1 // Copyright 2020 The Abseil Authors.
2 //
3 // Licensed under the Apache License, Version 2.0 (the "License");
4 // you may not use this file except in compliance with the License.
5 // You may obtain a copy of the License at
6 //
7 // https://www.apache.org/licenses/LICENSE-2.0
8 //
9 // Unless required by applicable law or agreed to in writing, software
10 // distributed under the License is distributed on an "AS IS" BASIS,
11 // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12 // See the License for the specific language governing permissions and
13 // limitations under the License.
14 //
15 // -----------------------------------------------------------------------------
16 // File: cord.h
17 // -----------------------------------------------------------------------------
18 //
19 // This file defines the `absl::Cord` data structure and operations on that data
20 // structure. A Cord is a string-like sequence of characters optimized for
21 // specific use cases. Unlike a `std::string`, which stores an array of
22 // contiguous characters, Cord data is stored in a structure consisting of
23 // separate, reference-counted "chunks."
24 //
25 // Because a Cord consists of these chunks, data can be added to or removed from
26 // a Cord during its lifetime. Chunks may also be shared between Cords. Unlike a
27 // `std::string`, a Cord can therefore accommodate data that changes over its
28 // lifetime, though it's not quite "mutable"; it can change only in the
29 // attachment, detachment, or rearrangement of chunks of its constituent data.
30 //
31 // A Cord provides some benefit over `std::string` under the following (albeit
32 // narrow) circumstances:
33 //
34 // * Cord data is designed to grow and shrink over a Cord's lifetime. Cord
35 // provides efficient insertions and deletions at the start and end of the
36 // character sequences, avoiding copies in those cases. Static data should
37 // generally be stored as strings.
38 // * External memory consisting of string-like data can be directly added to
39 // a Cord without requiring copies or allocations.
40 // * Cord data may be shared and copied cheaply. Cord provides a copy-on-write
41 // implementation and cheap sub-Cord operations. Copying a Cord is an O(1)
42 // operation.
43 //
44 // As a consequence to the above, Cord data is generally large. Small data
45 // should generally use strings, as construction of a Cord requires some
46 // overhead. Small Cords (<= 15 bytes) are represented inline, but most small
47 // Cords are expected to grow over their lifetimes.
48 //
49 // Note that because a Cord is made up of separate chunked data, random access
50 // to character data within a Cord is slower than within a `std::string`.
51 //
52 // Thread Safety
53 //
54 // Cord has the same thread-safety properties as many other types like
55 // std::string, std::vector<>, int, etc -- it is thread-compatible. In
56 // particular, if threads do not call non-const methods, then it is safe to call
57 // const methods without synchronization. Copying a Cord produces a new instance
58 // that can be used concurrently with the original in arbitrary ways.
59
60 #ifndef ABSL_STRINGS_CORD_H_
61 #define ABSL_STRINGS_CORD_H_
62
63 #include <algorithm>
64 #include <cstddef>
65 #include <cstdint>
66 #include <cstring>
67 #include <iosfwd>
68 #include <iterator>
69 #include <string>
70 #include <type_traits>
71
72 #include "absl/base/attributes.h"
73 #include "absl/base/config.h"
74 #include "absl/base/internal/endian.h"
75 #include "absl/base/internal/per_thread_tls.h"
76 #include "absl/base/macros.h"
77 #include "absl/base/port.h"
78 #include "absl/container/inlined_vector.h"
79 #include "absl/crc/internal/crc_cord_state.h"
80 #include "absl/functional/function_ref.h"
81 #include "absl/meta/type_traits.h"
82 #include "absl/strings/cord_analysis.h"
83 #include "absl/strings/cord_buffer.h"
84 #include "absl/strings/internal/cord_data_edge.h"
85 #include "absl/strings/internal/cord_internal.h"
86 #include "absl/strings/internal/cord_rep_btree.h"
87 #include "absl/strings/internal/cord_rep_btree_reader.h"
88 #include "absl/strings/internal/cord_rep_crc.h"
89 #include "absl/strings/internal/cord_rep_ring.h"
90 #include "absl/strings/internal/cordz_functions.h"
91 #include "absl/strings/internal/cordz_info.h"
92 #include "absl/strings/internal/cordz_statistics.h"
93 #include "absl/strings/internal/cordz_update_scope.h"
94 #include "absl/strings/internal/cordz_update_tracker.h"
95 #include "absl/strings/internal/resize_uninitialized.h"
96 #include "absl/strings/internal/string_constant.h"
97 #include "absl/strings/string_view.h"
98 #include "absl/types/optional.h"
99
100 namespace absl {
101 ABSL_NAMESPACE_BEGIN
102 class Cord;
103 class CordTestPeer;
104 template <typename Releaser>
105 Cord MakeCordFromExternal(absl::string_view, Releaser&&);
106 void CopyCordToString(const Cord& src, std::string* dst);
107
108 // Cord memory accounting modes
109 enum class CordMemoryAccounting {
110 // Counts the *approximate* number of bytes held in full or in part by this
111 // Cord (which may not remain the same between invocations). Cords that share
112 // memory could each be "charged" independently for the same shared memory.
113 // See also comment on `kTotalMorePrecise` on internally shared memory.
114 kTotal,
115
116 // Counts the *approximate* number of bytes held in full or in part by this
117 // Cord for the distinct memory held by this cord. This option is similar
118 // to `kTotal`, except that if the cord has multiple references to the same
119 // memory, that memory is only counted once.
120 //
121 // For example:
122 // absl::Cord cord;
123 // cord.append(some_other_cord);
124 // cord.append(some_other_cord);
125 // // Counts `some_other_cord` twice:
126 // cord.EstimatedMemoryUsage(kTotal);
127 // // Counts `some_other_cord` once:
128 // cord.EstimatedMemoryUsage(kTotalMorePrecise);
129 //
130 // The `kTotalMorePrecise` number is more expensive to compute as it requires
131 // deduplicating all memory references. Applications should prefer to use
132 // `kFairShare` or `kTotal` unless they really need a more precise estimate
133 // on "how much memory is potentially held / kept alive by this cord?"
134 kTotalMorePrecise,
135
136 // Counts the *approximate* number of bytes held in full or in part by this
137 // Cord weighted by the sharing ratio of that data. For example, if some data
138 // edge is shared by 4 different Cords, then each cord is attributed 1/4th of
139 // the total memory usage as a 'fair share' of the total memory usage.
140 kFairShare,
141 };
142
143 // Cord
144 //
145 // A Cord is a sequence of characters, designed to be more efficient than a
146 // `std::string` in certain circumstances: namely, large string data that needs
147 // to change over its lifetime or shared, especially when such data is shared
148 // across API boundaries.
149 //
150 // A Cord stores its character data in a structure that allows efficient prepend
151 // and append operations. This makes a Cord useful for large string data sent
152 // over in a wire format that may need to be prepended or appended at some point
153 // during the data exchange (e.g. HTTP, protocol buffers). For example, a
154 // Cord is useful for storing an HTTP request, and prepending an HTTP header to
155 // such a request.
156 //
157 // Cords should not be used for storing general string data, however. They
158 // require overhead to construct and are slower than strings for random access.
159 //
160 // The Cord API provides the following common API operations:
161 //
162 // * Create or assign Cords out of existing string data, memory, or other Cords
163 // * Append and prepend data to an existing Cord
164 // * Create new Sub-Cords from existing Cord data
165 // * Swap Cord data and compare Cord equality
166 // * Write out Cord data by constructing a `std::string`
167 //
168 // Additionally, the API provides iterator utilities to iterate through Cord
169 // data via chunks or character bytes.
170 //
171 class Cord {
172 private:
173 template <typename T>
174 using EnableIfString =
175 absl::enable_if_t<std::is_same<T, std::string>::value, int>;
176
177 public:
178 // Cord::Cord() Constructors.
179
180 // Creates an empty Cord.
181 constexpr Cord() noexcept;
182
183 // Creates a Cord from an existing Cord. Cord is copyable and efficiently
184 // movable. The moved-from state is valid but unspecified.
185 Cord(const Cord& src);
186 Cord(Cord&& src) noexcept;
187 Cord& operator=(const Cord& x);
188 Cord& operator=(Cord&& x) noexcept;
189
190 // Creates a Cord from a `src` string. This constructor is marked explicit to
191 // prevent implicit Cord constructions from arguments convertible to an
192 // `absl::string_view`.
193 explicit Cord(absl::string_view src);
194 Cord& operator=(absl::string_view src);
195
196 // Creates a Cord from a `std::string&&` rvalue. These constructors are
197 // templated to avoid ambiguities for types that are convertible to both
198 // `absl::string_view` and `std::string`, such as `const char*`.
199 template <typename T, EnableIfString<T> = 0>
200 explicit Cord(T&& src);
201 template <typename T, EnableIfString<T> = 0>
202 Cord& operator=(T&& src);
203
204 // Cord::~Cord()
205 //
206 // Destructs the Cord.
