blob: 7bf84302e5f576f38bf1a5ff8ed82e42812d8f68 [file] [log] [blame]
Austin Schuhe89fa2d2019-08-14 20:24:23 -07001/*
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 <map>
21// Used to select STL variant.
22#include "flatbuffers/base.h"
23// We use the basic binary writing functions from the regular FlatBuffers.
24#include "flatbuffers/util.h"
25
26#ifdef _MSC_VER
27# include <intrin.h>
28#endif
29
30#if defined(_MSC_VER)
31# pragma warning(push)
32# pragma warning(disable : 4127) // C4127: conditional expression is constant
33#endif
34
35namespace flexbuffers {
36
37class Reference;
38class Map;
39
40// These are used in the lower 2 bits of a type field to determine the size of
41// the elements (and or size field) of the item pointed to (e.g. vector).
42enum BitWidth {
43 BIT_WIDTH_8 = 0,
44 BIT_WIDTH_16 = 1,
45 BIT_WIDTH_32 = 2,
46 BIT_WIDTH_64 = 3,
47};
48
49// These are used as the upper 6 bits of a type field to indicate the actual
50// type.
51enum Type {
52 FBT_NULL = 0,
53 FBT_INT = 1,
54 FBT_UINT = 2,
55 FBT_FLOAT = 3,
James Kuszmaul8e62b022022-03-22 09:33:25 -070056 // Types above stored inline, types below (except FBT_BOOL) store an offset.
Austin Schuhe89fa2d2019-08-14 20:24:23 -070057 FBT_KEY = 4,
58 FBT_STRING = 5,
59 FBT_INDIRECT_INT = 6,
60 FBT_INDIRECT_UINT = 7,
61 FBT_INDIRECT_FLOAT = 8,
62 FBT_MAP = 9,
63 FBT_VECTOR = 10, // Untyped.
64 FBT_VECTOR_INT = 11, // Typed any size (stores no type table).
65 FBT_VECTOR_UINT = 12,
66 FBT_VECTOR_FLOAT = 13,
67 FBT_VECTOR_KEY = 14,
Austin Schuh272c6132020-11-14 16:37:52 -080068 // DEPRECATED, use FBT_VECTOR or FBT_VECTOR_KEY instead.
69 // Read test.cpp/FlexBuffersDeprecatedTest() for details on why.
70 FBT_VECTOR_STRING_DEPRECATED = 15,
Austin Schuhe89fa2d2019-08-14 20:24:23 -070071 FBT_VECTOR_INT2 = 16, // Typed tuple (no type table, no size field).
72 FBT_VECTOR_UINT2 = 17,
73 FBT_VECTOR_FLOAT2 = 18,
74 FBT_VECTOR_INT3 = 19, // Typed triple (no type table, no size field).
75 FBT_VECTOR_UINT3 = 20,
76 FBT_VECTOR_FLOAT3 = 21,
77 FBT_VECTOR_INT4 = 22, // Typed quad (no type table, no size field).
78 FBT_VECTOR_UINT4 = 23,
79 FBT_VECTOR_FLOAT4 = 24,
80 FBT_BLOB = 25,
81 FBT_BOOL = 26,
82 FBT_VECTOR_BOOL =
83 36, // To Allow the same type of conversion of type to vector type
James Kuszmaul8e62b022022-03-22 09:33:25 -070084
85 FBT_MAX_TYPE = 37
Austin Schuhe89fa2d2019-08-14 20:24:23 -070086};
87
88inline bool IsInline(Type t) { return t <= FBT_FLOAT || t == FBT_BOOL; }
89
90inline bool IsTypedVectorElementType(Type t) {
91 return (t >= FBT_INT && t <= FBT_STRING) || t == FBT_BOOL;
92}
93
94inline bool IsTypedVector(Type t) {
Austin Schuh272c6132020-11-14 16:37:52 -080095 return (t >= FBT_VECTOR_INT && t <= FBT_VECTOR_STRING_DEPRECATED) ||
Austin Schuhe89fa2d2019-08-14 20:24:23 -070096 t == FBT_VECTOR_BOOL;
97}
98
99inline bool IsFixedTypedVector(Type t) {
100 return t >= FBT_VECTOR_INT2 && t <= FBT_VECTOR_FLOAT4;
101}
102
103inline Type ToTypedVector(Type t, size_t fixed_len = 0) {
104 FLATBUFFERS_ASSERT(IsTypedVectorElementType(t));
105 switch (fixed_len) {
106 case 0: return static_cast<Type>(t - FBT_INT + FBT_VECTOR_INT);
107 case 2: return static_cast<Type>(t - FBT_INT + FBT_VECTOR_INT2);
108 case 3: return static_cast<Type>(t - FBT_INT + FBT_VECTOR_INT3);
109 case 4: return static_cast<Type>(t - FBT_INT + FBT_VECTOR_INT4);
110 default: FLATBUFFERS_ASSERT(0); return FBT_NULL;
111 }
112}
113
114inline Type ToTypedVectorElementType(Type t) {
115 FLATBUFFERS_ASSERT(IsTypedVector(t));
116 return static_cast<Type>(t - FBT_VECTOR_INT + FBT_INT);
117}
118
119inline Type ToFixedTypedVectorElementType(Type t, uint8_t *len) {
120 FLATBUFFERS_ASSERT(IsFixedTypedVector(t));
121 auto fixed_type = t - FBT_VECTOR_INT2;
122 *len = static_cast<uint8_t>(fixed_type / 3 +
123 2); // 3 types each, starting from length 2.
124 return static_cast<Type>(fixed_type % 3 + FBT_INT);
125}
126
127// TODO: implement proper support for 8/16bit floats, or decide not to
128// support them.
129typedef int16_t half;
130typedef int8_t quarter;
131
132// TODO: can we do this without conditionals using intrinsics or inline asm
133// on some platforms? Given branch prediction the method below should be
134// decently quick, but it is the most frequently executed function.
135// We could do an (unaligned) 64-bit read if we ifdef out the platforms for
136// which that doesn't work (or where we'd read into un-owned memory).
137template<typename R, typename T1, typename T2, typename T4, typename T8>
138R ReadSizedScalar(const uint8_t *data, uint8_t byte_width) {
139 return byte_width < 4
140 ? (byte_width < 2
141 ? static_cast<R>(flatbuffers::ReadScalar<T1>(data))
142 : static_cast<R>(flatbuffers::ReadScalar<T2>(data)))
143 : (byte_width < 8
144 ? static_cast<R>(flatbuffers::ReadScalar<T4>(data))
145 : static_cast<R>(flatbuffers::ReadScalar<T8>(data)));
146}
147
148inline int64_t ReadInt64(const uint8_t *data, uint8_t byte_width) {
149 return ReadSizedScalar<int64_t, int8_t, int16_t, int32_t, int64_t>(
150 data, byte_width);
151}
152
153inline uint64_t ReadUInt64(const uint8_t *data, uint8_t byte_width) {
154 // This is the "hottest" function (all offset lookups use this), so worth
155 // optimizing if possible.
156 // TODO: GCC apparently replaces memcpy by a rep movsb, but only if count is a
157 // constant, which here it isn't. Test if memcpy is still faster than
158 // the conditionals in ReadSizedScalar. Can also use inline asm.
Austin Schuh2dd86a92022-09-14 21:19:23 -0700159
Austin Schuhe89fa2d2019-08-14 20:24:23 -0700160 // clang-format off
James Kuszmaul8e62b022022-03-22 09:33:25 -0700161 #if defined(_MSC_VER) && defined(_M_X64) && !defined(_M_ARM64EC)
162 // This is 64-bit Windows only, __movsb does not work on 32-bit Windows.
Austin Schuhe89fa2d2019-08-14 20:24:23 -0700163 uint64_t u = 0;
164 __movsb(reinterpret_cast<uint8_t *>(&u),
165 reinterpret_cast<const uint8_t *>(data), byte_width);
166 return flatbuffers::EndianScalar(u);
167 #else
168 return ReadSizedScalar<uint64_t, uint8_t, uint16_t, uint32_t, uint64_t>(
169 data, byte_width);
170 #endif
171 // clang-format on
172}
173
174inline double ReadDouble(const uint8_t *data, uint8_t byte_width) {
175 return ReadSizedScalar<double, quarter, half, float, double>(data,
176 byte_width);
177}
178
179inline const uint8_t *Indirect(const uint8_t *offset, uint8_t byte_width) {
180 return offset - ReadUInt64(offset, byte_width);
181}
182
183template<typename T> const uint8_t *Indirect(const uint8_t *offset) {
184 return offset - flatbuffers::ReadScalar<T>(offset);
185}
186
187inline BitWidth WidthU(uint64_t u) {
188#define FLATBUFFERS_GET_FIELD_BIT_WIDTH(value, width) \
189 { \
190 if (!((u) & ~((1ULL << (width)) - 1ULL))) return BIT_WIDTH_##width; \
191 }
192 FLATBUFFERS_GET_FIELD_BIT_WIDTH(u, 8);
193 FLATBUFFERS_GET_FIELD_BIT_WIDTH(u, 16);
194 FLATBUFFERS_GET_FIELD_BIT_WIDTH(u, 32);
195#undef FLATBUFFERS_GET_FIELD_BIT_WIDTH
196 return BIT_WIDTH_64;
197}
198
199inline BitWidth WidthI(int64_t i) {
200 auto u = static_cast<uint64_t>(i) << 1;
201 return WidthU(i >= 0 ? u : ~u);
202}
203
204inline BitWidth WidthF(double f) {
205 return static_cast<double>(static_cast<float>(f)) == f ? BIT_WIDTH_32
206 : BIT_WIDTH_64;
207}
208
209// Base class of all types below.
210// Points into the data buffer and allows access to one type.
211class Object {
212 public:
213 Object(const uint8_t *data, uint8_t byte_width)
214 : data_(data), byte_width_(byte_width) {}
215
216 protected:
217 const uint8_t *data_;
218 uint8_t byte_width_;
219};
220
Austin Schuh272c6132020-11-14 16:37:52 -0800221// Object that has a size, obtained either from size prefix, or elsewhere.
Austin Schuhe89fa2d2019-08-14 20:24:23 -0700222class Sized : public Object {
223 public:
Austin Schuh272c6132020-11-14 16:37:52 -0800224 // Size prefix.
225 Sized(const uint8_t *data, uint8_t byte_width)
226 : Object(data, byte_width), size_(read_size()) {}
227 // Manual size.
228 Sized(const uint8_t *data, uint8_t byte_width, size_t sz)
229 : Object(data, byte_width), size_(sz) {}
230 size_t size() const { return size_; }
231 // Access size stored in `byte_width_` bytes before data_ pointer.
