 return true;
632 }
633 return false;
634 }
fromJSON(const Value & E,llvm::Optional<T> & Out)635 template <typename T> bool fromJSON(const Value &E, llvm::Optional<T> &Out) {
636 if (E.getAsNull()) {
637 Out = llvm::None;
638 return true;
639 }
640 T Result;
641 if (!fromJSON(E, Result))
642 return false;
643 Out = std::move(Result);
644 return true;
645 }
fromJSON(const Value & E,std::vector<T> & Out)646 template <typename T> bool fromJSON(const Value &E, std::vector<T> &Out) {
647 if (auto *A = E.getAsArray()) {
648 Out.clear();
649 Out.resize(A->size());
650 for (size_t I = 0; I < A->size(); ++I)
651 if (!fromJSON((*A)[I], Out[I]))
652 return false;
653 return true;
654 }
655 return false;
656 }
657 template <typename T>
fromJSON(const Value & E,std::map<std::string,T> & Out)658 bool fromJSON(const Value &E, std::map<std::string, T> &Out) {
659 if (auto *O = E.getAsObject()) {
660 Out.clear();
661 for (const auto &KV : *O)
662 if (!fromJSON(KV.second, Out[llvm::StringRef(KV.first)]))
663 return false;
664 return true;
665 }
666 return false;
667 }
668
669 // Allow serialization of Optional<T> for supported T.
toJSON(const llvm::Optional<T> & Opt)670 template <typename T> Value toJSON(const llvm::Optional<T> &Opt) {
671 return Opt ? Value(*Opt) : Value(nullptr);
672 }
673
674 /// Helper for mapping JSON objects onto protocol structs.
675 ///
676 /// Example:
677 /// \code
678 /// bool fromJSON(const Value &E, MyStruct &R) {
679 /// ObjectMapper O(E);
680 /// if (!O || !O.map("mandatory_field", R.MandatoryField))
681 /// return false;
682 /// O.map("optional_field", R.OptionalField);
683 /// return true;
684 /// }
685 /// \endcode
686 class ObjectMapper {
687 public:
ObjectMapper(const Value & E)688 ObjectMapper(const Value &E) : O(E.getAsObject()) {}
689
690 /// True if the expression is an object.
691 /// Must be checked before calling map().
692 operator bool() { return O; }
693
694 /// Maps a property to a field, if it exists.
map(StringRef Prop,T & Out)695 template <typename T> bool map(StringRef Prop, T &Out) {
696 assert(*this && "Must check this is an object before calling map()");
697 if (const Value *E = O->get(Prop))
698 return fromJSON(*E, Out);
699 return false;
700 }
701
702 /// Maps a property to a field, if it exists.
703 /// (Optional requires special handling, because missing keys are OK).
map(StringRef Prop,llvm::Optional<T> & Out)704 template <typename T> bool map(StringRef Prop, llvm::Optional<T> &Out) {
705 assert(*this && "Must check this is an object before calling map()");
706 if (const Value *E = O->get(Prop))
707 return fromJSON(*E, Out);
708 Out = llvm::None;
709 return true;
710 }
711
712 private:
713 const Object *O;
714 };
715
716 /// Parses the provided JSON source, or returns a ParseError.
717 /// The returned Value is self-contained and owns its strings (they do not refer
718 /// to the original source).
719 llvm::Expected<Value> parse(llvm::StringRef JSON);
720
721 class ParseError : public llvm::ErrorInfo<ParseError> {
722 const char *Msg;
723 unsigned Line, Column, Offset;
724
725 public:
726 static char ID;
ParseError(const char * Msg,unsigned Line,unsigned Column,unsigned Offset)727 ParseError(const char *Msg, unsigned Line, unsigned Column, unsigned Offset)
728 : Msg(Msg), Line(Line), Column(Column), Offset(Offset) {}
log(llvm::raw_ostream & OS)729 void log(llvm::raw_ostream &OS) const override {
730 OS << llvm::formatv("[{0}:{1}, byte={2}]: {3}", Line, Column, Offset, Msg);
731 }
convertToErrorCode()732 std::error_code convertToErrorCode() const override {
733 return llvm::inconvertibleErrorCode();
734 }
735 };
736
737 /// json::OStream allows writing well-formed JSON without materializing
738 /// all structures as json::Value ahead of time.
739 /// It's faster, lower-level, and less safe than OS << json::Value.
740 ///
741 /// Only one "top-level" object can be written to a stream.
742 /// Simplest usage involves passing lambdas (Blocks) to fill in containers:
743 ///
744 /// json::OStream J(OS);
745 /// J.array([&]{
746 /// for (const Event &E : Events)
747 /// J.object([&] {
748 /// J.attribute("timestamp", int64_t(E.Time));
749 /// J.attributeArray("participants", [&] {
750 /// for (const Participant &P : E.Participants)
751 /// J.value(P.toString());
752 /// });
753 /// });
754 /// });
755 ///
756 /// This would produce JSON like:
757 ///
758 /// [
759 /// {
760 /// "timestamp": 19287398741,
761 /// "participants": [
762 /// "King Kong",
763 /// "Miley Cyrus",
764 /// "Cleopatra"
765 /// ]
766 /// },
767 /// ...
