1 /*
2 * Copyright 2017 Google Inc. All rights reserved.
3 *
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at
7 *
8 * http://www.apache.org/licenses/LICENSE-2.0
9 *
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
15 */
16
17 #ifndef FLATBUFFERS_FLEXBUFFERS_H_
18 #define FLATBUFFERS_FLEXBUFFERS_H_
19
20 #include <algorithm>
21 #include <map>
22 // Used to select STL variant.
23 #include "flatbuffers/base.h"
24 // We use the basic binary writing functions from the regular FlatBuffers.
25 #include "flatbuffers/util.h"
26
27 #ifdef _MSC_VER
28 # include <intrin.h>
29 #endif
30
31 #if defined(_MSC_VER)
32 # pragma warning(push)
33 # pragma warning(disable : 4127) // C4127: conditional expression is constant
34 #endif
35
36 namespace flexbuffers {
37
38 class Reference;
39 class Map;
40
41 // These are used in the lower 2 bits of a type field to determine the size of
42 // the elements (and or size field) of the item pointed to (e.g. vector).
43 enum BitWidth {
44 BIT_WIDTH_8 = 0,
45 BIT_WIDTH_16 = 1,
46 BIT_WIDTH_32 = 2,
47 BIT_WIDTH_64 = 3,
48 };
49
50 // These are used as the upper 6 bits of a type field to indicate the actual
51 // type.
52 enum Type {
53 FBT_NULL = 0,
54 FBT_INT = 1,
55 FBT_UINT = 2,
56 FBT_FLOAT = 3,
57 // Types above stored inline, types below (except FBT_BOOL) store an offset.
58 FBT_KEY = 4,
59 FBT_STRING = 5,
60 FBT_INDIRECT_INT = 6,
61 FBT_INDIRECT_UINT = 7,
62 FBT_INDIRECT_FLOAT = 8,
63 FBT_MAP = 9,
64 FBT_VECTOR = 10, // Untyped.
65 FBT_VECTOR_INT = 11, // Typed any size (stores no type table).
66 FBT_VECTOR_UINT = 12,
67 FBT_VECTOR_FLOAT = 13,
68 FBT_VECTOR_KEY = 14,
69 // DEPRECATED, use FBT_VECTOR or FBT_VECTOR_KEY instead.
70 // Read test.cpp/FlexBuffersDeprecatedTest() for details on why.
71 FBT_VECTOR_STRING_DEPRECATED = 15,
72 FBT_VECTOR_INT2 = 16, // Typed tuple (no type table, no size field).
73 FBT_VECTOR_UINT2 = 17,
74 FBT_VECTOR_FLOAT2 = 18,
75 FBT_VECTOR_INT3 = 19, // Typed triple (no type table, no size field).
76 FBT_VECTOR_UINT3 = 20,
77 FBT_VECTOR_FLOAT3 = 21,
78 FBT_VECTOR_INT4 = 22, // Typed quad (no type table, no size field).
79 FBT_VECTOR_UINT4 = 23,
80 FBT_VECTOR_FLOAT4 = 24,
81 FBT_BLOB = 25,
82 FBT_BOOL = 26,
83 FBT_VECTOR_BOOL =
84 36, // To Allow the same type of conversion of type to vector type
85
86 FBT_MAX_TYPE = 37
87 };
88
IsInline(Type t)89 inline bool IsInline(Type t) { return t <= FBT_FLOAT || t == FBT_BOOL; }
90
IsTypedVectorElementType(Type t)91 inline bool IsTypedVectorElementType(Type t) {
92 return (t >= FBT_INT && t <= FBT_STRING) || t == FBT_BOOL;
93 }
94
IsTypedVector(Type t)95 inline bool IsTypedVector(Type t) {
96 return (t >= FBT_VECTOR_INT && t <= FBT_VECTOR_STRING_DEPRECATED) ||
97 t == FBT_VECTOR_BOOL;
98 }
99
IsFixedTypedVector(Type t)100 inline bool IsFixedTypedVector(Type t) {
101 return t >= FBT_VECTOR_INT2 && t <= FBT_VECTOR_FLOAT4;
102 }
103
104 inline Type ToTypedVector(Type t, size_t fixed_len = 0) {
105 FLATBUFFERS_ASSERT(IsTypedVectorElementType(t));
106 switch (fixed_len) {
107 case 0: return static_cast<Type>(t - FBT_INT + FBT_VECTOR_INT);
108 case 2: return static_cast<Type>(t - FBT_INT + FBT_VECTOR_INT2);
109 case 3: return static_cast<Type>(t - FBT_INT + FBT_VECTOR_INT3);
110 case 4: return static_cast<Type>(t - FBT_INT + FBT_VECTOR_INT4);
111 default: FLATBUFFERS_ASSERT(0); return FBT_NULL;
112 }
113 }
114
ToTypedVectorElementType(Type t)115 inline Type ToTypedVectorElementType(Type t) {
116 FLATBUFFERS_ASSERT(IsTypedVector(t));
117 return static_cast<Type>(t - FBT_VECTOR_INT + FBT_INT);
118 }
119
ToFixedTypedVectorElementType(Type t,uint8_t * len)120 inline Type ToFixedTypedVectorElementType(Type t, uint8_t *len) {
121 FLATBUFFERS_ASSERT(IsFixedTypedVector(t));
122 auto fixed_type = t - FBT_VECTOR_INT2;
123 *len = static_cast<uint8_t>(fixed_type / 3 +
124 2); // 3 types each, starting from length 2.
125 return static_cast<Type>(fixed_type % 3 + FBT_INT);
126 }
127
128 // TODO: implement proper support for 8/16bit floats, or decide not to
129 // support them.
130 typedef int16_t half;
131 typedef int8_t quarter;
132
133 // TODO: can we do this without conditionals using intrinsics or inline asm
134 // on some platforms? Given branch prediction the method below should be
135 // decently quick, but it is the most frequently executed function.
136 // We could do an (unaligned) 64-bit read if we ifdef out the platforms for
137 // which that doesn't work (or where we'd read into un-owned memory).
138 template<typename R, typename T1, typename T2, typename T4, typename T8>
ReadSizedScalar(const uint8_t * data,uint8_t byte_width)139 R ReadSizedScalar(const uint8_t *data, uint8_t byte_width) {
140 return byte_width < 4
141 ? (byte_width < 2
142 ? static_cast<R>(flatbuffers::ReadScalar<T1>(data))
143 : static_cast<R>(flatbuffers::ReadScalar<T2>(data)))
144 : (byte_width < 8
145 ? static_cast<R>(flatbuffers::ReadScalar<T4>(data))
146 : static_cast<R>(flatbuffers::ReadScalar<T8>(data)));
147 }
148
ReadInt64(const uint8_t * data,uint8_t byte_width)149 inline int64_t ReadInt64(const uint8_t *data, uint8_t byte_width) {
150 return ReadSizedScalar<int64_t, int8_t, int16_t, int32_t, int64_t>(
151 data, byte_width);
152 }
153
ReadUInt64(const uint8_t * data,uint8_t byte_width)154 inline uint64_t ReadUInt64(const uint8_t *data, uint8_t byte_width) {
155 // This is the "hottest" function (all offset lookups use this), so worth
156 // optimizing if possible.
157 // TODO: GCC apparently replaces memcpy by a rep movsb, but only if count is a
158 // constant, which here it isn't. Test if memcpy is still faster than
159 // the conditionals in ReadSizedScalar. Can also use inline asm.
160
161 // clang-format off
162 #if defined(_MSC_VER) && defined(_M_X64) && !defined(_M_ARM64EC)
163 // This is 64-bit Windows only, __movsb does not work on 32-bit Windows.
164 uint64_t u = 0;
165 __movsb(reinterpret_cast<uint8_t *>(&u),
166 reinterpret_cast<const uint8_t *>(data), byte_width);
167 return flatbuffers::EndianScalar(u);
168 #else
169 return ReadSizedScalar<uint64_t, uint8_t, uint16_t, uint32_t, uint64_t>(
170 data, byte_width);
171 #endif
172 // clang-format on
173 }
174
ReadDouble(const uint8_t * data,uint8_t byte_width)175 inline double ReadDouble(const uint8_t *data, uint8_t byte_width) {
176 return ReadSizedScalar<double, quarter, half, float, double>(data,
177 byte_width);
178 }
179
Indirect(const uint8_t * offset,uint8_t byte_width)180 inline const uint8_t *Indirect(const uint8_t *offset, uint8_t byte_width) {
181 return offset - ReadUInt64(offset, byte_width);
182 }
183
Indirect(const uint8_t * offset)184 template<typename T> const uint8_t *Indirect(const uint8_t *offset) {
185 return offset - flatbuffers::ReadScalar<T>(offset);
186 }
187
WidthU(uint64_t u)188 inline BitWidth WidthU(uint64_t u) {
189 #define FLATBUFFERS_GET_FIELD_BIT_WIDTH(value, width) \
190 { \
191 if (!((u) & ~((1ULL << (width)) - 1ULL))) return BIT_WIDTH_##width; \
192 }
193 FLATBUFFERS_GET_FIELD_BIT_WIDTH(u, 8);
194 FLATBUFFERS_GET_FIELD_BIT_WIDTH(u, 16);
195 FLATBUFFERS_GET_FIELD_BIT_WIDTH(u, 32);
196 #undef FLATBUFFERS_GET_FIELD_BIT_WIDTH
197 return BIT_WIDTH_64;
198 }
199
WidthI(int64_t i)200 inline BitWidth WidthI(int64_t i) {
201 auto u = static_cast<uint64_t>(i) << 1;
202 return WidthU(i >= 0 ? u : ~u);
203 }
204
WidthF(double f)205 inline BitWidth WidthF(double f) {
206 return static_cast<double>(static_cast<float>(f)) == f ? BIT_WIDTH_32
207 : BIT_WIDTH_64;
208 }
209
210 // Base class of all types below.
211 // Points into the data buffer and allows access to one type.
212 class Object {
213 public:
Object(const uint8_t * data,uint8_t byte_width)214 Object(const uint8_t *data, uint8_t byte_width)
215 : data_(data), byte_width_(byte_width) {}
216
217 protected:
218 const uint8_t *data_;
219 uint8_t byte_width_;
220 };
221
222 // Object that has a size, obtained either from size prefix, or elsewhere.
223 class Sized : public Object {
224 public:
225 // Size prefix.
Sized(const uint8_t * data,uint8_t byte_width)226 Sized(const uint8_t *data, uint8_t byte_width)
227 : Object(data, byte_width), size_(read_size()) {}
228 // Manual size.
Sized(const uint8_t * data,uint8_t byte_width,size_t sz)229 Sized(const uint8_t *data, uint8_t byte_width, size_t sz)
230 : Object(data, byte_width), size_(sz) {}
size()231 size_t size() const { return size_; }
232 // Access size stored in `byte_width_` bytes before data_ pointer.
read_size()233 size_t read_size() const {
234 return static_cast<size_t>(ReadUInt64(data_ - byte_width_, byte_width_));
235 }
236
237 protected:
238 size_t size_;
239 };
240
241 class String : public Sized {
242 public:
243 // Size prefix.
