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/cordz_functions.h"
90 #include "absl/strings/internal/cordz_info.h"
91 #include "absl/strings/internal/cordz_statistics.h"
92 #include "absl/strings/internal/cordz_update_scope.h"
93 #include "absl/strings/internal/cordz_update_tracker.h"
94 #include "absl/strings/internal/resize_uninitialized.h"
95 #include "absl/strings/internal/string_constant.h"
96 #include "absl/strings/string_view.h"
97 #include "absl/types/optional.h"
98
99 namespace absl {
100 ABSL_NAMESPACE_BEGIN
101 class Cord;
102 class CordTestPeer;
103 template <typename Releaser>
104 Cord MakeCordFromExternal(absl::string_view, Releaser&&);
105 void CopyCordToString(const Cord& src, std::string* dst);
106
107 // Cord memory accounting modes
108 enum class CordMemoryAccounting {
109 // Counts the *approximate* number of bytes held in full or in part by this
110 // Cord (which may not remain the same between invocations). Cords that share
111 // memory could each be "charged" independently for the same shared memory.
112 // See also comment on `kTotalMorePrecise` on internally shared memory.
113 kTotal,
114
115 // Counts the *approximate* number of bytes held in full or in part by this
116 // Cord for the distinct memory held by this cord. This option is similar
117 // to `kTotal`, except that if the cord has multiple references to the same
118 // memory, that memory is only counted once.
119 //
120 // For example:
121 // absl::Cord cord;
122 // cord.append(some_other_cord);
123 // cord.append(some_other_cord);
124 // // Counts `some_other_cord` twice:
125 // cord.EstimatedMemoryUsage(kTotal);
126 // // Counts `some_other_cord` once:
127 // cord.EstimatedMemoryUsage(kTotalMorePrecise);
128 //
129 // The `kTotalMorePrecise` number is more expensive to compute as it requires
130 // deduplicating all memory references. Applications should prefer to use
131 // `kFairShare` or `kTotal` unless they really need a more precise estimate
132 // on "how much memory is potentially held / kept alive by this cord?"
133 kTotalMorePrecise,
134
135 // Counts the *approximate* number of bytes held in full or in part by this
136 // Cord weighted by the sharing ratio of that data. For example, if some data
137 // edge is shared by 4 different Cords, then each cord is attributed 1/4th of
138 // the total memory usage as a 'fair share' of the total memory usage.
139 kFairShare,
140 };
141
142 // Cord
143 //
144 // A Cord is a sequence of characters, designed to be more efficient than a
145 // `std::string` in certain circumstances: namely, large string data that needs
146 // to change over its lifetime or shared, especially when such data is shared
147 // across API boundaries.
148 //
149 // A Cord stores its character data in a structure that allows efficient prepend
150 // and append operations. This makes a Cord useful for large string data sent
151 // over in a wire format that may need to be prepended or appended at some point
152 // during the data exchange (e.g. HTTP, protocol buffers). For example, a
153 // Cord is useful for storing an HTTP request, and prepending an HTTP header to
154 // such a request.
155 //
156 // Cords should not be used for storing general string data, however. They
157 // require overhead to construct and are slower than strings for random access.
158 //
159 // The Cord API provides the following common API operations:
160 //
161 // * Create or assign Cords out of existing string data, memory, or other Cords
162 // * Append and prepend data to an existing Cord
163 // * Create new Sub-Cords from existing Cord data
164 // * Swap Cord data and compare Cord equality
165 // * Write out Cord data by constructing a `std::string`
166 //
167 // Additionally, the API provides iterator utilities to iterate through Cord
168 // data via chunks or character bytes.
169 //
170 class Cord {
171 private:
172 template <typename T>
173 using EnableIfString =
174 absl::enable_if_t<std::is_same<T, std::string>::value, int>;
175
176 public:
177 // Cord::Cord() Constructors.
178
179 // Creates an empty Cord.
180 constexpr Cord() noexcept;
181
182 // Creates a Cord from an existing Cord. Cord is copyable and efficiently
183 // movable. The moved-from state is valid but unspecified.
184 Cord(const Cord& src);
185 Cord(Cord&& src) noexcept;
186 Cord& operator=(const Cord& x);
187 Cord& operator=(Cord&& x) noexcept;
188
189 // Creates a Cord from a `src` string. This constructor is marked explicit to
190 // prevent implicit Cord constructions from arguments convertible to an
191 // `absl::string_view`.
192 explicit Cord(absl::string_view src);
193 Cord& operator=(absl::string_view src);
194
195 // Creates a Cord from a `std::string&&` rvalue. These constructors are
196 // templated to avoid ambiguities for types that are convertible to both
197 // `absl::string_view` and `std::string`, such as `const char*`.
198 template <typename T, EnableIfString<T> = 0>
199 explicit Cord(T&& src);
200 template <typename T, EnableIfString<T> = 0>
201 Cord& operator=(T&& src);
202
203 // Cord::~Cord()
204 //
205 // Destructs the Cord.
~Cord()206 ~Cord() {
207 if (contents_.is_tree()) DestroyCordSlow();
208 }
209
210 // MakeCordFromExternal()
211 //
212 // Creates a Cord that takes ownership of external string memory. The
213 // contents of `data` are not copied to the Cord; instead, the external
214 // memory is added to the Cord and reference-counted. This data may not be
215 // changed for the life of the Cord, though it may be prepended or appended
216 // to.
217 //
218 // `MakeCordFromExternal()` takes a callable "releaser" that is invoked when
219 // the reference count for `data` reaches zero. As noted above, this data must
220 // remain live until the releaser is invoked. The callable releaser also must:
221 //
222 // * be move constructible
223 // * support `void operator()(absl::string_view) const` or `void operator()`
224 //
225 // Example:
226 //
227 // Cord MakeCord(BlockPool* pool) {
228 // Block* block = pool->NewBlock();
229 // FillBlock(block);
230 // return absl::MakeCordFromExternal(
231 // block->ToStringView(),
232 // [pool, block](absl::string_view v) {
233 // pool->FreeBlock(block, v);
234 // });
235 // }
236 //
237 // WARNING: Because a Cord can be reference-counted, it's likely a bug if your
238 // releaser doesn't do anything. For example, consider the following:
239 //
240 // void Foo(const char* buffer, int len) {
241 // auto c = absl::MakeCordFromExternal(absl::string_view(buffer, len),
242 // [](absl::string_view) {});
243 //
244 // // BUG: If Bar() copies its cord for any reason, including keeping a
245 // // substring of it, the lifetime of buffer might be extended beyond
246 // // when Foo() returns.
247 // Bar(c);
248 // }
249 template <typename Releaser>
250 friend Cord MakeCordFromExternal(absl::string_view data, Releaser&& releaser);
251
252 // Cord::Clear()
253 //
254 // Releases the Cord data. Any nodes that share data with other Cords, if
255 // applicable, will have their reference counts reduced by 1.
256 ABSL_ATTRIBUTE_REINITIALIZES void Clear();
257
258 // Cord::Append()
259 //
260 // Appends data to the Cord, which may come from another Cord or other string
261 // data.
262 void Append(const Cord& src);
263 void Append(Cord&& src);
264 void Append(absl::string_view src);
265 template <typename T, EnableIfString<T> = 0>
266 void Append(T&& src);
267
268 // Appends `buffer` to this cord, unless `buffer` has a zero length in which
269 // case this method has no effect on this cord instance.
270 // This method is guaranteed to consume `buffer`.
