forked from protocolbuffers/protobuf
/
map.h
1687 lines (1476 loc) · 58.2 KB
/
map.h
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
// Protocol Buffers - Google's data interchange format
// Copyright 2008 Google Inc. All rights reserved.
//
// Use of this source code is governed by a BSD-style
// license that can be found in the LICENSE file or at
// https://developers.google.com/open-source/licenses/bsd
// This file defines the map container and its helpers to support protobuf maps.
//
// The Map and MapIterator types are provided by this header file.
// Please avoid using other types defined here, unless they are public
// types within Map or MapIterator, such as Map::value_type.
#ifndef GOOGLE_PROTOBUF_MAP_H__
#define GOOGLE_PROTOBUF_MAP_H__
#include <algorithm>
#include <cstddef>
#include <functional>
#include <initializer_list>
#include <iterator>
#include <limits> // To support Visual Studio 2008
#include <string>
#include <type_traits>
#include <utility>
#if !defined(GOOGLE_PROTOBUF_NO_RDTSC) && defined(__APPLE__)
#include <time.h>
#endif
#include "google/protobuf/stubs/common.h"
#include "absl/base/attributes.h"
#include "absl/container/btree_map.h"
#include "absl/hash/hash.h"
#include "absl/meta/type_traits.h"
#include "absl/strings/string_view.h"
#include "google/protobuf/arena.h"
#include "google/protobuf/generated_enum_util.h"
#include "google/protobuf/internal_visibility.h"
#include "google/protobuf/map_type_handler.h"
#include "google/protobuf/port.h"
#include "google/protobuf/wire_format_lite.h"
#ifdef SWIG
#error "You cannot SWIG proto headers"
#endif
// Must be included last.
#include "google/protobuf/port_def.inc"
namespace google {
namespace protobuf {
template <typename Key, typename T>
class Map;
class MapIterator;
template <typename Enum>
struct is_proto_enum;
namespace internal {
template <typename Key, typename T>
class MapFieldLite;
template <typename Derived, typename Key, typename T,
WireFormatLite::FieldType key_wire_type,
WireFormatLite::FieldType value_wire_type>
class MapField;
struct MapTestPeer;
struct MapBenchmarkPeer;
template <typename Key, typename T>
class TypeDefinedMapFieldBase;
class DynamicMapField;
class GeneratedMessageReflection;
// The largest valid serialization for a message is INT_MAX, so we can't have
// more than 32-bits worth of elements.
using map_index_t = uint32_t;
// Internal type traits that can be used to define custom key/value types. These
// are only be specialized by protobuf internals, and never by users.
template <typename T, typename VoidT = void>
struct is_internal_map_key_type : std::false_type {};
template <typename T, typename VoidT = void>
struct is_internal_map_value_type : std::false_type {};
// re-implement std::allocator to use arena allocator for memory allocation.
// Used for Map implementation. Users should not use this class
// directly.
template <typename U>
class MapAllocator {
public:
using value_type = U;
using pointer = value_type*;
using const_pointer = const value_type*;
using reference = value_type&;
using const_reference = const value_type&;
using size_type = size_t;
using difference_type = ptrdiff_t;
constexpr MapAllocator() : arena_(nullptr) {}
explicit constexpr MapAllocator(Arena* arena) : arena_(arena) {}
template <typename X>
MapAllocator(const MapAllocator<X>& allocator) // NOLINT(runtime/explicit)
: arena_(allocator.arena()) {}
// MapAllocator does not support alignments beyond 8. Technically we should
// support up to std::max_align_t, but this fails with ubsan and tcmalloc
// debug allocation logic which assume 8 as default alignment.
static_assert(alignof(value_type) <= 8, "");
pointer allocate(size_type n, const void* /* hint */ = nullptr) {
// If arena is not given, malloc needs to be called which doesn't
// construct element object.
if (arena_ == nullptr) {
return static_cast<pointer>(::operator new(n * sizeof(value_type)));
} else {
return reinterpret_cast<pointer>(
Arena::CreateArray<uint8_t>(arena_, n * sizeof(value_type)));
}
}
void deallocate(pointer p, size_type n) {
if (arena_ == nullptr) {
internal::SizedDelete(p, n * sizeof(value_type));
}
}
#if !defined(GOOGLE_PROTOBUF_OS_APPLE) && !defined(GOOGLE_PROTOBUF_OS_NACL) && \
!defined(GOOGLE_PROTOBUF_OS_EMSCRIPTEN)
template <class NodeType, class... Args>
void construct(NodeType* p, Args&&... args) {
// Clang 3.6 doesn't compile static casting to void* directly. (Issue
// #1266) According C++ standard 5.2.9/1: "The static_cast operator shall
// not cast away constness". So first the maybe const pointer is casted to
// const void* and after the const void* is const casted.