String(const uint8_t * data,uint8_t byte_width)244 String(const uint8_t *data, uint8_t byte_width) : Sized(data, byte_width) {}
245 // Manual size.
String(const uint8_t * data,uint8_t byte_width,size_t sz)246 String(const uint8_t *data, uint8_t byte_width, size_t sz)
247 : Sized(data, byte_width, sz) {}
248
length()249 size_t length() const { return size(); }
c_str()250 const char *c_str() const { return reinterpret_cast<const char *>(data_); }
str()251 std::string str() const { return std::string(c_str(), size()); }
252
EmptyString()253 static String EmptyString() {
254 static const char *empty_string = "";
255 return String(reinterpret_cast<const uint8_t *>(empty_string), 1, 0);
256 }
IsTheEmptyString()257 bool IsTheEmptyString() const { return data_ == EmptyString().data_; }
258 };
259
260 class Blob : public Sized {
261 public:
Blob(const uint8_t * data_buf,uint8_t byte_width)262 Blob(const uint8_t *data_buf, uint8_t byte_width)
263 : Sized(data_buf, byte_width) {}
264
EmptyBlob()265 static Blob EmptyBlob() {
266 static const uint8_t empty_blob[] = { 0 /*len*/ };
267 return Blob(empty_blob + 1, 1);
268 }
IsTheEmptyBlob()269 bool IsTheEmptyBlob() const { return data_ == EmptyBlob().data_; }
data()270 const uint8_t *data() const { return data_; }
271 };
272
273 class Vector : public Sized {
274 public:
Vector(const uint8_t * data,uint8_t byte_width)275 Vector(const uint8_t *data, uint8_t byte_width) : Sized(data, byte_width) {}
276
277 Reference operator[](size_t i) const;
278
EmptyVector()279 static Vector EmptyVector() {
280 static const uint8_t empty_vector[] = { 0 /*len*/ };
281 return Vector(empty_vector + 1, 1);
282 }
IsTheEmptyVector()283 bool IsTheEmptyVector() const { return data_ == EmptyVector().data_; }
284 };
285
286 class TypedVector : public Sized {
287 public:
TypedVector(const uint8_t * data,uint8_t byte_width,Type element_type)288 TypedVector(const uint8_t *data, uint8_t byte_width, Type element_type)
289 : Sized(data, byte_width), type_(element_type) {}
290
291 Reference operator[](size_t i) const;
292
EmptyTypedVector()293 static TypedVector EmptyTypedVector() {
294 static const uint8_t empty_typed_vector[] = { 0 /*len*/ };
295 return TypedVector(empty_typed_vector + 1, 1, FBT_INT);
296 }
IsTheEmptyVector()297 bool IsTheEmptyVector() const {
298 return data_ == TypedVector::EmptyTypedVector().data_;
299 }
300
ElementType()301 Type ElementType() { return type_; }
302
303 friend Reference;
304
305 private:
306 Type type_;
307
308 friend Map;
309 };
310
311 class FixedTypedVector : public Object {
312 public:
FixedTypedVector(const uint8_t * data,uint8_t byte_width,Type element_type,uint8_t len)313 FixedTypedVector(const uint8_t *data, uint8_t byte_width, Type element_type,
314 uint8_t len)
315 : Object(data, byte_width), type_(element_type), len_(len) {}
316
317 Reference operator[](size_t i) const;
318
EmptyFixedTypedVector()319 static FixedTypedVector EmptyFixedTypedVector() {
320 static const uint8_t fixed_empty_vector[] = { 0 /* unused */ };
321 return FixedTypedVector(fixed_empty_vector, 1, FBT_INT, 0);
322 }
IsTheEmptyFixedTypedVector()323 bool IsTheEmptyFixedTypedVector() const {
324 return data_ == FixedTypedVector::EmptyFixedTypedVector().data_;
325 }
326
ElementType()327 Type ElementType() const { return type_; }
size()328 uint8_t size() const { return len_; }
329
330 private:
331 Type type_;
332 uint8_t len_;
333 };
334
335 class Map : public Vector {
336 public:
Map(const uint8_t * data,uint8_t byte_width)337 Map(const uint8_t *data, uint8_t byte_width) : Vector(data, byte_width) {}
338
339 Reference operator[](const char *key) const;
340 Reference operator[](const std::string &key) const;
341
Values()342 Vector Values() const { return Vector(data_, byte_width_); }
343
Keys()344 TypedVector Keys() const {
345 const size_t num_prefixed_fields = 3;
346 auto keys_offset = data_ - byte_width_ * num_prefixed_fields;
347 return TypedVector(Indirect(keys_offset, byte_width_),
348 static_cast<uint8_t>(
349 ReadUInt64(keys_offset + byte_width_, byte_width_)),
350 FBT_KEY);
351 }
352
EmptyMap()353 static Map EmptyMap() {
354 static const uint8_t empty_map[] = {
355 0 /*keys_len*/, 0 /*keys_offset*/, 1 /*keys_width*/, 0 /*len*/
356 };
357 return Map(empty_map + 4, 1);
358 }
359
IsTheEmptyMap()360 bool IsTheEmptyMap() const { return data_ == EmptyMap().data_; }
361 };
362
IndentString(std::string & s,int indent,const char * indent_string)363 inline void IndentString(std::string &s, int indent,
364 const char *indent_string) {
365 for (int i = 0; i < indent; i++) s += indent_string;
366 }
367
368 template<typename T>
AppendToString(std::string & s,T && v,bool keys_quoted,bool indented,int cur_indent,const char * indent_string)369 void AppendToString(std::string &s, T &&v, bool keys_quoted, bool indented,
370 int cur_indent, const char *indent_string) {
371 s += "[";
372 s += indented ? "\n" : " ";
373 for (size_t i = 0; i < v.size(); i++) {
374 if (i) {
375 s += ",";
376 s += indented ? "\n" : " ";
377 }
378 if (indented) IndentString(s, cur_indent, indent_string);
379 v[i].ToString(true, keys_quoted, s, indented, cur_indent,
380 indent_string);
381 }
382 if (indented) {
383 s += "\n";
384 IndentString(s, cur_indent - 1, indent_string);
385 } else {
386 s += " ";
387 }
388 s += "]";
389 }
390
391 template<typename T>
AppendToString(std::string & s,T && v,bool keys_quoted)392 void AppendToString(std::string &s, T &&v, bool keys_quoted) {
393 AppendToString(s, v, keys_quoted);
394 }
395
396
397 class Reference {
398 public:
Reference()399 Reference()
400 : data_(nullptr), parent_width_(0), byte_width_(0), type_(FBT_NULL) {}
401
Reference(const uint8_t * data,uint8_t parent_width,uint8_t byte_width,Type type)402 Reference(const uint8_t *data, uint8_t parent_width, uint8_t byte_width,
403 Type type)
404 : data_(data),
405 parent_width_(parent_width),
406 byte_width_(byte_width),
407 type_(type) {}
408
Reference(const uint8_t * data,uint8_t parent_width,uint8_t packed_type)409 Reference(const uint8_t *data, uint8_t parent_width, uint8_t packed_type)
410 : data_(data),
411 parent_width_(parent_width),
412 byte_width_(static_cast<uint8_t>(1 << (packed_type & 3))),
413 type_(static_cast<Type>(packed_type >> 2)) {}
414
GetType()415 Type GetType() const { return type_; }
416
IsNull()417 bool IsNull() const { return type_ == FBT_NULL; }
IsBool()418 bool IsBool() const { return type_ == FBT_BOOL; }
IsInt()419 bool IsInt() const { return type_ == FBT_INT || type_ == FBT_INDIRECT_INT; }
IsUInt()420 bool IsUInt() const {
421 return type_ == FBT_UINT || type_ == FBT_INDIRECT_UINT;
422 }
IsIntOrUint()423 bool IsIntOrUint() const { return IsInt() || IsUInt(); }
IsFloat()424 bool IsFloat() const {
425 return type_ == FBT_FLOAT || type_ == FBT_INDIRECT_FLOAT;
426 }
IsNumeric()427 bool IsNumeric() const { return IsIntOrUint() || IsFloat(); }
IsString()428 bool IsString() const { return type_ == FBT_STRING; }
IsKey()429 bool IsKey() const { return type_ == FBT_KEY; }
IsVector()430 bool IsVector() const { return type_ == FBT_VECTOR || type_ == FBT_MAP; }
IsUntypedVector()431 bool IsUntypedVector() const { return type_ == FBT_VECTOR; }
IsTypedVector()432 bool IsTypedVector() const { return flexbuffers::IsTypedVector(type_); }
IsFixedTypedVector()433 bool IsFixedTypedVector() const {
434 return flexbuffers::IsFixedTypedVector(type_);
435 }
IsAnyVector()436 bool IsAnyVector() const {
437 return (IsTypedVector() || IsFixedTypedVector() || IsVector());
438 }
IsMap()439 bool IsMap() const { return type_ == FBT_MAP; }
IsBlob()440 bool IsBlob() const { return type_ == FBT_BLOB; }
AsBool()441 bool AsBool() const {
442 return (type_ == FBT_BOOL ? ReadUInt64(data_, parent_width_)
443 : AsUInt64()) != 0;
444 }
445
446 // Reads any type as a int64_t. Never fails, does most sensible conversion.
447 // Truncates floats, strings are attempted to be parsed for a number,
448 // vectors/maps return their size. Returns 0 if all else fails.
AsInt64()449 int64_t AsInt64() const {
450 if (type_ == FBT_INT) {
451 // A fast path for the common case.
452 return ReadInt64(data_, parent_width_);
453 } else
454 switch (type_) {
455 case FBT_INDIRECT_INT: return ReadInt64(Indirect(), byte_width_);
456 case FBT_UINT: return ReadUInt64(data_, parent_width_);
457 case FBT_INDIRECT_UINT: return ReadUInt64(Indirect(), byte_width_);
458 case FBT_FLOAT:
459 return static_cast<int64_t>(ReadDouble(data_, parent_width_));
460 case FBT_INDIRECT_FLOAT:
461 return static_cast<int64_t>(ReadDouble(Indirect(), byte_width_));
462 case FBT_NULL: return 0;
463 case FBT_STRING: return flatbuffers::StringToInt(AsString().c_str());
464 case FBT_VECTOR: return static_cast<int64_t>(AsVector().size());
465 case FBT_BOOL: return ReadInt64(data_, parent_width_);
466 default:
467 // Convert other things to int.
468 return 0;
469 }
470 }
471
472 // TODO: could specialize these to not use AsInt64() if that saves
473 // extension ops in generated code, and use a faster op than ReadInt64.