232 size_t read_size() const {
Austin Schuhe89fa2d2019-08-14 20:24:23 -0700233 return static_cast<size_t>(ReadUInt64(data_ - byte_width_, byte_width_));
234 }
Austin Schuh272c6132020-11-14 16:37:52 -0800235
236 protected:
237 size_t size_;
Austin Schuhe89fa2d2019-08-14 20:24:23 -0700238};
239
240class String : public Sized {
241 public:
Austin Schuh272c6132020-11-14 16:37:52 -0800242 // Size prefix.
Austin Schuhe89fa2d2019-08-14 20:24:23 -0700243 String(const uint8_t *data, uint8_t byte_width) : Sized(data, byte_width) {}
Austin Schuh272c6132020-11-14 16:37:52 -0800244 // Manual size.
245 String(const uint8_t *data, uint8_t byte_width, size_t sz)
246 : Sized(data, byte_width, sz) {}
Austin Schuhe89fa2d2019-08-14 20:24:23 -0700247
248 size_t length() const { return size(); }
249 const char *c_str() const { return reinterpret_cast<const char *>(data_); }
Austin Schuh272c6132020-11-14 16:37:52 -0800250 std::string str() const { return std::string(c_str(), size()); }
Austin Schuhe89fa2d2019-08-14 20:24:23 -0700251
252 static String EmptyString() {
Austin Schuh272c6132020-11-14 16:37:52 -0800253 static const char *empty_string = "";
254 return String(reinterpret_cast<const uint8_t *>(empty_string), 1, 0);
Austin Schuhe89fa2d2019-08-14 20:24:23 -0700255 }
256 bool IsTheEmptyString() const { return data_ == EmptyString().data_; }
257};
258
259class Blob : public Sized {
260 public:
261 Blob(const uint8_t *data_buf, uint8_t byte_width)
262 : Sized(data_buf, byte_width) {}
263
264 static Blob EmptyBlob() {
265 static const uint8_t empty_blob[] = { 0 /*len*/ };
266 return Blob(empty_blob + 1, 1);
267 }
268 bool IsTheEmptyBlob() const { return data_ == EmptyBlob().data_; }
269 const uint8_t *data() const { return data_; }
270};
271
272class Vector : public Sized {
273 public:
274 Vector(const uint8_t *data, uint8_t byte_width) : Sized(data, byte_width) {}
275
276 Reference operator[](size_t i) const;
277
278 static Vector EmptyVector() {
279 static const uint8_t empty_vector[] = { 0 /*len*/ };
280 return Vector(empty_vector + 1, 1);
281 }
282 bool IsTheEmptyVector() const { return data_ == EmptyVector().data_; }
283};
284
285class TypedVector : public Sized {
286 public:
287 TypedVector(const uint8_t *data, uint8_t byte_width, Type element_type)
288 : Sized(data, byte_width), type_(element_type) {}
289
290 Reference operator[](size_t i) const;
291
292 static TypedVector EmptyTypedVector() {
293 static const uint8_t empty_typed_vector[] = { 0 /*len*/ };
294 return TypedVector(empty_typed_vector + 1, 1, FBT_INT);
295 }
296 bool IsTheEmptyVector() const {
297 return data_ == TypedVector::EmptyTypedVector().data_;
298 }
299
300 Type ElementType() { return type_; }
301
Austin Schuh272c6132020-11-14 16:37:52 -0800302 friend Reference;
303
Austin Schuhe89fa2d2019-08-14 20:24:23 -0700304 private:
305 Type type_;
306
307 friend Map;
308};
309
310class FixedTypedVector : public Object {
311 public:
312 FixedTypedVector(const uint8_t *data, uint8_t byte_width, Type element_type,
313 uint8_t len)
314 : Object(data, byte_width), type_(element_type), len_(len) {}
315
316 Reference operator[](size_t i) const;
317
318 static FixedTypedVector EmptyFixedTypedVector() {
319 static const uint8_t fixed_empty_vector[] = { 0 /* unused */ };
320 return FixedTypedVector(fixed_empty_vector, 1, FBT_INT, 0);
321 }
322 bool IsTheEmptyFixedTypedVector() const {
323 return data_ == FixedTypedVector::EmptyFixedTypedVector().data_;
324 }
325
James Kuszmaul8e62b022022-03-22 09:33:25 -0700326 Type ElementType() const { return type_; }
327 uint8_t size() const { return len_; }
Austin Schuhe89fa2d2019-08-14 20:24:23 -0700328
329 private:
330 Type type_;
331 uint8_t len_;
332};
333
334class Map : public Vector {
335 public:
336 Map(const uint8_t *data, uint8_t byte_width) : Vector(data, byte_width) {}
337
338 Reference operator[](const char *key) const;
339 Reference operator[](const std::string &key) const;
340
341 Vector Values() const { return Vector(data_, byte_width_); }
342
343 TypedVector Keys() const {
344 const size_t num_prefixed_fields = 3;
345 auto keys_offset = data_ - byte_width_ * num_prefixed_fields;
346 return TypedVector(Indirect(keys_offset, byte_width_),
347 static_cast<uint8_t>(
348 ReadUInt64(keys_offset + byte_width_, byte_width_)),
349 FBT_KEY);
350 }
351
352 static Map EmptyMap() {
353 static const uint8_t empty_map[] = {
354 0 /*keys_len*/, 0 /*keys_offset*/, 1 /*keys_width*/, 0 /*len*/
355 };
356 return Map(empty_map + 4, 1);
357 }
358
359 bool IsTheEmptyMap() const { return data_ == EmptyMap().data_; }
360};
361
362template<typename T>
363void AppendToString(std::string &s, T &&v, bool keys_quoted) {
Austin Schuh272c6132020-11-14 16:37:52 -0800364 s += "[ ";
365 for (size_t i = 0; i < v.size(); i++) {
366 if (i) s += ", ";
367 v[i].ToString(true, keys_quoted, s);
368 }
369 s += " ]";
Austin Schuhe89fa2d2019-08-14 20:24:23 -0700370}
371
372class Reference {
373 public:
Austin Schuh272c6132020-11-14 16:37:52 -0800374 Reference()
James Kuszmaul8e62b022022-03-22 09:33:25 -0700375 : data_(nullptr), parent_width_(0), byte_width_(0), type_(FBT_NULL) {}
Austin Schuh272c6132020-11-14 16:37:52 -0800376
Austin Schuhe89fa2d2019-08-14 20:24:23 -0700377 Reference(const uint8_t *data, uint8_t parent_width, uint8_t byte_width,
378 Type type)
379 : data_(data),
380 parent_width_(parent_width),
381 byte_width_(byte_width),
382 type_(type) {}
383
384 Reference(const uint8_t *data, uint8_t parent_width, uint8_t packed_type)
385 : data_(data), parent_width_(parent_width) {
386 byte_width_ = 1U << static_cast<BitWidth>(packed_type & 3);
387 type_ = static_cast<Type>(packed_type >> 2);
388 }
389
390 Type GetType() const { return type_; }
391
392 bool IsNull() const { return type_ == FBT_NULL; }
393 bool IsBool() const { return type_ == FBT_BOOL; }
394 bool IsInt() const { return type_ == FBT_INT || type_ == FBT_INDIRECT_INT; }
395 bool IsUInt() const {
396 return type_ == FBT_UINT || type_ == FBT_INDIRECT_UINT;
397 }
398 bool IsIntOrUint() const { return IsInt() || IsUInt(); }
399 bool IsFloat() const {
400 return type_ == FBT_FLOAT || type_ == FBT_INDIRECT_FLOAT;
401 }
402 bool IsNumeric() const { return IsIntOrUint() || IsFloat(); }
403 bool IsString() const { return type_ == FBT_STRING; }
404 bool IsKey() const { return type_ == FBT_KEY; }
405 bool IsVector() const { return type_ == FBT_VECTOR || type_ == FBT_MAP; }
Austin Schuh272c6132020-11-14 16:37:52 -0800406 bool IsUntypedVector() const { return type_ == FBT_VECTOR; }
Austin Schuhe89fa2d2019-08-14 20:24:23 -0700407 bool IsTypedVector() const { return flexbuffers::IsTypedVector(type_); }
Austin Schuh272c6132020-11-14 16:37:52 -0800408 bool IsFixedTypedVector() const {
409 return flexbuffers::IsFixedTypedVector(type_);
410 }
411 bool IsAnyVector() const {
412 return (IsTypedVector() || IsFixedTypedVector() || IsVector());
413 }
Austin Schuhe89fa2d2019-08-14 20:24:23 -0700414 bool IsMap() const { return type_ == FBT_MAP; }
415 bool IsBlob() const { return type_ == FBT_BLOB; }
Austin Schuhe89fa2d2019-08-14 20:24:23 -0700416 bool AsBool() const {
417 return (type_ == FBT_BOOL ? ReadUInt64(data_, parent_width_)
Austin Schuh272c6132020-11-14 16:37:52 -0800418 : AsUInt64()) != 0;
Austin Schuhe89fa2d2019-08-14 20:24:23 -0700419 }
420
421 // Reads any type as a int64_t. Never fails, does most sensible conversion.
422 // Truncates floats, strings are attempted to be parsed for a number,
423 // vectors/maps return their size. Returns 0 if all else fails.
424 int64_t AsInt64() const {
425 if (type_ == FBT_INT) {
426 // A fast path for the common case.
427 return ReadInt64(data_, parent_width_);
428 } else
429 switch (type_) {
430 case FBT_INDIRECT_INT: return ReadInt64(Indirect(), byte_width_);
431 case FBT_UINT: return ReadUInt64(data_, parent_width_);
432 case FBT_INDIRECT_UINT: return ReadUInt64(Indirect(), byte_width_);
433 case FBT_FLOAT:
434 return static_cast<int64_t>(ReadDouble(data_, parent_width_));
435 case FBT_INDIRECT_FLOAT:
436 return static_cast<int64_t>(ReadDouble(Indirect(), byte_width_));
437 case FBT_NULL: return 0;
438 case FBT_STRING: return flatbuffers::StringToInt(AsString().c_str());
439 case FBT_VECTOR: return static_cast<int64_t>(AsVector().size());
440 case FBT_BOOL: return ReadInt64(data_, parent_width_);
441 default:
442 // Convert other things to int.
443 return 0;
444 }
445 }
446
447 // TODO: could specialize these to not use AsInt64() if that saves
448 // extension ops in generated code, and use a faster op than ReadInt64.
449 int32_t AsInt32() const { return static_cast<int32_t>(AsInt64()); }
450 int16_t AsInt16() const { return static_cast<int16_t>(AsInt64()); }
451 int8_t AsInt8() const { return static_cast<int8_t>(AsInt64()); }
452
453 uint64_t AsUInt64() const {
454 if (type_ == FBT_UINT) {
455 // A fast path for the common case.