271 void Append(CordBuffer buffer);
272
273 // Returns a CordBuffer, re-using potential existing capacity in this cord.
274 //
275 // Cord instances may have additional unused capacity in the last (or first)
276 // nodes of the underlying tree to facilitate amortized growth. This method
277 // allows applications to explicitly use this spare capacity if available,
278 // or create a new CordBuffer instance otherwise.
279 // If this cord has a final non-shared node with at least `min_capacity`
280 // available, then this method will return that buffer including its data
281 // contents. I.e.; the returned buffer will have a non-zero length, and
282 // a capacity of at least `buffer.length + min_capacity`. Otherwise, this
283 // method will return `CordBuffer::CreateWithDefaultLimit(capacity)`.
284 //
285 // Below an example of using GetAppendBuffer. Notice that in this example we
286 // use `GetAppendBuffer()` only on the first iteration. As we know nothing
287 // about any initial extra capacity in `cord`, we may be able to use the extra
288 // capacity. But as we add new buffers with fully utilized contents after that
289 // we avoid calling `GetAppendBuffer()` on subsequent iterations: while this
290 // works fine, it results in an unnecessary inspection of cord contents:
291 //
292 // void AppendRandomDataToCord(absl::Cord &cord, size_t n) {
293 // bool first = true;
294 // while (n > 0) {
295 // CordBuffer buffer = first ? cord.GetAppendBuffer(n)
296 // : CordBuffer::CreateWithDefaultLimit(n);
297 // absl::Span<char> data = buffer.available_up_to(n);
298 // FillRandomValues(data.data(), data.size());
299 // buffer.IncreaseLengthBy(data.size());
300 // cord.Append(std::move(buffer));
301 // n -= data.size();
302 // first = false;
303 // }
304 // }
305 CordBuffer GetAppendBuffer(size_t capacity, size_t min_capacity = 16);
306
307 // Returns a CordBuffer, re-using potential existing capacity in this cord.
308 //
309 // This function is identical to `GetAppendBuffer`, except that in the case
310 // where a new `CordBuffer` is allocated, it is allocated using the provided
311 // custom limit instead of the default limit. `GetAppendBuffer` will default
312 // to `CordBuffer::CreateWithDefaultLimit(capacity)` whereas this method
313 // will default to `CordBuffer::CreateWithCustomLimit(block_size, capacity)`.
314 // This method is equivalent to `GetAppendBuffer` if `block_size` is zero.
315 // See the documentation for `CreateWithCustomLimit` for more details on the
316 // restrictions and legal values for `block_size`.
317 CordBuffer GetCustomAppendBuffer(size_t block_size, size_t capacity,
318 size_t min_capacity = 16);
319
320 // Cord::Prepend()
321 //
322 // Prepends data to the Cord, which may come from another Cord or other string
323 // data.
324 void Prepend(const Cord& src);
325 void Prepend(absl::string_view src);
326 template <typename T, EnableIfString<T> = 0>
327 void Prepend(T&& src);
328
329 // Prepends `buffer` to this cord, unless `buffer` has a zero length in which
330 // case this method has no effect on this cord instance.
331 // This method is guaranteed to consume `buffer`.
332 void Prepend(CordBuffer buffer);
333
334 // Cord::RemovePrefix()
335 //
336 // Removes the first `n` bytes of a Cord.
337 void RemovePrefix(size_t n);
338 void RemoveSuffix(size_t n);
339
340 // Cord::Subcord()
341 //
342 // Returns a new Cord representing the subrange [pos, pos + new_size) of
343 // *this. If pos >= size(), the result is empty(). If
344 // (pos + new_size) >= size(), the result is the subrange [pos, size()).
345 Cord Subcord(size_t pos, size_t new_size) const;
346
347 // Cord::swap()
348 //
349 // Swaps the contents of the Cord with `other`.
350 void swap(Cord& other) noexcept;
351
352 // swap()
353 //
354 // Swaps the contents of two Cords.
swap(Cord & x,Cord & y)355 friend void swap(Cord& x, Cord& y) noexcept { x.swap(y); }
356
357 // Cord::size()
358 //
359 // Returns the size of the Cord.
360 size_t size() const;
361
362 // Cord::empty()
363 //
364 // Determines whether the given Cord is empty, returning `true` if so.
365 bool empty() const;
366
367 // Cord::EstimatedMemoryUsage()
368 //
369 // Returns the *approximate* number of bytes held by this cord.
370 // See CordMemoryAccounting for more information on the accounting method.
371 size_t EstimatedMemoryUsage(CordMemoryAccounting accounting_method =
372 CordMemoryAccounting::kTotal) const;
373
374 // Cord::Compare()
375 //
376 // Compares 'this' Cord with rhs. This function and its relatives treat Cords
377 // as sequences of unsigned bytes. The comparison is a straightforward
378 // lexicographic comparison. `Cord::Compare()` returns values as follows:
379 //
380 // -1 'this' Cord is smaller
381 // 0 two Cords are equal
382 // 1 'this' Cord is larger
383 int Compare(absl::string_view rhs) const;
384 int Compare(const Cord& rhs) const;
385
386 // Cord::StartsWith()
387 //
388 // Determines whether the Cord starts with the passed string data `rhs`.
389 bool StartsWith(const Cord& rhs) const;
390 bool StartsWith(absl::string_view rhs) const;
391
392 // Cord::EndsWith()
393 //
394 // Determines whether the Cord ends with the passed string data `rhs`.
395 bool EndsWith(absl::string_view rhs) const;
396 bool EndsWith(const Cord& rhs) const;
397
398 // Cord::Contains()
399 //
400 // Determines whether the Cord contains the passed string data `rhs`.
401 bool Contains(absl::string_view rhs) const;
402 bool Contains(const Cord& rhs) const;
403
404 // Cord::operator std::string()
405 //
406 // Converts a Cord into a `std::string()`. This operator is marked explicit to
407 // prevent unintended Cord usage in functions that take a string.
408 explicit operator std::string() const;
409
410 // CopyCordToString()
411 //
412 // Copies the contents of a `src` Cord into a `*dst` string.
413 //
414 // This function optimizes the case of reusing the destination string since it
415 // can reuse previously allocated capacity. However, this function does not
416 // guarantee that pointers previously returned by `dst->data()` remain valid
417 // even if `*dst` had enough capacity to hold `src`. If `*dst` is a new
418 // object, prefer to simply use the conversion operator to `std::string`.
419 friend void CopyCordToString(const Cord& src, std::string* dst);
420
421 class CharIterator;
422
423 //----------------------------------------------------------------------------
424 // Cord::ChunkIterator
425 //----------------------------------------------------------------------------
426 //
427 // A `Cord::ChunkIterator` allows iteration over the constituent chunks of its
428 // Cord. Such iteration allows you to perform non-const operations on the data
429 // of a Cord without modifying it.
430 //
431 // Generally, you do not instantiate a `Cord::ChunkIterator` directly;
432 // instead, you create one implicitly through use of the `Cord::Chunks()`
433 // member function.
434 //
435 // The `Cord::ChunkIterator` has the following properties:
436 //
437 // * The iterator is invalidated after any non-const operation on the
438 // Cord object over which it iterates.
439 // * The `string_view` returned by dereferencing a valid, non-`end()`
440 // iterator is guaranteed to be non-empty.
441 // * Two `ChunkIterator` objects can be compared equal if and only if they
442 // remain valid and iterate over the same Cord.
443 // * The iterator in this case is a proxy iterator; the `string_view`
444 // returned by the iterator does not live inside the Cord, and its
445 // lifetime is limited to the lifetime of the iterator itself. To help
446 // prevent lifetime issues, `ChunkIterator::reference` is not a true
447 // reference type and is equivalent to `value_type`.