~Cord()207 ~Cord() {
208 if (contents_.is_tree()) DestroyCordSlow();
209 }
210
211 // MakeCordFromExternal()
212 //
213 // Creates a Cord that takes ownership of external string memory. The
214 // contents of `data` are not copied to the Cord; instead, the external
215 // memory is added to the Cord and reference-counted. This data may not be
216 // changed for the life of the Cord, though it may be prepended or appended
217 // to.
218 //
219 // `MakeCordFromExternal()` takes a callable "releaser" that is invoked when
220 // the reference count for `data` reaches zero. As noted above, this data must
221 // remain live until the releaser is invoked. The callable releaser also must:
222 //
223 // * be move constructible
224 // * support `void operator()(absl::string_view) const` or `void operator()`
225 //
226 // Example:
227 //
228 // Cord MakeCord(BlockPool* pool) {
229 // Block* block = pool->NewBlock();
230 // FillBlock(block);
231 // return absl::MakeCordFromExternal(
232 // block->ToStringView(),
233 // [pool, block](absl::string_view v) {
234 // pool->FreeBlock(block, v);
235 // });
236 // }
237 //
238 // WARNING: Because a Cord can be reference-counted, it's likely a bug if your
239 // releaser doesn't do anything. For example, consider the following:
240 //
241 // void Foo(const char* buffer, int len) {
242 // auto c = absl::MakeCordFromExternal(absl::string_view(buffer, len),
243 // [](absl::string_view) {});
244 //
245 // // BUG: If Bar() copies its cord for any reason, including keeping a
246 // // substring of it, the lifetime of buffer might be extended beyond
247 // // when Foo() returns.
248 // Bar(c);
249 // }
250 template <typename Releaser>
251 friend Cord MakeCordFromExternal(absl::string_view data, Releaser&& releaser);
252
253 // Cord::Clear()
254 //
255 // Releases the Cord data. Any nodes that share data with other Cords, if
256 // applicable, will have their reference counts reduced by 1.
257 ABSL_ATTRIBUTE_REINITIALIZES void Clear();
258
259 // Cord::Append()
260 //
261 // Appends data to the Cord, which may come from another Cord or other string
262 // data.
263 void Append(const Cord& src);
264 void Append(Cord&& src);
265 void Append(absl::string_view src);
266 template <typename T, EnableIfString<T> = 0>
267 void Append(T&& src);
268
269 // Appends `buffer` to this cord, unless `buffer` has a zero length in which
270 // case this method has no effect on this cord instance.
271 // This method is guaranteed to consume `buffer`.
272 void Append(CordBuffer buffer);
273
274 // Returns a CordBuffer, re-using potential existing capacity in this cord.
275 //
276 // Cord instances may have additional unused capacity in the last (or first)
277 // nodes of the underlying tree to facilitate amortized growth. This method
278 // allows applications to explicitly use this spare capacity if available,
279 // or create a new CordBuffer instance otherwise.
280 // If this cord has a final non-shared node with at least `min_capacity`
281 // available, then this method will return that buffer including its data
282 // contents. I.e.; the returned buffer will have a non-zero length, and
283 // a capacity of at least `buffer.length + min_capacity`. Otherwise, this
284 // method will return `CordBuffer::CreateWithDefaultLimit(capacity)`.
285 //
286 // Below an example of using GetAppendBuffer. Notice that in this example we
287 // use `GetAppendBuffer()` only on the first iteration. As we know nothing
288 // about any initial extra capacity in `cord`, we may be able to use the extra
289 // capacity. But as we add new buffers with fully utilized contents after that
290 // we avoid calling `GetAppendBuffer()` on subsequent iterations: while this
291 // works fine, it results in an unnecessary inspection of cord contents:
292 //
293 // void AppendRandomDataToCord(absl::Cord &cord, size_t n) {
294 // bool first = true;
295 // while (n > 0) {
296 // CordBuffer buffer = first ? cord.GetAppendBuffer(n)
297 // : CordBuffer::CreateWithDefaultLimit(n);
298 // absl::Span<char> data = buffer.available_up_to(n);
299 // FillRandomValues(data.data(), data.size());
300 // buffer.IncreaseLengthBy(data.size());
301 // cord.Append(std::move(buffer));
302 // n -= data.size();
303 // first = false;
304 // }
305 // }
306 CordBuffer GetAppendBuffer(size_t capacity, size_t min_capacity = 16);
307
308 // Returns a CordBuffer, re-using potential existing capacity in this cord.
309 //
310 // This function is identical to `GetAppendBuffer`, except that in the case
311 // where a new `CordBuffer` is allocated, it is allocated using the provided
312 // custom limit instead of the default limit. `GetAppendBuffer` will default
313 // to `CordBuffer::CreateWithDefaultLimit(capacity)` whereas this method
314 // will default to `CordBuffer::CreateWithCustomLimit(block_size, capacity)`.
315 // This method is equivalent to `GetAppendBuffer` if `block_size` is zero.
316 // See the documentation for `CreateWithCustomLimit` for more details on the
317 // restrictions and legal values for `block_size`.
318 CordBuffer GetCustomAppendBuffer(size_t block_size, size_t capacity,
319 size_t min_capacity = 16);
320
321 // Cord::Prepend()
322 //
323 // Prepends data to the Cord, which may come from another Cord or other string
324 // data.
325 void Prepend(const Cord& src);
326 void Prepend(absl::string_view src);
327 template <typename T, EnableIfString<T> = 0>
328 void Prepend(T&& src);
329
330 // Prepends `buffer` to this cord, unless `buffer` has a zero length in which
331 // case this method has no effect on this cord instance.
332 // This method is guaranteed to consume `buffer`.
333 void Prepend(CordBuffer buffer);
334
335 // Cord::RemovePrefix()
336 //
337 // Removes the first `n` bytes of a Cord.
338 void RemovePrefix(size_t n);
339 void RemoveSuffix(size_t n);
340
341 // Cord::Subcord()
342 //
343 // Returns a new Cord representing the subrange [pos, pos + new_size) of
344 // *this. If pos >= size(), the result is empty(). If
345 // (pos + new_size) >= size(), the result is the subrange [pos, size()).
346 Cord Subcord(size_t pos, size_t new_size) const;
347
348 // Cord::swap()
349 //
350 // Swaps the contents of the Cord with `other`.
351 void swap(Cord& other) noexcept;
352
353 // swap()
354 //
355 // Swaps the contents of two Cords.
swap(Cord & x,Cord & y)356 friend void swap(Cord& x, Cord& y) noexcept { x.swap(y); }
357
358 // Cord::size()
359 //
360 // Returns the size of the Cord.
361 size_t size() const;
362
363 // Cord::empty()
364 //
365 // Determines whether the given Cord is empty, returning `true` is so.
366 bool empty() const;
367
368 // Cord::EstimatedMemoryUsage()
369 //
370 // Returns the *approximate* number of bytes held by this cord.
371 // See CordMemoryAccounting for more information on the accounting method.
372 size_t EstimatedMemoryUsage(CordMemoryAccounting accounting_method =
373 CordMemoryAccounting::kTotal) const;
374
375 // Cord::Compare()
376 //
377 // Compares 'this' Cord with rhs. This function and its relatives treat Cords
378 // as sequences of unsigned bytes. The comparison is a straightforward
379 // lexicographic comparison. `Cord::Compare()` returns values as follows:
380 //
381 // -1 'this' Cord is smaller
382 // 0 two Cords are equal
383 // 1 'this' Cord is larger
384 int Compare(absl::string_view rhs) const;
385 int Compare(const Cord& rhs) const;
386
387 // Cord::StartsWith()
388 //
389 // Determines whether the Cord starts with the passed string data `rhs`.
390 bool StartsWith(const Cord& rhs) const;
391 bool StartsWith(absl::string_view rhs) const;
392
393 // Cord::EndsWith()
394 //
395 // Determines whether the Cord ends with the passed string data `rhs`.
396 bool EndsWith(absl::string_view rhs) const;
397 bool EndsWith(const Cord& rhs) const;
398
399 // Cord::operator std::string()
400 //
401 // Converts a Cord into a `std::string()`. This operator is marked explicit to
402 // prevent unintended Cord usage in functions that take a string.
403 explicit operator std::string() const;
404
405 // CopyCordToString()
406 //
407 // Copies the contents of a `src` Cord into a `*dst` string.
408 //
409 // This function optimizes the case of reusing the destination string since it
410 // can reuse previously allocated capacity. However, this function does not
411 // guarantee that pointers previously returned by `dst->data()` remain valid
412 // even if `*dst` had enough capacity to hold `src`. If `*dst` is a new
413 // object, prefer to simply use the conversion operator to `std::string`.
414 friend void CopyCordToString(const Cord& src, std::string* dst);
415
416 class CharIterator;
417
418 //----------------------------------------------------------------------------
419 // Cord::ChunkIterator
420 //----------------------------------------------------------------------------
421 //
422 // A `Cord::ChunkIterator` allows iteration over the constituent chunks of its
423 // Cord. Such iteration allows you to perform non-const operations on the data
424 // of a Cord without modifying it.