new (const_cast<void*>(static_cast<const void*>(p)))
NodeType(std::forward<Args>(args)...);
}
template <class NodeType>
void destroy(NodeType* p) {
p->~NodeType();
}
#else
void construct(pointer p, const_reference t) { new (p) value_type(t); }
void destroy(pointer p) { p->~value_type(); }
#endif
template <typename X>
struct rebind {
using other = MapAllocator<X>;
};
template <typename X>
bool operator==(const MapAllocator<X>& other) const {
return arena_ == other.arena_;
}
template <typename X>
bool operator!=(const MapAllocator<X>& other) const {
return arena_ != other.arena_;
}
// To support Visual Studio 2008
size_type max_size() const {
// parentheses around (std::...:max) prevents macro warning of max()
return (std::numeric_limits<size_type>::max)();
}
// To support gcc-4.4, which does not properly
// support templated friend classes
Arena* arena() const { return arena_; }
private:
using DestructorSkippable_ = void;
Arena* arena_;
};
// To save on binary size and simplify generic uses of the map types we collapse
// signed/unsigned versions of the same sized integer to the unsigned version.
template <typename T, typename = void>
struct KeyForBaseImpl {
using type = T;
};
template <typename T>
struct KeyForBaseImpl<T, std::enable_if_t<std::is_integral<T>::value &&
std::is_signed<T>::value>> {
using type = std::make_unsigned_t<T>;
};
template <typename T>
using KeyForBase = typename KeyForBaseImpl<T>::type;
// Default case: Not transparent.
// We use std::hash<key_type>/std::less<key_type> and all the lookup functions
// only accept `key_type`.
template <typename key_type>
struct TransparentSupport {
// We hash all the scalars as uint64_t so that we can implement the same hash
// function for VariantKey. This way we can have MapKey provide the same hash
// as the underlying value would have.
using hash = std::hash<
std::conditional_t<std::is_scalar<key_type>::value, uint64_t, key_type>>;
static bool Equals(const key_type& a, const key_type& b) { return a == b; }
template <typename K>
using key_arg = key_type;
using ViewType = std::conditional_t<std::is_scalar<key_type>::value, key_type,
const key_type&>;
static ViewType ToView(const key_type& v) { return v; }
};
// We add transparent support for std::string keys. We use
// std::hash<absl::string_view> as it supports the input types we care about.
// The lookup functions accept arbitrary `K`. This will include any key type
// that is convertible to absl::string_view.
template <>
struct TransparentSupport<std::string> {
// If the element is not convertible to absl::string_view, try to convert to
// std::string first, and then fallback to support for converting from
// std::string_view. The ranked overload pattern is used to specify our
// order of preference.
struct Rank0 {};
struct Rank1 : Rank0 {};
struct Rank2 : Rank1 {};
template <typename T, typename = std::enable_if_t<
std::is_convertible<T, absl::string_view>::value>>
static absl::string_view ImplicitConvertImpl(T&& str, Rank2) {
absl::string_view ref = str;
return ref;
}
template <typename T, typename = std::enable_if_t<
std::is_convertible<T, const std::string&>::value>>
static absl::string_view ImplicitConvertImpl(T&& str, Rank1) {
const std::string& ref = str;
return ref;
}
template <typename T>
static absl::string_view ImplicitConvertImpl(T&& str, Rank0) {
return {str.data(), str.size()};
}
template <typename T>
static absl::string_view ImplicitConvert(T&& str) {
return ImplicitConvertImpl(std::forward<T>(str), Rank2{});
}
struct hash : public absl::Hash<absl::string_view> {
using is_transparent = void;
template <typename T>
size_t operator()(T&& str) const {
return absl::Hash<absl::string_view>::operator()(
ImplicitConvert(std::forward<T>(str)));
}
};
template <typename T, typename U>
static bool Equals(T&& t, U&& u) {
return ImplicitConvert(std::forward<T>(t)) ==
ImplicitConvert(std::forward<U>(u));
}
template <typename K>
using key_arg = K;
using ViewType = absl::string_view;
template <typename T>
static ViewType ToView(const T& v) {
return ImplicitConvert(v);
}
};
enum class MapNodeSizeInfoT : uint32_t;
inline uint16_t SizeFromInfo(MapNodeSizeInfoT node_size_info) {
return static_cast<uint16_t>(static_cast<uint32_t>(node_size_info) >> 16);
}
inline uint16_t ValueOffsetFromInfo(MapNodeSizeInfoT node_size_info) {
return static_cast<uint16_t>(static_cast<uint32_t>(node_size_info) >> 0);
}
constexpr MapNodeSizeInfoT MakeNodeInfo(uint16_t size, uint16_t value_offset) {
return static_cast<MapNodeSizeInfoT>((static_cast<uint32_t>(size) << 16) |
value_offset);
}
struct NodeBase {
// Align the node to allow KeyNode to predict the location of the key.