448 // * The iterator keeps state that can grow for Cords that contain many
449 // nodes and are imbalanced due to sharing. Prefer to pass this type by
450 // const reference instead of by value.
451 class ChunkIterator {
452 public:
453 using iterator_category = std::input_iterator_tag;
454 using value_type = absl::string_view;
455 using difference_type = ptrdiff_t;
456 using pointer = const value_type*;
457 using reference = value_type;
458
459 ChunkIterator() = default;
460
461 ChunkIterator& operator++();
462 ChunkIterator operator++(int);
463 bool operator==(const ChunkIterator& other) const;
464 bool operator!=(const ChunkIterator& other) const;
465 reference operator*() const;
466 pointer operator->() const;
467
468 friend class Cord;
469 friend class CharIterator;
470
471 private:
472 using CordRep = absl::cord_internal::CordRep;
473 using CordRepBtree = absl::cord_internal::CordRepBtree;
474 using CordRepBtreeReader = absl::cord_internal::CordRepBtreeReader;
475
476 // Constructs a `begin()` iterator from `tree`. `tree` must not be null.
477 explicit ChunkIterator(cord_internal::CordRep* tree);
478
479 // Constructs a `begin()` iterator from `cord`.
480 explicit ChunkIterator(const Cord* cord);
481
482 // Initializes this instance from a tree. Invoked by constructors.
483 void InitTree(cord_internal::CordRep* tree);
484
485 // Removes `n` bytes from `current_chunk_`. Expects `n` to be smaller than
486 // `current_chunk_.size()`.
487 void RemoveChunkPrefix(size_t n);
488 Cord AdvanceAndReadBytes(size_t n);
489 void AdvanceBytes(size_t n);
490
491 // Btree specific operator++
492 ChunkIterator& AdvanceBtree();
493 void AdvanceBytesBtree(size_t n);
494
495 // A view into bytes of the current `CordRep`. It may only be a view to a
496 // suffix of bytes if this is being used by `CharIterator`.
497 absl::string_view current_chunk_;
498 // The current leaf, or `nullptr` if the iterator points to short data.
499 // If the current chunk is a substring node, current_leaf_ points to the
500 // underlying flat or external node.
501 absl::cord_internal::CordRep* current_leaf_ = nullptr;
502 // The number of bytes left in the `Cord` over which we are iterating.
503 size_t bytes_remaining_ = 0;
504
505 // Cord reader for cord btrees. Empty if not traversing a btree.
506 CordRepBtreeReader btree_reader_;
507 };
508
509 // Cord::chunk_begin()
510 //
511 // Returns an iterator to the first chunk of the `Cord`.
512 //
513 // Generally, prefer using `Cord::Chunks()` within a range-based for loop for
514 // iterating over the chunks of a Cord. This method may be useful for getting
515 // a `ChunkIterator` where range-based for-loops are not useful.
516 //
517 // Example:
518 //
519 // absl::Cord::ChunkIterator FindAsChunk(const absl::Cord& c,
520 // absl::string_view s) {
521 // return std::find(c.chunk_begin(), c.chunk_end(), s);
522 // }
523 ChunkIterator chunk_begin() const ABSL_ATTRIBUTE_LIFETIME_BOUND;
524
525 // Cord::chunk_end()
526 //
527 // Returns an iterator one increment past the last chunk of the `Cord`.
528 //
529 // Generally, prefer using `Cord::Chunks()` within a range-based for loop for
530 // iterating over the chunks of a Cord. This method may be useful for getting
531 // a `ChunkIterator` where range-based for-loops may not be available.
532 ChunkIterator chunk_end() const ABSL_ATTRIBUTE_LIFETIME_BOUND;
533
534 //----------------------------------------------------------------------------
535 // Cord::ChunkRange
536 //----------------------------------------------------------------------------
537 //
538 // `ChunkRange` is a helper class for iterating over the chunks of the `Cord`,
539 // producing an iterator which can be used within a range-based for loop.
540 // Construction of a `ChunkRange` will return an iterator pointing to the
541 // first chunk of the Cord. Generally, do not construct a `ChunkRange`
542 // directly; instead, prefer to use the `Cord::Chunks()` method.
543 //
544 // Implementation note: `ChunkRange` is simply a convenience wrapper over
545 // `Cord::chunk_begin()` and `Cord::chunk_end()`.
546 class ChunkRange {
547 public:
548 // Fulfill minimum c++ container requirements [container.requirements]
549 // These (partial) container type definitions allow ChunkRange to be used
550 // in various utilities expecting a subset of [container.requirements].
551 // For example, the below enables using `::testing::ElementsAre(...)`
552 using value_type = absl::string_view;
553 using reference = value_type&;
554 using const_reference = const value_type&;
555 using iterator = ChunkIterator;
556 using const_iterator = ChunkIterator;
557
ChunkRange(const Cord * cord)558 explicit ChunkRange(const Cord* cord) : cord_(cord) {}
559
560 ChunkIterator begin() const;
561 ChunkIterator end() const;
562
563 private:
564 const Cord* cord_;
565 };
566
567 // Cord::Chunks()
568 //
569 // Returns a `Cord::ChunkRange` for iterating over the chunks of a `Cord` with
570 // a range-based for-loop. For most iteration tasks on a Cord, use
571 // `Cord::Chunks()` to retrieve this iterator.
572 //
573 // Example:
574 //
575 // void ProcessChunks(const Cord& cord) {
576 // for (absl::string_view chunk : cord.Chunks()) { ... }
577 // }
578 //
579 // Note that the ordinary caveats of temporary lifetime extension apply:
580 //
581 // void Process() {
582 // for (absl::string_view chunk : CordFactory().Chunks()) {
583 // // The temporary Cord returned by CordFactory has been destroyed!
584 // }
585 // }
586 ChunkRange Chunks() const ABSL_ATTRIBUTE_LIFETIME_BOUND;
587
588 //----------------------------------------------------------------------------
589 // Cord::CharIterator
590 //----------------------------------------------------------------------------
591 //
592 // A `Cord::CharIterator` allows iteration over the constituent characters of
593 // a `Cord`.
594 //
595 // Generally, you do not instantiate a `Cord::CharIterator` directly; instead,
596 // you create one implicitly through use of the `Cord::Chars()` member
597 // function.
598 //
599 // A `Cord::CharIterator` has the following properties:
600 //
601 // * The iterator is invalidated after any non-const operation on the
602 // Cord object over which it iterates.
603 // * Two `CharIterator` objects can be compared equal if and only if they
604 // remain valid and iterate over the same Cord.
605 // * The iterator keeps state that can grow for Cords that contain many
606 // nodes and are imbalanced due to sharing. Prefer to pass this type by
607 // const reference instead of by value.
608 // * This type cannot act as a forward iterator because a `Cord` can reuse
609 // sections of memory. This fact violates the requirement for forward
610 // iterators to compare equal if dereferencing them returns the same
611 // object.