425 //
426 // Generally, you do not instantiate a `Cord::ChunkIterator` directly;
427 // instead, you create one implicitly through use of the `Cord::Chunks()`
428 // member function.
429 //
430 // The `Cord::ChunkIterator` has the following properties:
431 //
432 // * The iterator is invalidated after any non-const operation on the
433 // Cord object over which it iterates.
434 // * The `string_view` returned by dereferencing a valid, non-`end()`
435 // iterator is guaranteed to be non-empty.
436 // * Two `ChunkIterator` objects can be compared equal if and only if they
437 // remain valid and iterate over the same Cord.
438 // * The iterator in this case is a proxy iterator; the `string_view`
439 // returned by the iterator does not live inside the Cord, and its
440 // lifetime is limited to the lifetime of the iterator itself. To help
441 // prevent lifetime issues, `ChunkIterator::reference` is not a true
442 // reference type and is equivalent to `value_type`.
443 // * The iterator keeps state that can grow for Cords that contain many
444 // nodes and are imbalanced due to sharing. Prefer to pass this type by
445 // const reference instead of by value.
446 class ChunkIterator {
447 public:
448 using iterator_category = std::input_iterator_tag;
449 using value_type = absl::string_view;
450 using difference_type = ptrdiff_t;
451 using pointer = const value_type*;
452 using reference = value_type;
453
454 ChunkIterator() = default;
455
456 ChunkIterator& operator++();
457 ChunkIterator operator++(int);
458 bool operator==(const ChunkIterator& other) const;
459 bool operator!=(const ChunkIterator& other) const;
460 reference operator*() const;
461 pointer operator->() const;
462
463 friend class Cord;
464 friend class CharIterator;
465
466 private:
467 using CordRep = absl::cord_internal::CordRep;
468 using CordRepBtree = absl::cord_internal::CordRepBtree;
469 using CordRepBtreeReader = absl::cord_internal::CordRepBtreeReader;
470
471 // Constructs a `begin()` iterator from `tree`. `tree` must not be null.
472 explicit ChunkIterator(cord_internal::CordRep* tree);
473
474 // Constructs a `begin()` iterator from `cord`.
475 explicit ChunkIterator(const Cord* cord);
476
477 // Initializes this instance from a tree. Invoked by constructors.
478 void InitTree(cord_internal::CordRep* tree);
479
480 // Removes `n` bytes from `current_chunk_`. Expects `n` to be smaller than
481 // `current_chunk_.size()`.
482 void RemoveChunkPrefix(size_t n);
483 Cord AdvanceAndReadBytes(size_t n);
484 void AdvanceBytes(size_t n);
485
486 // Btree specific operator++
487 ChunkIterator& AdvanceBtree();
488 void AdvanceBytesBtree(size_t n);
489
490 // A view into bytes of the current `CordRep`. It may only be a view to a
491 // suffix of bytes if this is being used by `CharIterator`.
492 absl::string_view current_chunk_;
493 // The current leaf, or `nullptr` if the iterator points to short data.
494 // If the current chunk is a substring node, current_leaf_ points to the
495 // underlying flat or external node.
496 absl::cord_internal::CordRep* current_leaf_ = nullptr;
497 // The number of bytes left in the `Cord` over which we are iterating.
498 size_t bytes_remaining_ = 0;
499
500 // Cord reader for cord btrees. Empty if not traversing a btree.
501 CordRepBtreeReader btree_reader_;
502 };
503
504 // Cord::chunk_begin()
505 //
506 // Returns an iterator to the first chunk of the `Cord`.
507 //
508 // Generally, prefer using `Cord::Chunks()` within a range-based for loop for
509 // iterating over the chunks of a Cord. This method may be useful for getting
510 // a `ChunkIterator` where range-based for-loops are not useful.
511 //
512 // Example:
513 //
514 // absl::Cord::ChunkIterator FindAsChunk(const absl::Cord& c,
515 // absl::string_view s) {
516 // return std::find(c.chunk_begin(), c.chunk_end(), s);
517 // }
518 ChunkIterator chunk_begin() const;
519
520 // Cord::chunk_end()
521 //
522 // Returns an iterator one increment past the last chunk of the `Cord`.
523 //
524 // Generally, prefer using `Cord::Chunks()` within a range-based for loop for
525 // iterating over the chunks of a Cord. This method may be useful for getting
526 // a `ChunkIterator` where range-based for-loops may not be available.
527 ChunkIterator chunk_end() const;
528
529 //----------------------------------------------------------------------------
530 // Cord::ChunkRange
531 //----------------------------------------------------------------------------
532 //
533 // `ChunkRange` is a helper class for iterating over the chunks of the `Cord`,
534 // producing an iterator which can be used within a range-based for loop.
535 // Construction of a `ChunkRange` will return an iterator pointing to the
536 // first chunk of the Cord. Generally, do not construct a `ChunkRange`
537 // directly; instead, prefer to use the `Cord::Chunks()` method.
538 //
539 // Implementation note: `ChunkRange` is simply a convenience wrapper over
540 // `Cord::chunk_begin()` and `Cord::chunk_end()`.
541 class ChunkRange {
542 public:
543 // Fulfill minimum c++ container requirements [container.requirements]
544 // These (partial) container type definitions allow ChunkRange to be used
545 // in various utilities expecting a subset of [container.requirements].
546 // For example, the below enables using `::testing::ElementsAre(...)`
547 using value_type = absl::string_view;
548 using reference = value_type&;
549 using const_reference = const value_type&;
550 using iterator = ChunkIterator;
551 using const_iterator = ChunkIterator;
552
ChunkRange(const Cord * cord)553 explicit ChunkRange(const Cord* cord) : cord_(cord) {}
554
555 ChunkIterator begin() const;
556 ChunkIterator end() const;
557
558 private:
559 const Cord* cord_;
560 };
561
562 // Cord::Chunks()
563 //
564 // Returns a `Cord::ChunkRange` for iterating over the chunks of a `Cord` with
565 // a range-based for-loop. For most iteration tasks on a Cord, use
566 // `Cord::Chunks()` to retrieve this iterator.
567 //
568 // Example:
569 //
570 // void ProcessChunks(const Cord& cord) {
571 // for (absl::string_view chunk : cord.Chunks()) { ... }
572 // }
573 //
574 // Note that the ordinary caveats of temporary lifetime extension apply:
575 //
576 // void Process() {
577 // for (absl::string_view chunk : CordFactory().Chunks()) {
578 // // The temporary Cord returned by CordFactory has been destroyed!
579 // }
580 // }
581 ChunkRange Chunks() const;
582
583 //----------------------------------------------------------------------------
584 // Cord::CharIterator
585 //----------------------------------------------------------------------------
586 //
587 // A `Cord::CharIterator` allows iteration over the constituent characters of
588 // a `Cord`.
589 //
590 // Generally, you do not instantiate a `Cord::CharIterator` directly; instead,
591 // you create one implicitly through use of the `Cord::Chars()` member
592 // function.
593 //
594 // A `Cord::CharIterator` has the following properties:
595 //
596 // * The iterator is invalidated after any non-const operation on the
597 // Cord object over which it iterates.
598 // * Two `CharIterator` objects can be compared equal if and only if they
599 // remain valid and iterate over the same Cord.
600 // * The iterator keeps state that can grow for Cords that contain many
601 // nodes and are imbalanced due to sharing. Prefer to pass this type by
602 // const reference instead of by value.
603 // * This type cannot act as a forward iterator because a `Cord` can reuse
604 // sections of memory. This fact violates the requirement for forward
605 // iterators to compare equal if dereferencing them returns the same
606 // object.
607 class CharIterator {
608 public:
609 using iterator_category = std::input_iterator_tag;
610 using value_type = char;
611 using difference_type = ptrdiff_t;
612 using pointer = const char*;
613 using reference = const char&;
614
615 CharIterator() = default;
616
617 CharIterator& operator++();
618 CharIterator operator++(int);
619 bool operator==(const CharIterator& other) const;
620 bool operator!=(const CharIterator& other) const;
621 reference operator*() const;
622 pointer operator->() const;
623
624 friend Cord;
625
626 private:
CharIterator(const Cord * cord)627 explicit CharIterator(const Cord* cord) : chunk_iterator_(cord) {}
628
629 ChunkIterator chunk_iterator_;
630 };
631
632 // Cord::AdvanceAndRead()
633 //
634 // Advances the `Cord::CharIterator` by `n_bytes` and returns the bytes
635 // advanced as a separate `Cord`. `n_bytes` must be less than or equal to the
636 // number of bytes within the Cord; otherwise, behavior is undefined. It is
637 // valid to pass `char_end()` and `0`.
638 static Cord AdvanceAndRead(CharIterator* it, size_t n_bytes);
639
640 // Cord::Advance()
641 //
642 // Advances the `Cord::CharIterator` by `n_bytes`. `n_bytes` must be less than
643 // or equal to the number of bytes remaining within the Cord; otherwise,
644 // behavior is undefined. It is valid to pass `char_end()` and `0`.