456 return ReadUInt64(data_, parent_width_);
457 } else
458 switch (type_) {
459 case FBT_INDIRECT_UINT: return ReadUInt64(Indirect(), byte_width_);
460 case FBT_INT: return ReadInt64(data_, parent_width_);
461 case FBT_INDIRECT_INT: return ReadInt64(Indirect(), byte_width_);
462 case FBT_FLOAT:
463 return static_cast<uint64_t>(ReadDouble(data_, parent_width_));
464 case FBT_INDIRECT_FLOAT:
465 return static_cast<uint64_t>(ReadDouble(Indirect(), byte_width_));
466 case FBT_NULL: return 0;
467 case FBT_STRING: return flatbuffers::StringToUInt(AsString().c_str());
468 case FBT_VECTOR: return static_cast<uint64_t>(AsVector().size());
469 case FBT_BOOL: return ReadUInt64(data_, parent_width_);
470 default:
471 // Convert other things to uint.
472 return 0;
473 }
474 }
475
476 uint32_t AsUInt32() const { return static_cast<uint32_t>(AsUInt64()); }
477 uint16_t AsUInt16() const { return static_cast<uint16_t>(AsUInt64()); }
478 uint8_t AsUInt8() const { return static_cast<uint8_t>(AsUInt64()); }
479
480 double AsDouble() const {
481 if (type_ == FBT_FLOAT) {
482 // A fast path for the common case.
483 return ReadDouble(data_, parent_width_);
484 } else
485 switch (type_) {
486 case FBT_INDIRECT_FLOAT: return ReadDouble(Indirect(), byte_width_);
487 case FBT_INT:
488 return static_cast<double>(ReadInt64(data_, parent_width_));
489 case FBT_UINT:
490 return static_cast<double>(ReadUInt64(data_, parent_width_));
491 case FBT_INDIRECT_INT:
492 return static_cast<double>(ReadInt64(Indirect(), byte_width_));
493 case FBT_INDIRECT_UINT:
494 return static_cast<double>(ReadUInt64(Indirect(), byte_width_));
495 case FBT_NULL: return 0.0;
Austin Schuh272c6132020-11-14 16:37:52 -0800496 case FBT_STRING: {
497 double d;
498 flatbuffers::StringToNumber(AsString().c_str(), &d);
499 return d;
500 }
Austin Schuhe89fa2d2019-08-14 20:24:23 -0700501 case FBT_VECTOR: return static_cast<double>(AsVector().size());
502 case FBT_BOOL:
503 return static_cast<double>(ReadUInt64(data_, parent_width_));
504 default:
505 // Convert strings and other things to float.
506 return 0;
507 }
508 }
509
510 float AsFloat() const { return static_cast<float>(AsDouble()); }
511
512 const char *AsKey() const {
Austin Schuh272c6132020-11-14 16:37:52 -0800513 if (type_ == FBT_KEY || type_ == FBT_STRING) {
Austin Schuhe89fa2d2019-08-14 20:24:23 -0700514 return reinterpret_cast<const char *>(Indirect());
515 } else {
516 return "";
517 }
518 }
519
Austin Schuh272c6132020-11-14 16:37:52 -0800520 // This function returns the empty string if you try to read something that
521 // is not a string or key.
Austin Schuhe89fa2d2019-08-14 20:24:23 -0700522 String AsString() const {
523 if (type_ == FBT_STRING) {
524 return String(Indirect(), byte_width_);
Austin Schuh272c6132020-11-14 16:37:52 -0800525 } else if (type_ == FBT_KEY) {
526 auto key = Indirect();
527 return String(key, byte_width_,
528 strlen(reinterpret_cast<const char *>(key)));
Austin Schuhe89fa2d2019-08-14 20:24:23 -0700529 } else {
530 return String::EmptyString();
531 }
532 }
533
534 // Unlike AsString(), this will convert any type to a std::string.
535 std::string ToString() const {
536 std::string s;
537 ToString(false, false, s);
538 return s;
539 }
540
541 // Convert any type to a JSON-like string. strings_quoted determines if
542 // string values at the top level receive "" quotes (inside other values
543 // they always do). keys_quoted determines if keys are quoted, at any level.
544 // TODO(wvo): add further options to have indentation/newlines.
545 void ToString(bool strings_quoted, bool keys_quoted, std::string &s) const {
546 if (type_ == FBT_STRING) {
547 String str(Indirect(), byte_width_);
548 if (strings_quoted) {
549 flatbuffers::EscapeString(str.c_str(), str.length(), &s, true, false);
550 } else {
551 s.append(str.c_str(), str.length());
552 }
553 } else if (IsKey()) {
554 auto str = AsKey();
555 if (keys_quoted) {
556 flatbuffers::EscapeString(str, strlen(str), &s, true, false);
557 } else {
558 s += str;
559 }
560 } else if (IsInt()) {
561 s += flatbuffers::NumToString(AsInt64());
562 } else if (IsUInt()) {
563 s += flatbuffers::NumToString(AsUInt64());
564 } else if (IsFloat()) {
565 s += flatbuffers::NumToString(AsDouble());
566 } else if (IsNull()) {
567 s += "null";
568 } else if (IsBool()) {
569 s += AsBool() ? "true" : "false";
570 } else if (IsMap()) {
571 s += "{ ";
572 auto m = AsMap();
573 auto keys = m.Keys();
574 auto vals = m.Values();
575 for (size_t i = 0; i < keys.size(); i++) {
James Kuszmaul8e62b022022-03-22 09:33:25 -0700576 bool kq = keys_quoted;
577 if (!kq) {
578 // FlexBuffers keys may contain arbitrary characters, only allow
579 // unquoted if it looks like an "identifier":
580 const char *p = keys[i].AsKey();
581 if (!flatbuffers::is_alpha(*p) && *p != '_') {
Austin Schuh2dd86a92022-09-14 21:19:23 -0700582 kq = true;
James Kuszmaul8e62b022022-03-22 09:33:25 -0700583 } else {
584 while (*++p) {
585 if (!flatbuffers::is_alnum(*p) && *p != '_') {
586 kq = true;
587 break;
588 }
589 }
590 }
591 }
592 keys[i].ToString(true, kq, s);
Austin Schuhe89fa2d2019-08-14 20:24:23 -0700593 s += ": ";
594 vals[i].ToString(true, keys_quoted, s);
595 if (i < keys.size() - 1) s += ", ";
596 }
597 s += " }";
598 } else if (IsVector()) {
599 AppendToString<Vector>(s, AsVector(), keys_quoted);
600 } else if (IsTypedVector()) {
601 AppendToString<TypedVector>(s, AsTypedVector(), keys_quoted);
602 } else if (IsFixedTypedVector()) {
603 AppendToString<FixedTypedVector>(s, AsFixedTypedVector(), keys_quoted);
604 } else if (IsBlob()) {
605 auto blob = AsBlob();
Austin Schuh272c6132020-11-14 16:37:52 -0800606 flatbuffers::EscapeString(reinterpret_cast<const char *>(blob.data()),
607 blob.size(), &s, true, false);
Austin Schuhe89fa2d2019-08-14 20:24:23 -0700608 } else {
609 s += "(?)";
610 }
611 }
612
613 // This function returns the empty blob if you try to read a not-blob.
614 // Strings can be viewed as blobs too.
615 Blob AsBlob() const {
616 if (type_ == FBT_BLOB || type_ == FBT_STRING) {
617 return Blob(Indirect(), byte_width_);
618 } else {
619 return Blob::EmptyBlob();
620 }
621 }
622
623 // This function returns the empty vector if you try to read a not-vector.
624 // Maps can be viewed as vectors too.
625 Vector AsVector() const {
626 if (type_ == FBT_VECTOR || type_ == FBT_MAP) {
627 return Vector(Indirect(), byte_width_);
628 } else {
629 return Vector::EmptyVector();
630 }
631 }
632
633 TypedVector AsTypedVector() const {
634 if (IsTypedVector()) {
Austin Schuh272c6132020-11-14 16:37:52 -0800635 auto tv =
636 TypedVector(Indirect(), byte_width_, ToTypedVectorElementType(type_));
637 if (tv.type_ == FBT_STRING) {
638 // These can't be accessed as strings, since we don't know the bit-width
639 // of the size field, see the declaration of
640 // FBT_VECTOR_STRING_DEPRECATED above for details.
641 // We change the type here to be keys, which are a subtype of strings,
642 // and will ignore the size field. This will truncate strings with
643 // embedded nulls.
644 tv.type_ = FBT_KEY;
645 }
646 return tv;
Austin Schuhe89fa2d2019-08-14 20:24:23 -0700647 } else {
648 return TypedVector::EmptyTypedVector();
649 }
650 }
651
652 FixedTypedVector AsFixedTypedVector() const {
653 if (IsFixedTypedVector()) {
654 uint8_t len = 0;
655 auto vtype = ToFixedTypedVectorElementType(type_, &len);
656 return FixedTypedVector(Indirect(), byte_width_, vtype, len);
657 } else {
658 return FixedTypedVector::EmptyFixedTypedVector();
659 }
660 }
661
662 Map AsMap() const {
663 if (type_ == FBT_MAP) {
664 return Map(Indirect(), byte_width_);
665 } else {
666 return Map::EmptyMap();
667 }
668 }
669
670 template<typename T> T As() const;
671
672 // Experimental: Mutation functions.
673 // These allow scalars in an already created buffer to be updated in-place.
674 // Since by default scalars are stored in the smallest possible space,
675 // the new value may not fit, in which case these functions return false.
676 // To avoid this, you can construct the values you intend to mutate using
677 // Builder::ForceMinimumBitWidth.