612 class CharIterator {
613 public:
614 using iterator_category = std::input_iterator_tag;
615 using value_type = char;
616 using difference_type = ptrdiff_t;
617 using pointer = const char*;
618 using reference = const char&;
619
620 CharIterator() = default;
621
622 CharIterator& operator++();
623 CharIterator operator++(int);
624 bool operator==(const CharIterator& other) const;
625 bool operator!=(const CharIterator& other) const;
626 reference operator*() const;
627 pointer operator->() const;
628
629 friend Cord;
630
631 private:
CharIterator(const Cord * cord)632 explicit CharIterator(const Cord* cord) : chunk_iterator_(cord) {}
633
634 ChunkIterator chunk_iterator_;
635 };
636
637 // Cord::AdvanceAndRead()
638 //
639 // Advances the `Cord::CharIterator` by `n_bytes` and returns the bytes
640 // advanced as a separate `Cord`. `n_bytes` must be less than or equal to the
641 // number of bytes within the Cord; otherwise, behavior is undefined. It is
642 // valid to pass `char_end()` and `0`.
643 static Cord AdvanceAndRead(CharIterator* it, size_t n_bytes);
644
645 // Cord::Advance()
646 //
647 // Advances the `Cord::CharIterator` by `n_bytes`. `n_bytes` must be less than
648 // or equal to the number of bytes remaining within the Cord; otherwise,
649 // behavior is undefined. It is valid to pass `char_end()` and `0`.
650 static void Advance(CharIterator* it, size_t n_bytes);
651
652 // Cord::ChunkRemaining()
653 //
654 // Returns the longest contiguous view starting at the iterator's position.
655 //
656 // `it` must be dereferenceable.
657 static absl::string_view ChunkRemaining(const CharIterator& it);
658
659 // Cord::char_begin()
660 //
661 // Returns an iterator to the first character of the `Cord`.
662 //
663 // Generally, prefer using `Cord::Chars()` within a range-based for loop for
664 // iterating over the chunks of a Cord. This method may be useful for getting
665 // a `CharIterator` where range-based for-loops may not be available.
666 CharIterator char_begin() const ABSL_ATTRIBUTE_LIFETIME_BOUND;
667
668 // Cord::char_end()
669 //
670 // Returns an iterator to one past the last character of the `Cord`.
671 //
672 // Generally, prefer using `Cord::Chars()` within a range-based for loop for
673 // iterating over the chunks of a Cord. This method may be useful for getting
674 // a `CharIterator` where range-based for-loops are not useful.
675 CharIterator char_end() const ABSL_ATTRIBUTE_LIFETIME_BOUND;
676
677 // Cord::CharRange
678 //
679 // `CharRange` is a helper class for iterating over the characters of a
680 // producing an iterator which can be used within a range-based for loop.
681 // Construction of a `CharRange` will return an iterator pointing to the first
682 // character of the Cord. Generally, do not construct a `CharRange` directly;
683 // instead, prefer to use the `Cord::Chars()` method shown below.
684 //
685 // Implementation note: `CharRange` is simply a convenience wrapper over
686 // `Cord::char_begin()` and `Cord::char_end()`.
687 class CharRange {
688 public:
689 // Fulfill minimum c++ container requirements [container.requirements]
690 // These (partial) container type definitions allow CharRange to be used
691 // in various utilities expecting a subset of [container.requirements].
692 // For example, the below enables using `::testing::ElementsAre(...)`
693 using value_type = char;
694 using reference = value_type&;
695 using const_reference = const value_type&;
696 using iterator = CharIterator;
697 using const_iterator = CharIterator;
698
CharRange(const Cord * cord)699 explicit CharRange(const Cord* cord) : cord_(cord) {}
700
701 CharIterator begin() const;
702 CharIterator end() const;
703
704 private:
705 const Cord* cord_;
706 };
707
708 // Cord::Chars()
709 //
710 // Returns a `Cord::CharRange` for iterating over the characters of a `Cord`
711 // with a range-based for-loop. For most character-based iteration tasks on a
712 // Cord, use `Cord::Chars()` to retrieve this iterator.
713 //
714 // Example:
715 //
716 // void ProcessCord(const Cord& cord) {
717 // for (char c : cord.Chars()) { ... }
718 // }
719 //
720 // Note that the ordinary caveats of temporary lifetime extension apply:
721 //
722 // void Process() {
723 // for (char c : CordFactory().Chars()) {
724 // // The temporary Cord returned by CordFactory has been destroyed!
725 // }
726 // }
727 CharRange Chars() const ABSL_ATTRIBUTE_LIFETIME_BOUND;
728
729 // Cord::operator[]
730 //
731 // Gets the "i"th character of the Cord and returns it, provided that
732 // 0 <= i < Cord.size().
733 //
734 // NOTE: This routine is reasonably efficient. It is roughly
735 // logarithmic based on the number of chunks that make up the cord. Still,
736 // if you need to iterate over the contents of a cord, you should
737 // use a CharIterator/ChunkIterator rather than call operator[] or Get()
738 // repeatedly in a loop.
739 char operator[](size_t i) const;
740
741 // Cord::TryFlat()
742 //
743 // If this cord's representation is a single flat array, returns a
744 // string_view referencing that array. Otherwise returns nullopt.
745 absl::optional<absl::string_view> TryFlat() const
746 ABSL_ATTRIBUTE_LIFETIME_BOUND;
747
748 // Cord::Flatten()
749 //
750 // Flattens the cord into a single array and returns a view of the data.
751 //
752 // If the cord was already flat, the contents are not modified.
753 absl::string_view Flatten() ABSL_ATTRIBUTE_LIFETIME_BOUND;
754
755 // Cord::Find()
756 //
757 // Returns an iterator to the first occurrance of the substring `needle`.
758 //
759 // If the substring `needle` does not occur, `Cord::char_end()` is returned.
760 CharIterator Find(absl::string_view needle) const;
761 CharIterator Find(const absl::Cord& needle) const;
762
763 // Supports absl::Cord as a sink object for absl::Format().
AbslFormatFlush(absl::Cord * cord,absl::string_view part)764 friend void AbslFormatFlush(absl::Cord* cord, absl::string_view part) {
765 cord->Append(part);
766 }
767
768 // Support automatic stringification with absl::StrCat and absl::StrFormat.
769 template <typename Sink>
AbslStringify(Sink & sink,const absl::Cord & cord)770 friend void AbslStringify(Sink& sink, const absl::Cord& cord) {
771 for (absl::string_view chunk : cord.Chunks()) {
772 sink.Append(chunk);
773 }
774 }
775
776 // Cord::SetExpectedChecksum()
777 //
778 // Stores a checksum value with this non-empty cord instance, for later
779 // retrieval.
780 //
781 // The expected checksum is a number stored out-of-band, alongside the data.
782 // It is preserved across copies and assignments, but any mutations to a cord
783 // will cause it to lose its expected checksum.
784 //
785 // The expected checksum is not part of a Cord's value, and does not affect
786 // operations such as equality or hashing.
787 //
788 // This field is intended to store a CRC32C checksum for later validation, to
789 // help support end-to-end checksum workflows. However, the Cord API itself
790 // does no CRC validation, and assigns no meaning to this number.
791 //
792 // This call has no effect if this cord is empty.
793 void SetExpectedChecksum(uint32_t crc);
794
795 // Returns this cord's expected checksum, if it has one. Otherwise, returns
796 // nullopt.
797 absl::optional<uint32_t> ExpectedChecksum() const;
798
799 template <typename H>
AbslHashValue(H hash_state,const absl::Cord & c)800 friend H AbslHashValue(H hash_state, const absl::Cord& c) {
801 absl::optional<absl::string_view> maybe_flat = c.TryFlat();
802 if (maybe_flat.has_value()) {
803 return H::combine(std::move(hash_state), *maybe_flat);
804 }
805 return c.HashFragmented(std::move(hash_state));
806 }
807
808 // Create a Cord with the contents of StringConstant<T>::value.