AsInt32()474 int32_t AsInt32() const { return static_cast<int32_t>(AsInt64()); }
AsInt16()475 int16_t AsInt16() const { return static_cast<int16_t>(AsInt64()); }
AsInt8()476 int8_t AsInt8() const { return static_cast<int8_t>(AsInt64()); }
477
AsUInt64()478 uint64_t AsUInt64() const {
479 if (type_ == FBT_UINT) {
480 // A fast path for the common case.
481 return ReadUInt64(data_, parent_width_);
482 } else
483 switch (type_) {
484 case FBT_INDIRECT_UINT: return ReadUInt64(Indirect(), byte_width_);
485 case FBT_INT: return ReadInt64(data_, parent_width_);
486 case FBT_INDIRECT_INT: return ReadInt64(Indirect(), byte_width_);
487 case FBT_FLOAT:
488 return static_cast<uint64_t>(ReadDouble(data_, parent_width_));
489 case FBT_INDIRECT_FLOAT:
490 return static_cast<uint64_t>(ReadDouble(Indirect(), byte_width_));
491 case FBT_NULL: return 0;
492 case FBT_STRING: return flatbuffers::StringToUInt(AsString().c_str());
493 case FBT_VECTOR: return static_cast<uint64_t>(AsVector().size());
494 case FBT_BOOL: return ReadUInt64(data_, parent_width_);
495 default:
496 // Convert other things to uint.
497 return 0;
498 }
499 }
500
AsUInt32()501 uint32_t AsUInt32() const { return static_cast<uint32_t>(AsUInt64()); }
AsUInt16()502 uint16_t AsUInt16() const { return static_cast<uint16_t>(AsUInt64()); }
AsUInt8()503 uint8_t AsUInt8() const { return static_cast<uint8_t>(AsUInt64()); }
504
AsDouble()505 double AsDouble() const {
506 if (type_ == FBT_FLOAT) {
507 // A fast path for the common case.
508 return ReadDouble(data_, parent_width_);
509 } else
510 switch (type_) {
511 case FBT_INDIRECT_FLOAT: return ReadDouble(Indirect(), byte_width_);
512 case FBT_INT:
513 return static_cast<double>(ReadInt64(data_, parent_width_));
514 case FBT_UINT:
515 return static_cast<double>(ReadUInt64(data_, parent_width_));
516 case FBT_INDIRECT_INT:
517 return static_cast<double>(ReadInt64(Indirect(), byte_width_));
518 case FBT_INDIRECT_UINT:
519 return static_cast<double>(ReadUInt64(Indirect(), byte_width_));
520 case FBT_NULL: return 0.0;
521 case FBT_STRING: {
522 double d;
523 flatbuffers::StringToNumber(AsString().c_str(), &d);
524 return d;
525 }
526 case FBT_VECTOR: return static_cast<double>(AsVector().size());
527 case FBT_BOOL:
528 return static_cast<double>(ReadUInt64(data_, parent_width_));
529 default:
530 // Convert strings and other things to float.
531 return 0;
532 }
533 }
534
AsFloat()535 float AsFloat() const { return static_cast<float>(AsDouble()); }
536
AsKey()537 const char *AsKey() const {
538 if (type_ == FBT_KEY || type_ == FBT_STRING) {
539 return reinterpret_cast<const char *>(Indirect());
540 } else {
541 return "";
542 }
543 }
544
545 // This function returns the empty string if you try to read something that
546 // is not a string or key.
AsString()547 String AsString() const {
548 if (type_ == FBT_STRING) {
549 return String(Indirect(), byte_width_);
550 } else if (type_ == FBT_KEY) {
551 auto key = Indirect();
552 return String(key, byte_width_,
553 strlen(reinterpret_cast<const char *>(key)));
554 } else {
555 return String::EmptyString();
556 }
557 }
558
559 // Unlike AsString(), this will convert any type to a std::string.
ToString()560 std::string ToString() const {
561 std::string s;
562 ToString(false, false, s);
563 return s;
564 }
565
566 // Convert any type to a JSON-like string. strings_quoted determines if
567 // string values at the top level receive "" quotes (inside other values
568 // they always do). keys_quoted determines if keys are quoted, at any level.
ToString(bool strings_quoted,bool keys_quoted,std::string & s)569 void ToString(bool strings_quoted, bool keys_quoted, std::string &s) const {
570 ToString(strings_quoted, keys_quoted, s, false, 0, "");
571 }
572
573 // This version additionally allow you to specify if you want indentation.
ToString(bool strings_quoted,bool keys_quoted,std::string & s,bool indented,int cur_indent,const char * indent_string)574 void ToString(bool strings_quoted, bool keys_quoted, std::string &s,
575 bool indented, int cur_indent, const char *indent_string) const {
576 if (type_ == FBT_STRING) {
577 String str(Indirect(), byte_width_);
578 if (strings_quoted) {
579 flatbuffers::EscapeString(str.c_str(), str.length(), &s, true, false);
580 } else {
581 s.append(str.c_str(), str.length());
582 }
583 } else if (IsKey()) {
584 auto str = AsKey();
585 if (keys_quoted) {
586 flatbuffers::EscapeString(str, strlen(str), &s, true, false);
587 } else {
588 s += str;
589 }
590 } else if (IsInt()) {
591 s += flatbuffers::NumToString(AsInt64());
592 } else if (IsUInt()) {
593 s += flatbuffers::NumToString(AsUInt64());
594 } else if (IsFloat()) {
595 s += flatbuffers::NumToString(AsDouble());
596 } else if (IsNull()) {
597 s += "null";
598 } else if (IsBool()) {
599 s += AsBool() ? "true" : "false";
600 } else if (IsMap()) {
601 s += "{";
602 s += indented ? "\n" : " ";
603 auto m = AsMap();
604 auto keys = m.Keys();
605 auto vals = m.Values();
606 for (size_t i = 0; i < keys.size(); i++) {
607 bool kq = keys_quoted;
608 if (!kq) {
609 // FlexBuffers keys may contain arbitrary characters, only allow
610 // unquoted if it looks like an "identifier":
611 const char *p = keys[i].AsKey();
612 if (!flatbuffers::is_alpha(*p) && *p != '_') {
613 kq = true;
614 } else {
615 while (*++p) {
616 if (!flatbuffers::is_alnum(*p) && *p != '_') {
617 kq = true;
618 break;
619 }
620 }
621 }
622 }
623 if (indented) IndentString(s, cur_indent + 1, indent_string);
624 keys[i].ToString(true, kq, s);
625 s += ": ";
626 vals[i].ToString(true, keys_quoted, s, indented, cur_indent + 1, indent_string);
627 if (i < keys.size() - 1) {
628 s += ",";
629 if (!indented) s += " ";
630 }
631 if (indented) s += "\n";
632 }
633 if (!indented) s += " ";
634 if (indented) IndentString(s, cur_indent, indent_string);
635 s += "}";
636 } else if (IsVector()) {
637 AppendToString<Vector>(s, AsVector(), keys_quoted, indented,
638 cur_indent + 1, indent_string);
639 } else if (IsTypedVector()) {
640 AppendToString<TypedVector>(s, AsTypedVector(), keys_quoted, indented,
641 cur_indent + 1, indent_string);
642 } else if (IsFixedTypedVector()) {
643 AppendToString<FixedTypedVector>(s, AsFixedTypedVector(), keys_quoted,
644 indented, cur_indent + 1, indent_string);
645 } else if (IsBlob()) {
646 auto blob = AsBlob();
647 flatbuffers::EscapeString(reinterpret_cast<const char *>(blob.data()),
648 blob.size(), &s, true, false);
649 } else {
650 s += "(?)";
651 }
652 }
653
654 // This function returns the empty blob if you try to read a not-blob.
655 // Strings can be viewed as blobs too.
AsBlob()656 Blob AsBlob() const {
657 if (type_ == FBT_BLOB || type_ == FBT_STRING) {
658 return Blob(Indirect(), byte_width_);
659 } else {
660 return Blob::EmptyBlob();
661 }
662 }
663
664 // This function returns the empty vector if you try to read a not-vector.
665 // Maps can be viewed as vectors too.
AsVector()666 Vector AsVector() const {
667 if (type_ == FBT_VECTOR || type_ == FBT_MAP) {
668 return Vector(Indirect(), byte_width_);
669 } else {
670 return Vector::EmptyVector();
671 }
672 }
673
AsTypedVector()674 TypedVector AsTypedVector() const {
675 if (IsTypedVector()) {
676 auto tv =
677 TypedVector(Indirect(), byte_width_, ToTypedVectorElementType(type_));
678 if (tv.type_ == FBT_STRING) {
679 // These can't be accessed as strings, since we don't know the bit-width
680 // of the size field, see the declaration of
681 // FBT_VECTOR_STRING_DEPRECATED above for details.
682 // We change the type here to be keys, which are a subtype of strings,
683 // and will ignore the size field. This will truncate strings with
684 // embedded nulls.
685 tv.type_ = FBT_KEY;
686 }
687 return tv;
688 } else {
689 return TypedVector::EmptyTypedVector();
690 }
691 }
692
AsFixedTypedVector()693 FixedTypedVector AsFixedTypedVector() const {
694 if (IsFixedTypedVector()) {
695 uint8_t len = 0;
696 auto vtype = ToFixedTypedVectorElementType(type_, &len);
697 return FixedTypedVector(Indirect(), byte_width_, vtype, len);
698 } else {
699 return FixedTypedVector::EmptyFixedTypedVector();
700 }
701 }
702
AsMap()703 Map AsMap() const {
704 if (type_ == FBT_MAP) {
705 return Map(Indirect(), byte_width_);
706 } else {
707 return Map::EmptyMap();
708 }
709 }
710
711 template<typename T> T As() const;
712
713 // Experimental: Mutation functions.
714 // These allow scalars in an already created buffer to be updated in-place.
715 // Since by default scalars are stored in the smallest possible space,
716 // the new value may not fit, in which case these functions return false.
717 // To avoid this, you can construct the values you intend to mutate using
718 // Builder::ForceMinimumBitWidth.