678 bool MutateInt(int64_t i) {
679 if (type_ == FBT_INT) {
680 return Mutate(data_, i, parent_width_, WidthI(i));
681 } else if (type_ == FBT_INDIRECT_INT) {
682 return Mutate(Indirect(), i, byte_width_, WidthI(i));
683 } else if (type_ == FBT_UINT) {
684 auto u = static_cast<uint64_t>(i);
685 return Mutate(data_, u, parent_width_, WidthU(u));
686 } else if (type_ == FBT_INDIRECT_UINT) {
687 auto u = static_cast<uint64_t>(i);
688 return Mutate(Indirect(), u, byte_width_, WidthU(u));
689 } else {
690 return false;
691 }
692 }
693
694 bool MutateBool(bool b) {
695 return type_ == FBT_BOOL && Mutate(data_, b, parent_width_, BIT_WIDTH_8);
696 }
697
698 bool MutateUInt(uint64_t u) {
699 if (type_ == FBT_UINT) {
700 return Mutate(data_, u, parent_width_, WidthU(u));
701 } else if (type_ == FBT_INDIRECT_UINT) {
702 return Mutate(Indirect(), u, byte_width_, WidthU(u));
703 } else if (type_ == FBT_INT) {
704 auto i = static_cast<int64_t>(u);
705 return Mutate(data_, i, parent_width_, WidthI(i));
706 } else if (type_ == FBT_INDIRECT_INT) {
707 auto i = static_cast<int64_t>(u);
708 return Mutate(Indirect(), i, byte_width_, WidthI(i));
709 } else {
710 return false;
711 }
712 }
713
714 bool MutateFloat(float f) {
715 if (type_ == FBT_FLOAT) {
716 return MutateF(data_, f, parent_width_, BIT_WIDTH_32);
717 } else if (type_ == FBT_INDIRECT_FLOAT) {
718 return MutateF(Indirect(), f, byte_width_, BIT_WIDTH_32);
719 } else {
720 return false;
721 }
722 }
723
724 bool MutateFloat(double d) {
725 if (type_ == FBT_FLOAT) {
726 return MutateF(data_, d, parent_width_, WidthF(d));
727 } else if (type_ == FBT_INDIRECT_FLOAT) {
728 return MutateF(Indirect(), d, byte_width_, WidthF(d));
729 } else {
730 return false;
731 }
732 }
733
734 bool MutateString(const char *str, size_t len) {
735 auto s = AsString();
736 if (s.IsTheEmptyString()) return false;
737 // This is very strict, could allow shorter strings, but that creates
738 // garbage.
739 if (s.length() != len) return false;
740 memcpy(const_cast<char *>(s.c_str()), str, len);
741 return true;
742 }
743 bool MutateString(const char *str) { return MutateString(str, strlen(str)); }
744 bool MutateString(const std::string &str) {
745 return MutateString(str.data(), str.length());
746 }
747
748 private:
749 const uint8_t *Indirect() const {
750 return flexbuffers::Indirect(data_, parent_width_);
751 }
752
753 template<typename T>
754 bool Mutate(const uint8_t *dest, T t, size_t byte_width,
755 BitWidth value_width) {
756 auto fits = static_cast<size_t>(static_cast<size_t>(1U) << value_width) <=
757 byte_width;
758 if (fits) {
759 t = flatbuffers::EndianScalar(t);
760 memcpy(const_cast<uint8_t *>(dest), &t, byte_width);
761 }
762 return fits;
763 }
764
765 template<typename T>
766 bool MutateF(const uint8_t *dest, T t, size_t byte_width,
767 BitWidth value_width) {
768 if (byte_width == sizeof(double))
769 return Mutate(dest, static_cast<double>(t), byte_width, value_width);
770 if (byte_width == sizeof(float))
771 return Mutate(dest, static_cast<float>(t), byte_width, value_width);
772 FLATBUFFERS_ASSERT(false);
773 return false;
774 }
775
James Kuszmaul8e62b022022-03-22 09:33:25 -0700776 friend class Verifier;
777
Austin Schuhe89fa2d2019-08-14 20:24:23 -0700778 const uint8_t *data_;
779 uint8_t parent_width_;
780 uint8_t byte_width_;
781 Type type_;
782};
783
784// Template specialization for As().
785template<> inline bool Reference::As<bool>() const { return AsBool(); }
786
787template<> inline int8_t Reference::As<int8_t>() const { return AsInt8(); }
788template<> inline int16_t Reference::As<int16_t>() const { return AsInt16(); }
789template<> inline int32_t Reference::As<int32_t>() const { return AsInt32(); }
790template<> inline int64_t Reference::As<int64_t>() const { return AsInt64(); }
791
792template<> inline uint8_t Reference::As<uint8_t>() const { return AsUInt8(); }
Austin Schuh272c6132020-11-14 16:37:52 -0800793template<> inline uint16_t Reference::As<uint16_t>() const {
794 return AsUInt16();
795}
796template<> inline uint32_t Reference::As<uint32_t>() const {
797 return AsUInt32();
798}
799template<> inline uint64_t Reference::As<uint64_t>() const {
800 return AsUInt64();
801}
Austin Schuhe89fa2d2019-08-14 20:24:23 -0700802
803template<> inline double Reference::As<double>() const { return AsDouble(); }
804template<> inline float Reference::As<float>() const { return AsFloat(); }
805
806template<> inline String Reference::As<String>() const { return AsString(); }
807template<> inline std::string Reference::As<std::string>() const {
808 return AsString().str();
809}
810
811template<> inline Blob Reference::As<Blob>() const { return AsBlob(); }
812template<> inline Vector Reference::As<Vector>() const { return AsVector(); }
813template<> inline TypedVector Reference::As<TypedVector>() const {
814 return AsTypedVector();
815}
816template<> inline FixedTypedVector Reference::As<FixedTypedVector>() const {
817 return AsFixedTypedVector();
818}
819template<> inline Map Reference::As<Map>() const { return AsMap(); }
820
821inline uint8_t PackedType(BitWidth bit_width, Type type) {
822 return static_cast<uint8_t>(bit_width | (type << 2));
823}
824
825inline uint8_t NullPackedType() { return PackedType(BIT_WIDTH_8, FBT_NULL); }
826
827// Vector accessors.
828// Note: if you try to access outside of bounds, you get a Null value back
829// instead. Normally this would be an assert, but since this is "dynamically
830// typed" data, you may not want that (someone sends you a 2d vector and you
831// wanted 3d).
832// The Null converts seamlessly into a default value for any other type.
833// TODO(wvo): Could introduce an #ifdef that makes this into an assert?
834inline Reference Vector::operator[](size_t i) const {
835 auto len = size();
836 if (i >= len) return Reference(nullptr, 1, NullPackedType());
837 auto packed_type = (data_ + len * byte_width_)[i];
838 auto elem = data_ + i * byte_width_;
839 return Reference(elem, byte_width_, packed_type);
840}
841
842inline Reference TypedVector::operator[](size_t i) const {
843 auto len = size();
844 if (i >= len) return Reference(nullptr, 1, NullPackedType());
845 auto elem = data_ + i * byte_width_;
846 return Reference(elem, byte_width_, 1, type_);
847}
848
849inline Reference FixedTypedVector::operator[](size_t i) const {
850 if (i >= len_) return Reference(nullptr, 1, NullPackedType());
851 auto elem = data_ + i * byte_width_;
852 return Reference(elem, byte_width_, 1, type_);
853}
854
855template<typename T> int KeyCompare(const void *key, const void *elem) {
856 auto str_elem = reinterpret_cast<const char *>(
857 Indirect<T>(reinterpret_cast<const uint8_t *>(elem)));
858 auto skey = reinterpret_cast<const char *>(key);
859 return strcmp(skey, str_elem);
860}
861
862inline Reference Map::operator[](const char *key) const {
863 auto keys = Keys();
864 // We can't pass keys.byte_width_ to the comparison function, so we have
865 // to pick the right one ahead of time.
866 int (*comp)(const void *, const void *) = nullptr;
867 switch (keys.byte_width_) {
868 case 1: comp = KeyCompare<uint8_t>; break;
869 case 2: comp = KeyCompare<uint16_t>; break;
870 case 4: comp = KeyCompare<uint32_t>; break;
871 case 8: comp = KeyCompare<uint64_t>; break;
James Kuszmaul8e62b022022-03-22 09:33:25 -0700872 default: FLATBUFFERS_ASSERT(false); return Reference();
Austin Schuhe89fa2d2019-08-14 20:24:23 -0700873 }
874 auto res = std::bsearch(key, keys.data_, keys.size(), keys.byte_width_, comp);
875 if (!res) return Reference(nullptr, 1, NullPackedType());
876 auto i = (reinterpret_cast<uint8_t *>(res) - keys.data_) / keys.byte_width_;
877 return (*static_cast<const Vector *>(this))[i];
878}
879
880inline Reference Map::operator[](const std::string &key) const {
881 return (*this)[key.c_str()];
882}
883
884inline Reference GetRoot(const uint8_t *buffer, size_t size) {
885 // See Finish() below for the serialization counterpart of this.
886 // The root starts at the end of the buffer, so we parse backwards from there.
887 auto end = buffer + size;
888 auto byte_width = *--end;
889 auto packed_type = *--end;
890 end -= byte_width; // The root data item.
891 return Reference(end, byte_width, packed_type);
892}
893
894inline Reference GetRoot(const std::vector<uint8_t> &buffer) {
James Kuszmaul8e62b022022-03-22 09:33:25 -0700895 return GetRoot(buffer.data(), buffer.size());
Austin Schuhe89fa2d2019-08-14 20:24:23 -0700896}
897
898// Flags that configure how the Builder behaves.
899// The "Share" flags determine if the Builder automatically tries to pool
900// this type. Pooling can reduce the size of serialized data if there are
901// multiple maps of the same kind, at the expense of slightly slower
902// serialization (the cost of lookups) and more memory use (std::set).
903// By default this is on for keys, but off for strings.
904// Turn keys off if you have e.g. only one map.
905// Turn strings on if you expect many non-unique string values.
906// Additionally, sharing key vectors can save space if you have maps with
907// identical field populations.
908enum BuilderFlag {
909 BUILDER_FLAG_NONE = 0,
910 BUILDER_FLAG_SHARE_KEYS = 1,
911 BUILDER_FLAG_SHARE_STRINGS = 2,
912 BUILDER_FLAG_SHARE_KEYS_AND_STRINGS = 3,
913 BUILDER_FLAG_SHARE_KEY_VECTORS = 4,
914 BUILDER_FLAG_SHARE_ALL = 7,
915};
916
917class Builder FLATBUFFERS_FINAL_CLASS {
918 public:
919 Builder(size_t initial_size = 256,
920 BuilderFlag flags = BUILDER_FLAG_SHARE_KEYS)
921 : buf_(initial_size),
922 finished_(false),
Austin Schuh58b9b472020-11-25 19:12:44 -0800923 has_duplicate_keys_(false),
Austin Schuhe89fa2d2019-08-14 20:24:23 -0700924 flags_(flags),
925 force_min_bit_width_(BIT_WIDTH_8),
926 key_pool(KeyOffsetCompare(buf_)),
927 string_pool(StringOffsetCompare(buf_)) {
928 buf_.clear();
929 }
930
Austin Schuh58b9b472020-11-25 19:12:44 -0800931#ifdef FLATBUFFERS_DEFAULT_DECLARATION
932 Builder(Builder &&) = default;
933 Builder &operator=(Builder &&) = default;
934#endif
935
Austin Schuhe89fa2d2019-08-14 20:24:23 -0700936 /// @brief Get the serialized buffer (after you call `Finish()`).
937 /// @return Returns a vector owned by this class.
938 const std::vector<uint8_t> &GetBuffer() const {
939 Finished();
940 return buf_;
941 }
942
943 // Size of the buffer. Does not include unfinished values.