768 /// ]
769 ///
770 /// The lower level begin/end methods (arrayBegin()) are more flexible but
771 /// care must be taken to pair them correctly:
772 ///
773 /// json::OStream J(OS);
774 // J.arrayBegin();
775 /// for (const Event &E : Events) {
776 /// J.objectBegin();
777 /// J.attribute("timestamp", int64_t(E.Time));
778 /// J.attributeBegin("participants");
779 /// for (const Participant &P : E.Participants)
780 /// J.value(P.toString());
781 /// J.attributeEnd();
782 /// J.objectEnd();
783 /// }
784 /// J.arrayEnd();
785 ///
786 /// If the call sequence isn't valid JSON, asserts will fire in debug mode.
787 /// This can be mismatched begin()/end() pairs, trying to emit attributes inside
788 /// an array, and so on.
789 /// With asserts disabled, this is undefined behavior.
790 class OStream {
791 public:
792 using Block = llvm::function_ref<void()>;
793 // If IndentSize is nonzero, output is pretty-printed.
794 explicit OStream(llvm::raw_ostream &OS, unsigned IndentSize = 0)
OS(OS)795 : OS(OS), IndentSize(IndentSize) {
796 Stack.emplace_back();
797 }
~OStream()798 ~OStream() {
799 assert(Stack.size() == 1 && "Unmatched begin()/end()");
800 assert(Stack.back().Ctx == Singleton);
801 assert(Stack.back().HasValue && "Did not write top-level value");
802 }
803
804 /// Flushes the underlying ostream. OStream does not buffer internally.
flush()805 void flush() { OS.flush(); }
806
807 // High level functions to output a value.
808 // Valid at top-level (exactly once), in an attribute value (exactly once),
809 // or in an array (any number of times).
810
811 /// Emit a self-contained value (number, string, vector<string> etc).
812 void value(const Value &V);
813 /// Emit an array whose elements are emitted in the provided Block.
array(Block Contents)814 void array(Block Contents) {
815 arrayBegin();
816 Contents();
817 arrayEnd();
818 }
819 /// Emit an object whose elements are emitted in the provided Block.
object(Block Contents)820 void object(Block Contents) {
821 objectBegin();
822 Contents();
823 objectEnd();
824 }
825
826 // High level functions to output object attributes.
827 // Valid only within an object (any number of times).
828
829 /// Emit an attribute whose value is self-contained (number, vector<int> etc).
attribute(llvm::StringRef Key,const Value & Contents)830 void attribute(llvm::StringRef Key, const Value& Contents) {
831 attributeImpl(Key, [&] { value(Contents); });
832 }
833 /// Emit an attribute whose value is an array with elements from the Block.
attributeArray(llvm::StringRef Key,Block Contents)834 void attributeArray(llvm::StringRef Key, Block Contents) {
835 attributeImpl(Key, [&] { array(Contents); });
836 }
837 /// Emit an attribute whose value is an object with attributes from the Block.
attributeObject(llvm::StringRef Key,Block Contents)838 void attributeObject(llvm::StringRef Key, Block Contents) {
839 attributeImpl(Key, [&] { object(Contents); });
840 }
841
842 // Low-level begin/end functions to output arrays, objects, and attributes.
843 // Must be correctly paired. Allowed contexts are as above.
844
845 void arrayBegin();
846 void arrayEnd();
847 void objectBegin();
848 void objectEnd();
849 void attributeBegin(llvm::StringRef Key);
850 void attributeEnd();
851
852 private:
attributeImpl(llvm::StringRef Key,Block Contents)853 void attributeImpl(llvm::StringRef Key, Block Contents) {
854 attributeBegin(Key);
855 Contents();
856 attributeEnd();
857 }
858
859 void valueBegin();
860 void newline();
861
862 enum Context {
863 Singleton, // Top level, or object attribute.
864 Array,
865 Object,
866 };
867 struct State {
868 Context Ctx = Singleton;
869 bool HasValue = false;
870 };
871 llvm::SmallVector<State, 16> Stack; // Never empty.
872 llvm::raw_ostream &OS;
873 unsigned IndentSize;
874 unsigned Indent = 0;
875 };
876
877 /// Serializes this Value to JSON, writing it to the provided stream.
878 /// The formatting is compact (no extra whitespace) and deterministic.
879 /// For pretty-printing, use the formatv() format_provider below.
880 inline llvm::raw_ostream &operator<<(llvm::raw_ostream &OS, const Value &V) {
881 OStream(OS).value(V);
882 return OS;
883 }
884 } // namespace json
885
886 /// Allow printing json::Value with formatv().
887 /// The default style is basic/compact formatting, like operator<<.
888 /// A format string like formatv("{0:2}", Value) pretty-prints with indent 2.
889 template <> struct format_provider<llvm::json::Value> {
890 static void format(const llvm::json::Value &, raw_ostream &, StringRef);
891 };
892 } // namespace llvm
893
894 #endif
895