MutateInt(int64_t i)719 bool MutateInt(int64_t i) {
720 if (type_ == FBT_INT) {
721 return Mutate(data_, i, parent_width_, WidthI(i));
722 } else if (type_ == FBT_INDIRECT_INT) {
723 return Mutate(Indirect(), i, byte_width_, WidthI(i));
724 } else if (type_ == FBT_UINT) {
725 auto u = static_cast<uint64_t>(i);
726 return Mutate(data_, u, parent_width_, WidthU(u));
727 } else if (type_ == FBT_INDIRECT_UINT) {
728 auto u = static_cast<uint64_t>(i);
729 return Mutate(Indirect(), u, byte_width_, WidthU(u));
730 } else {
731 return false;
732 }
733 }
734
MutateBool(bool b)735 bool MutateBool(bool b) {
736 return type_ == FBT_BOOL && Mutate(data_, b, parent_width_, BIT_WIDTH_8);
737 }
738
MutateUInt(uint64_t u)739 bool MutateUInt(uint64_t u) {
740 if (type_ == FBT_UINT) {
741 return Mutate(data_, u, parent_width_, WidthU(u));
742 } else if (type_ == FBT_INDIRECT_UINT) {
743 return Mutate(Indirect(), u, byte_width_, WidthU(u));
744 } else if (type_ == FBT_INT) {
745 auto i = static_cast<int64_t>(u);
746 return Mutate(data_, i, parent_width_, WidthI(i));
747 } else if (type_ == FBT_INDIRECT_INT) {
748 auto i = static_cast<int64_t>(u);
749 return Mutate(Indirect(), i, byte_width_, WidthI(i));
750 } else {
751 return false;
752 }
753 }
754
MutateFloat(float f)755 bool MutateFloat(float f) {
756 if (type_ == FBT_FLOAT) {
757 return MutateF(data_, f, parent_width_, BIT_WIDTH_32);
758 } else if (type_ == FBT_INDIRECT_FLOAT) {
759 return MutateF(Indirect(), f, byte_width_, BIT_WIDTH_32);
760 } else {
761 return false;
762 }
763 }
764
MutateFloat(double d)765 bool MutateFloat(double d) {
766 if (type_ == FBT_FLOAT) {
767 return MutateF(data_, d, parent_width_, WidthF(d));
768 } else if (type_ == FBT_INDIRECT_FLOAT) {
769 return MutateF(Indirect(), d, byte_width_, WidthF(d));
770 } else {
771 return false;
772 }
773 }
774
MutateString(const char * str,size_t len)775 bool MutateString(const char *str, size_t len) {
776 auto s = AsString();
777 if (s.IsTheEmptyString()) return false;
778 // This is very strict, could allow shorter strings, but that creates
779 // garbage.
780 if (s.length() != len) return false;
781 memcpy(const_cast<char *>(s.c_str()), str, len);
782 return true;
783 }
MutateString(const char * str)784 bool MutateString(const char *str) { return MutateString(str, strlen(str)); }
MutateString(const std::string & str)785 bool MutateString(const std::string &str) {
786 return MutateString(str.data(), str.length());
787 }
788
789 private:
Indirect()790 const uint8_t *Indirect() const {
791 return flexbuffers::Indirect(data_, parent_width_);
792 }
793
794 template<typename T>
Mutate(const uint8_t * dest,T t,size_t byte_width,BitWidth value_width)795 bool Mutate(const uint8_t *dest, T t, size_t byte_width,
796 BitWidth value_width) {
797 auto fits = static_cast<size_t>(static_cast<size_t>(1U) << value_width) <=
798 byte_width;
799 if (fits) {
800 t = flatbuffers::EndianScalar(t);
801 memcpy(const_cast<uint8_t *>(dest), &t, byte_width);
802 }
803 return fits;
804 }
805
806 template<typename T>
MutateF(const uint8_t * dest,T t,size_t byte_width,BitWidth value_width)807 bool MutateF(const uint8_t *dest, T t, size_t byte_width,
808 BitWidth value_width) {
809 if (byte_width == sizeof(double))
810 return Mutate(dest, static_cast<double>(t), byte_width, value_width);
811 if (byte_width == sizeof(float))
812 return Mutate(dest, static_cast<float>(t), byte_width, value_width);
813 FLATBUFFERS_ASSERT(false);
814 return false;
815 }
816
817 friend class Verifier;
818
819 const uint8_t *data_;
820 uint8_t parent_width_;
821 uint8_t byte_width_;
822 Type type_;
823 };
824
825 // Template specialization for As().
826 template<> inline bool Reference::As<bool>() const { return AsBool(); }
827
828 template<> inline int8_t Reference::As<int8_t>() const { return AsInt8(); }
829 template<> inline int16_t Reference::As<int16_t>() const { return AsInt16(); }
830 template<> inline int32_t Reference::As<int32_t>() const { return AsInt32(); }
831 template<> inline int64_t Reference::As<int64_t>() const { return AsInt64(); }
832
833 template<> inline uint8_t Reference::As<uint8_t>() const { return AsUInt8(); }
834 template<> inline uint16_t Reference::As<uint16_t>() const {
835 return AsUInt16();
836 }
837 template<> inline uint32_t Reference::As<uint32_t>() const {
838 return AsUInt32();
839 }
840 template<> inline uint64_t Reference::As<uint64_t>() const {
841 return AsUInt64();
842 }
843
844 template<> inline double Reference::As<double>() const { return AsDouble(); }
845 template<> inline float Reference::As<float>() const { return AsFloat(); }
846
847 template<> inline String Reference::As<String>() const { return AsString(); }
848 template<> inline std::string Reference::As<std::string>() const {
849 return AsString().str();
850 }
851
852 template<> inline Blob Reference::As<Blob>() const { return AsBlob(); }
853 template<> inline Vector Reference::As<Vector>() const { return AsVector(); }
854 template<> inline TypedVector Reference::As<TypedVector>() const {
855 return AsTypedVector();
856 }
857 template<> inline FixedTypedVector Reference::As<FixedTypedVector>() const {
858 return AsFixedTypedVector();
859 }
860 template<> inline Map Reference::As<Map>() const { return AsMap(); }
861
PackedType(BitWidth bit_width,Type type)862 inline uint8_t PackedType(BitWidth bit_width, Type type) {
863 return static_cast<uint8_t>(bit_width | (type << 2));
864 }
865
NullPackedType()866 inline uint8_t NullPackedType() { return PackedType(BIT_WIDTH_8, FBT_NULL); }
867
868 // Vector accessors.
869 // Note: if you try to access outside of bounds, you get a Null value back
870 // instead. Normally this would be an assert, but since this is "dynamically
871 // typed" data, you may not want that (someone sends you a 2d vector and you
872 // wanted 3d).
873 // The Null converts seamlessly into a default value for any other type.
874 // TODO(wvo): Could introduce an #ifdef that makes this into an assert?
875 inline Reference Vector::operator[](size_t i) const {
876 auto len = size();
877 if (i >= len) return Reference(nullptr, 1, NullPackedType());
878 auto packed_type = (data_ + len * byte_width_)[i];
879 auto elem = data_ + i * byte_width_;
880 return Reference(elem, byte_width_, packed_type);
881 }
882
883 inline Reference TypedVector::operator[](size_t i) const {
884 auto len = size();
885 if (i >= len) return Reference(nullptr, 1, NullPackedType());
886 auto elem = data_ + i * byte_width_;
887 return Reference(elem, byte_width_, 1, type_);
888 }
889
890 inline Reference FixedTypedVector::operator[](size_t i) const {
891 if (i >= len_) return Reference(nullptr, 1, NullPackedType());
892 auto elem = data_ + i * byte_width_;
893 return Reference(elem, byte_width_, 1, type_);
894 }
895
KeyCompare(const void * key,const void * elem)896 template<typename T> int KeyCompare(const void *key, const void *elem) {
897 auto str_elem = reinterpret_cast<const char *>(
898 Indirect<T>(reinterpret_cast<const uint8_t *>(elem)));
899 auto skey = reinterpret_cast<const char *>(key);
900 return strcmp(skey, str_elem);
901 }
902
903 inline Reference Map::operator[](const char *key) const {
904 auto keys = Keys();
905 // We can't pass keys.byte_width_ to the comparison function, so we have
906 // to pick the right one ahead of time.
907 int (*comp)(const void *, const void *) = nullptr;
908 switch (keys.byte_width_) {
909 case 1: comp = KeyCompare<uint8_t>; break;
910 case 2: comp = KeyCompare<uint16_t>; break;
911 case 4: comp = KeyCompare<uint32_t>; break;
912 case 8: comp = KeyCompare<uint64_t>; break;
913 default: FLATBUFFERS_ASSERT(false); return Reference();
914 }
915 auto res = std::bsearch(key, keys.data_, keys.size(), keys.byte_width_, comp);
916 if (!res) return Reference(nullptr, 1, NullPackedType());
917 auto i = (reinterpret_cast<uint8_t *>(res) - keys.data_) / keys.byte_width_;
918 return (*static_cast<const Vector *>(this))[i];
919 }
920
921 inline Reference Map::operator[](const std::string &key) const {
922 return (*this)[key.c_str()];
923 }
924
GetRoot(const uint8_t * buffer,size_t size)925 inline Reference GetRoot(const uint8_t *buffer, size_t size) {
926 // See Finish() below for the serialization counterpart of this.
927 // The root starts at the end of the buffer, so we parse backwards from there.
928 auto end = buffer + size;
929 auto byte_width = *--end;
930 auto packed_type = *--end;
931 end -= byte_width; // The root data item.
932 return Reference(end, byte_width, packed_type);
933 }
934
GetRoot(const std::vector<uint8_t> & buffer)935 inline Reference GetRoot(const std::vector<uint8_t> &buffer) {
936 return GetRoot(buffer.data(), buffer.size());
937 }
938
939 // Flags that configure how the Builder behaves.
940 // The "Share" flags determine if the Builder automatically tries to pool
941 // this type. Pooling can reduce the size of serialized data if there are
942 // multiple maps of the same kind, at the expense of slightly slower
943 // serialization (the cost of lookups) and more memory use (std::set).
944 // By default this is on for keys, but off for strings.
945 // Turn keys off if you have e.g. only one map.
946 // Turn strings on if you expect many non-unique string values.
947 // Additionally, sharing key vectors can save space if you have maps with
948 // identical field populations.
949 enum BuilderFlag {
950 BUILDER_FLAG_NONE = 0,
951 BUILDER_FLAG_SHARE_KEYS = 1,
952 BUILDER_FLAG_SHARE_STRINGS = 2,
953 BUILDER_FLAG_SHARE_KEYS_AND_STRINGS = 3,
954 BUILDER_FLAG_SHARE_KEY_VECTORS = 4,
955 BUILDER_FLAG_SHARE_ALL = 7,
956 };
957
958 class Builder FLATBUFFERS_FINAL_CLASS {
959 public:
960 Builder(size_t initial_size = 256,
961 BuilderFlag flags = BUILDER_FLAG_SHARE_KEYS)
buf_(initial_size)962 : buf_(initial_size),
963 finished_(false),
964 has_duplicate_keys_(false),
965 flags_(flags),
966 force_min_bit_width_(BIT_WIDTH_8),
967 key_pool(KeyOffsetCompare(buf_)),
968 string_pool(StringOffsetCompare(buf_)) {
969 buf_.clear();
970 }
971
972 #ifdef FLATBUFFERS_DEFAULT_DECLARATION
973 Builder(Builder &&) = default;
974 Builder &operator=(Builder &&) = default;
975 #endif
976
977 /// @brief Get the serialized buffer (after you call `Finish()`).
978 /// @return Returns a vector owned by this class.
GetBuffer()979 const std::vector<uint8_t> &GetBuffer() const {
980 Finished();
981 return buf_;
982 }
983
984 // Size of the buffer. Does not include unfinished values.
GetSize()985 size_t GetSize() const { return buf_.size(); }
986
987 // Reset all state so we can re-use the buffer.