// This way sizeof(NodeBase) contains any possible padding it was going to
// have between NodeBase and the key.
alignas(kMaxMessageAlignment) NodeBase* next;
void* GetVoidKey() { return this + 1; }
const void* GetVoidKey() const { return this + 1; }
void* GetVoidValue(MapNodeSizeInfoT size_info) {
return reinterpret_cast<char*>(this) + ValueOffsetFromInfo(size_info);
}
};
inline NodeBase* EraseFromLinkedList(NodeBase* item, NodeBase* head) {
if (head == item) {
return head->next;
} else {
head->next = EraseFromLinkedList(item, head->next);
return head;
}
}
constexpr size_t MapTreeLengthThreshold() { return 8; }
inline bool TableEntryIsTooLong(NodeBase* node) {
const size_t kMaxLength = MapTreeLengthThreshold();
size_t count = 0;
do {
++count;
node = node->next;
} while (node != nullptr);
// Invariant: no linked list ever is more than kMaxLength in length.
ABSL_DCHECK_LE(count, kMaxLength);
return count >= kMaxLength;
}
// Similar to the public MapKey, but specialized for the internal
// implementation.
struct VariantKey {
// We make this value 16 bytes to make it cheaper to pass in the ABI.
// Can't overload string_view this way, so we unpack the fields.
// data==nullptr means this is a number and `integral` is the value.
// data!=nullptr means this is a string and `integral` is the size.
const char* data;
uint64_t integral;
explicit VariantKey(uint64_t v) : data(nullptr), integral(v) {}
explicit VariantKey(absl::string_view v)
: data(v.data()), integral(v.size()) {
// We use `data` to discriminate between the types, so make sure it is never
// null here.
if (data == nullptr) data = "";
}
size_t Hash() const {
return data == nullptr ? std::hash<uint64_t>{}(integral)
: absl::Hash<absl::string_view>{}(
absl::string_view(data, integral));
}
friend bool operator<(const VariantKey& left, const VariantKey& right) {
ABSL_DCHECK_EQ(left.data == nullptr, right.data == nullptr);
if (left.integral != right.integral) {
// If they are numbers with different value, or strings with different
// size, check the number only.
return left.integral < right.integral;
}
if (left.data == nullptr) {
// If they are numbers they have the same value, so return.
return false;
}
// They are strings of the same size, so check the bytes.
return memcmp(left.data, right.data, left.integral) < 0;
}
};
// This is to be specialized by MapKey.
template <typename T>
struct RealKeyToVariantKey {
VariantKey operator()(T value) const { return VariantKey(value); }
};
template <>
struct RealKeyToVariantKey<std::string> {
template <typename T>
VariantKey operator()(const T& value) const {
return VariantKey(TransparentSupport<std::string>::ImplicitConvert(value));
}
};
// We use a single kind of tree for all maps. This reduces code duplication.
using TreeForMap =
absl::btree_map<VariantKey, NodeBase*, std::less<VariantKey>,
MapAllocator<std::pair<const VariantKey, NodeBase*>>>;
// Type safe tagged pointer.
// We convert to/from nodes and trees using the operations below.
// They ensure that the tags are used correctly.