944 size_t GetSize() const { return buf_.size(); }
945
946 // Reset all state so we can re-use the buffer.
947 void Clear() {
948 buf_.clear();
949 stack_.clear();
950 finished_ = false;
951 // flags_ remains as-is;
952 force_min_bit_width_ = BIT_WIDTH_8;
953 key_pool.clear();
954 string_pool.clear();
955 }
956
957 // All value constructing functions below have two versions: one that
958 // takes a key (for placement inside a map) and one that doesn't (for inside
959 // vectors and elsewhere).
960
961 void Null() { stack_.push_back(Value()); }
962 void Null(const char *key) {
963 Key(key);
964 Null();
965 }
966
967 void Int(int64_t i) { stack_.push_back(Value(i, FBT_INT, WidthI(i))); }
968 void Int(const char *key, int64_t i) {
969 Key(key);
970 Int(i);
971 }
972
973 void UInt(uint64_t u) { stack_.push_back(Value(u, FBT_UINT, WidthU(u))); }
974 void UInt(const char *key, uint64_t u) {
975 Key(key);
976 UInt(u);
977 }
978
979 void Float(float f) { stack_.push_back(Value(f)); }
980 void Float(const char *key, float f) {
981 Key(key);
982 Float(f);
983 }
984
985 void Double(double f) { stack_.push_back(Value(f)); }
986 void Double(const char *key, double d) {
987 Key(key);
988 Double(d);
989 }
990
991 void Bool(bool b) { stack_.push_back(Value(b)); }
992 void Bool(const char *key, bool b) {
993 Key(key);
994 Bool(b);
995 }
996
997 void IndirectInt(int64_t i) { PushIndirect(i, FBT_INDIRECT_INT, WidthI(i)); }
998 void IndirectInt(const char *key, int64_t i) {
999 Key(key);
1000 IndirectInt(i);
1001 }
1002
1003 void IndirectUInt(uint64_t u) {
1004 PushIndirect(u, FBT_INDIRECT_UINT, WidthU(u));
1005 }
1006 void IndirectUInt(const char *key, uint64_t u) {
1007 Key(key);
1008 IndirectUInt(u);
1009 }
1010
1011 void IndirectFloat(float f) {
1012 PushIndirect(f, FBT_INDIRECT_FLOAT, BIT_WIDTH_32);
1013 }
1014 void IndirectFloat(const char *key, float f) {
1015 Key(key);
1016 IndirectFloat(f);
1017 }
1018
1019 void IndirectDouble(double f) {
1020 PushIndirect(f, FBT_INDIRECT_FLOAT, WidthF(f));
1021 }
1022 void IndirectDouble(const char *key, double d) {
1023 Key(key);
1024 IndirectDouble(d);
1025 }
1026
1027 size_t Key(const char *str, size_t len) {
1028 auto sloc = buf_.size();
1029 WriteBytes(str, len + 1);
1030 if (flags_ & BUILDER_FLAG_SHARE_KEYS) {
1031 auto it = key_pool.find(sloc);
1032 if (it != key_pool.end()) {
1033 // Already in the buffer. Remove key we just serialized, and use
1034 // existing offset instead.
1035 buf_.resize(sloc);
1036 sloc = *it;
1037 } else {
1038 key_pool.insert(sloc);
1039 }
1040 }
1041 stack_.push_back(Value(static_cast<uint64_t>(sloc), FBT_KEY, BIT_WIDTH_8));
1042 return sloc;
1043 }
1044
1045 size_t Key(const char *str) { return Key(str, strlen(str)); }
1046 size_t Key(const std::string &str) { return Key(str.c_str(), str.size()); }
1047
1048 size_t String(const char *str, size_t len) {
1049 auto reset_to = buf_.size();
1050 auto sloc = CreateBlob(str, len, 1, FBT_STRING);
1051 if (flags_ & BUILDER_FLAG_SHARE_STRINGS) {
1052 StringOffset so(sloc, len);
1053 auto it = string_pool.find(so);
1054 if (it != string_pool.end()) {
1055 // Already in the buffer. Remove string we just serialized, and use
1056 // existing offset instead.
1057 buf_.resize(reset_to);
1058 sloc = it->first;
1059 stack_.back().u_ = sloc;
1060 } else {
1061 string_pool.insert(so);
1062 }
1063 }
1064 return sloc;
1065 }
1066 size_t String(const char *str) { return String(str, strlen(str)); }
1067 size_t String(const std::string &str) {
1068 return String(str.c_str(), str.size());
1069 }
1070 void String(const flexbuffers::String &str) {
1071 String(str.c_str(), str.length());
1072 }
1073
1074 void String(const char *key, const char *str) {
1075 Key(key);
1076 String(str);
1077 }
1078 void String(const char *key, const std::string &str) {
1079 Key(key);
1080 String(str);
1081 }
1082 void String(const char *key, const flexbuffers::String &str) {
1083 Key(key);
1084 String(str);
1085 }
1086
1087 size_t Blob(const void *data, size_t len) {
1088 return CreateBlob(data, len, 0, FBT_BLOB);
1089 }
1090 size_t Blob(const std::vector<uint8_t> &v) {
James Kuszmaul8e62b022022-03-22 09:33:25 -07001091 return CreateBlob(v.data(), v.size(), 0, FBT_BLOB);
1092 }
1093
1094 void Blob(const char *key, const void *data, size_t len) {
1095 Key(key);
1096 Blob(data, len);
1097 }
1098 void Blob(const char *key, const std::vector<uint8_t> &v) {
1099 Key(key);
1100 Blob(v);
Austin Schuhe89fa2d2019-08-14 20:24:23 -07001101 }
1102
1103 // TODO(wvo): support all the FlexBuffer types (like flexbuffers::String),
1104 // e.g. Vector etc. Also in overloaded versions.
1105 // Also some FlatBuffers types?
1106
1107 size_t StartVector() { return stack_.size(); }
1108 size_t StartVector(const char *key) {
1109 Key(key);
1110 return stack_.size();
1111 }
1112 size_t StartMap() { return stack_.size(); }
1113 size_t StartMap(const char *key) {
1114 Key(key);
1115 return stack_.size();
1116 }
1117
James Kuszmaul8e62b022022-03-22 09:33:25 -07001118 // TODO(wvo): allow this to specify an alignment greater than the natural
Austin Schuhe89fa2d2019-08-14 20:24:23 -07001119 // alignment.
1120 size_t EndVector(size_t start, bool typed, bool fixed) {
1121 auto vec = CreateVector(start, stack_.size() - start, 1, typed, fixed);
1122 // Remove temp elements and return vector.
1123 stack_.resize(start);
1124 stack_.push_back(vec);
1125 return static_cast<size_t>(vec.u_);
1126 }
1127
1128 size_t EndMap(size_t start) {
1129 // We should have interleaved keys and values on the stack.
1130 // Make sure it is an even number:
1131 auto len = stack_.size() - start;
1132 FLATBUFFERS_ASSERT(!(len & 1));
1133 len /= 2;
1134 // Make sure keys are all strings:
1135 for (auto key = start; key < stack_.size(); key += 2) {
1136 FLATBUFFERS_ASSERT(stack_[key].type_ == FBT_KEY);
1137 }
Austin Schuh272c6132020-11-14 16:37:52 -08001138 // Now sort values, so later we can do a binary search lookup.
Austin Schuhe89fa2d2019-08-14 20:24:23 -07001139 // We want to sort 2 array elements at a time.
1140 struct TwoValue {
1141 Value key;
1142 Value val;
1143 };
1144 // TODO(wvo): strict aliasing?
1145 // TODO(wvo): allow the caller to indicate the data is already sorted
1146 // for maximum efficiency? With an assert to check sortedness to make sure
1147 // we're not breaking binary search.
1148 // Or, we can track if the map is sorted as keys are added which would be
1149 // be quite cheap (cheaper than checking it here), so we can skip this
1150 // step automatically when appliccable, and encourage people to write in
1151 // sorted fashion.
1152 // std::sort is typically already a lot faster on sorted data though.
James Kuszmaul8e62b022022-03-22 09:33:25 -07001153 auto dict = reinterpret_cast<TwoValue *>(stack_.data() + start);
1154 std::sort(
1155 dict, dict + len, [&](const TwoValue &a, const TwoValue &b) -> bool {
1156 auto as = reinterpret_cast<const char *>(buf_.data() + a.key.u_);
1157 auto bs = reinterpret_cast<const char *>(buf_.data() + b.key.u_);
1158 auto comp = strcmp(as, bs);
1159 // We want to disallow duplicate keys, since this results in a
1160 // map where values cannot be found.
1161 // But we can't assert here (since we don't want to fail on
1162 // random JSON input) or have an error mechanism.
1163 // Instead, we set has_duplicate_keys_ in the builder to
1164 // signal this.
1165 // TODO: Have to check for pointer equality, as some sort
1166 // implementation apparently call this function with the same
1167 // element?? Why?
1168 if (!comp && &a != &b) has_duplicate_keys_ = true;
1169 return comp < 0;
1170 });
Austin Schuhe89fa2d2019-08-14 20:24:23 -07001171 // First create a vector out of all keys.
1172 // TODO(wvo): if kBuilderFlagShareKeyVectors is true, see if we can share
1173 // the first vector.
1174 auto keys = CreateVector(start, len, 2, true, false);
1175 auto vec = CreateVector(start + 1, len, 2, false, false, &keys);
1176 // Remove temp elements and return map.
1177 stack_.resize(start);
1178 stack_.push_back(vec);
1179 return static_cast<size_t>(vec.u_);
1180 }
1181
Austin Schuh58b9b472020-11-25 19:12:44 -08001182 // Call this after EndMap to see if the map had any duplicate keys.
1183 // Any map with such keys won't be able to retrieve all values.
1184 bool HasDuplicateKeys() const { return has_duplicate_keys_; }
1185
Austin Schuhe89fa2d2019-08-14 20:24:23 -07001186 template<typename F> size_t Vector(F f) {
1187 auto start = StartVector();
1188 f();
1189 return EndVector(start, false, false);
1190 }
1191 template<typename F, typename T> size_t Vector(F f, T &state) {
1192 auto start = StartVector();
1193 f(state);
1194 return EndVector(start, false, false);
1195 }
1196 template<typename F> size_t Vector(const char *key, F f) {
1197 auto start = StartVector(key);
1198 f();
1199 return EndVector(start, false, false);
1200 }
1201 template<typename F, typename T>
1202 size_t Vector(const char *key, F f, T &state) {
1203 auto start = StartVector(key);
1204 f(state);
1205 return EndVector(start, false, false);
1206 }
1207
1208 template<typename T> void Vector(const T *elems, size_t len) {
1209 if (flatbuffers::is_scalar<T>::value) {
1210 // This path should be a lot quicker and use less space.