Clear()988 void Clear() {
989 buf_.clear();
990 stack_.clear();
991 finished_ = false;
992 // flags_ remains as-is;
993 force_min_bit_width_ = BIT_WIDTH_8;
994 key_pool.clear();
995 string_pool.clear();
996 }
997
998 // All value constructing functions below have two versions: one that
999 // takes a key (for placement inside a map) and one that doesn't (for inside
1000 // vectors and elsewhere).
1001
Null()1002 void Null() { stack_.push_back(Value()); }
Null(const char * key)1003 void Null(const char *key) {
1004 Key(key);
1005 Null();
1006 }
1007
Int(int64_t i)1008 void Int(int64_t i) { stack_.push_back(Value(i, FBT_INT, WidthI(i))); }
Int(const char * key,int64_t i)1009 void Int(const char *key, int64_t i) {
1010 Key(key);
1011 Int(i);
1012 }
1013
UInt(uint64_t u)1014 void UInt(uint64_t u) { stack_.push_back(Value(u, FBT_UINT, WidthU(u))); }
UInt(const char * key,uint64_t u)1015 void UInt(const char *key, uint64_t u) {
1016 Key(key);
1017 UInt(u);
1018 }
1019
Float(float f)1020 void Float(float f) { stack_.push_back(Value(f)); }
Float(const char * key,float f)1021 void Float(const char *key, float f) {
1022 Key(key);
1023 Float(f);
1024 }
1025
Double(double f)1026 void Double(double f) { stack_.push_back(Value(f)); }
Double(const char * key,double d)1027 void Double(const char *key, double d) {
1028 Key(key);
1029 Double(d);
1030 }
1031
Bool(bool b)1032 void Bool(bool b) { stack_.push_back(Value(b)); }
Bool(const char * key,bool b)1033 void Bool(const char *key, bool b) {
1034 Key(key);
1035 Bool(b);
1036 }
1037
IndirectInt(int64_t i)1038 void IndirectInt(int64_t i) { PushIndirect(i, FBT_INDIRECT_INT, WidthI(i)); }
IndirectInt(const char * key,int64_t i)1039 void IndirectInt(const char *key, int64_t i) {
1040 Key(key);
1041 IndirectInt(i);
1042 }
1043
IndirectUInt(uint64_t u)1044 void IndirectUInt(uint64_t u) {
1045 PushIndirect(u, FBT_INDIRECT_UINT, WidthU(u));
1046 }
IndirectUInt(const char * key,uint64_t u)1047 void IndirectUInt(const char *key, uint64_t u) {
1048 Key(key);
1049 IndirectUInt(u);
1050 }
1051
IndirectFloat(float f)1052 void IndirectFloat(float f) {
1053 PushIndirect(f, FBT_INDIRECT_FLOAT, BIT_WIDTH_32);
1054 }
IndirectFloat(const char * key,float f)1055 void IndirectFloat(const char *key, float f) {
1056 Key(key);
1057 IndirectFloat(f);
1058 }
1059
IndirectDouble(double f)1060 void IndirectDouble(double f) {
1061 PushIndirect(f, FBT_INDIRECT_FLOAT, WidthF(f));
1062 }
IndirectDouble(const char * key,double d)1063 void IndirectDouble(const char *key, double d) {
1064 Key(key);
1065 IndirectDouble(d);
1066 }
1067
Key(const char * str,size_t len)1068 size_t Key(const char *str, size_t len) {
1069 auto sloc = buf_.size();
1070 WriteBytes(str, len + 1);
1071 if (flags_ & BUILDER_FLAG_SHARE_KEYS) {
1072 auto it = key_pool.find(sloc);
1073 if (it != key_pool.end()) {
1074 // Already in the buffer. Remove key we just serialized, and use
1075 // existing offset instead.
1076 buf_.resize(sloc);
1077 sloc = *it;
1078 } else {
1079 key_pool.insert(sloc);
1080 }
1081 }
1082 stack_.push_back(Value(static_cast<uint64_t>(sloc), FBT_KEY, BIT_WIDTH_8));
1083 return sloc;
1084 }
1085
Key(const char * str)1086 size_t Key(const char *str) { return Key(str, strlen(str)); }
Key(const std::string & str)1087 size_t Key(const std::string &str) { return Key(str.c_str(), str.size()); }
1088
String(const char * str,size_t len)1089 size_t String(const char *str, size_t len) {
1090 auto reset_to = buf_.size();
1091 auto sloc = CreateBlob(str, len, 1, FBT_STRING);
1092 if (flags_ & BUILDER_FLAG_SHARE_STRINGS) {
1093 StringOffset so(sloc, len);
1094 auto it = string_pool.find(so);
1095 if (it != string_pool.end()) {
1096 // Already in the buffer. Remove string we just serialized, and use
1097 // existing offset instead.
1098 buf_.resize(reset_to);
1099 sloc = it->first;
1100 stack_.back().u_ = sloc;
1101 } else {
1102 string_pool.insert(so);
1103 }
1104 }
1105 return sloc;
1106 }
String(const char * str)1107 size_t String(const char *str) { return String(str, strlen(str)); }
String(const std::string & str)1108 size_t String(const std::string &str) {
1109 return String(str.c_str(), str.size());
1110 }
String(const flexbuffers::String & str)1111 void String(const flexbuffers::String &str) {
1112 String(str.c_str(), str.length());
1113 }
1114
String(const char * key,const char * str)1115 void String(const char *key, const char *str) {
1116 Key(key);
1117 String(str);
1118 }
String(const char * key,const std::string & str)1119 void String(const char *key, const std::string &str) {
1120 Key(key);
1121 String(str);
1122 }
String(const char * key,const flexbuffers::String & str)1123 void String(const char *key, const flexbuffers::String &str) {
1124 Key(key);
1125 String(str);
1126 }
1127
Blob(const void * data,size_t len)1128 size_t Blob(const void *data, size_t len) {
1129 return CreateBlob(data, len, 0, FBT_BLOB);
1130 }
Blob(const std::vector<uint8_t> & v)1131 size_t Blob(const std::vector<uint8_t> &v) {
1132 return CreateBlob(v.data(), v.size(), 0, FBT_BLOB);
1133 }
1134
Blob(const char * key,const void * data,size_t len)1135 void Blob(const char *key, const void *data, size_t len) {
1136 Key(key);
1137 Blob(data, len);
1138 }
Blob(const char * key,const std::vector<uint8_t> & v)1139 void Blob(const char *key, const std::vector<uint8_t> &v) {
1140 Key(key);
1141 Blob(v);
1142 }
1143
1144 // TODO(wvo): support all the FlexBuffer types (like flexbuffers::String),
1145 // e.g. Vector etc. Also in overloaded versions.
1146 // Also some FlatBuffers types?
1147
StartVector()1148 size_t StartVector() { return stack_.size(); }
StartVector(const char * key)1149 size_t StartVector(const char *key) {
1150 Key(key);
1151 return stack_.size();
1152 }
StartMap()1153 size_t StartMap() { return stack_.size(); }
StartMap(const char * key)1154 size_t StartMap(const char *key) {
1155 Key(key);
1156 return stack_.size();
1157 }
1158
1159 // TODO(wvo): allow this to specify an alignment greater than the natural
1160 // alignment.
EndVector(size_t start,bool typed,bool fixed)1161 size_t EndVector(size_t start, bool typed, bool fixed) {
1162 auto vec = CreateVector(start, stack_.size() - start, 1, typed, fixed);
1163 // Remove temp elements and return vector.
1164 stack_.resize(start);
1165 stack_.push_back(vec);
1166 return static_cast<size_t>(vec.u_);
1167 }
1168
EndMap(size_t start)1169 size_t EndMap(size_t start) {
1170 // We should have interleaved keys and values on the stack.
1171 auto len = MapElementCount(start);
1172 // Make sure keys are all strings:
1173 for (auto key = start; key < stack_.size(); key += 2) {
1174 FLATBUFFERS_ASSERT(stack_[key].type_ == FBT_KEY);
1175 }
1176 // Now sort values, so later we can do a binary search lookup.
1177 // We want to sort 2 array elements at a time.
1178 struct TwoValue {
1179 Value key;
1180 Value val;
1181 };
1182 // TODO(wvo): strict aliasing?
1183 // TODO(wvo): allow the caller to indicate the data is already sorted
1184 // for maximum efficiency? With an assert to check sortedness to make sure
1185 // we're not breaking binary search.
1186 // Or, we can track if the map is sorted as keys are added which would be
1187 // be quite cheap (cheaper than checking it here), so we can skip this
1188 // step automatically when appliccable, and encourage people to write in
1189 // sorted fashion.
1190 // std::sort is typically already a lot faster on sorted data though.
1191 auto dict = reinterpret_cast<TwoValue *>(stack_.data() + start);
1192 std::sort(
1193 dict, dict + len, [&](const TwoValue &a, const TwoValue &b) -> bool {
1194 auto as = reinterpret_cast<const char *>(buf_.data() + a.key.u_);
1195 auto bs = reinterpret_cast<const char *>(buf_.data() + b.key.u_);
1196 auto comp = strcmp(as, bs);
1197 // We want to disallow duplicate keys, since this results in a
1198 // map where values cannot be found.
1199 // But we can't assert here (since we don't want to fail on
1200 // random JSON input) or have an error mechanism.
1201 // Instead, we set has_duplicate_keys_ in the builder to
1202 // signal this.
1203 // TODO: Have to check for pointer equality, as some sort
1204 // implementation apparently call this function with the same
1205 // element?? Why?
1206 if (!comp && &a != &b) has_duplicate_keys_ = true;
1207 return comp < 0;
1208 });
1209 // First create a vector out of all keys.
1210 // TODO(wvo): if kBuilderFlagShareKeyVectors is true, see if we can share
1211 // the first vector.
1212 auto keys = CreateVector(start, len, 2, true, false);
1213 auto vec = CreateVector(start + 1, len, 2, false, false, &keys);
1214 // Remove temp elements and return map.
1215 stack_.resize(start);
1216 stack_.push_back(vec);
1217 return static_cast<size_t>(vec.u_);
1218 }
1219
1220 // Call this after EndMap to see if the map had any duplicate keys.
1221 // Any map with such keys won't be able to retrieve all values.
HasDuplicateKeys()1222 bool HasDuplicateKeys() const { return has_duplicate_keys_; }
1223
Vector(F f)1224 template<typename F> size_t Vector(F f) {
1225 auto start = StartVector();
1226 f();
1227 return EndVector(start, false, false);
1228 }
Vector(F f,T & state)1229 template<typename F, typename T> size_t Vector(F f, T &state) {
1230 auto start = StartVector();
1231 f(state);
1232 return EndVector(start, false, false);
1233 }
Vector(const char * key,F f)1234 template<typename F> size_t Vector(const char *key, F f) {
1235 auto start = StartVector(key);
1236 f();
1237 return EndVector(start, false, false);
1238 }
1239 template<typename F, typename T>
Vector(const char * key,F f,T & state)1240 size_t Vector(const char *key, F f, T &state) {
1241 auto start = StartVector(key);
1242 f(state);
1243 return EndVector(start, false, false);
1244 }
1245
Vector(const T * elems,size_t len)1246 template<typename T> void Vector(const T *elems, size_t len) {
1247 if (flatbuffers::is_scalar<T>::value) {
1248 // This path should be a lot quicker and use less space.