645 static void Advance(CharIterator* it, size_t n_bytes);
646
647 // Cord::ChunkRemaining()
648 //
649 // Returns the longest contiguous view starting at the iterator's position.
650 //
651 // `it` must be dereferenceable.
652 static absl::string_view ChunkRemaining(const CharIterator& it);
653
654 // Cord::char_begin()
655 //
656 // Returns an iterator to the first character of the `Cord`.
657 //
658 // Generally, prefer using `Cord::Chars()` within a range-based for loop for
659 // iterating over the chunks of a Cord. This method may be useful for getting
660 // a `CharIterator` where range-based for-loops may not be available.
661 CharIterator char_begin() const;
662
663 // Cord::char_end()
664 //
665 // Returns an iterator to one past the last character of the `Cord`.
666 //
667 // Generally, prefer using `Cord::Chars()` within a range-based for loop for
668 // iterating over the chunks of a Cord. This method may be useful for getting
669 // a `CharIterator` where range-based for-loops are not useful.
670 CharIterator char_end() const;
671
672 // Cord::CharRange
673 //
674 // `CharRange` is a helper class for iterating over the characters of a
675 // producing an iterator which can be used within a range-based for loop.
676 // Construction of a `CharRange` will return an iterator pointing to the first
677 // character of the Cord. Generally, do not construct a `CharRange` directly;
678 // instead, prefer to use the `Cord::Chars()` method shown below.
679 //
680 // Implementation note: `CharRange` is simply a convenience wrapper over
681 // `Cord::char_begin()` and `Cord::char_end()`.
682 class CharRange {
683 public:
684 // Fulfill minimum c++ container requirements [container.requirements]
685 // These (partial) container type definitions allow CharRange to be used
686 // in various utilities expecting a subset of [container.requirements].
687 // For example, the below enables using `::testing::ElementsAre(...)`
688 using value_type = char;
689 using reference = value_type&;
690 using const_reference = const value_type&;
691 using iterator = CharIterator;
692 using const_iterator = CharIterator;
693
CharRange(const Cord * cord)694 explicit CharRange(const Cord* cord) : cord_(cord) {}
695
696 CharIterator begin() const;
697 CharIterator end() const;
698
699 private:
700 const Cord* cord_;
701 };
702
703 // Cord::Chars()
704 //
705 // Returns a `Cord::CharRange` for iterating over the characters of a `Cord`
706 // with a range-based for-loop. For most character-based iteration tasks on a
707 // Cord, use `Cord::Chars()` to retrieve this iterator.
708 //
709 // Example:
710 //
711 // void ProcessCord(const Cord& cord) {
712 // for (char c : cord.Chars()) { ... }
713 // }
714 //
715 // Note that the ordinary caveats of temporary lifetime extension apply:
716 //
717 // void Process() {
718 // for (char c : CordFactory().Chars()) {
719 // // The temporary Cord returned by CordFactory has been destroyed!
720 // }
721 // }
722 CharRange Chars() const;
723
724 // Cord::operator[]
725 //
726 // Gets the "i"th character of the Cord and returns it, provided that
727 // 0 <= i < Cord.size().
728 //
729 // NOTE: This routine is reasonably efficient. It is roughly
730 // logarithmic based on the number of chunks that make up the cord. Still,
731 // if you need to iterate over the contents of a cord, you should
732 // use a CharIterator/ChunkIterator rather than call operator[] or Get()
733 // repeatedly in a loop.
734 char operator[](size_t i) const;
735
736 // Cord::TryFlat()
737 //
738 // If this cord's representation is a single flat array, returns a
739 // string_view referencing that array. Otherwise returns nullopt.
740 absl::optional<absl::string_view> TryFlat() const;
741
742 // Cord::Flatten()
743 //
744 // Flattens the cord into a single array and returns a view of the data.
745 //
746 // If the cord was already flat, the contents are not modified.
747 absl::string_view Flatten();
748
749 // Supports absl::Cord as a sink object for absl::Format().
AbslFormatFlush(absl::Cord * cord,absl::string_view part)750 friend void AbslFormatFlush(absl::Cord* cord, absl::string_view part) {
751 cord->Append(part);
752 }
753
754 // Cord::SetExpectedChecksum()
755 //
756 // Stores a checksum value with this non-empty cord instance, for later
757 // retrieval.
758 //
759 // The expected checksum is a number stored out-of-band, alongside the data.
760 // It is preserved across copies and assignments, but any mutations to a cord
761 // will cause it to lose its expected checksum.
762 //
763 // The expected checksum is not part of a Cord's value, and does not affect
764 // operations such as equality or hashing.
765 //
766 // This field is intended to store a CRC32C checksum for later validation, to
767 // help support end-to-end checksum workflows. However, the Cord API itself
768 // does no CRC validation, and assigns no meaning to this number.
769 //
770 // This call has no effect if this cord is empty.
771 void SetExpectedChecksum(uint32_t crc);
772
773 // Returns this cord's expected checksum, if it has one. Otherwise, returns
774 // nullopt.
775 absl::optional<uint32_t> ExpectedChecksum() const;
776
777 template <typename H>
AbslHashValue(H hash_state,const absl::Cord & c)778 friend H AbslHashValue(H hash_state, const absl::Cord& c) {
779 absl::optional<absl::string_view> maybe_flat = c.TryFlat();
780 if (maybe_flat.has_value()) {
781 return H::combine(std::move(hash_state), *maybe_flat);
782 }
783 return c.HashFragmented(std::move(hash_state));
784 }
785
786 // Create a Cord with the contents of StringConstant<T>::value.
787 // No allocations will be done and no data will be copied.
788 // This is an INTERNAL API and subject to change or removal. This API can only
789 // be used by spelling absl::strings_internal::MakeStringConstant, which is
790 // also an internal API.
791 template <typename T>
792 // NOLINTNEXTLINE(google-explicit-constructor)
793 constexpr Cord(strings_internal::StringConstant<T>);
794
795 private:
796 using CordRep = absl::cord_internal::CordRep;
797 using CordRepFlat = absl::cord_internal::CordRepFlat;
798 using CordzInfo = cord_internal::CordzInfo;
799 using CordzUpdateScope = cord_internal::CordzUpdateScope;
800 using CordzUpdateTracker = cord_internal::CordzUpdateTracker;
801 using InlineData = cord_internal::InlineData;
802 using MethodIdentifier = CordzUpdateTracker::MethodIdentifier;
803
804 // Creates a cord instance with `method` representing the originating
805 // public API call causing the cord to be created.
806 explicit Cord(absl::string_view src, MethodIdentifier method);
807
808 friend class CordTestPeer;
809 friend bool operator==(const Cord& lhs, const Cord& rhs);
810 friend bool operator==(const Cord& lhs, absl::string_view rhs);
811
812 friend const CordzInfo* GetCordzInfoForTesting(const Cord& cord);
813
814 // Calls the provided function once for each cord chunk, in order. Unlike
815 // Chunks(), this API will not allocate memory.
816 void ForEachChunk(absl::FunctionRef<void(absl::string_view)>) const;
817
818 // Allocates new contiguous storage for the contents of the cord. This is
819 // called by Flatten() when the cord was not already flat.
820 absl::string_view FlattenSlowPath();
821
822 // Actual cord contents are hidden inside the following simple
823 // class so that we can isolate the bulk of cord.cc from changes
824 // to the representation.
825 //
826 // InlineRep holds either a tree pointer, or an array of kMaxInline bytes.
827 class InlineRep {
828 public:
829 static constexpr unsigned char kMaxInline = cord_internal::kMaxInline;
830 static_assert(kMaxInline >= sizeof(absl::cord_internal::CordRep*), "");
831
InlineRep()832 constexpr InlineRep() : data_() {}
InlineRep(InlineData::DefaultInitType init)833 explicit InlineRep(InlineData::DefaultInitType init) : data_(init) {}
834 InlineRep(const InlineRep& src);
835 InlineRep(InlineRep&& src);
836 InlineRep& operator=(const InlineRep& src);
837 InlineRep& operator=(InlineRep&& src) noexcept;
838
839 explicit constexpr InlineRep(absl::string_view sv, CordRep* rep);
840
841 void Swap(InlineRep* rhs);
842 bool empty() const;
843 size_t size() const;
844 const char* data() const; // Returns nullptr if holding pointer
845 void set_data(const char* data, size_t n); // Discards pointer, if any
846 char* set_data(size_t n); // Write data to the result
847 // Returns nullptr if holding bytes
848 absl::cord_internal::CordRep* tree() const;
849 absl::cord_internal::CordRep* as_tree() const;
850 const char* as_chars() const;
851 // Returns non-null iff was holding a pointer
852 absl::cord_internal::CordRep* clear();
853 // Converts to pointer if necessary.
854 void reduce_size(size_t n); // REQUIRES: holding data
855 void remove_prefix(size_t n); // REQUIRES: holding data
856 void AppendArray(absl::string_view src, MethodIdentifier method);
857 absl::string_view FindFlatStartPiece() const;
858
859 // Creates a CordRepFlat instance from the current inlined data with `extra'
860 // bytes of desired additional capacity.