809 // No allocations will be done and no data will be copied.
810 // This is an INTERNAL API and subject to change or removal. This API can only
811 // be used by spelling absl::strings_internal::MakeStringConstant, which is
812 // also an internal API.
813 template <typename T>
814 // NOLINTNEXTLINE(google-explicit-constructor)
815 constexpr Cord(strings_internal::StringConstant<T>);
816
817 private:
818 using CordRep = absl::cord_internal::CordRep;
819 using CordRepFlat = absl::cord_internal::CordRepFlat;
820 using CordzInfo = cord_internal::CordzInfo;
821 using CordzUpdateScope = cord_internal::CordzUpdateScope;
822 using CordzUpdateTracker = cord_internal::CordzUpdateTracker;
823 using InlineData = cord_internal::InlineData;
824 using MethodIdentifier = CordzUpdateTracker::MethodIdentifier;
825
826 // Creates a cord instance with `method` representing the originating
827 // public API call causing the cord to be created.
828 explicit Cord(absl::string_view src, MethodIdentifier method);
829
830 friend class CordTestPeer;
831 friend bool operator==(const Cord& lhs, const Cord& rhs);
832 friend bool operator==(const Cord& lhs, absl::string_view rhs);
833
834 friend const CordzInfo* GetCordzInfoForTesting(const Cord& cord);
835
836 // Calls the provided function once for each cord chunk, in order. Unlike
837 // Chunks(), this API will not allocate memory.
838 void ForEachChunk(absl::FunctionRef<void(absl::string_view)>) const;
839
840 // Allocates new contiguous storage for the contents of the cord. This is
841 // called by Flatten() when the cord was not already flat.
842 absl::string_view FlattenSlowPath();
843
844 // Actual cord contents are hidden inside the following simple
845 // class so that we can isolate the bulk of cord.cc from changes
846 // to the representation.
847 //
848 // InlineRep holds either a tree pointer, or an array of kMaxInline bytes.
849 class InlineRep {
850 public:
851 static constexpr unsigned char kMaxInline = cord_internal::kMaxInline;
852 static_assert(kMaxInline >= sizeof(absl::cord_internal::CordRep*), "");
853
InlineRep()854 constexpr InlineRep() : data_() {}
InlineRep(InlineData::DefaultInitType init)855 explicit InlineRep(InlineData::DefaultInitType init) : data_(init) {}
856 InlineRep(const InlineRep& src);
857 InlineRep(InlineRep&& src);
858 InlineRep& operator=(const InlineRep& src);
859 InlineRep& operator=(InlineRep&& src) noexcept;
860
861 explicit constexpr InlineRep(absl::string_view sv, CordRep* rep);
862
863 void Swap(InlineRep* rhs);
864 size_t size() const;
865 const char* data() const; // Returns nullptr if holding pointer
866 void set_data(const char* data, size_t n); // Discards pointer, if any
867 char* set_data(size_t n); // Write data to the result
868 // Returns nullptr if holding bytes
869 absl::cord_internal::CordRep* tree() const;
870 absl::cord_internal::CordRep* as_tree() const;
871 const char* as_chars() const;
872 // Returns non-null iff was holding a pointer
873 absl::cord_internal::CordRep* clear();
874 // Converts to pointer if necessary.
875 void reduce_size(size_t n); // REQUIRES: holding data
876 void remove_prefix(size_t n); // REQUIRES: holding data
877 void AppendArray(absl::string_view src, MethodIdentifier method);
878 absl::string_view FindFlatStartPiece() const;
879
880 // Creates a CordRepFlat instance from the current inlined data with `extra'
881 // bytes of desired additional capacity.
882 CordRepFlat* MakeFlatWithExtraCapacity(size_t extra);
883
884 // Sets the tree value for this instance. `rep` must not be null.
885 // Requires the current instance to hold a tree, and a lock to be held on
886 // any CordzInfo referenced by this instance. The latter is enforced through
887 // the CordzUpdateScope argument. If the current instance is sampled, then
888 // the CordzInfo instance is updated to reference the new `rep` value.
889 void SetTree(CordRep* rep, const CordzUpdateScope& scope);
890
891 // Identical to SetTree(), except that `rep` is allowed to be null, in
892 // which case the current instance is reset to an empty value.
893 void SetTreeOrEmpty(CordRep* rep, const CordzUpdateScope& scope);
894
895 // Sets the tree value for this instance, and randomly samples this cord.
896 // This function disregards existing contents in `data_`, and should be
897 // called when a Cord is 'promoted' from an 'uninitialized' or 'inlined'
898 // value to a non-inlined (tree / ring) value.
899 void EmplaceTree(CordRep* rep, MethodIdentifier method);
900
901 // Identical to EmplaceTree, except that it copies the parent stack from
902 // the provided `parent` data if the parent is sampled.
903 void EmplaceTree(CordRep* rep, const InlineData& parent,
904 MethodIdentifier method);
905
906 // Commits the change of a newly created, or updated `rep` root value into
907 // this cord. `old_rep` indicates the old (inlined or tree) value of the
908 // cord, and determines if the commit invokes SetTree() or EmplaceTree().
909 void CommitTree(const CordRep* old_rep, CordRep* rep,
910 const CordzUpdateScope& scope, MethodIdentifier method);
911
912 void AppendTreeToInlined(CordRep* tree, MethodIdentifier method);
913 void AppendTreeToTree(CordRep* tree, MethodIdentifier method);
914 void AppendTree(CordRep* tree, MethodIdentifier method);
915 void PrependTreeToInlined(CordRep* tree, MethodIdentifier method);
916 void PrependTreeToTree(CordRep* tree, MethodIdentifier method);
917 void PrependTree(CordRep* tree, MethodIdentifier method);
918
IsSame(const InlineRep & other)919 bool IsSame(const InlineRep& other) const { return data_ == other.data_; }
920
CopyTo(std::string * dst)921 void CopyTo(std::string* dst) const {
922 // memcpy is much faster when operating on a known size. On most supported
923 // platforms, the small string optimization is large enough that resizing
924 // to 15 bytes does not cause a memory allocation.
925 absl::strings_internal::STLStringResizeUninitialized(dst, kMaxInline);
926 data_.copy_max_inline_to(&(*dst)[0]);
927 // erase is faster than resize because the logic for memory allocation is
928 // not needed.
929 dst->erase(inline_size());
930 }
931
932 // Copies the inline contents into `dst`. Assumes the cord is not empty.
933 void CopyToArray(char* dst) const;
934
is_tree()935 bool is_tree() const { return data_.is_tree(); }
936
937 // Returns true if the Cord is being profiled by cordz.
is_profiled()938 bool is_profiled() const { return data_.is_tree() && data_.is_profiled(); }
939
940 // Returns the available inlined capacity, or 0 if is_tree() == true.
remaining_inline_capacity()941 size_t remaining_inline_capacity() const {
942 return data_.is_tree() ? 0 : kMaxInline - data_.inline_size();
943 }
944
945 // Returns the profiled CordzInfo, or nullptr if not sampled.
cordz_info()946 absl::cord_internal::CordzInfo* cordz_info() const {
947 return data_.cordz_info();
948 }
949
950 // Sets the profiled CordzInfo. `cordz_info` must not be null.
set_cordz_info(cord_internal::CordzInfo * cordz_info)951 void set_cordz_info(cord_internal::CordzInfo* cordz_info) {
952 assert(cordz_info != nullptr);
953 data_.set_cordz_info(cordz_info);
954 }
955
956 // Resets the current cordz_info to null / empty.
clear_cordz_info()957 void clear_cordz_info() { data_.clear_cordz_info(); }
958
959 private:
960 friend class Cord;
961
962 void AssignSlow(const InlineRep& src);
963 // Unrefs the tree and stops profiling.