// There are three states:
// - x == 0: the entry is empty
// - x != 0 && (x&1) == 0: the entry is a node list
// - x != 0 && (x&1) == 1: the entry is a tree
enum class TableEntryPtr : uintptr_t;
inline bool TableEntryIsEmpty(TableEntryPtr entry) {
return entry == TableEntryPtr{};
}
inline bool TableEntryIsTree(TableEntryPtr entry) {
return (static_cast<uintptr_t>(entry) & 1) == 1;
}
inline bool TableEntryIsList(TableEntryPtr entry) {
return !TableEntryIsTree(entry);
}
inline bool TableEntryIsNonEmptyList(TableEntryPtr entry) {
return !TableEntryIsEmpty(entry) && TableEntryIsList(entry);
}
inline NodeBase* TableEntryToNode(TableEntryPtr entry) {
ABSL_DCHECK(TableEntryIsList(entry));
return reinterpret_cast<NodeBase*>(static_cast<uintptr_t>(entry));
}
inline TableEntryPtr NodeToTableEntry(NodeBase* node) {
ABSL_DCHECK((reinterpret_cast<uintptr_t>(node) & 1) == 0);
return static_cast<TableEntryPtr>(reinterpret_cast<uintptr_t>(node));
}
inline TreeForMap* TableEntryToTree(TableEntryPtr entry) {
ABSL_DCHECK(TableEntryIsTree(entry));
return reinterpret_cast<TreeForMap*>(static_cast<uintptr_t>(entry) - 1);
}
inline TableEntryPtr TreeToTableEntry(TreeForMap* node) {
ABSL_DCHECK((reinterpret_cast<uintptr_t>(node) & 1) == 0);
return static_cast<TableEntryPtr>(reinterpret_cast<uintptr_t>(node) | 1);
}
// This captures all numeric types.
inline size_t MapValueSpaceUsedExcludingSelfLong(bool) { return 0; }
inline size_t MapValueSpaceUsedExcludingSelfLong(const std::string& str) {
return StringSpaceUsedExcludingSelfLong(str);
}
template <typename T,
typename = decltype(std::declval<const T&>().SpaceUsedLong())>
size_t MapValueSpaceUsedExcludingSelfLong(const T& message) {
return message.SpaceUsedLong() - sizeof(T);
}
constexpr size_t kGlobalEmptyTableSize = 1;
PROTOBUF_EXPORT extern const TableEntryPtr
kGlobalEmptyTable[kGlobalEmptyTableSize];
template <typename Map,
typename = typename std::enable_if<
!std::is_scalar<typename Map::key_type>::value ||
!std::is_scalar<typename Map::mapped_type>::value>::type>
size_t SpaceUsedInValues(const Map* map) {
size_t size = 0;
for (const auto& v : *map) {
size += internal::MapValueSpaceUsedExcludingSelfLong(v.first) +
internal::MapValueSpaceUsedExcludingSelfLong(v.second);
}
return size;
}
inline size_t SpaceUsedInValues(const void*) { return 0; }
class UntypedMapBase;
class UntypedMapIterator {
public:
// Invariants:
// node_ is always correct. This is handy because the most common
// operations are operator* and operator-> and they only use node_.
// When node_ is set to a non-null value, all the other non-const fields
// are updated to be correct also, but those fields can become stale
// if the underlying map is modified. When those fields are needed they
// are rechecked, and updated if necessary.
UntypedMapIterator() : node_(nullptr), m_(nullptr), bucket_index_(0) {}
explicit UntypedMapIterator(const UntypedMapBase* m);
UntypedMapIterator(NodeBase* n, const UntypedMapBase* m, map_index_t index)
: node_(n), m_(m), bucket_index_(index) {}
// Advance through buckets, looking for the first that isn't empty.
// If nothing non-empty is found then leave node_ == nullptr.
void SearchFrom(map_index_t start_bucket);
// The definition of operator== is handled by the derived type. If we were
// to do it in this class it would allow comparing iterators of different
// map types.
bool Equals(const UntypedMapIterator& other) const {
return node_ == other.node_;
}
// The definition of operator++ is handled in the derived type. We would not
// be able to return the right type from here.
void PlusPlus() {
if (node_->next == nullptr) {
SearchFrom(bucket_index_ + 1);
} else {
node_ = node_->next;
}
}
NodeBase* node_;
const UntypedMapBase* m_;
map_index_t bucket_index_;
};
// Base class for all Map instantiations.
// This class holds all the data and provides the basic functionality shared
// among all instantiations.
// Having an untyped base class helps generic consumers (like the table-driven
// parser) by having non-template code that can handle all instantiations.
class PROTOBUF_EXPORT UntypedMapBase {
using Allocator = internal::MapAllocator<void*>;
using Tree = internal::TreeForMap;
public:
using size_type = size_t;
explicit constexpr UntypedMapBase(Arena* arena)
: num_elements_(0),
num_buckets_(internal::kGlobalEmptyTableSize),
seed_(0),
index_of_first_non_null_(internal::kGlobalEmptyTableSize),
table_(const_cast<TableEntryPtr*>(internal::kGlobalEmptyTable)),
alloc_(arena) {}
UntypedMapBase(const UntypedMapBase&) = delete;
UntypedMapBase& operator=(const UntypedMapBase&) = delete;
protected:
// 16 bytes is the minimum useful size for the array cache in the arena.