1211 ScalarVector(elems, len, false);
1212 } else {
1213 auto start = StartVector();
1214 for (size_t i = 0; i < len; i++) Add(elems[i]);
1215 EndVector(start, false, false);
1216 }
1217 }
1218 template<typename T>
1219 void Vector(const char *key, const T *elems, size_t len) {
1220 Key(key);
1221 Vector(elems, len);
1222 }
1223 template<typename T> void Vector(const std::vector<T> &vec) {
James Kuszmaul8e62b022022-03-22 09:33:25 -07001224 Vector(vec.data(), vec.size());
Austin Schuhe89fa2d2019-08-14 20:24:23 -07001225 }
1226
1227 template<typename F> size_t TypedVector(F f) {
1228 auto start = StartVector();
1229 f();
1230 return EndVector(start, true, false);
1231 }
1232 template<typename F, typename T> size_t TypedVector(F f, T &state) {
1233 auto start = StartVector();
1234 f(state);
1235 return EndVector(start, true, false);
1236 }
1237 template<typename F> size_t TypedVector(const char *key, F f) {
1238 auto start = StartVector(key);
1239 f();
1240 return EndVector(start, true, false);
1241 }
1242 template<typename F, typename T>
1243 size_t TypedVector(const char *key, F f, T &state) {
1244 auto start = StartVector(key);
1245 f(state);
1246 return EndVector(start, true, false);
1247 }
1248
1249 template<typename T> size_t FixedTypedVector(const T *elems, size_t len) {
1250 // We only support a few fixed vector lengths. Anything bigger use a
1251 // regular typed vector.
1252 FLATBUFFERS_ASSERT(len >= 2 && len <= 4);
1253 // And only scalar values.
1254 static_assert(flatbuffers::is_scalar<T>::value, "Unrelated types");
1255 return ScalarVector(elems, len, true);
1256 }
1257
1258 template<typename T>
1259 size_t FixedTypedVector(const char *key, const T *elems, size_t len) {
1260 Key(key);
1261 return FixedTypedVector(elems, len);
1262 }
1263
1264 template<typename F> size_t Map(F f) {
1265 auto start = StartMap();
1266 f();
1267 return EndMap(start);
1268 }
1269 template<typename F, typename T> size_t Map(F f, T &state) {
1270 auto start = StartMap();
1271 f(state);
1272 return EndMap(start);
1273 }
1274 template<typename F> size_t Map(const char *key, F f) {
1275 auto start = StartMap(key);
1276 f();
1277 return EndMap(start);
1278 }
1279 template<typename F, typename T> size_t Map(const char *key, F f, T &state) {
1280 auto start = StartMap(key);
1281 f(state);
1282 return EndMap(start);
1283 }
1284 template<typename T> void Map(const std::map<std::string, T> &map) {
1285 auto start = StartMap();
1286 for (auto it = map.begin(); it != map.end(); ++it)
1287 Add(it->first.c_str(), it->second);
1288 EndMap(start);
1289 }
1290
Austin Schuh272c6132020-11-14 16:37:52 -08001291 // If you wish to share a value explicitly (a value not shared automatically
1292 // through one of the BUILDER_FLAG_SHARE_* flags) you can do so with these
1293 // functions. Or if you wish to turn those flags off for performance reasons
1294 // and still do some explicit sharing. For example:
1295 // builder.IndirectDouble(M_PI);
1296 // auto id = builder.LastValue(); // Remember where we stored it.
1297 // .. more code goes here ..
1298 // builder.ReuseValue(id); // Refers to same double by offset.
1299 // LastValue works regardless of whether the value has a key or not.
1300 // Works on any data type.
1301 struct Value;
1302 Value LastValue() { return stack_.back(); }
1303 void ReuseValue(Value v) { stack_.push_back(v); }
1304 void ReuseValue(const char *key, Value v) {
1305 Key(key);
1306 ReuseValue(v);
1307 }
1308
Austin Schuhe89fa2d2019-08-14 20:24:23 -07001309 // Overloaded Add that tries to call the correct function above.
1310 void Add(int8_t i) { Int(i); }
1311 void Add(int16_t i) { Int(i); }
1312 void Add(int32_t i) { Int(i); }
1313 void Add(int64_t i) { Int(i); }
1314 void Add(uint8_t u) { UInt(u); }
1315 void Add(uint16_t u) { UInt(u); }
1316 void Add(uint32_t u) { UInt(u); }
1317 void Add(uint64_t u) { UInt(u); }
1318 void Add(float f) { Float(f); }
1319 void Add(double d) { Double(d); }
1320 void Add(bool b) { Bool(b); }
1321 void Add(const char *str) { String(str); }
1322 void Add(const std::string &str) { String(str); }
1323 void Add(const flexbuffers::String &str) { String(str); }
1324
1325 template<typename T> void Add(const std::vector<T> &vec) { Vector(vec); }
1326
1327 template<typename T> void Add(const char *key, const T &t) {
1328 Key(key);
1329 Add(t);
1330 }
1331
1332 template<typename T> void Add(const std::map<std::string, T> &map) {
1333 Map(map);
1334 }
1335
1336 template<typename T> void operator+=(const T &t) { Add(t); }
1337
1338 // This function is useful in combination with the Mutate* functions above.
1339 // It forces elements of vectors and maps to have a minimum size, such that
1340 // they can later be updated without failing.
1341 // Call with no arguments to reset.
1342 void ForceMinimumBitWidth(BitWidth bw = BIT_WIDTH_8) {
1343 force_min_bit_width_ = bw;
1344 }
1345
1346 void Finish() {
1347 // If you hit this assert, you likely have objects that were never included
1348 // in a parent. You need to have exactly one root to finish a buffer.
1349 // Check your Start/End calls are matched, and all objects are inside
1350 // some other object.
1351 FLATBUFFERS_ASSERT(stack_.size() == 1);
1352
1353 // Write root value.
1354 auto byte_width = Align(stack_[0].ElemWidth(buf_.size(), 0));
1355 WriteAny(stack_[0], byte_width);
1356 // Write root type.
1357 Write(stack_[0].StoredPackedType(), 1);
1358 // Write root size. Normally determined by parent, but root has no parent :)
1359 Write(byte_width, 1);
1360
1361 finished_ = true;
1362 }
1363
1364 private:
1365 void Finished() const {
1366 // If you get this assert, you're attempting to get access a buffer
1367 // which hasn't been finished yet. Be sure to call
1368 // Builder::Finish with your root object.
1369 FLATBUFFERS_ASSERT(finished_);
1370 }
1371
1372 // Align to prepare for writing a scalar with a certain size.
1373 uint8_t Align(BitWidth alignment) {
1374 auto byte_width = 1U << alignment;
1375 buf_.insert(buf_.end(), flatbuffers::PaddingBytes(buf_.size(), byte_width),
1376 0);
1377 return static_cast<uint8_t>(byte_width);
1378 }
1379
1380 void WriteBytes(const void *val, size_t size) {
1381 buf_.insert(buf_.end(), reinterpret_cast<const uint8_t *>(val),
1382 reinterpret_cast<const uint8_t *>(val) + size);
1383 }
1384
1385 template<typename T> void Write(T val, size_t byte_width) {
1386 FLATBUFFERS_ASSERT(sizeof(T) >= byte_width);
1387 val = flatbuffers::EndianScalar(val);
1388 WriteBytes(&val, byte_width);
1389 }
1390
1391 void WriteDouble(double f, uint8_t byte_width) {
1392 switch (byte_width) {
1393 case 8: Write(f, byte_width); break;
1394 case 4: Write(static_cast<float>(f), byte_width); break;
1395 // case 2: Write(static_cast<half>(f), byte_width); break;
1396 // case 1: Write(static_cast<quarter>(f), byte_width); break;
1397 default: FLATBUFFERS_ASSERT(0);
1398 }
1399 }
1400
1401 void WriteOffset(uint64_t o, uint8_t byte_width) {
1402 auto reloff = buf_.size() - o;
1403 FLATBUFFERS_ASSERT(byte_width == 8 || reloff < 1ULL << (byte_width * 8));
1404 Write(reloff, byte_width);
1405 }
1406
1407 template<typename T> void PushIndirect(T val, Type type, BitWidth bit_width) {
1408 auto byte_width = Align(bit_width);
1409 auto iloc = buf_.size();
1410 Write(val, byte_width);
1411 stack_.push_back(Value(static_cast<uint64_t>(iloc), type, bit_width));
1412 }
1413
1414 static BitWidth WidthB(size_t byte_width) {
1415 switch (byte_width) {
1416 case 1: return BIT_WIDTH_8;
1417 case 2: return BIT_WIDTH_16;
1418 case 4: return BIT_WIDTH_32;
1419 case 8: return BIT_WIDTH_64;
1420 default: FLATBUFFERS_ASSERT(false); return BIT_WIDTH_64;
1421 }
1422 }
1423
1424 template<typename T> static Type GetScalarType() {
1425 static_assert(flatbuffers::is_scalar<T>::value, "Unrelated types");
Austin Schuh2dd86a92022-09-14 21:19:23 -07001426 return flatbuffers::is_floating_point<T>::value
1427 ? FBT_FLOAT
1428 : flatbuffers::is_same<T, bool>::value
1429 ? FBT_BOOL
1430 : (flatbuffers::is_unsigned<T>::value ? FBT_UINT
1431 : FBT_INT);
Austin Schuhe89fa2d2019-08-14 20:24:23 -07001432 }
1433
Austin Schuh272c6132020-11-14 16:37:52 -08001434 public:
1435 // This was really intended to be private, except for LastValue/ReuseValue.
Austin Schuhe89fa2d2019-08-14 20:24:23 -07001436 struct Value {
1437 union {
1438 int64_t i_;
1439 uint64_t u_;
1440 double f_;
1441 };
1442
1443 Type type_;
1444
1445 // For scalars: of itself, for vector: of its elements, for string: length.
1446 BitWidth min_bit_width_;
1447
1448 Value() : i_(0), type_(FBT_NULL), min_bit_width_(BIT_WIDTH_8) {}
1449
1450 Value(bool b)
1451 : u_(static_cast<uint64_t>(b)),
1452 type_(FBT_BOOL),
1453 min_bit_width_(BIT_WIDTH_8) {}
1454
1455 Value(int64_t i, Type t, BitWidth bw)
1456 : i_(i), type_(t), min_bit_width_(bw) {}
1457 Value(uint64_t u, Type t, BitWidth bw)
1458 : u_(u), type_(t), min_bit_width_(bw) {}
1459
Austin Schuh272c6132020-11-14 16:37:52 -08001460 Value(float f)
1461 : f_(static_cast<double>(f)),
1462 type_(FBT_FLOAT),
1463 min_bit_width_(BIT_WIDTH_32) {}
Austin Schuhe89fa2d2019-08-14 20:24:23 -07001464 Value(double f) : f_(f), type_(FBT_FLOAT), min_bit_width_(WidthF(f)) {}
1465
1466 uint8_t StoredPackedType(BitWidth parent_bit_width_ = BIT_WIDTH_8) const {
1467 return PackedType(StoredWidth(parent_bit_width_), type_);
1468 }
1469
1470 BitWidth ElemWidth(size_t buf_size, size_t elem_index) const {
1471 if (IsInline(type_)) {
1472 return min_bit_width_;
1473 } else {
1474 // We have an absolute offset, but want to store a relative offset
1475 // elem_index elements beyond the current buffer end. Since whether
1476 // the relative offset fits in a certain byte_width depends on
1477 // the size of the elements before it (and their alignment), we have
1478 // to test for each size in turn.