1249 ScalarVector(elems, len, false);
1250 } else {
1251 auto start = StartVector();
1252 for (size_t i = 0; i < len; i++) Add(elems[i]);
1253 EndVector(start, false, false);
1254 }
1255 }
1256 template<typename T>
Vector(const char * key,const T * elems,size_t len)1257 void Vector(const char *key, const T *elems, size_t len) {
1258 Key(key);
1259 Vector(elems, len);
1260 }
Vector(const std::vector<T> & vec)1261 template<typename T> void Vector(const std::vector<T> &vec) {
1262 Vector(vec.data(), vec.size());
1263 }
1264
TypedVector(F f)1265 template<typename F> size_t TypedVector(F f) {
1266 auto start = StartVector();
1267 f();
1268 return EndVector(start, true, false);
1269 }
TypedVector(F f,T & state)1270 template<typename F, typename T> size_t TypedVector(F f, T &state) {
1271 auto start = StartVector();
1272 f(state);
1273 return EndVector(start, true, false);
1274 }
TypedVector(const char * key,F f)1275 template<typename F> size_t TypedVector(const char *key, F f) {
1276 auto start = StartVector(key);
1277 f();
1278 return EndVector(start, true, false);
1279 }
1280 template<typename F, typename T>
TypedVector(const char * key,F f,T & state)1281 size_t TypedVector(const char *key, F f, T &state) {
1282 auto start = StartVector(key);
1283 f(state);
1284 return EndVector(start, true, false);
1285 }
1286
FixedTypedVector(const T * elems,size_t len)1287 template<typename T> size_t FixedTypedVector(const T *elems, size_t len) {
1288 // We only support a few fixed vector lengths. Anything bigger use a
1289 // regular typed vector.
1290 FLATBUFFERS_ASSERT(len >= 2 && len <= 4);
1291 // And only scalar values.
1292 static_assert(flatbuffers::is_scalar<T>::value, "Unrelated types");
1293 return ScalarVector(elems, len, true);
1294 }
1295
1296 template<typename T>
FixedTypedVector(const char * key,const T * elems,size_t len)1297 size_t FixedTypedVector(const char *key, const T *elems, size_t len) {
1298 Key(key);
1299 return FixedTypedVector(elems, len);
1300 }
1301
Map(F f)1302 template<typename F> size_t Map(F f) {
1303 auto start = StartMap();
1304 f();
1305 return EndMap(start);
1306 }
Map(F f,T & state)1307 template<typename F, typename T> size_t Map(F f, T &state) {
1308 auto start = StartMap();
1309 f(state);
1310 return EndMap(start);
1311 }
Map(const char * key,F f)1312 template<typename F> size_t Map(const char *key, F f) {
1313 auto start = StartMap(key);
1314 f();
1315 return EndMap(start);
1316 }
Map(const char * key,F f,T & state)1317 template<typename F, typename T> size_t Map(const char *key, F f, T &state) {
1318 auto start = StartMap(key);
1319 f(state);
1320 return EndMap(start);
1321 }
Map(const std::map<std::string,T> & map)1322 template<typename T> void Map(const std::map<std::string, T> &map) {
1323 auto start = StartMap();
1324 for (auto it = map.begin(); it != map.end(); ++it)
1325 Add(it->first.c_str(), it->second);
1326 EndMap(start);
1327 }
1328
MapElementCount(size_t start)1329 size_t MapElementCount(size_t start) {
1330 // Make sure it is an even number:
1331 auto len = stack_.size() - start;
1332 FLATBUFFERS_ASSERT(!(len & 1));
1333 len /= 2;
1334 return len;
1335 }
1336
1337 // If you wish to share a value explicitly (a value not shared automatically
1338 // through one of the BUILDER_FLAG_SHARE_* flags) you can do so with these
1339 // functions. Or if you wish to turn those flags off for performance reasons
1340 // and still do some explicit sharing. For example:
1341 // builder.IndirectDouble(M_PI);
1342 // auto id = builder.LastValue(); // Remember where we stored it.
1343 // .. more code goes here ..
1344 // builder.ReuseValue(id); // Refers to same double by offset.
1345 // LastValue works regardless of whether the value has a key or not.
1346 // Works on any data type.
1347 struct Value;
LastValue()1348 Value LastValue() { return stack_.back(); }
ReuseValue(Value v)1349 void ReuseValue(Value v) { stack_.push_back(v); }
ReuseValue(const char * key,Value v)1350 void ReuseValue(const char *key, Value v) {
1351 Key(key);
1352 ReuseValue(v);
1353 }
1354
1355 // Undo the last element serialized. Call once for a value and once for a
1356 // key.
Undo()1357 void Undo() {
1358 stack_.pop_back();
1359 }
1360
1361 // Overloaded Add that tries to call the correct function above.
Add(int8_t i)1362 void Add(int8_t i) { Int(i); }
Add(int16_t i)1363 void Add(int16_t i) { Int(i); }
Add(int32_t i)1364 void Add(int32_t i) { Int(i); }
Add(int64_t i)1365 void Add(int64_t i) { Int(i); }
Add(uint8_t u)1366 void Add(uint8_t u) { UInt(u); }
Add(uint16_t u)1367 void Add(uint16_t u) { UInt(u); }
Add(uint32_t u)1368 void Add(uint32_t u) { UInt(u); }
Add(uint64_t u)1369 void Add(uint64_t u) { UInt(u); }
Add(float f)1370 void Add(float f) { Float(f); }
Add(double d)1371 void Add(double d) { Double(d); }
Add(bool b)1372 void Add(bool b) { Bool(b); }
Add(const char * str)1373 void Add(const char *str) { String(str); }
Add(const std::string & str)1374 void Add(const std::string &str) { String(str); }
Add(const flexbuffers::String & str)1375 void Add(const flexbuffers::String &str) { String(str); }
1376
Add(const std::vector<T> & vec)1377 template<typename T> void Add(const std::vector<T> &vec) { Vector(vec); }
1378
Add(const char * key,const T & t)1379 template<typename T> void Add(const char *key, const T &t) {
1380 Key(key);
1381 Add(t);
1382 }
1383
Add(const std::map<std::string,T> & map)1384 template<typename T> void Add(const std::map<std::string, T> &map) {
1385 Map(map);
1386 }
1387
1388 template<typename T> void operator+=(const T &t) { Add(t); }
1389
1390 // This function is useful in combination with the Mutate* functions above.
1391 // It forces elements of vectors and maps to have a minimum size, such that
1392 // they can later be updated without failing.
1393 // Call with no arguments to reset.
1394 void ForceMinimumBitWidth(BitWidth bw = BIT_WIDTH_8) {
1395 force_min_bit_width_ = bw;
1396 }
1397
Finish()1398 void Finish() {
1399 // If you hit this assert, you likely have objects that were never included
1400 // in a parent. You need to have exactly one root to finish a buffer.
1401 // Check your Start/End calls are matched, and all objects are inside
1402 // some other object.
1403 FLATBUFFERS_ASSERT(stack_.size() == 1);
1404
1405 // Write root value.
1406 auto byte_width = Align(stack_[0].ElemWidth(buf_.size(), 0));
1407 WriteAny(stack_[0], byte_width);
1408 // Write root type.
1409 Write(stack_[0].StoredPackedType(), 1);
1410 // Write root size. Normally determined by parent, but root has no parent :)
1411 Write(byte_width, 1);
1412
1413 finished_ = true;
1414 }
1415
1416 private:
Finished()1417 void Finished() const {
1418 // If you get this assert, you're attempting to get access a buffer
1419 // which hasn't been finished yet. Be sure to call
1420 // Builder::Finish with your root object.
1421 FLATBUFFERS_ASSERT(finished_);
1422 }
1423
1424 // Align to prepare for writing a scalar with a certain size.
Align(BitWidth alignment)1425 uint8_t Align(BitWidth alignment) {
1426 auto byte_width = 1U << alignment;
1427 buf_.insert(buf_.end(), flatbuffers::PaddingBytes(buf_.size(), byte_width),
1428 0);
1429 return static_cast<uint8_t>(byte_width);
1430 }
1431
WriteBytes(const void * val,size_t size)1432 void WriteBytes(const void *val, size_t size) {
1433 buf_.insert(buf_.end(), reinterpret_cast<const uint8_t *>(val),
1434 reinterpret_cast<const uint8_t *>(val) + size);
1435 }
1436
Write(T val,size_t byte_width)1437 template<typename T> void Write(T val, size_t byte_width) {
1438 FLATBUFFERS_ASSERT(sizeof(T) >= byte_width);
1439 val = flatbuffers::EndianScalar(val);
1440 WriteBytes(&val, byte_width);
1441 }
1442
WriteDouble(double f,uint8_t byte_width)1443 void WriteDouble(double f, uint8_t byte_width) {
1444 switch (byte_width) {
1445 case 8: Write(f, byte_width); break;
1446 case 4: Write(static_cast<float>(f), byte_width); break;
1447 // case 2: Write(static_cast<half>(f), byte_width); break;
1448 // case 1: Write(static_cast<quarter>(f), byte_width); break;
1449 default: FLATBUFFERS_ASSERT(0);
1450 }
1451 }
1452
WriteOffset(uint64_t o,uint8_t byte_width)1453 void WriteOffset(uint64_t o, uint8_t byte_width) {
1454 auto reloff = buf_.size() - o;
1455 FLATBUFFERS_ASSERT(byte_width == 8 || reloff < 1ULL << (byte_width * 8));
1456 Write(reloff, byte_width);
1457 }
1458
PushIndirect(T val,Type type,BitWidth bit_width)1459 template<typename T> void PushIndirect(T val, Type type, BitWidth bit_width) {
1460 auto byte_width = Align(bit_width);
1461 auto iloc = buf_.size();
1462 Write(val, byte_width);
1463 stack_.push_back(Value(static_cast<uint64_t>(iloc), type, bit_width));
1464 }
1465
WidthB(size_t byte_width)1466 static BitWidth WidthB(size_t byte_width) {
1467 switch (byte_width) {
1468 case 1: return BIT_WIDTH_8;
1469 case 2: return BIT_WIDTH_16;
1470 case 4: return BIT_WIDTH_32;
1471 case 8: return BIT_WIDTH_64;
1472 default: FLATBUFFERS_ASSERT(false); return BIT_WIDTH_64;
1473 }
1474 }
1475
GetScalarType()1476 template<typename T> static Type GetScalarType() {
1477 static_assert(flatbuffers::is_scalar<T>::value, "Unrelated types");
1478 return flatbuffers::is_floating_point<T>::value ? FBT_FLOAT
1479 : flatbuffers::is_same<T, bool>::value
1480 ? FBT_BOOL
1481 : (flatbuffers::is_unsigned<T>::value ? FBT_UINT : FBT_INT);
1482 }
1483
1484 public:
1485 // This was really intended to be private, except for LastValue/ReuseValue.