enum { kMinTableSize = 16 / sizeof(void*) };
public:
Arena* arena() const { return this->alloc_.arena(); }
void InternalSwap(UntypedMapBase* other) {
std::swap(num_elements_, other->num_elements_);
std::swap(num_buckets_, other->num_buckets_);
std::swap(seed_, other->seed_);
std::swap(index_of_first_non_null_, other->index_of_first_non_null_);
std::swap(table_, other->table_);
std::swap(alloc_, other->alloc_);
}
static size_type max_size() {
return std::numeric_limits<map_index_t>::max();
}
size_type size() const { return num_elements_; }
bool empty() const { return size() == 0; }
UntypedMapIterator begin() const { return UntypedMapIterator(this); }
// We make this a static function to reduce the cost in MapField.
// All the end iterators are singletons anyway.
static UntypedMapIterator EndIterator() { return {}; }
protected:
friend class TcParser;
friend struct MapTestPeer;
friend struct MapBenchmarkPeer;
friend class UntypedMapIterator;
struct NodeAndBucket {
NodeBase* node;
map_index_t bucket;
};
// Returns whether we should insert after the head of the list. For
// non-optimized builds, we randomly decide whether to insert right at the
// head of the list or just after the head. This helps add a little bit of
// non-determinism to the map ordering.
bool ShouldInsertAfterHead(void* node) {
#ifdef NDEBUG
(void)node;
return false;
#else
// Doing modulo with a prime mixes the bits more.
return (reinterpret_cast<uintptr_t>(node) ^ seed_) % 13 > 6;
#endif
}
// Helper for InsertUnique. Handles the case where bucket b is a
// not-too-long linked list.
void InsertUniqueInList(map_index_t b, NodeBase* node) {
if (!TableEntryIsEmpty(b) && ShouldInsertAfterHead(node)) {
auto* first = TableEntryToNode(table_[b]);
node->next = first->next;
first->next = node;
} else {
node->next = TableEntryToNode(table_[b]);
table_[b] = NodeToTableEntry(node);
}
}
bool TableEntryIsEmpty(map_index_t b) const {
return internal::TableEntryIsEmpty(table_[b]);
}
bool TableEntryIsNonEmptyList(map_index_t b) const {
return internal::TableEntryIsNonEmptyList(table_[b]);
}
bool TableEntryIsTree(map_index_t b) const {
return internal::TableEntryIsTree(table_[b]);
}
bool TableEntryIsList(map_index_t b) const {
return internal::TableEntryIsList(table_[b]);
}
// Return whether table_[b] is a linked list that seems awfully long.
// Requires table_[b] to point to a non-empty linked list.
bool TableEntryIsTooLong(map_index_t b) {
return internal::TableEntryIsTooLong(TableEntryToNode(table_[b]));
}
// Return a power of two no less than max(kMinTableSize, n).
// Assumes either n < kMinTableSize or n is a power of two.
map_index_t TableSize(map_index_t n) {
return n < static_cast<map_index_t>(kMinTableSize)
? static_cast<map_index_t>(kMinTableSize)
: n;
}
template <typename T>
using AllocFor = absl::allocator_traits<Allocator>::template rebind_alloc<T>;
// Alignment of the nodes is the same as alignment of NodeBase.
NodeBase* AllocNode(MapNodeSizeInfoT size_info) {
return AllocNode(SizeFromInfo(size_info));
}
NodeBase* AllocNode(size_t node_size) {
PROTOBUF_ASSUME(node_size % sizeof(NodeBase) == 0);
return AllocFor<NodeBase>(alloc_).allocate(node_size / sizeof(NodeBase));
}
void DeallocNode(NodeBase* node, MapNodeSizeInfoT size_info) {
DeallocNode(node, SizeFromInfo(size_info));
}
void DeallocNode(NodeBase* node, size_t node_size) {
PROTOBUF_ASSUME(node_size % sizeof(NodeBase) == 0);
AllocFor<NodeBase>(alloc_).deallocate(node, node_size / sizeof(NodeBase));
}
void DeleteTable(TableEntryPtr* table, map_index_t n) {
if (auto* a = arena()) {
a->ReturnArrayMemory(table, n * sizeof(TableEntryPtr));
} else {
internal::SizedDelete(table, n * sizeof(TableEntryPtr));
}
}
NodeBase* DestroyTree(Tree* tree);
using GetKey = VariantKey (*)(NodeBase*);
void InsertUniqueInTree(map_index_t b, GetKey get_key, NodeBase* node);
void TransferTree(Tree* tree, GetKey get_key);
TableEntryPtr ConvertToTree(NodeBase* node, GetKey get_key);
void EraseFromTree(map_index_t b, typename Tree::iterator tree_it);
map_index_t VariantBucketNumber(VariantKey key) const;
map_index_t BucketNumberFromHash(uint64_t h) const {
// We xor the hash value against the random seed so that we effectively
// have a random hash function.