964 void UnrefTree();
965
ResetToEmpty()966 void ResetToEmpty() { data_ = {}; }
967
set_inline_size(size_t size)968 void set_inline_size(size_t size) { data_.set_inline_size(size); }
inline_size()969 size_t inline_size() const { return data_.inline_size(); }
970
971 // Empty cords that carry a checksum have a CordRepCrc node with a null
972 // child node. The code can avoid lots of special cases where it would
973 // otherwise transition from tree to inline storage if we just remove the
974 // CordRepCrc node before mutations. Must never be called inside a
975 // CordzUpdateScope since it untracks the cordz info.
976 void MaybeRemoveEmptyCrcNode();
977
978 cord_internal::InlineData data_;
979 };
980 InlineRep contents_;
981
982 // Helper for GetFlat() and TryFlat().
983 static bool GetFlatAux(absl::cord_internal::CordRep* rep,
984 absl::string_view* fragment);
985
986 // Helper for ForEachChunk().
987 static void ForEachChunkAux(
988 absl::cord_internal::CordRep* rep,
989 absl::FunctionRef<void(absl::string_view)> callback);
990
991 // The destructor for non-empty Cords.
992 void DestroyCordSlow();
993
994 // Out-of-line implementation of slower parts of logic.
995 void CopyToArraySlowPath(char* dst) const;
996 int CompareSlowPath(absl::string_view rhs, size_t compared_size,
997 size_t size_to_compare) const;
998 int CompareSlowPath(const Cord& rhs, size_t compared_size,
999 size_t size_to_compare) const;
1000 bool EqualsImpl(absl::string_view rhs, size_t size_to_compare) const;
1001 bool EqualsImpl(const Cord& rhs, size_t size_to_compare) const;
1002 int CompareImpl(const Cord& rhs) const;
1003
1004 template <typename ResultType, typename RHS>
1005 friend ResultType GenericCompare(const Cord& lhs, const RHS& rhs,
1006 size_t size_to_compare);
1007 static absl::string_view GetFirstChunk(const Cord& c);
1008 static absl::string_view GetFirstChunk(absl::string_view sv);
1009
1010 // Returns a new reference to contents_.tree(), or steals an existing
1011 // reference if called on an rvalue.
1012 absl::cord_internal::CordRep* TakeRep() const&;
1013 absl::cord_internal::CordRep* TakeRep() &&;
1014
1015 // Helper for Append().
1016 template <typename C>
1017 void AppendImpl(C&& src);
1018
1019 // Appends / Prepends `src` to this instance, using precise sizing.
1020 // This method does explicitly not attempt to use any spare capacity
1021 // in any pending last added private owned flat.
1022 // Requires `src` to be <= kMaxFlatLength.
1023 void AppendPrecise(absl::string_view src, MethodIdentifier method);
1024 void PrependPrecise(absl::string_view src, MethodIdentifier method);
1025
1026 CordBuffer GetAppendBufferSlowPath(size_t block_size, size_t capacity,
1027 size_t min_capacity);
1028
1029 // Prepends the provided data to this instance. `method` contains the public
1030 // API method for this action which is tracked for Cordz sampling purposes.
1031 void PrependArray(absl::string_view src, MethodIdentifier method);
1032
1033 // Assigns the value in 'src' to this instance, 'stealing' its contents.
1034 // Requires src.length() > kMaxBytesToCopy.
1035 Cord& AssignLargeString(std::string&& src);
1036
1037 // Helper for AbslHashValue().
1038 template <typename H>
HashFragmented(H hash_state)1039 H HashFragmented(H hash_state) const {
1040 typename H::AbslInternalPiecewiseCombiner combiner;
1041 ForEachChunk([&combiner, &hash_state](absl::string_view chunk) {
1042 hash_state = combiner.add_buffer(std::move(hash_state), chunk.data(),
1043 chunk.size());
1044 });
1045 return H::combine(combiner.finalize(std::move(hash_state)), size());
1046 }
1047
1048 friend class CrcCord;
1049 void SetCrcCordState(crc_internal::CrcCordState state);
1050 const crc_internal::CrcCordState* MaybeGetCrcCordState() const;
1051
1052 CharIterator FindImpl(CharIterator it, absl::string_view needle) const;
1053 };
1054
1055 ABSL_NAMESPACE_END
1056 } // namespace absl
1057
1058 namespace absl {
1059 ABSL_NAMESPACE_BEGIN
1060
1061 // allow a Cord to be logged
1062 extern std::ostream& operator<<(std::ostream& out, const Cord& cord);
1063
1064 // ------------------------------------------------------------------
1065 // Internal details follow. Clients should ignore.
1066
1067 namespace cord_internal {
1068
1069 // Does non-template-specific `CordRepExternal` initialization.
1070 // Requires `data` to be non-empty.
1071 void InitializeCordRepExternal(absl::string_view data, CordRepExternal* rep);
1072
1073 // Creates a new `CordRep` that owns `data` and `releaser` and returns a pointer
1074 // to it. Requires `data` to be non-empty.
1075 template <typename Releaser>
1076 // NOLINTNEXTLINE - suppress clang-tidy raw pointer return.
NewExternalRep(absl::string_view data,Releaser && releaser)1077 CordRep* NewExternalRep(absl::string_view data, Releaser&& releaser) {
1078 assert(!data.empty());
1079 using ReleaserType = absl::decay_t<Releaser>;
1080 CordRepExternal* rep = new CordRepExternalImpl<ReleaserType>(
1081 std::forward<Releaser>(releaser), 0);
1082 InitializeCordRepExternal(data, rep);
1083 return rep;
1084 }
1085
1086 // Overload for function reference types that dispatches using a function
1087 // pointer because there are no `alignof()` or `sizeof()` a function reference.
1088 // NOLINTNEXTLINE - suppress clang-tidy raw pointer return.