1486 struct Value {
1487 union {
1488 int64_t i_;
1489 uint64_t u_;
1490 double f_;
1491 };
1492
1493 Type type_;
1494
1495 // For scalars: of itself, for vector: of its elements, for string: length.
1496 BitWidth min_bit_width_;
1497
ValueValue1498 Value() : i_(0), type_(FBT_NULL), min_bit_width_(BIT_WIDTH_8) {}
1499
ValueValue1500 Value(bool b)
1501 : u_(static_cast<uint64_t>(b)),
1502 type_(FBT_BOOL),
1503 min_bit_width_(BIT_WIDTH_8) {}
1504
ValueValue1505 Value(int64_t i, Type t, BitWidth bw)
1506 : i_(i), type_(t), min_bit_width_(bw) {}
ValueValue1507 Value(uint64_t u, Type t, BitWidth bw)
1508 : u_(u), type_(t), min_bit_width_(bw) {}
1509
ValueValue1510 Value(float f)
1511 : f_(static_cast<double>(f)),
1512 type_(FBT_FLOAT),
1513 min_bit_width_(BIT_WIDTH_32) {}
ValueValue1514 Value(double f) : f_(f), type_(FBT_FLOAT), min_bit_width_(WidthF(f)) {}
1515
1516 uint8_t StoredPackedType(BitWidth parent_bit_width_ = BIT_WIDTH_8) const {
1517 return PackedType(StoredWidth(parent_bit_width_), type_);
1518 }
1519
ElemWidthValue1520 BitWidth ElemWidth(size_t buf_size, size_t elem_index) const {
1521 if (IsInline(type_)) {
1522 return min_bit_width_;
1523 } else {
1524 // We have an absolute offset, but want to store a relative offset
1525 // elem_index elements beyond the current buffer end. Since whether
1526 // the relative offset fits in a certain byte_width depends on
1527 // the size of the elements before it (and their alignment), we have
1528 // to test for each size in turn.
1529 for (size_t byte_width = 1;
1530 byte_width <= sizeof(flatbuffers::largest_scalar_t);
1531 byte_width *= 2) {
1532 // Where are we going to write this offset?
1533 auto offset_loc = buf_size +
1534 flatbuffers::PaddingBytes(buf_size, byte_width) +
1535 elem_index * byte_width;
1536 // Compute relative offset.
1537 auto offset = offset_loc - u_;
1538 // Does it fit?
1539 auto bit_width = WidthU(offset);
1540 if (static_cast<size_t>(static_cast<size_t>(1U) << bit_width) ==
1541 byte_width)
1542 return bit_width;
1543 }
1544 FLATBUFFERS_ASSERT(false); // Must match one of the sizes above.
1545 return BIT_WIDTH_64;
1546 }
1547 }
1548
1549 BitWidth StoredWidth(BitWidth parent_bit_width_ = BIT_WIDTH_8) const {
1550 if (IsInline(type_)) {
1551 return (std::max)(min_bit_width_, parent_bit_width_);
1552 } else {
1553 return min_bit_width_;
1554 }
1555 }
1556 };
1557
1558 private:
WriteAny(const Value & val,uint8_t byte_width)1559 void WriteAny(const Value &val, uint8_t byte_width) {
1560 switch (val.type_) {
1561 case FBT_NULL:
1562 case FBT_INT: Write(val.i_, byte_width); break;
1563 case FBT_BOOL:
1564 case FBT_UINT: Write(val.u_, byte_width); break;
1565 case FBT_FLOAT: WriteDouble(val.f_, byte_width); break;
1566 default: WriteOffset(val.u_, byte_width); break;
1567 }
1568 }
1569
CreateBlob(const void * data,size_t len,size_t trailing,Type type)1570 size_t CreateBlob(const void *data, size_t len, size_t trailing, Type type) {
1571 auto bit_width = WidthU(len);
1572 auto byte_width = Align(bit_width);
1573 Write<uint64_t>(len, byte_width);
1574 auto sloc = buf_.size();
1575 WriteBytes(data, len + trailing);
1576 stack_.push_back(Value(static_cast<uint64_t>(sloc), type, bit_width));
1577 return sloc;
1578 }
1579
1580 template<typename T>
ScalarVector(const T * elems,size_t len,bool fixed)1581 size_t ScalarVector(const T *elems, size_t len, bool fixed) {
1582 auto vector_type = GetScalarType<T>();
1583 auto byte_width = sizeof(T);
1584 auto bit_width = WidthB(byte_width);
1585 // If you get this assert, you're trying to write a vector with a size
1586 // field that is bigger than the scalars you're trying to write (e.g. a
1587 // byte vector > 255 elements). For such types, write a "blob" instead.
1588 // TODO: instead of asserting, could write vector with larger elements
1589 // instead, though that would be wasteful.
1590 FLATBUFFERS_ASSERT(WidthU(len) <= bit_width);
1591 Align(bit_width);
1592 if (!fixed) Write<uint64_t>(len, byte_width);
1593 auto vloc = buf_.size();
1594 for (size_t i = 0; i < len; i++) Write(elems[i], byte_width);
1595 stack_.push_back(Value(static_cast<uint64_t>(vloc),
1596 ToTypedVector(vector_type, fixed ? len : 0),
1597 bit_width));
1598 return vloc;
1599 }
1600
1601 Value CreateVector(size_t start, size_t vec_len, size_t step, bool typed,
1602 bool fixed, const Value *keys = nullptr) {
1603 FLATBUFFERS_ASSERT(
1604 !fixed ||
1605 typed); // typed=false, fixed=true combination is not supported.
1606 // Figure out smallest bit width we can store this vector with.
1607 auto bit_width = (std::max)(force_min_bit_width_, WidthU(vec_len));
1608 auto prefix_elems = 1;
1609 if (keys) {
1610 // If this vector is part of a map, we will pre-fix an offset to the keys
1611 // to this vector.
1612 bit_width = (std::max)(bit_width, keys->ElemWidth(buf_.size(), 0));
1613 prefix_elems += 2;
1614 }
1615 Type vector_type = FBT_KEY;
1616 // Check bit widths and types for all elements.
1617 for (size_t i = start; i < stack_.size(); i += step) {
1618 auto elem_width =
1619 stack_[i].ElemWidth(buf_.size(), i - start + prefix_elems);
1620 bit_width = (std::max)(bit_width, elem_width);
1621 if (typed) {
1622 if (i == start) {
1623 vector_type = stack_[i].type_;
1624 } else {
1625 // If you get this assert, you are writing a typed vector with
1626 // elements that are not all the same type.
1627 FLATBUFFERS_ASSERT(vector_type == stack_[i].type_);
1628 }
1629 }
1630 }
1631 // If you get this assert, your typed types are not one of:
1632 // Int / UInt / Float / Key.
1633 FLATBUFFERS_ASSERT(!typed || IsTypedVectorElementType(vector_type));
1634 auto byte_width = Align(bit_width);
1635 // Write vector. First the keys width/offset if available, and size.
1636 if (keys) {
1637 WriteOffset(keys->u_, byte_width);
1638 Write<uint64_t>(1ULL << keys->min_bit_width_, byte_width);
1639 }
1640 if (!fixed) Write<uint64_t>(vec_len, byte_width);
1641 // Then the actual data.
1642 auto vloc = buf_.size();
1643 for (size_t i = start; i < stack_.size(); i += step) {
1644 WriteAny(stack_[i], byte_width);
1645 }
1646 // Then the types.
1647 if (!typed) {
1648 for (size_t i = start; i < stack_.size(); i += step) {
1649 buf_.push_back(stack_[i].StoredPackedType(bit_width));
1650 }
1651 }
1652 return Value(static_cast<uint64_t>(vloc),
1653 keys ? FBT_MAP
1654 : (typed ? ToTypedVector(vector_type, fixed ? vec_len : 0)
1655 : FBT_VECTOR),
1656 bit_width);
1657 }
1658
1659 // You shouldn't really be copying instances of this class.
1660 Builder(const Builder &);
1661 Builder &operator=(const Builder &);
1662
1663 std::vector<uint8_t> buf_;
1664 std::vector<Value> stack_;
1665
1666 bool finished_;
1667 bool has_duplicate_keys_;
1668
1669 BuilderFlag flags_;
1670
1671 BitWidth force_min_bit_width_;
1672
1673 struct KeyOffsetCompare {
KeyOffsetCompareKeyOffsetCompare1674 explicit KeyOffsetCompare(const std::vector<uint8_t> &buf) : buf_(&buf) {}
operatorKeyOffsetCompare1675 bool operator()(size_t a, size_t b) const {
1676 auto stra = reinterpret_cast<const char *>(buf_->data() + a);
1677 auto strb = reinterpret_cast<const char *>(buf_->data() + b);
1678 return strcmp(stra, strb) < 0;
1679 }
1680 const std::vector<uint8_t> *buf_;
1681 };
1682
1683 typedef std::pair<size_t, size_t> StringOffset;
1684 struct StringOffsetCompare {
StringOffsetCompareStringOffsetCompare1685 explicit StringOffsetCompare(const std::vector<uint8_t> &buf)
1686 : buf_(&buf) {}
operatorStringOffsetCompare1687 bool operator()(const StringOffset &a, const StringOffset &b) const {
1688 auto stra = buf_->data() + a.first;
1689 auto strb = buf_->data() + b.first;
1690 auto cr = memcmp(stra, strb, (std::min)(a.second, b.second) + 1);
1691 return cr < 0 || (cr == 0 && a.second < b.second);
1692 }
1693 const std::vector<uint8_t> *buf_;
1694 };
1695
1696 typedef std::set<size_t, KeyOffsetCompare> KeyOffsetMap;
1697 typedef std::set<StringOffset, StringOffsetCompare> StringOffsetMap;
1698
1699 KeyOffsetMap key_pool;
1700 StringOffsetMap string_pool;
1701
1702 friend class Verifier;
1703 };
1704
1705 // Helper class to verify the integrity of a FlexBuffer
1706 class Verifier FLATBUFFERS_FINAL_CLASS {
1707 public:
1708 Verifier(const uint8_t *buf, size_t buf_len,
1709 // Supplying this vector likely results in faster verification
1710 // of larger buffers with many shared keys/strings, but
1711 // comes at the cost of using additional memory the same size of
1712 // the buffer being verified, so it is by default off.