// We use absl::Hash to do bit mixing for uniform bucket selection.
return absl::HashOf(h ^ seed_) & (num_buckets_ - 1);
}
TableEntryPtr* CreateEmptyTable(map_index_t n) {
ABSL_DCHECK_GE(n, map_index_t{kMinTableSize});
ABSL_DCHECK_EQ(n & (n - 1), 0u);
TableEntryPtr* result = AllocFor<TableEntryPtr>(alloc_).allocate(n);
memset(result, 0, n * sizeof(result[0]));
return result;
}
// Return a randomish value.
map_index_t Seed() const {
uint64_t s = 0;
#if !defined(GOOGLE_PROTOBUF_NO_RDTSC)
#if defined(__APPLE__)
// Use a commpage-based fast time function on Apple environments (MacOS,
// iOS, tvOS, watchOS, etc).
s = clock_gettime_nsec_np(CLOCK_UPTIME_RAW);
#elif defined(__x86_64__) && defined(__GNUC__)
uint32_t hi, lo;
asm volatile("rdtsc" : "=a"(lo), "=d"(hi));
s = ((static_cast<uint64_t>(hi) << 32) | lo);
#elif defined(__aarch64__) && defined(__GNUC__)
// There is no rdtsc on ARMv8. CNTVCT_EL0 is the virtual counter of the
// system timer. It runs at a different frequency than the CPU's, but is
// the best source of time-based entropy we get.
uint64_t virtual_timer_value;
asm volatile("mrs %0, cntvct_el0" : "=r"(virtual_timer_value));
s = virtual_timer_value;
#endif
#endif // !defined(GOOGLE_PROTOBUF_NO_RDTSC)
// Add entropy from the address of the map and the address of the table
// array.
return static_cast<map_index_t>(
absl::HashOf(s, table_, static_cast<const void*>(this)));
}
enum {
kKeyIsString = 1 << 0,
kValueIsString = 1 << 1,
kValueIsProto = 1 << 2,
kUseDestructFunc = 1 << 3,
};
template <typename Key, typename Value>
static constexpr uint8_t MakeDestroyBits() {
uint8_t result = 0;
if (!std::is_trivially_destructible<Key>::value) {
if (std::is_same<Key, std::string>::value) {
result |= kKeyIsString;
} else {
return kUseDestructFunc;
}
}
if (!std::is_trivially_destructible<Value>::value) {
if (std::is_same<Value, std::string>::value) {
result |= kValueIsString;
} else if (std::is_base_of<MessageLite, Value>::value) {
result |= kValueIsProto;
} else {
return kUseDestructFunc;
}
}
return result;
}
struct ClearInput {
MapNodeSizeInfoT size_info;
uint8_t destroy_bits;
bool reset_table;
void (*destroy_node)(NodeBase*);
};
template <typename Node>
static void DestroyNode(NodeBase* node) {
static_cast<Node*>(node)->~Node();
}
template <typename Node>
static constexpr ClearInput MakeClearInput(bool reset) {
constexpr auto bits =
MakeDestroyBits<typename Node::key_type, typename Node::mapped_type>();
return ClearInput{Node::size_info(), bits, reset,
bits & kUseDestructFunc ? DestroyNode<Node> : nullptr};
}
void ClearTable(ClearInput input);
NodeAndBucket FindFromTree(map_index_t b, VariantKey key,
Tree::iterator* it) const;
// Space used for the table, trees, and nodes.
// Does not include the indirect space used. Eg the data of a std::string.