861 CordRepFlat* MakeFlatWithExtraCapacity(size_t extra);
862
863 // Sets the tree value for this instance. `rep` must not be null.
864 // Requires the current instance to hold a tree, and a lock to be held on
865 // any CordzInfo referenced by this instance. The latter is enforced through
866 // the CordzUpdateScope argument. If the current instance is sampled, then
867 // the CordzInfo instance is updated to reference the new `rep` value.
868 void SetTree(CordRep* rep, const CordzUpdateScope& scope);
869
870 // Identical to SetTree(), except that `rep` is allowed to be null, in
871 // which case the current instance is reset to an empty value.
872 void SetTreeOrEmpty(CordRep* rep, const CordzUpdateScope& scope);
873
874 // Sets the tree value for this instance, and randomly samples this cord.
875 // This function disregards existing contents in `data_`, and should be
876 // called when a Cord is 'promoted' from an 'uninitialized' or 'inlined'
877 // value to a non-inlined (tree / ring) value.
878 void EmplaceTree(CordRep* rep, MethodIdentifier method);
879
880 // Identical to EmplaceTree, except that it copies the parent stack from
881 // the provided `parent` data if the parent is sampled.
882 void EmplaceTree(CordRep* rep, const InlineData& parent,
883 MethodIdentifier method);
884
885 // Commits the change of a newly created, or updated `rep` root value into
886 // this cord. `old_rep` indicates the old (inlined or tree) value of the
887 // cord, and determines if the commit invokes SetTree() or EmplaceTree().
888 void CommitTree(const CordRep* old_rep, CordRep* rep,
889 const CordzUpdateScope& scope, MethodIdentifier method);
890
891 void AppendTreeToInlined(CordRep* tree, MethodIdentifier method);
892 void AppendTreeToTree(CordRep* tree, MethodIdentifier method);
893 void AppendTree(CordRep* tree, MethodIdentifier method);
894 void PrependTreeToInlined(CordRep* tree, MethodIdentifier method);
895 void PrependTreeToTree(CordRep* tree, MethodIdentifier method);
896 void PrependTree(CordRep* tree, MethodIdentifier method);
897
IsSame(const InlineRep & other)898 bool IsSame(const InlineRep& other) const { return data_ == other.data_; }
899
CopyTo(std::string * dst)900 void CopyTo(std::string* dst) const {
901 // memcpy is much faster when operating on a known size. On most supported
902 // platforms, the small string optimization is large enough that resizing
903 // to 15 bytes does not cause a memory allocation.
904 absl::strings_internal::STLStringResizeUninitialized(dst, kMaxInline);
905 data_.copy_max_inline_to(&(*dst)[0]);
906 // erase is faster than resize because the logic for memory allocation is
907 // not needed.
908 dst->erase(inline_size());
909 }
910
911 // Copies the inline contents into `dst`. Assumes the cord is not empty.
912 void CopyToArray(char* dst) const;
913
is_tree()914 bool is_tree() const { return data_.is_tree(); }
915
916 // Returns true if the Cord is being profiled by cordz.
is_profiled()917 bool is_profiled() const { return data_.is_tree() && data_.is_profiled(); }
918
919 // Returns the available inlined capacity, or 0 if is_tree() == true.
remaining_inline_capacity()920 size_t remaining_inline_capacity() const {
921 return data_.is_tree() ? 0 : kMaxInline - data_.inline_size();
922 }
923
924 // Returns the profiled CordzInfo, or nullptr if not sampled.
cordz_info()925 absl::cord_internal::CordzInfo* cordz_info() const {
926 return data_.cordz_info();
927 }
928
929 // Sets the profiled CordzInfo. `cordz_info` must not be null.
set_cordz_info(cord_internal::CordzInfo * cordz_info)930 void set_cordz_info(cord_internal::CordzInfo* cordz_info) {
931 assert(cordz_info != nullptr);
932 data_.set_cordz_info(cordz_info);
933 }
934
935 // Resets the current cordz_info to null / empty.
clear_cordz_info()936 void clear_cordz_info() { data_.clear_cordz_info(); }
937
938 private:
939 friend class Cord;
940
941 void AssignSlow(const InlineRep& src);
942 // Unrefs the tree and stops profiling.
943 void UnrefTree();
944
ResetToEmpty()945 void ResetToEmpty() { data_ = {}; }
946
set_inline_size(size_t size)947 void set_inline_size(size_t size) { data_.set_inline_size(size); }
inline_size()948 size_t inline_size() const { return data_.inline_size(); }
949
950 // Empty cords that carry a checksum have a CordRepCrc node with a null
951 // child node. The code can avoid lots of special cases where it would
952 // otherwise transition from tree to inline storage if we just remove the
953 // CordRepCrc node before mutations. Must never be called inside a
954 // CordzUpdateScope since it untracks the cordz info.
955 void MaybeRemoveEmptyCrcNode();
956
957 cord_internal::InlineData data_;
958 };
959 InlineRep contents_;
960
961 // Helper for GetFlat() and TryFlat().
962 static bool GetFlatAux(absl::cord_internal::CordRep* rep,
963 absl::string_view* fragment);
964
965 // Helper for ForEachChunk().
966 static void ForEachChunkAux(
967 absl::cord_internal::CordRep* rep,
968 absl::FunctionRef<void(absl::string_view)> callback);
969
970 // The destructor for non-empty Cords.
971 void DestroyCordSlow();
972
973 // Out-of-line implementation of slower parts of logic.
974 void CopyToArraySlowPath(char* dst) const;
975 int CompareSlowPath(absl::string_view rhs, size_t compared_size,
976 size_t size_to_compare) const;
977 int CompareSlowPath(const Cord& rhs, size_t compared_size,
978 size_t size_to_compare) const;
979 bool EqualsImpl(absl::string_view rhs, size_t size_to_compare) const;
980 bool EqualsImpl(const Cord& rhs, size_t size_to_compare) const;
981 int CompareImpl(const Cord& rhs) const;
982
983 template <typename ResultType, typename RHS>
984 friend ResultType GenericCompare(const Cord& lhs, const RHS& rhs,
985 size_t size_to_compare);
986 static absl::string_view GetFirstChunk(const Cord& c);
987 static absl::string_view GetFirstChunk(absl::string_view sv);
988
989 // Returns a new reference to contents_.tree(), or steals an existing
990 // reference if called on an rvalue.
991 absl::cord_internal::CordRep* TakeRep() const&;
992 absl::cord_internal::CordRep* TakeRep() &&;
993
994 // Helper for Append().
995 template <typename C>
996 void AppendImpl(C&& src);
997
998 // Appends / Prepends `src` to this instance, using precise sizing.
999 // This method does explicitly not attempt to use any spare capacity
1000 // in any pending last added private owned flat.
1001 // Requires `src` to be <= kMaxFlatLength.
1002 void AppendPrecise(absl::string_view src, MethodIdentifier method);
1003 void PrependPrecise(absl::string_view src, MethodIdentifier method);
1004
1005 CordBuffer GetAppendBufferSlowPath(size_t block_size, size_t capacity,
1006 size_t min_capacity);
1007
1008 // Prepends the provided data to this instance. `method` contains the public
1009 // API method for this action which is tracked for Cordz sampling purposes.
1010 void PrependArray(absl::string_view src, MethodIdentifier method);
1011
1012 // Assigns the value in 'src' to this instance, 'stealing' its contents.
1013 // Requires src.length() > kMaxBytesToCopy.
1014 Cord& AssignLargeString(std::string&& src);
1015
1016 // Helper for AbslHashValue().
1017 template <typename H>
HashFragmented(H hash_state)1018 H HashFragmented(H hash_state) const {
1019 typename H::AbslInternalPiecewiseCombiner combiner;
1020 ForEachChunk([&combiner, &hash_state](absl::string_view chunk) {
1021 hash_state = combiner.add_buffer(std::move(hash_state), chunk.data(),
1022 chunk.size());
1023 });
1024 return H::combine(combiner.finalize(std::move(hash_state)), size());
1025 }
1026
1027 friend class CrcCord;
1028 void SetCrcCordState(crc_internal::CrcCordState state);
1029 const crc_internal::CrcCordState* MaybeGetCrcCordState() const;
1030 };
1031
1032 ABSL_NAMESPACE_END
1033 } // namespace absl
1034
1035 namespace absl {
1036 ABSL_NAMESPACE_BEGIN
1037
1038 // allow a Cord to be logged
1039 extern std::ostream& operator<<(std::ostream& out, const Cord& cord);
1040
1041 // ------------------------------------------------------------------
1042 // Internal details follow. Clients should ignore.
1043
1044 namespace cord_internal {
1045
1046 // Does non-template-specific `CordRepExternal` initialization.
1047 // Requires `data` to be non-empty.
1048 void InitializeCordRepExternal(absl::string_view data, CordRepExternal* rep);
1049
1050 // Creates a new `CordRep` that owns `data` and `releaser` and returns a pointer
1051 // to it. Requires `data` to be non-empty.