NewExternalRep(absl::string_view data,void (& releaser)(absl::string_view))1089 inline CordRep* NewExternalRep(absl::string_view data,
1090 void (&releaser)(absl::string_view)) {
1091 return NewExternalRep(data, &releaser);
1092 }
1093
1094 } // namespace cord_internal
1095
1096 template <typename Releaser>
MakeCordFromExternal(absl::string_view data,Releaser && releaser)1097 Cord MakeCordFromExternal(absl::string_view data, Releaser&& releaser) {
1098 Cord cord;
1099 if (ABSL_PREDICT_TRUE(!data.empty())) {
1100 cord.contents_.EmplaceTree(::absl::cord_internal::NewExternalRep(
1101 data, std::forward<Releaser>(releaser)),
1102 Cord::MethodIdentifier::kMakeCordFromExternal);
1103 } else {
1104 using ReleaserType = absl::decay_t<Releaser>;
1105 cord_internal::InvokeReleaser(
1106 cord_internal::Rank0{}, ReleaserType(std::forward<Releaser>(releaser)),
1107 data);
1108 }
1109 return cord;
1110 }
1111
InlineRep(absl::string_view sv,CordRep * rep)1112 constexpr Cord::InlineRep::InlineRep(absl::string_view sv, CordRep* rep)
1113 : data_(sv, rep) {}
1114
InlineRep(const Cord::InlineRep & src)1115 inline Cord::InlineRep::InlineRep(const Cord::InlineRep& src)
1116 : data_(InlineData::kDefaultInit) {
1117 if (CordRep* tree = src.tree()) {
1118 EmplaceTree(CordRep::Ref(tree), src.data_,
1119 CordzUpdateTracker::kConstructorCord);
1120 } else {
1121 data_ = src.data_;
1122 }
1123 }
1124
InlineRep(Cord::InlineRep && src)1125 inline Cord::InlineRep::InlineRep(Cord::InlineRep&& src) : data_(src.data_) {
1126 src.ResetToEmpty();
1127 }
1128
1129 inline Cord::InlineRep& Cord::InlineRep::operator=(const Cord::InlineRep& src) {
1130 if (this == &src) {
1131 return *this;
1132 }
1133 if (!is_tree() && !src.is_tree()) {
1134 data_ = src.data_;
1135 return *this;
1136 }
1137 AssignSlow(src);
1138 return *this;
1139 }
1140
1141 inline Cord::InlineRep& Cord::InlineRep::operator=(
1142 Cord::InlineRep&& src) noexcept {
1143 if (is_tree()) {
1144 UnrefTree();
1145 }
1146 data_ = src.data_;
1147 src.ResetToEmpty();
1148 return *this;
1149 }
1150
Swap(Cord::InlineRep * rhs)1151 inline void Cord::InlineRep::Swap(Cord::InlineRep* rhs) {
1152 if (rhs == this) {
1153 return;
1154 }
1155 std::swap(data_, rhs->data_);
1156 }
1157
data()1158 inline const char* Cord::InlineRep::data() const {
1159 return is_tree() ? nullptr : data_.as_chars();
1160 }
1161
as_chars()1162 inline const char* Cord::InlineRep::as_chars() const {
1163 assert(!data_.is_tree());
1164 return data_.as_chars();
1165 }
1166
as_tree()1167 inline absl::cord_internal::CordRep* Cord::InlineRep::as_tree() const {
1168 assert(data_.is_tree());
1169 return data_.as_tree();
1170 }
1171
tree()1172 inline absl::cord_internal::CordRep* Cord::InlineRep::tree() const {
1173 if (is_tree()) {
1174 return as_tree();
1175 } else {
1176 return nullptr;
1177 }
1178 }
1179
size()1180 inline size_t Cord::InlineRep::size() const {
1181 return is_tree() ? as_tree()->length : inline_size();
1182 }
1183
MakeFlatWithExtraCapacity(size_t extra)1184 inline cord_internal::CordRepFlat* Cord::InlineRep::MakeFlatWithExtraCapacity(
1185 size_t extra) {
1186 static_assert(cord_internal::kMinFlatLength >= sizeof(data_), "");
1187 size_t len = data_.inline_size();
1188 auto* result = CordRepFlat::New(len + extra);
1189 result->length = len;
1190 data_.copy_max_inline_to(result->Data());
1191 return result;
1192 }
1193
EmplaceTree(CordRep * rep,MethodIdentifier method)1194 inline void Cord::InlineRep::EmplaceTree(CordRep* rep,
1195 MethodIdentifier method) {
1196 assert(rep);
1197 data_.make_tree(rep);
1198 CordzInfo::MaybeTrackCord(data_, method);
1199 }
1200
EmplaceTree(CordRep * rep,const InlineData & parent,MethodIdentifier method)1201 inline void Cord::InlineRep::EmplaceTree(CordRep* rep, const InlineData& parent,
1202 MethodIdentifier method) {
1203 data_.make_tree(rep);
1204 CordzInfo::MaybeTrackCord(data_, parent, method);
1205 }
1206
SetTree(CordRep * rep,const CordzUpdateScope & scope)1207 inline void Cord::InlineRep::SetTree(CordRep* rep,
1208 const CordzUpdateScope& scope) {
1209 assert(rep);
1210 assert(data_.is_tree());
1211 data_.set_tree(rep);
1212 scope.SetCordRep(rep);
1213 }
1214
SetTreeOrEmpty(CordRep * rep,const CordzUpdateScope & scope)1215 inline void Cord::InlineRep::SetTreeOrEmpty(CordRep* rep,
1216 const CordzUpdateScope& scope) {
1217 assert(data_.is_tree());
1218 if (rep) {
1219 data_.set_tree(rep);
1220 } else {
1221 data_ = {};
1222 }
1223 scope.SetCordRep(rep);
1224 }
1225
CommitTree(const CordRep * old_rep,CordRep * rep,const CordzUpdateScope & scope,MethodIdentifier method)1226 inline void Cord::InlineRep::CommitTree(const CordRep* old_rep, CordRep* rep,
1227 const CordzUpdateScope& scope,
1228 MethodIdentifier method) {
1229 if (old_rep) {
1230 SetTree(rep, scope);
1231 } else {
1232 EmplaceTree(rep, method);
1233 }
1234 }
1235
clear()1236 inline absl::cord_internal::CordRep* Cord::InlineRep::clear() {
1237 if (is_tree()) {
1238 CordzInfo::MaybeUntrackCord(cordz_info());
1239 }
1240 absl::cord_internal::CordRep* result = tree();
1241 ResetToEmpty();
1242 return result;
1243 }
1244
CopyToArray(char * dst)1245 inline void Cord::InlineRep::CopyToArray(char* dst) const {
1246 assert(!is_tree());
1247 size_t n = inline_size();
1248 assert(n != 0);
1249 cord_internal::SmallMemmove(dst, data_.as_chars(), n);
1250 }
1251
MaybeRemoveEmptyCrcNode()1252 inline void Cord::InlineRep::MaybeRemoveEmptyCrcNode() {
1253 CordRep* rep = tree();
1254 if (rep == nullptr || ABSL_PREDICT_TRUE(rep->length > 0)) {
1255 return;
1256 }
1257 assert(rep->IsCrc());
1258 assert(rep->crc()->child == nullptr);
1259 CordzInfo::MaybeUntrackCord(cordz_info());
1260 CordRep::Unref(rep);
1261 ResetToEmpty();
1262 }
1263
Cord()1264 constexpr inline Cord::Cord() noexcept {}
1265
Cord(absl::string_view src)1266 inline Cord::Cord(absl::string_view src)
1267 : Cord(src, CordzUpdateTracker::kConstructorString) {}
1268
1269 template <typename T>
Cord(strings_internal::StringConstant<T>)1270 constexpr Cord::Cord(strings_internal::StringConstant<T>)
1271 : contents_(strings_internal::StringConstant<T>::value,
1272 strings_internal::StringConstant<T>::value.size() <=
1273 cord_internal::kMaxInline
1274 ? nullptr
1275 : &cord_internal::ConstInitExternalStorage<
1276 strings_internal::StringConstant<T>>::value) {}
1277
1278 inline Cord& Cord::operator=(const Cord& x) {
1279 contents_ = x.contents_;
1280 return *this;
1281 }
1282
1283 template <typename T, Cord::EnableIfString<T>>
1284 Cord& Cord::operator=(T&& src) {