size_t SpaceUsedInTable(size_t sizeof_node) const;
map_index_t num_elements_;
map_index_t num_buckets_;
map_index_t seed_;
map_index_t index_of_first_non_null_;
TableEntryPtr* table_; // an array with num_buckets_ entries
Allocator alloc_;
};
inline UntypedMapIterator::UntypedMapIterator(const UntypedMapBase* m) : m_(m) {
if (m_->index_of_first_non_null_ == m_->num_buckets_) {
bucket_index_ = 0;
node_ = nullptr;
} else {
bucket_index_ = m_->index_of_first_non_null_;
TableEntryPtr entry = m_->table_[bucket_index_];
node_ = PROTOBUF_PREDICT_TRUE(TableEntryIsList(entry))
? TableEntryToNode(entry)
: TableEntryToTree(entry)->begin()->second;
PROTOBUF_ASSUME(node_ != nullptr);
}
}
inline void UntypedMapIterator::SearchFrom(map_index_t start_bucket) {
ABSL_DCHECK(m_->index_of_first_non_null_ == m_->num_buckets_ ||
!m_->TableEntryIsEmpty(m_->index_of_first_non_null_));
for (map_index_t i = start_bucket; i < m_->num_buckets_; ++i) {
TableEntryPtr entry = m_->table_[i];
if (entry == TableEntryPtr{}) continue;
bucket_index_ = i;
if (PROTOBUF_PREDICT_TRUE(TableEntryIsList(entry))) {
node_ = TableEntryToNode(entry);
} else {
TreeForMap* tree = TableEntryToTree(entry);
ABSL_DCHECK(!tree->empty());
node_ = tree->begin()->second;
}
return;
}
node_ = nullptr;
bucket_index_ = 0;
}
// Base class used by TcParser to extract the map object from a map field.
// We keep it here to avoid a dependency into map_field.h from the main TcParser
// code, since that would bring in Message too.
class MapFieldBaseForParse {
public:
const UntypedMapBase& GetMap() const {
return vtable_->get_map(*this, false);
}
UntypedMapBase* MutableMap() {
return &const_cast<UntypedMapBase&>(vtable_->get_map(*this, true));
}
protected:
struct VTable {
const UntypedMapBase& (*get_map)(const MapFieldBaseForParse&,
bool is_mutable);
};
explicit constexpr MapFieldBaseForParse(const VTable* vtable)
: vtable_(vtable) {}
~MapFieldBaseForParse() = default;
const VTable* vtable_;
};
// The value might be of different signedness, so use memcpy to extract it.
template <typename T, std::enable_if_t<std::is_integral<T>::value, int> = 0>
T ReadKey(const void* ptr) {
T out;
memcpy(&out, ptr, sizeof(T));
return out;
}
template <typename T, std::enable_if_t<!std::is_integral<T>::value, int> = 0>
const T& ReadKey(const void* ptr) {
return *reinterpret_cast<const T*>(ptr);
}
template <typename Key>
struct KeyNode : NodeBase {
static constexpr size_t kOffset = sizeof(NodeBase);
decltype(auto) key() const { return ReadKey<Key>(GetVoidKey()); }
};
// KeyMapBase is a chaining hash map with the additional feature that some
// buckets can be converted to use an ordered container. This ensures O(lg n)
// bounds on find, insert, and erase, while avoiding the overheads of ordered
// containers most of the time.
//
// The implementation doesn't need the full generality of unordered_map,
// and it doesn't have it. More bells and whistles can be added as needed.
// Some implementation details:
// 1. The number of buckets is a power of two.
// 2. As is typical for hash_map and such, the Keys and Values are always
// stored in linked list nodes. Pointers to elements are never invalidated
// until the element is deleted.
// 3. The trees' payload type is pointer to linked-list node. Tree-converting
// a bucket doesn't copy Key-Value pairs.
// 4. Once we've tree-converted a bucket, it is never converted back unless the
// bucket is completely emptied out. Note that the items a tree contains may
// wind up assigned to trees or lists upon a rehash.
// 5. Mutations to a map do not invalidate the map's iterators, pointers to
// elements, or references to elements.
// 6. Except for erase(iterator), any non-const method can reorder iterators.
// 7. Uses VariantKey when using the Tree representation, which holds all
// possible key types as a variant value.
template <typename Key>
class KeyMapBase : public UntypedMapBase {
static_assert(!std::is_signed<Key>::value || !std::is_integral<Key>::value,
"");
using TS = TransparentSupport<Key>;
public:
using hasher = typename TS::hash;
using UntypedMapBase::UntypedMapBase;
protected:
using KeyNode = internal::KeyNode<Key>;
// Trees. The payload type is a copy of Key, so that we can query the tree
// with Keys that are not in any particular data structure.