1052 template <typename Releaser>
1053 // NOLINTNEXTLINE - suppress clang-tidy raw pointer return.
NewExternalRep(absl::string_view data,Releaser && releaser)1054 CordRep* NewExternalRep(absl::string_view data, Releaser&& releaser) {
1055 assert(!data.empty());
1056 using ReleaserType = absl::decay_t<Releaser>;
1057 CordRepExternal* rep = new CordRepExternalImpl<ReleaserType>(
1058 std::forward<Releaser>(releaser), 0);
1059 InitializeCordRepExternal(data, rep);
1060 return rep;
1061 }
1062
1063 // Overload for function reference types that dispatches using a function
1064 // pointer because there are no `alignof()` or `sizeof()` a function reference.
1065 // NOLINTNEXTLINE - suppress clang-tidy raw pointer return.
NewExternalRep(absl::string_view data,void (& releaser)(absl::string_view))1066 inline CordRep* NewExternalRep(absl::string_view data,
1067 void (&releaser)(absl::string_view)) {
1068 return NewExternalRep(data, &releaser);
1069 }
1070
1071 } // namespace cord_internal
1072
1073 template <typename Releaser>
MakeCordFromExternal(absl::string_view data,Releaser && releaser)1074 Cord MakeCordFromExternal(absl::string_view data, Releaser&& releaser) {
1075 Cord cord;
1076 if (ABSL_PREDICT_TRUE(!data.empty())) {
1077 cord.contents_.EmplaceTree(::absl::cord_internal::NewExternalRep(
1078 data, std::forward<Releaser>(releaser)),
1079 Cord::MethodIdentifier::kMakeCordFromExternal);
1080 } else {
1081 using ReleaserType = absl::decay_t<Releaser>;
1082 cord_internal::InvokeReleaser(
1083 cord_internal::Rank0{}, ReleaserType(std::forward<Releaser>(releaser)),
1084 data);
1085 }
1086 return cord;
1087 }
1088
InlineRep(absl::string_view sv,CordRep * rep)1089 constexpr Cord::InlineRep::InlineRep(absl::string_view sv, CordRep* rep)
1090 : data_(sv, rep) {}
1091
InlineRep(const Cord::InlineRep & src)1092 inline Cord::InlineRep::InlineRep(const Cord::InlineRep& src)
1093 : data_(InlineData::kDefaultInit) {
1094 if (CordRep* tree = src.tree()) {
1095 EmplaceTree(CordRep::Ref(tree), src.data_,
1096 CordzUpdateTracker::kConstructorCord);
1097 } else {
1098 data_ = src.data_;
1099 }
1100 }
1101
InlineRep(Cord::InlineRep && src)1102 inline Cord::InlineRep::InlineRep(Cord::InlineRep&& src) : data_(src.data_) {
1103 src.ResetToEmpty();
1104 }
1105
1106 inline Cord::InlineRep& Cord::InlineRep::operator=(const Cord::InlineRep& src) {
1107 if (this == &src) {
1108 return *this;
1109 }
1110 if (!is_tree() && !src.is_tree()) {
1111 data_ = src.data_;
1112 return *this;
1113 }
1114 AssignSlow(src);
1115 return *this;
1116 }
1117
1118 inline Cord::InlineRep& Cord::InlineRep::operator=(
1119 Cord::InlineRep&& src) noexcept {
1120 if (is_tree()) {
1121 UnrefTree();
1122 }
1123 data_ = src.data_;
1124 src.ResetToEmpty();
1125 return *this;
1126 }
1127
Swap(Cord::InlineRep * rhs)1128 inline void Cord::InlineRep::Swap(Cord::InlineRep* rhs) {
1129 if (rhs == this) {
1130 return;
1131 }
1132 std::swap(data_, rhs->data_);
1133 }
1134
data()1135 inline const char* Cord::InlineRep::data() const {
1136 return is_tree() ? nullptr : data_.as_chars();
1137 }
1138
as_chars()1139 inline const char* Cord::InlineRep::as_chars() const {
1140 assert(!data_.is_tree());
1141 return data_.as_chars();
1142 }
1143
as_tree()1144 inline absl::cord_internal::CordRep* Cord::InlineRep::as_tree() const {
1145 assert(data_.is_tree());
1146 return data_.as_tree();
1147 }
1148
tree()1149 inline absl::cord_internal::CordRep* Cord::InlineRep::tree() const {
1150 if (is_tree()) {
1151 return as_tree();
1152 } else {
1153 return nullptr;
1154 }
1155 }
1156
empty()1157 inline bool Cord::InlineRep::empty() const { return data_.is_empty(); }
1158
size()1159 inline size_t Cord::InlineRep::size() const {
1160 return is_tree() ? as_tree()->length : inline_size();
1161 }
1162
MakeFlatWithExtraCapacity(size_t extra)1163 inline cord_internal::CordRepFlat* Cord::InlineRep::MakeFlatWithExtraCapacity(
1164 size_t extra) {
1165 static_assert(cord_internal::kMinFlatLength >= sizeof(data_), "");
1166 size_t len = data_.inline_size();
1167 auto* result = CordRepFlat::New(len + extra);
1168 result->length = len;
1169 data_.copy_max_inline_to(result->Data());
1170 return result;
1171 }
1172
EmplaceTree(CordRep * rep,MethodIdentifier method)1173 inline void Cord::InlineRep::EmplaceTree(CordRep* rep,
1174 MethodIdentifier method) {
1175 assert(rep);
1176 data_.make_tree(rep);
1177 CordzInfo::MaybeTrackCord(data_, method);
1178 }
1179
EmplaceTree(CordRep * rep,const InlineData & parent,MethodIdentifier method)1180 inline void Cord::InlineRep::EmplaceTree(CordRep* rep, const InlineData& parent,
1181 MethodIdentifier method) {
1182 data_.make_tree(rep);
1183 CordzInfo::MaybeTrackCord(data_, parent, method);
1184 }
1185
SetTree(CordRep * rep,const CordzUpdateScope & scope)1186 inline void Cord::InlineRep::SetTree(CordRep* rep,
1187 const CordzUpdateScope& scope) {
1188 assert(rep);
1189 assert(data_.is_tree());
1190 data_.set_tree(rep);
1191 scope.SetCordRep(rep);
1192 }
1193
SetTreeOrEmpty(CordRep * rep,const CordzUpdateScope & scope)1194 inline void Cord::InlineRep::SetTreeOrEmpty(CordRep* rep,
1195 const CordzUpdateScope& scope) {
1196 assert(data_.is_tree());
1197 if (rep) {
1198 data_.set_tree(rep);
1199 } else {
1200 data_ = {};
1201 }
1202 scope.SetCordRep(rep);
1203 }
1204
CommitTree(const CordRep * old_rep,CordRep * rep,const CordzUpdateScope & scope,MethodIdentifier method)1205 inline void Cord::InlineRep::CommitTree(const CordRep* old_rep, CordRep* rep,
1206 const CordzUpdateScope& scope,
1207 MethodIdentifier method) {
1208 if (old_rep) {
1209 SetTree(rep, scope);
1210 } else {
1211 EmplaceTree(rep, method);
1212 }
1213 }
1214
clear()1215 inline absl::cord_internal::CordRep* Cord::InlineRep::clear() {
1216 if (is_tree()) {
1217 CordzInfo::MaybeUntrackCord(cordz_info());
1218 }
1219 absl::cord_internal::CordRep* result = tree();
1220 ResetToEmpty();
1221 return result;
1222 }
1223
CopyToArray(char * dst)1224 inline void Cord::InlineRep::CopyToArray(char* dst) const {
1225 assert(!is_tree());
1226 size_t n = inline_size();
1227 assert(n != 0);
1228 cord_internal::SmallMemmove(dst, data_.as_chars(), n);
1229 }
1230
MaybeRemoveEmptyCrcNode()1231 inline void Cord::InlineRep::MaybeRemoveEmptyCrcNode() {
1232 CordRep* rep = tree();
1233 if (rep == nullptr || ABSL_PREDICT_TRUE(rep->length > 0)) {
1234 return;
1235 }
1236 assert(rep->IsCrc());
1237 assert(rep->crc()->child == nullptr);
1238 CordzInfo::MaybeUntrackCord(cordz_info());
1239 CordRep::Unref(rep);
1240 ResetToEmpty();
1241 }
1242
Cord()1243 constexpr inline Cord::Cord() noexcept {}
1244
Cord(absl::string_view src)1245 inline Cord::Cord(absl::string_view src)
1246 : Cord(src, CordzUpdateTracker::kConstructorString) {}
1247
1248 template <typename T>
Cord(strings_internal::StringConstant<T>)1249 constexpr Cord::Cord(strings_internal::StringConstant<T>)
1250 : contents_(strings_internal::StringConstant<T>::value,
1251 strings_internal::StringConstant<T>::value.size() <=
1252 cord_internal::kMaxInline
1253 ? nullptr
1254 : &cord_internal::ConstInitExternalStorage<
1255 strings_internal::StringConstant<T>>::value) {}
1256
1257 inline Cord& Cord::operator=(const Cord& x) {
1258 contents_ = x.contents_;
1259 return *this;
1260 }
1261
1262 template <typename T, Cord::EnableIfString<T>>