1479 for (size_t byte_width = 1;
1480 byte_width <= sizeof(flatbuffers::largest_scalar_t);
1481 byte_width *= 2) {
1482 // Where are we going to write this offset?
1483 auto offset_loc = buf_size +
1484 flatbuffers::PaddingBytes(buf_size, byte_width) +
1485 elem_index * byte_width;
1486 // Compute relative offset.
1487 auto offset = offset_loc - u_;
1488 // Does it fit?
1489 auto bit_width = WidthU(offset);
1490 if (static_cast<size_t>(static_cast<size_t>(1U) << bit_width) ==
1491 byte_width)
1492 return bit_width;
1493 }
1494 FLATBUFFERS_ASSERT(false); // Must match one of the sizes above.
1495 return BIT_WIDTH_64;
1496 }
1497 }
1498
1499 BitWidth StoredWidth(BitWidth parent_bit_width_ = BIT_WIDTH_8) const {
1500 if (IsInline(type_)) {
1501 return (std::max)(min_bit_width_, parent_bit_width_);
1502 } else {
1503 return min_bit_width_;
1504 }
1505 }
1506 };
1507
Austin Schuh272c6132020-11-14 16:37:52 -08001508 private:
Austin Schuhe89fa2d2019-08-14 20:24:23 -07001509 void WriteAny(const Value &val, uint8_t byte_width) {
1510 switch (val.type_) {
1511 case FBT_NULL:
1512 case FBT_INT: Write(val.i_, byte_width); break;
1513 case FBT_BOOL:
1514 case FBT_UINT: Write(val.u_, byte_width); break;
1515 case FBT_FLOAT: WriteDouble(val.f_, byte_width); break;
1516 default: WriteOffset(val.u_, byte_width); break;
1517 }
1518 }
1519
1520 size_t CreateBlob(const void *data, size_t len, size_t trailing, Type type) {
1521 auto bit_width = WidthU(len);
1522 auto byte_width = Align(bit_width);
1523 Write<uint64_t>(len, byte_width);
1524 auto sloc = buf_.size();
1525 WriteBytes(data, len + trailing);
1526 stack_.push_back(Value(static_cast<uint64_t>(sloc), type, bit_width));
1527 return sloc;
1528 }
1529
1530 template<typename T>
1531 size_t ScalarVector(const T *elems, size_t len, bool fixed) {
1532 auto vector_type = GetScalarType<T>();
1533 auto byte_width = sizeof(T);
1534 auto bit_width = WidthB(byte_width);
1535 // If you get this assert, you're trying to write a vector with a size
1536 // field that is bigger than the scalars you're trying to write (e.g. a
1537 // byte vector > 255 elements). For such types, write a "blob" instead.
1538 // TODO: instead of asserting, could write vector with larger elements
1539 // instead, though that would be wasteful.
1540 FLATBUFFERS_ASSERT(WidthU(len) <= bit_width);
Austin Schuh272c6132020-11-14 16:37:52 -08001541 Align(bit_width);
Austin Schuhe89fa2d2019-08-14 20:24:23 -07001542 if (!fixed) Write<uint64_t>(len, byte_width);
1543 auto vloc = buf_.size();
1544 for (size_t i = 0; i < len; i++) Write(elems[i], byte_width);
1545 stack_.push_back(Value(static_cast<uint64_t>(vloc),
1546 ToTypedVector(vector_type, fixed ? len : 0),
1547 bit_width));
1548 return vloc;
1549 }
1550
1551 Value CreateVector(size_t start, size_t vec_len, size_t step, bool typed,
1552 bool fixed, const Value *keys = nullptr) {
Austin Schuh272c6132020-11-14 16:37:52 -08001553 FLATBUFFERS_ASSERT(
1554 !fixed ||
1555 typed); // typed=false, fixed=true combination is not supported.
Austin Schuhe89fa2d2019-08-14 20:24:23 -07001556 // Figure out smallest bit width we can store this vector with.
1557 auto bit_width = (std::max)(force_min_bit_width_, WidthU(vec_len));
1558 auto prefix_elems = 1;
1559 if (keys) {
1560 // If this vector is part of a map, we will pre-fix an offset to the keys
1561 // to this vector.
1562 bit_width = (std::max)(bit_width, keys->ElemWidth(buf_.size(), 0));
1563 prefix_elems += 2;
1564 }
1565 Type vector_type = FBT_KEY;
1566 // Check bit widths and types for all elements.
1567 for (size_t i = start; i < stack_.size(); i += step) {
Austin Schuh272c6132020-11-14 16:37:52 -08001568 auto elem_width =
1569 stack_[i].ElemWidth(buf_.size(), i - start + prefix_elems);
Austin Schuhe89fa2d2019-08-14 20:24:23 -07001570 bit_width = (std::max)(bit_width, elem_width);
1571 if (typed) {
1572 if (i == start) {
1573 vector_type = stack_[i].type_;
1574 } else {
1575 // If you get this assert, you are writing a typed vector with
1576 // elements that are not all the same type.
1577 FLATBUFFERS_ASSERT(vector_type == stack_[i].type_);
1578 }
1579 }
1580 }
James Kuszmaul8e62b022022-03-22 09:33:25 -07001581 // If you get this assert, your typed types are not one of:
Austin Schuhe89fa2d2019-08-14 20:24:23 -07001582 // Int / UInt / Float / Key.
James Kuszmaul8e62b022022-03-22 09:33:25 -07001583 FLATBUFFERS_ASSERT(!typed || IsTypedVectorElementType(vector_type));
Austin Schuhe89fa2d2019-08-14 20:24:23 -07001584 auto byte_width = Align(bit_width);
1585 // Write vector. First the keys width/offset if available, and size.
1586 if (keys) {
1587 WriteOffset(keys->u_, byte_width);
1588 Write<uint64_t>(1ULL << keys->min_bit_width_, byte_width);
1589 }
1590 if (!fixed) Write<uint64_t>(vec_len, byte_width);
1591 // Then the actual data.
1592 auto vloc = buf_.size();
1593 for (size_t i = start; i < stack_.size(); i += step) {
1594 WriteAny(stack_[i], byte_width);
1595 }
1596 // Then the types.
1597 if (!typed) {
1598 for (size_t i = start; i < stack_.size(); i += step) {
1599 buf_.push_back(stack_[i].StoredPackedType(bit_width));
1600 }
1601 }
1602 return Value(static_cast<uint64_t>(vloc),
1603 keys ? FBT_MAP
1604 : (typed ? ToTypedVector(vector_type, fixed ? vec_len : 0)
1605 : FBT_VECTOR),
1606 bit_width);
1607 }
1608
1609 // You shouldn't really be copying instances of this class.
1610 Builder(const Builder &);
1611 Builder &operator=(const Builder &);
1612
1613 std::vector<uint8_t> buf_;
1614 std::vector<Value> stack_;
1615
1616 bool finished_;
Austin Schuh58b9b472020-11-25 19:12:44 -08001617 bool has_duplicate_keys_;
Austin Schuhe89fa2d2019-08-14 20:24:23 -07001618
1619 BuilderFlag flags_;
1620
1621 BitWidth force_min_bit_width_;
1622
1623 struct KeyOffsetCompare {
1624 explicit KeyOffsetCompare(const std::vector<uint8_t> &buf) : buf_(&buf) {}
1625 bool operator()(size_t a, size_t b) const {
James Kuszmaul8e62b022022-03-22 09:33:25 -07001626 auto stra = reinterpret_cast<const char *>(buf_->data() + a);
1627 auto strb = reinterpret_cast<const char *>(buf_->data() + b);
Austin Schuhe89fa2d2019-08-14 20:24:23 -07001628 return strcmp(stra, strb) < 0;
1629 }
1630 const std::vector<uint8_t> *buf_;
1631 };
1632
1633 typedef std::pair<size_t, size_t> StringOffset;
1634 struct StringOffsetCompare {
Austin Schuh272c6132020-11-14 16:37:52 -08001635 explicit StringOffsetCompare(const std::vector<uint8_t> &buf)
1636 : buf_(&buf) {}
Austin Schuhe89fa2d2019-08-14 20:24:23 -07001637 bool operator()(const StringOffset &a, const StringOffset &b) const {
James Kuszmaul8e62b022022-03-22 09:33:25 -07001638 auto stra = buf_->data() + a.first;
1639 auto strb = buf_->data() + b.first;
1640 auto cr = memcmp(stra, strb, (std::min)(a.second, b.second) + 1);
1641 return cr < 0 || (cr == 0 && a.second < b.second);
Austin Schuhe89fa2d2019-08-14 20:24:23 -07001642 }
1643 const std::vector<uint8_t> *buf_;
1644 };
1645
1646 typedef std::set<size_t, KeyOffsetCompare> KeyOffsetMap;
1647 typedef std::set<StringOffset, StringOffsetCompare> StringOffsetMap;
1648
1649 KeyOffsetMap key_pool;
1650 StringOffsetMap string_pool;
James Kuszmaul8e62b022022-03-22 09:33:25 -07001651
1652 friend class Verifier;
Austin Schuhe89fa2d2019-08-14 20:24:23 -07001653};
1654
James Kuszmaul8e62b022022-03-22 09:33:25 -07001655// Helper class to verify the integrity of a FlexBuffer
1656class Verifier FLATBUFFERS_FINAL_CLASS {
1657 public:
1658 Verifier(const uint8_t *buf, size_t buf_len,
1659 // Supplying this vector likely results in faster verification
1660 // of larger buffers with many shared keys/strings, but
1661 // comes at the cost of using additional memory the same size of
1662 // the buffer being verified, so it is by default off.