// The value is a void* pointing to Node. We use void* instead of Node* to
// avoid code bloat. That way there is only one instantiation of the tree
// class per key type.
using Tree = internal::TreeForMap;
using TreeIterator = typename Tree::iterator;
public:
hasher hash_function() const { return {}; }
protected:
friend class TcParser;
friend struct MapTestPeer;
friend struct MapBenchmarkPeer;
PROTOBUF_NOINLINE void erase_no_destroy(map_index_t b, KeyNode* node) {
TreeIterator tree_it;
const bool is_list = revalidate_if_necessary(b, node, &tree_it);
if (is_list) {
ABSL_DCHECK(TableEntryIsNonEmptyList(b));
auto* head = TableEntryToNode(table_[b]);
head = EraseFromLinkedList(node, head);
table_[b] = NodeToTableEntry(head);
} else {
EraseFromTree(b, tree_it);
}
--num_elements_;
if (PROTOBUF_PREDICT_FALSE(b == index_of_first_non_null_)) {
while (index_of_first_non_null_ < num_buckets_ &&
TableEntryIsEmpty(index_of_first_non_null_)) {
++index_of_first_non_null_;
}
}
}
NodeAndBucket FindHelper(typename TS::ViewType k,
TreeIterator* it = nullptr) const {
map_index_t b = BucketNumber(k);
if (TableEntryIsNonEmptyList(b)) {
auto* node = internal::TableEntryToNode(table_[b]);
do {
if (TS::Equals(static_cast<KeyNode*>(node)->key(), k)) {
return {node, b};
} else {
node = node->next;
}
} while (node != nullptr);
} else if (TableEntryIsTree(b)) {
return FindFromTree(b, internal::RealKeyToVariantKey<Key>{}(k), it);
}
return {nullptr, b};
}
// Insert the given node.
// If the key is a duplicate, it inserts the new node and returns the old one.
// Gives ownership to the caller.
// If the key is unique, it returns `nullptr`.
KeyNode* InsertOrReplaceNode(KeyNode* node) {
KeyNode* to_erase = nullptr;
auto p = this->FindHelper(node->key());
if (p.node != nullptr) {
erase_no_destroy(p.bucket, static_cast<KeyNode*>(p.node));
to_erase = static_cast<KeyNode*>(p.node);
} else if (ResizeIfLoadIsOutOfRange(num_elements_ + 1)) {
p = FindHelper(node->key());
}
const map_index_t b = p.bucket; // bucket number
InsertUnique(b, node);
++num_elements_;
return to_erase;
}
// Insert the given Node in bucket b. If that would make bucket b too big,
// and bucket b is not a tree, create a tree for buckets b.
// Requires count(*KeyPtrFromNodePtr(node)) == 0 and that b is the correct
// bucket. num_elements_ is not modified.
void InsertUnique(map_index_t b, KeyNode* node) {
ABSL_DCHECK(index_of_first_non_null_ == num_buckets_ ||
!TableEntryIsEmpty(index_of_first_non_null_));
// In practice, the code that led to this point may have already
// determined whether we are inserting into an empty list, a short list,
// or whatever. But it's probably cheap enough to recompute that here;
// it's likely that we're inserting into an empty or short list.
ABSL_DCHECK(FindHelper(node->key()).node == nullptr);
if (TableEntryIsEmpty(b)) {
InsertUniqueInList(b, node);
index_of_first_non_null_ = (std::min)(index_of_first_non_null_, b);
} else if (TableEntryIsNonEmptyList(b) && !TableEntryIsTooLong(b)) {
InsertUniqueInList(b, node);
} else {
InsertUniqueInTree(b, NodeToVariantKey, node);
}
}
static VariantKey NodeToVariantKey(NodeBase* node) {
return internal::RealKeyToVariantKey<Key>{}(
static_cast<KeyNode*>(node)->key());
}
// Have it a separate function for testing.
static size_type CalculateHiCutoff(size_type num_buckets) {
// We want the high cutoff to follow this rules:
// - When num_buckets_ == kGlobalEmptyTableSize, then make it 0 to force an
// allocation.
// - When num_buckets_ < 8, then make it num_buckets_ to avoid
// a reallocation. A large load factor is not that important on small
// tables and saves memory.
// - Otherwise, make it 75% of num_buckets_.
return num_buckets - num_buckets / 16 * 4 - num_buckets % 2;
}
// Returns whether it did resize. Currently this is only used when
// num_elements_ increases, though it could be used in other situations.
// It checks for load too low as well as load too high: because any number
// of erases can occur between inserts, the load could be as low as 0 here.
// Resizing to a lower size is not always helpful, but failing to do so can
// destroy the expected big-O bounds for some operations. By having the
// policy that sometimes we resize down as well as up, clients can easily
// keep O(size()) = O(number of buckets) if they want that.
bool ResizeIfLoadIsOutOfRange(size_type new_size) {
const size_type hi_cutoff = CalculateHiCutoff(num_buckets_);