1263 Cord& Cord::operator=(T&& src) {
1264 if (src.size() <= cord_internal::kMaxBytesToCopy) {
1265 return operator=(absl::string_view(src));
1266 } else {
1267 return AssignLargeString(std::forward<T>(src));
1268 }
1269 }
1270
Cord(const Cord & src)1271 inline Cord::Cord(const Cord& src) : contents_(src.contents_) {}
1272
Cord(Cord && src)1273 inline Cord::Cord(Cord&& src) noexcept : contents_(std::move(src.contents_)) {}
1274
swap(Cord & other)1275 inline void Cord::swap(Cord& other) noexcept {
1276 contents_.Swap(&other.contents_);
1277 }
1278
1279 inline Cord& Cord::operator=(Cord&& x) noexcept {
1280 contents_ = std::move(x.contents_);
1281 return *this;
1282 }
1283
1284 extern template Cord::Cord(std::string&& src);
1285
size()1286 inline size_t Cord::size() const {
1287 // Length is 1st field in str.rep_
1288 return contents_.size();
1289 }
1290
empty()1291 inline bool Cord::empty() const { return size() == 0; }
1292
EstimatedMemoryUsage(CordMemoryAccounting accounting_method)1293 inline size_t Cord::EstimatedMemoryUsage(
1294 CordMemoryAccounting accounting_method) const {
1295 size_t result = sizeof(Cord);
1296 if (const absl::cord_internal::CordRep* rep = contents_.tree()) {
1297 switch (accounting_method) {
1298 case CordMemoryAccounting::kFairShare:
1299 result += cord_internal::GetEstimatedFairShareMemoryUsage(rep);
1300 break;
1301 case CordMemoryAccounting::kTotalMorePrecise:
1302 result += cord_internal::GetMorePreciseMemoryUsage(rep);
1303 break;
1304 case CordMemoryAccounting::kTotal:
1305 result += cord_internal::GetEstimatedMemoryUsage(rep);
1306 break;
1307 }
1308 }
1309 return result;
1310 }
1311
TryFlat()1312 inline absl::optional<absl::string_view> Cord::TryFlat() const {
1313 absl::cord_internal::CordRep* rep = contents_.tree();
1314 if (rep == nullptr) {
1315 return absl::string_view(contents_.data(), contents_.size());
1316 }
1317 absl::string_view fragment;
1318 if (GetFlatAux(rep, &fragment)) {
1319 return fragment;
1320 }
1321 return absl::nullopt;
1322 }
1323
Flatten()1324 inline absl::string_view Cord::Flatten() {
1325 absl::cord_internal::CordRep* rep = contents_.tree();
1326 if (rep == nullptr) {
1327 return absl::string_view(contents_.data(), contents_.size());
1328 } else {
1329 absl::string_view already_flat_contents;
1330 if (GetFlatAux(rep, &already_flat_contents)) {
1331 return already_flat_contents;
1332 }
1333 }
1334 return FlattenSlowPath();
1335 }
1336
Append(absl::string_view src)1337 inline void Cord::Append(absl::string_view src) {
1338 contents_.AppendArray(src, CordzUpdateTracker::kAppendString);
1339 }
1340
Prepend(absl::string_view src)1341 inline void Cord::Prepend(absl::string_view src) {
1342 PrependArray(src, CordzUpdateTracker::kPrependString);
1343 }
1344
Append(CordBuffer buffer)1345 inline void Cord::Append(CordBuffer buffer) {
1346 if (ABSL_PREDICT_FALSE(buffer.length() == 0)) return;
1347 absl::string_view short_value;
1348 if (CordRep* rep = buffer.ConsumeValue(short_value)) {
1349 contents_.AppendTree(rep, CordzUpdateTracker::kAppendCordBuffer);
1350 } else {
1351 AppendPrecise(short_value, CordzUpdateTracker::kAppendCordBuffer);
1352 }
1353 }
1354
Prepend(CordBuffer buffer)1355 inline void Cord::Prepend(CordBuffer buffer) {
1356 if (ABSL_PREDICT_FALSE(buffer.length() == 0)) return;
1357 absl::string_view short_value;
1358 if (CordRep* rep = buffer.ConsumeValue(short_value)) {
1359 contents_.PrependTree(rep, CordzUpdateTracker::kPrependCordBuffer);
1360 } else {
1361 PrependPrecise(short_value, CordzUpdateTracker::kPrependCordBuffer);
1362 }
1363 }
1364
GetAppendBuffer(size_t capacity,size_t min_capacity)1365 inline CordBuffer Cord::GetAppendBuffer(size_t capacity, size_t min_capacity) {
1366 if (empty()) return CordBuffer::CreateWithDefaultLimit(capacity);
1367 return GetAppendBufferSlowPath(0, capacity, min_capacity);
1368 }
1369
GetCustomAppendBuffer(size_t block_size,size_t capacity,size_t min_capacity)1370 inline CordBuffer Cord::GetCustomAppendBuffer(size_t block_size,
1371 size_t capacity,
1372 size_t min_capacity) {
1373 if (empty()) {
1374 return block_size ? CordBuffer::CreateWithCustomLimit(block_size, capacity)
1375 : CordBuffer::CreateWithDefaultLimit(capacity);
1376 }
1377 return GetAppendBufferSlowPath(block_size, capacity, min_capacity);
1378 }
1379
1380 extern template void Cord::Append(std::string&& src);
1381 extern template void Cord::Prepend(std::string&& src);
1382
Compare(const Cord & rhs)1383 inline int Cord::Compare(const Cord& rhs) const {
1384 if (!contents_.is_tree() && !rhs.contents_.is_tree()) {
1385 return contents_.data_.Compare(rhs.contents_.data_);
1386 }
1387
1388 return CompareImpl(rhs);
1389 }
1390
1391 // Does 'this' cord start/end with rhs
StartsWith(const Cord & rhs)1392 inline bool Cord::StartsWith(const Cord& rhs) const {
1393 if (contents_.IsSame(rhs.contents_)) return true;
1394 size_t rhs_size = rhs.size();
1395 if (size() < rhs_size) return false;
1396 return EqualsImpl(rhs, rhs_size);
1397 }
1398
StartsWith(absl::string_view rhs)1399 inline bool Cord::StartsWith(absl::string_view rhs) const {
1400 size_t rhs_size = rhs.size();
1401 if (size() < rhs_size) return false;
1402 return EqualsImpl(rhs, rhs_size);
1403 }
1404
InitTree(cord_internal::CordRep * tree)1405 inline void Cord::ChunkIterator::InitTree(cord_internal::CordRep* tree) {
1406 tree = cord_internal::SkipCrcNode(tree);
1407 if (tree->tag == cord_internal::BTREE) {
1408 current_chunk_ = btree_reader_.Init(tree->btree());
1409 } else {
1410 current_leaf_ = tree;
1411 current_chunk_ = cord_internal::EdgeData(tree);
1412 }
1413 }
1414
ChunkIterator(cord_internal::CordRep * tree)1415 inline Cord::ChunkIterator::ChunkIterator(cord_internal::CordRep* tree) {
1416 bytes_remaining_ = tree->length;
1417 InitTree(tree);
1418 }
1419
ChunkIterator(const Cord * cord)1420 inline Cord::ChunkIterator::ChunkIterator(const Cord* cord) {
1421 if (CordRep* tree = cord->contents_.tree()) {
1422 bytes_remaining_ = tree->length;
1423 if (ABSL_PREDICT_TRUE(bytes_remaining_ != 0)) {
1424 InitTree(tree);
1425 } else {
1426 current_chunk_ = {};
1427 }
1428 } else {
1429 bytes_remaining_ = cord->contents_.inline_size();
1430 current_chunk_ = {cord->contents_.data(), bytes_remaining_};
1431 }
1432 }
1433
AdvanceBtree()1434 inline Cord::ChunkIterator& Cord::ChunkIterator::AdvanceBtree() {
1435 current_chunk_ = btree_reader_.Next();
1436 return *this;
1437 }
1438
AdvanceBytesBtree(size_t n)1439 inline void Cord::ChunkIterator::AdvanceBytesBtree(size_t n) {
1440 assert(n >= current_chunk_.size());
1441 bytes_remaining_ -= n;
1442 if (bytes_remaining_) {
1443 if (n == current_chunk_.size()) {
1444 current_chunk_ = btree_reader_.Next();
1445 } else {
1446 size_t offset = btree_reader_.length() - bytes_remaining_;
1447 current_chunk_ = btree_reader_.Seek(offset);
1448 }
1449 } else {
1450 current_chunk_ = {};
1451 }
1452 }
1453
1454 inline Cord::ChunkIterator& Cord::ChunkIterator::operator++() {
1455 ABSL_HARDENING_ASSERT(bytes_remaining_ > 0 &&
1456 "Attempted to iterate past `end()`");
1457 assert(bytes_remaining_ >= current_chunk_.size());
1458 bytes_remaining_ -= current_chunk_.size();
1459 if (bytes_remaining_ > 0) {
1460 if (btree_reader_) {
1461 return AdvanceBtree();
1462 } else {
1463 assert(!current_chunk_.empty()); // Called on invalid iterator.