1663 std::vector<uint8_t> *reuse_tracker = nullptr,
1664 bool _check_alignment = true, size_t max_depth = 64)
1665 : buf_(buf),
1666 size_(buf_len),
1667 depth_(0),
1668 max_depth_(max_depth),
1669 num_vectors_(0),
1670 max_vectors_(buf_len),
1671 check_alignment_(_check_alignment),
1672 reuse_tracker_(reuse_tracker) {
1673 FLATBUFFERS_ASSERT(size_ < FLATBUFFERS_MAX_BUFFER_SIZE);
1674 if (reuse_tracker_) {
1675 reuse_tracker_->clear();
1676 reuse_tracker_->resize(size_, PackedType(BIT_WIDTH_8, FBT_NULL));
1677 }
1678 }
1679
1680 private:
1681 // Central location where any verification failures register.
1682 bool Check(bool ok) const {
1683 // clang-format off
1684 #ifdef FLATBUFFERS_DEBUG_VERIFICATION_FAILURE
1685 FLATBUFFERS_ASSERT(ok);
1686 #endif
1687 // clang-format on
1688 return ok;
1689 }
1690
1691 // Verify any range within the buffer.
1692 bool VerifyFrom(size_t elem, size_t elem_len) const {
1693 return Check(elem_len < size_ && elem <= size_ - elem_len);
1694 }
1695 bool VerifyBefore(size_t elem, size_t elem_len) const {
1696 return Check(elem_len <= elem);
1697 }
1698
1699 bool VerifyFromPointer(const uint8_t *p, size_t len) {
1700 auto o = static_cast<size_t>(p - buf_);
1701 return VerifyFrom(o, len);
1702 }
1703 bool VerifyBeforePointer(const uint8_t *p, size_t len) {
1704 auto o = static_cast<size_t>(p - buf_);
1705 return VerifyBefore(o, len);
1706 }
1707
1708 bool VerifyByteWidth(size_t width) {
1709 return Check(width == 1 || width == 2 || width == 4 || width == 8);
1710 }
1711
1712 bool VerifyType(int type) { return Check(type >= 0 && type < FBT_MAX_TYPE); }
1713
1714 bool VerifyOffset(uint64_t off, const uint8_t *p) {
1715 return Check(off <= static_cast<uint64_t>(size_)) &&
1716 off <= static_cast<uint64_t>(p - buf_);
1717 }
1718
1719 bool VerifyAlignment(const uint8_t *p, size_t size) const {
1720 auto o = static_cast<size_t>(p - buf_);
1721 return Check((o & (size - 1)) == 0 || !check_alignment_);
1722 }
1723
1724// Macro, since we want to escape from parent function & use lazy args.
1725#define FLEX_CHECK_VERIFIED(P, PACKED_TYPE) \
1726 if (reuse_tracker_) { \
1727 auto packed_type = PACKED_TYPE; \
1728 auto existing = (*reuse_tracker_)[P - buf_]; \
1729 if (existing == packed_type) return true; \
1730 /* Fail verification if already set with different type! */ \
1731 if (!Check(existing == 0)) return false; \
1732 (*reuse_tracker_)[P - buf_] = packed_type; \
1733 }
1734
1735 bool VerifyVector(Reference r, const uint8_t *p, Type elem_type) {
1736 // Any kind of nesting goes thru this function, so guard against that
1737 // here, both with simple nesting checks, and the reuse tracker if on.
1738 depth_++;
1739 num_vectors_++;
1740 if (!Check(depth_ <= max_depth_ && num_vectors_ <= max_vectors_))
1741 return false;
1742 auto size_byte_width = r.byte_width_;
James Kuszmaul8e62b022022-03-22 09:33:25 -07001743 if (!VerifyBeforePointer(p, size_byte_width)) return false;
Austin Schuh2dd86a92022-09-14 21:19:23 -07001744 FLEX_CHECK_VERIFIED(p - size_byte_width,
1745 PackedType(Builder::WidthB(size_byte_width), r.type_));
James Kuszmaul8e62b022022-03-22 09:33:25 -07001746 auto sized = Sized(p, size_byte_width);
1747 auto num_elems = sized.size();
1748 auto elem_byte_width = r.type_ == FBT_STRING || r.type_ == FBT_BLOB
1749 ? uint8_t(1)
1750 : r.byte_width_;
1751 auto max_elems = SIZE_MAX / elem_byte_width;
1752 if (!Check(num_elems < max_elems))
1753 return false; // Protect against byte_size overflowing.
1754 auto byte_size = num_elems * elem_byte_width;
1755 if (!VerifyFromPointer(p, byte_size)) return false;
1756 if (elem_type == FBT_NULL) {
1757 // Verify type bytes after the vector.
1758 if (!VerifyFromPointer(p + byte_size, num_elems)) return false;
1759 auto v = Vector(p, size_byte_width);
1760 for (size_t i = 0; i < num_elems; i++)
1761 if (!VerifyRef(v[i])) return false;
1762 } else if (elem_type == FBT_KEY) {
1763 auto v = TypedVector(p, elem_byte_width, FBT_KEY);
1764 for (size_t i = 0; i < num_elems; i++)
1765 if (!VerifyRef(v[i])) return false;
1766 } else {
1767 FLATBUFFERS_ASSERT(IsInline(elem_type));
1768 }
1769 depth_--;
1770 return true;
1771 }
1772
1773 bool VerifyKeys(const uint8_t *p, uint8_t byte_width) {
1774 // The vector part of the map has already been verified.
1775 const size_t num_prefixed_fields = 3;
1776 if (!VerifyBeforePointer(p, byte_width * num_prefixed_fields)) return false;
1777 p -= byte_width * num_prefixed_fields;
1778 auto off = ReadUInt64(p, byte_width);
1779 if (!VerifyOffset(off, p)) return false;
1780 auto key_byte_with =
1781 static_cast<uint8_t>(ReadUInt64(p + byte_width, byte_width));
1782 if (!VerifyByteWidth(key_byte_with)) return false;
1783 return VerifyVector(Reference(p, byte_width, key_byte_with, FBT_VECTOR_KEY),
1784 p - off, FBT_KEY);
1785 }
1786
1787 bool VerifyKey(const uint8_t *p) {
1788 FLEX_CHECK_VERIFIED(p, PackedType(BIT_WIDTH_8, FBT_KEY));
1789 while (p < buf_ + size_)
1790 if (*p++) return true;
1791 return false;
1792 }
1793
1794#undef FLEX_CHECK_VERIFIED
1795
1796 bool VerifyTerminator(const String &s) {
1797 return VerifyFromPointer(reinterpret_cast<const uint8_t *>(s.c_str()),
1798 s.size() + 1);
1799 }
1800
1801 bool VerifyRef(Reference r) {
1802 // r.parent_width_ and r.data_ already verified.
1803 if (!VerifyByteWidth(r.byte_width_) || !VerifyType(r.type_)) {
1804 return false;
1805 }
1806 if (IsInline(r.type_)) {
1807 // Inline scalars, don't require further verification.
1808 return true;
1809 }
1810 // All remaining types are an offset.
1811 auto off = ReadUInt64(r.data_, r.parent_width_);
1812 if (!VerifyOffset(off, r.data_)) return false;
1813 auto p = r.Indirect();
1814 if (!VerifyAlignment(p, r.byte_width_)) return false;
1815 switch (r.type_) {
1816 case FBT_INDIRECT_INT:
1817 case FBT_INDIRECT_UINT:
1818 case FBT_INDIRECT_FLOAT: return VerifyFromPointer(p, r.byte_width_);
1819 case FBT_KEY: return VerifyKey(p);
1820 case FBT_MAP:
1821 return VerifyVector(r, p, FBT_NULL) && VerifyKeys(p, r.byte_width_);
1822 case FBT_VECTOR: return VerifyVector(r, p, FBT_NULL);
1823 case FBT_VECTOR_INT: return VerifyVector(r, p, FBT_INT);
1824 case FBT_VECTOR_BOOL:
1825 case FBT_VECTOR_UINT: return VerifyVector(r, p, FBT_UINT);
1826 case FBT_VECTOR_FLOAT: return VerifyVector(r, p, FBT_FLOAT);
1827 case FBT_VECTOR_KEY: return VerifyVector(r, p, FBT_KEY);
1828 case FBT_VECTOR_STRING_DEPRECATED:
1829 // Use of FBT_KEY here intentional, see elsewhere.
1830 return VerifyVector(r, p, FBT_KEY);
1831 case FBT_BLOB: return VerifyVector(r, p, FBT_UINT);
1832 case FBT_STRING:
1833 return VerifyVector(r, p, FBT_UINT) &&
1834 VerifyTerminator(String(p, r.byte_width_));
1835 case FBT_VECTOR_INT2:
1836 case FBT_VECTOR_UINT2:
1837 case FBT_VECTOR_FLOAT2:
1838 case FBT_VECTOR_INT3:
1839 case FBT_VECTOR_UINT3:
1840 case FBT_VECTOR_FLOAT3:
1841 case FBT_VECTOR_INT4:
1842 case FBT_VECTOR_UINT4:
1843 case FBT_VECTOR_FLOAT4: {
1844 uint8_t len = 0;
1845 auto vtype = ToFixedTypedVectorElementType(r.type_, &len);
1846 if (!VerifyType(vtype)) return false;
1847 return VerifyFromPointer(p, r.byte_width_ * len);
1848 }
1849 default: return false;
1850 }
1851 }
1852
1853 public:
1854 bool VerifyBuffer() {
1855 if (!Check(size_ >= 3)) return false;
1856 auto end = buf_ + size_;
1857 auto byte_width = *--end;
1858 auto packed_type = *--end;
1859 return VerifyByteWidth(byte_width) && Check(end - buf_ >= byte_width) &&
1860 VerifyRef(Reference(end - byte_width, byte_width, packed_type));
1861 }
1862
1863 private:
1864 const uint8_t *buf_;
1865 size_t size_;
1866 size_t depth_;
1867 const size_t max_depth_;
1868 size_t num_vectors_;
1869 const size_t max_vectors_;
1870 bool check_alignment_;
1871 std::vector<uint8_t> *reuse_tracker_;
1872};
1873
Austin Schuh2dd86a92022-09-14 21:19:23 -07001874// Utility function that constructs the Verifier for you, see above for
James Kuszmaul8e62b022022-03-22 09:33:25 -07001875// parameters.
1876inline bool VerifyBuffer(const uint8_t *buf, size_t buf_len,
1877 std::vector<uint8_t> *reuse_tracker = nullptr) {
1878 Verifier verifier(buf, buf_len, reuse_tracker);
1879 return verifier.VerifyBuffer();
1880}
1881
Austin Schuhe89fa2d2019-08-14 20:24:23 -07001882} // namespace flexbuffers
1883
Austin Schuh272c6132020-11-14 16:37:52 -08001884#if defined(_MSC_VER)
1885# pragma warning(pop)
1886#endif
Austin Schuhe89fa2d2019-08-14 20:24:23 -07001887
1888#endif // FLATBUFFERS_FLEXBUFFERS_H_