1713 std::vector<uint8_t> *reuse_tracker = nullptr,
1714 bool _check_alignment = true, size_t max_depth = 64)
buf_(buf)1715 : buf_(buf),
1716 size_(buf_len),
1717 depth_(0),
1718 max_depth_(max_depth),
1719 num_vectors_(0),
1720 max_vectors_(buf_len),
1721 check_alignment_(_check_alignment),
1722 reuse_tracker_(reuse_tracker) {
1723 FLATBUFFERS_ASSERT(static_cast<int32_t>(size_) < FLATBUFFERS_MAX_BUFFER_SIZE);
1724 if (reuse_tracker_) {
1725 reuse_tracker_->clear();
1726 reuse_tracker_->resize(size_, PackedType(BIT_WIDTH_8, FBT_NULL));
1727 }
1728 }
1729
1730 private:
1731 // Central location where any verification failures register.
Check(bool ok)1732 bool Check(bool ok) const {
1733 // clang-format off
1734 #ifdef FLATBUFFERS_DEBUG_VERIFICATION_FAILURE
1735 FLATBUFFERS_ASSERT(ok);
1736 #endif
1737 // clang-format on
1738 return ok;
1739 }
1740
1741 // Verify any range within the buffer.
VerifyFrom(size_t elem,size_t elem_len)1742 bool VerifyFrom(size_t elem, size_t elem_len) const {
1743 return Check(elem_len < size_ && elem <= size_ - elem_len);
1744 }
VerifyBefore(size_t elem,size_t elem_len)1745 bool VerifyBefore(size_t elem, size_t elem_len) const {
1746 return Check(elem_len <= elem);
1747 }
1748
VerifyFromPointer(const uint8_t * p,size_t len)1749 bool VerifyFromPointer(const uint8_t *p, size_t len) {
1750 auto o = static_cast<size_t>(p - buf_);
1751 return VerifyFrom(o, len);
1752 }
VerifyBeforePointer(const uint8_t * p,size_t len)1753 bool VerifyBeforePointer(const uint8_t *p, size_t len) {
1754 auto o = static_cast<size_t>(p - buf_);
1755 return VerifyBefore(o, len);
1756 }
1757
VerifyByteWidth(size_t width)1758 bool VerifyByteWidth(size_t width) {
1759 return Check(width == 1 || width == 2 || width == 4 || width == 8);
1760 }
1761
VerifyType(int type)1762 bool VerifyType(int type) { return Check(type >= 0 && type < FBT_MAX_TYPE); }
1763
VerifyOffset(uint64_t off,const uint8_t * p)1764 bool VerifyOffset(uint64_t off, const uint8_t *p) {
1765 return Check(off <= static_cast<uint64_t>(size_)) &&
1766 off <= static_cast<uint64_t>(p - buf_);
1767 }
1768
VerifyAlignment(const uint8_t * p,size_t size)1769 bool VerifyAlignment(const uint8_t *p, size_t size) const {
1770 auto o = static_cast<size_t>(p - buf_);
1771 return Check((o & (size - 1)) == 0 || !check_alignment_);
1772 }
1773
1774 // Macro, since we want to escape from parent function & use lazy args.
1775 #define FLEX_CHECK_VERIFIED(P, PACKED_TYPE) \
1776 if (reuse_tracker_) { \
1777 auto packed_type = PACKED_TYPE; \
1778 auto existing = (*reuse_tracker_)[P - buf_]; \
1779 if (existing == packed_type) return true; \
1780 /* Fail verification if already set with different type! */ \
1781 if (!Check(existing == 0)) return false; \
1782 (*reuse_tracker_)[P - buf_] = packed_type; \
1783 }
1784
VerifyVector(Reference r,const uint8_t * p,Type elem_type)1785 bool VerifyVector(Reference r, const uint8_t *p, Type elem_type) {
1786 // Any kind of nesting goes thru this function, so guard against that
1787 // here, both with simple nesting checks, and the reuse tracker if on.
1788 depth_++;
1789 num_vectors_++;
1790 if (!Check(depth_ <= max_depth_ && num_vectors_ <= max_vectors_))
1791 return false;
1792 auto size_byte_width = r.byte_width_;
1793 if (!VerifyBeforePointer(p, size_byte_width)) return false;
1794 FLEX_CHECK_VERIFIED(p - size_byte_width,
1795 PackedType(Builder::WidthB(size_byte_width), r.type_));
1796 auto sized = Sized(p, size_byte_width);
1797 auto num_elems = sized.size();
1798 auto elem_byte_width = r.type_ == FBT_STRING || r.type_ == FBT_BLOB
1799 ? uint8_t(1)
1800 : r.byte_width_;
1801 auto max_elems = SIZE_MAX / elem_byte_width;
1802 if (!Check(num_elems < max_elems))
1803 return false; // Protect against byte_size overflowing.
1804 auto byte_size = num_elems * elem_byte_width;
1805 if (!VerifyFromPointer(p, byte_size)) return false;
1806 if (elem_type == FBT_NULL) {
1807 // Verify type bytes after the vector.
1808 if (!VerifyFromPointer(p + byte_size, num_elems)) return false;
1809 auto v = Vector(p, size_byte_width);
1810 for (size_t i = 0; i < num_elems; i++)
1811 if (!VerifyRef(v[i])) return false;
1812 } else if (elem_type == FBT_KEY) {
1813 auto v = TypedVector(p, elem_byte_width, FBT_KEY);
1814 for (size_t i = 0; i < num_elems; i++)
1815 if (!VerifyRef(v[i])) return false;
1816 } else {
1817 FLATBUFFERS_ASSERT(IsInline(elem_type));
1818 }
1819 depth_--;
1820 return true;
1821 }
1822
VerifyKeys(const uint8_t * p,uint8_t byte_width)1823 bool VerifyKeys(const uint8_t *p, uint8_t byte_width) {
1824 // The vector part of the map has already been verified.
1825 const size_t num_prefixed_fields = 3;
1826 if (!VerifyBeforePointer(p, byte_width * num_prefixed_fields)) return false;
1827 p -= byte_width * num_prefixed_fields;
1828 auto off = ReadUInt64(p, byte_width);
1829 if (!VerifyOffset(off, p)) return false;
1830 auto key_byte_with =
1831 static_cast<uint8_t>(ReadUInt64(p + byte_width, byte_width));
1832 if (!VerifyByteWidth(key_byte_with)) return false;
1833 return VerifyVector(Reference(p, byte_width, key_byte_with, FBT_VECTOR_KEY),
1834 p - off, FBT_KEY);
1835 }
1836
VerifyKey(const uint8_t * p)1837 bool VerifyKey(const uint8_t *p) {
1838 FLEX_CHECK_VERIFIED(p, PackedType(BIT_WIDTH_8, FBT_KEY));
1839 while (p < buf_ + size_)
1840 if (*p++) return true;
1841 return false;
1842 }
1843
1844 #undef FLEX_CHECK_VERIFIED
1845
VerifyTerminator(const String & s)1846 bool VerifyTerminator(const String &s) {
1847 return VerifyFromPointer(reinterpret_cast<const uint8_t *>(s.c_str()),
1848 s.size() + 1);
1849 }
1850
VerifyRef(Reference r)1851 bool VerifyRef(Reference r) {
1852 // r.parent_width_ and r.data_ already verified.
1853 if (!VerifyByteWidth(r.byte_width_) || !VerifyType(r.type_)) {
1854 return false;
1855 }
1856 if (IsInline(r.type_)) {
1857 // Inline scalars, don't require further verification.
1858 return true;
1859 }
1860 // All remaining types are an offset.
1861 auto off = ReadUInt64(r.data_, r.parent_width_);
1862 if (!VerifyOffset(off, r.data_)) return false;
1863 auto p = r.Indirect();
1864 if (!VerifyAlignment(p, r.byte_width_)) return false;
1865 switch (r.type_) {
1866 case FBT_INDIRECT_INT:
1867 case FBT_INDIRECT_UINT:
1868 case FBT_INDIRECT_FLOAT: return VerifyFromPointer(p, r.byte_width_);
1869 case FBT_KEY: return VerifyKey(p);
1870 case FBT_MAP:
1871 return VerifyVector(r, p, FBT_NULL) && VerifyKeys(p, r.byte_width_);
1872 case FBT_VECTOR: return VerifyVector(r, p, FBT_NULL);
1873 case FBT_VECTOR_INT: return VerifyVector(r, p, FBT_INT);
1874 case FBT_VECTOR_BOOL:
1875 case FBT_VECTOR_UINT: return VerifyVector(r, p, FBT_UINT);
1876 case FBT_VECTOR_FLOAT: return VerifyVector(r, p, FBT_FLOAT);
1877 case FBT_VECTOR_KEY: return VerifyVector(r, p, FBT_KEY);
1878 case FBT_VECTOR_STRING_DEPRECATED:
1879 // Use of FBT_KEY here intentional, see elsewhere.
1880 return VerifyVector(r, p, FBT_KEY);
1881 case FBT_BLOB: return VerifyVector(r, p, FBT_UINT);
1882 case FBT_STRING:
1883 return VerifyVector(r, p, FBT_UINT) &&
1884 VerifyTerminator(String(p, r.byte_width_));
1885 case FBT_VECTOR_INT2:
1886 case FBT_VECTOR_UINT2:
1887 case FBT_VECTOR_FLOAT2:
1888 case FBT_VECTOR_INT3:
1889 case FBT_VECTOR_UINT3:
1890 case FBT_VECTOR_FLOAT3:
1891 case FBT_VECTOR_INT4:
1892 case FBT_VECTOR_UINT4:
1893 case FBT_VECTOR_FLOAT4: {
1894 uint8_t len = 0;
1895 auto vtype = ToFixedTypedVectorElementType(r.type_, &len);
1896 if (!VerifyType(vtype)) return false;
1897 return VerifyFromPointer(p, static_cast<size_t>(r.byte_width_) * len);
1898 }
1899 default: return false;
1900 }
1901 }
1902
1903 public:
VerifyBuffer()1904 bool VerifyBuffer() {
1905 if (!Check(size_ >= 3)) return false;
1906 auto end = buf_ + size_;
1907 auto byte_width = *--end;
1908 auto packed_type = *--end;
1909 return VerifyByteWidth(byte_width) && Check(end - buf_ >= byte_width) &&
1910 VerifyRef(Reference(end - byte_width, byte_width, packed_type));
1911 }
1912
1913 private:
1914 const uint8_t *buf_;
1915 size_t size_;
1916 size_t depth_;
1917 const size_t max_depth_;
1918 size_t num_vectors_;
1919 const size_t max_vectors_;
1920 bool check_alignment_;
1921 std::vector<uint8_t> *reuse_tracker_;
1922 };
1923
1924 // Utility function that constructs the Verifier for you, see above for
1925 // parameters.
1926 inline bool VerifyBuffer(const uint8_t *buf, size_t buf_len,
1927 std::vector<uint8_t> *reuse_tracker = nullptr) {
1928 Verifier verifier(buf, buf_len, reuse_tracker);
1929 return verifier.VerifyBuffer();
1930 }
1931
1932 } // namespace flexbuffers
1933
1934 #if defined(_MSC_VER)
1935 # pragma warning(pop)
1936 #endif
1937
1938 #endif // FLATBUFFERS_FLEXBUFFERS_H_
1939