1464 }
1465 current_chunk_ = {};
1466 }
1467 return *this;
1468 }
1469
1470 inline Cord::ChunkIterator Cord::ChunkIterator::operator++(int) {
1471 ChunkIterator tmp(*this);
1472 operator++();
1473 return tmp;
1474 }
1475
1476 inline bool Cord::ChunkIterator::operator==(const ChunkIterator& other) const {
1477 return bytes_remaining_ == other.bytes_remaining_;
1478 }
1479
1480 inline bool Cord::ChunkIterator::operator!=(const ChunkIterator& other) const {
1481 return !(*this == other);
1482 }
1483
1484 inline Cord::ChunkIterator::reference Cord::ChunkIterator::operator*() const {
1485 ABSL_HARDENING_ASSERT(bytes_remaining_ != 0);
1486 return current_chunk_;
1487 }
1488
1489 inline Cord::ChunkIterator::pointer Cord::ChunkIterator::operator->() const {
1490 ABSL_HARDENING_ASSERT(bytes_remaining_ != 0);
1491 return ¤t_chunk_;
1492 }
1493
RemoveChunkPrefix(size_t n)1494 inline void Cord::ChunkIterator::RemoveChunkPrefix(size_t n) {
1495 assert(n < current_chunk_.size());
1496 current_chunk_.remove_prefix(n);
1497 bytes_remaining_ -= n;
1498 }
1499
AdvanceBytes(size_t n)1500 inline void Cord::ChunkIterator::AdvanceBytes(size_t n) {
1501 assert(bytes_remaining_ >= n);
1502 if (ABSL_PREDICT_TRUE(n < current_chunk_.size())) {
1503 RemoveChunkPrefix(n);
1504 } else if (n != 0) {
1505 if (btree_reader_) {
1506 AdvanceBytesBtree(n);
1507 } else {
1508 bytes_remaining_ = 0;
1509 }
1510 }
1511 }
1512
chunk_begin()1513 inline Cord::ChunkIterator Cord::chunk_begin() const {
1514 return ChunkIterator(this);
1515 }
1516
chunk_end()1517 inline Cord::ChunkIterator Cord::chunk_end() const { return ChunkIterator(); }
1518
begin()1519 inline Cord::ChunkIterator Cord::ChunkRange::begin() const {
1520 return cord_->chunk_begin();
1521 }
1522
end()1523 inline Cord::ChunkIterator Cord::ChunkRange::end() const {
1524 return cord_->chunk_end();
1525 }
1526
Chunks()1527 inline Cord::ChunkRange Cord::Chunks() const { return ChunkRange(this); }
1528
1529 inline Cord::CharIterator& Cord::CharIterator::operator++() {
1530 if (ABSL_PREDICT_TRUE(chunk_iterator_->size() > 1)) {
1531 chunk_iterator_.RemoveChunkPrefix(1);
1532 } else {
1533 ++chunk_iterator_;
1534 }
1535 return *this;
1536 }
1537
1538 inline Cord::CharIterator Cord::CharIterator::operator++(int) {
1539 CharIterator tmp(*this);
1540 operator++();
1541 return tmp;
1542 }
1543
1544 inline bool Cord::CharIterator::operator==(const CharIterator& other) const {
1545 return chunk_iterator_ == other.chunk_iterator_;
1546 }
1547
1548 inline bool Cord::CharIterator::operator!=(const CharIterator& other) const {
1549 return !(*this == other);
1550 }
1551
1552 inline Cord::CharIterator::reference Cord::CharIterator::operator*() const {
1553 return *chunk_iterator_->data();
1554 }
1555
1556 inline Cord::CharIterator::pointer Cord::CharIterator::operator->() const {
1557 return chunk_iterator_->data();
1558 }
1559
AdvanceAndRead(CharIterator * it,size_t n_bytes)1560 inline Cord Cord::AdvanceAndRead(CharIterator* it, size_t n_bytes) {
1561 assert(it != nullptr);
1562 return it->chunk_iterator_.AdvanceAndReadBytes(n_bytes);
1563 }
1564
Advance(CharIterator * it,size_t n_bytes)1565 inline void Cord::Advance(CharIterator* it, size_t n_bytes) {
1566 assert(it != nullptr);
1567 it->chunk_iterator_.AdvanceBytes(n_bytes);
1568 }
1569
ChunkRemaining(const CharIterator & it)1570 inline absl::string_view Cord::ChunkRemaining(const CharIterator& it) {
1571 return *it.chunk_iterator_;
1572 }
1573
char_begin()1574 inline Cord::CharIterator Cord::char_begin() const {
1575 return CharIterator(this);
1576 }
1577
char_end()1578 inline Cord::CharIterator Cord::char_end() const { return CharIterator(); }
1579
begin()1580 inline Cord::CharIterator Cord::CharRange::begin() const {
1581 return cord_->char_begin();
1582 }
1583
end()1584 inline Cord::CharIterator Cord::CharRange::end() const {
1585 return cord_->char_end();
1586 }
1587
Chars()1588 inline Cord::CharRange Cord::Chars() const { return CharRange(this); }
1589
ForEachChunk(absl::FunctionRef<void (absl::string_view)> callback)1590 inline void Cord::ForEachChunk(
1591 absl::FunctionRef<void(absl::string_view)> callback) const {
1592 absl::cord_internal::CordRep* rep = contents_.tree();
1593 if (rep == nullptr) {
1594 callback(absl::string_view(contents_.data(), contents_.size()));
1595 } else {
1596 ForEachChunkAux(rep, callback);
1597 }
1598 }
1599
1600 // Nonmember Cord-to-Cord relational operators.
1601 inline bool operator==(const Cord& lhs, const Cord& rhs) {
1602 if (lhs.contents_.IsSame(rhs.contents_)) return true;
1603 size_t rhs_size = rhs.size();
1604 if (lhs.size() != rhs_size) return false;
1605 return lhs.EqualsImpl(rhs, rhs_size);
1606 }
1607
1608 inline bool operator!=(const Cord& x, const Cord& y) { return !(x == y); }
1609 inline bool operator<(const Cord& x, const Cord& y) { return x.Compare(y) < 0; }
1610 inline bool operator>(const Cord& x, const Cord& y) { return x.Compare(y) > 0; }
1611 inline bool operator<=(const Cord& x, const Cord& y) {
1612 return x.Compare(y) <= 0;
1613 }
1614 inline bool operator>=(const Cord& x, const Cord& y) {
1615 return x.Compare(y) >= 0;
1616 }
1617
1618 // Nonmember Cord-to-absl::string_view relational operators.
1619 //
1620 // Due to implicit conversions, these also enable comparisons of Cord with
1621 // with std::string, ::string, and const char*.
1622 inline bool operator==(const Cord& lhs, absl::string_view rhs) {
1623 size_t lhs_size = lhs.size();
1624 size_t rhs_size = rhs.size();
1625 if (lhs_size != rhs_size) return false;
1626 return lhs.EqualsImpl(rhs, rhs_size);
1627 }
1628
1629 inline bool operator==(absl::string_view x, const Cord& y) { return y == x; }
1630 inline bool operator!=(const Cord& x, absl::string_view y) { return !(x == y); }
1631 inline bool operator!=(absl::string_view x, const Cord& y) { return !(x == y); }
1632 inline bool operator<(const Cord& x, absl::string_view y) {
1633 return x.Compare(y) < 0;
1634 }
1635 inline bool operator<(absl::string_view x, const Cord& y) {
1636 return y.Compare(x) > 0;
1637 }
1638 inline bool operator>(const Cord& x, absl::string_view y) { return y < x; }
1639 inline bool operator>(absl::string_view x, const Cord& y) { return y < x; }
1640 inline bool operator<=(const Cord& x, absl::string_view y) { return !(y < x); }
1641 inline bool operator<=(absl::string_view x, const Cord& y) { return !(y < x); }
1642 inline bool operator>=(const Cord& x, absl::string_view y) { return !(x < y); }
1643 inline bool operator>=(absl::string_view x, const Cord& y) { return !(x < y); }
1644
1645 // Some internals exposed to test code.
1646 namespace strings_internal {
1647 class CordTestAccess {
1648 public:
1649 static size_t FlatOverhead();
1650 static size_t MaxFlatLength();
1651 static size_t SizeofCordRepExternal();
1652 static size_t SizeofCordRepSubstring();
1653 static size_t FlatTagToLength(uint8_t tag);
1654 static uint8_t LengthToTag(size_t s);
1655 };
1656 } // namespace strings_internal
1657 ABSL_NAMESPACE_END
1658 } // namespace absl
1659
1660 #endif // ABSL_STRINGS_CORD_H_
1661