1285 if (src.size() <= cord_internal::kMaxBytesToCopy) {
1286 return operator=(absl::string_view(src));
1287 } else {
1288 return AssignLargeString(std::forward<T>(src));
1289 }
1290 }
1291
Cord(const Cord & src)1292 inline Cord::Cord(const Cord& src) : contents_(src.contents_) {}
1293
Cord(Cord && src)1294 inline Cord::Cord(Cord&& src) noexcept : contents_(std::move(src.contents_)) {}
1295
swap(Cord & other)1296 inline void Cord::swap(Cord& other) noexcept {
1297 contents_.Swap(&other.contents_);
1298 }
1299
1300 inline Cord& Cord::operator=(Cord&& x) noexcept {
1301 contents_ = std::move(x.contents_);
1302 return *this;
1303 }
1304
1305 extern template Cord::Cord(std::string&& src);
1306
size()1307 inline size_t Cord::size() const {
1308 // Length is 1st field in str.rep_
1309 return contents_.size();
1310 }
1311
empty()1312 inline bool Cord::empty() const { return size() == 0; }
1313
EstimatedMemoryUsage(CordMemoryAccounting accounting_method)1314 inline size_t Cord::EstimatedMemoryUsage(
1315 CordMemoryAccounting accounting_method) const {
1316 size_t result = sizeof(Cord);
1317 if (const absl::cord_internal::CordRep* rep = contents_.tree()) {
1318 switch (accounting_method) {
1319 case CordMemoryAccounting::kFairShare:
1320 result += cord_internal::GetEstimatedFairShareMemoryUsage(rep);
1321 break;
1322 case CordMemoryAccounting::kTotalMorePrecise:
1323 result += cord_internal::GetMorePreciseMemoryUsage(rep);
1324 break;
1325 case CordMemoryAccounting::kTotal:
1326 result += cord_internal::GetEstimatedMemoryUsage(rep);
1327 break;
1328 }
1329 }
1330 return result;
1331 }
1332
TryFlat()1333 inline absl::optional<absl::string_view> Cord::TryFlat() const {
1334 absl::cord_internal::CordRep* rep = contents_.tree();
1335 if (rep == nullptr) {
1336 return absl::string_view(contents_.data(), contents_.size());
1337 }
1338 absl::string_view fragment;
1339 if (GetFlatAux(rep, &fragment)) {
1340 return fragment;
1341 }
1342 return absl::nullopt;
1343 }
1344
Flatten()1345 inline absl::string_view Cord::Flatten() {
1346 absl::cord_internal::CordRep* rep = contents_.tree();
1347 if (rep == nullptr) {
1348 return absl::string_view(contents_.data(), contents_.size());
1349 } else {
1350 absl::string_view already_flat_contents;
1351 if (GetFlatAux(rep, &already_flat_contents)) {
1352 return already_flat_contents;
1353 }
1354 }
1355 return FlattenSlowPath();
1356 }
1357
Append(absl::string_view src)1358 inline void Cord::Append(absl::string_view src) {
1359 contents_.AppendArray(src, CordzUpdateTracker::kAppendString);
1360 }
1361
Prepend(absl::string_view src)1362 inline void Cord::Prepend(absl::string_view src) {
1363 PrependArray(src, CordzUpdateTracker::kPrependString);
1364 }
1365
Append(CordBuffer buffer)1366 inline void Cord::Append(CordBuffer buffer) {
1367 if (ABSL_PREDICT_FALSE(buffer.length() == 0)) return;
1368 absl::string_view short_value;
1369 if (CordRep* rep = buffer.ConsumeValue(short_value)) {
1370 contents_.AppendTree(rep, CordzUpdateTracker::kAppendCordBuffer);
1371 } else {
1372 AppendPrecise(short_value, CordzUpdateTracker::kAppendCordBuffer);
1373 }
1374 }
1375
Prepend(CordBuffer buffer)1376 inline void Cord::Prepend(CordBuffer buffer) {
1377 if (ABSL_PREDICT_FALSE(buffer.length() == 0)) return;
1378 absl::string_view short_value;
1379 if (CordRep* rep = buffer.ConsumeValue(short_value)) {
1380 contents_.PrependTree(rep, CordzUpdateTracker::kPrependCordBuffer);
1381 } else {
1382 PrependPrecise(short_value, CordzUpdateTracker::kPrependCordBuffer);
1383 }
1384 }
1385
GetAppendBuffer(size_t capacity,size_t min_capacity)1386 inline CordBuffer Cord::GetAppendBuffer(size_t capacity, size_t min_capacity) {
1387 if (empty()) return CordBuffer::CreateWithDefaultLimit(capacity);
1388 return GetAppendBufferSlowPath(0, capacity, min_capacity);
1389 }
1390
GetCustomAppendBuffer(size_t block_size,size_t capacity,size_t min_capacity)1391 inline CordBuffer Cord::GetCustomAppendBuffer(size_t block_size,
1392 size_t capacity,
1393 size_t min_capacity) {
1394 if (empty()) {
1395 return block_size ? CordBuffer::CreateWithCustomLimit(block_size, capacity)
1396 : CordBuffer::CreateWithDefaultLimit(capacity);
1397 }
1398 return GetAppendBufferSlowPath(block_size, capacity, min_capacity);
1399 }
1400
1401 extern template void Cord::Append(std::string&& src);
1402 extern template void Cord::Prepend(std::string&& src);
1403
Compare(const Cord & rhs)1404 inline int Cord::Compare(const Cord& rhs) const {
1405 if (!contents_.is_tree() && !rhs.contents_.is_tree()) {
1406 return contents_.data_.Compare(rhs.contents_.data_);
1407 }
1408
1409 return CompareImpl(rhs);
1410 }
1411
1412 // Does 'this' cord start/end with rhs
StartsWith(const Cord & rhs)1413 inline bool Cord::StartsWith(const Cord& rhs) const {
1414 if (contents_.IsSame(rhs.contents_)) return true;
1415 size_t rhs_size = rhs.size();
1416 if (size() < rhs_size) return false;
1417 return EqualsImpl(rhs, rhs_size);
1418 }
1419
StartsWith(absl::string_view rhs)1420 inline bool Cord::StartsWith(absl::string_view rhs) const {
1421 size_t rhs_size = rhs.size();
1422 if (size() < rhs_size) return false;
1423 return EqualsImpl(rhs, rhs_size);
1424 }
1425
InitTree(cord_internal::CordRep * tree)1426 inline void Cord::ChunkIterator::InitTree(cord_internal::CordRep* tree) {
1427 tree = cord_internal::SkipCrcNode(tree);
1428 if (tree->tag == cord_internal::BTREE) {
1429 current_chunk_ = btree_reader_.Init(tree->btree());
1430 } else {
1431 current_leaf_ = tree;
1432 current_chunk_ = cord_internal::EdgeData(tree);
1433 }
1434 }
1435
ChunkIterator(cord_internal::CordRep * tree)1436 inline Cord::ChunkIterator::ChunkIterator(cord_internal::CordRep* tree) {
1437 bytes_remaining_ = tree->length;
1438 InitTree(tree);
1439 }
1440
ChunkIterator(const Cord * cord)1441 inline Cord::ChunkIterator::ChunkIterator(const Cord* cord) {
1442 if (CordRep* tree = cord->contents_.tree()) {
1443 bytes_remaining_ = tree->length;
1444 if (ABSL_PREDICT_TRUE(bytes_remaining_ != 0)) {
1445 InitTree(tree);
1446 } else {
1447 current_chunk_ = {};
1448 }
1449 } else {
1450 bytes_remaining_ = cord->contents_.inline_size();
1451 current_chunk_ = {cord->contents_.data(), bytes_remaining_};
1452 }
1453 }
1454
AdvanceBtree()1455 inline Cord::ChunkIterator& Cord::ChunkIterator::AdvanceBtree() {
1456 current_chunk_ = btree_reader_.Next();
1457 return *this;
1458 }
1459
AdvanceBytesBtree(size_t n)1460 inline void Cord::ChunkIterator::AdvanceBytesBtree(size_t n) {
1461 assert(n >= current_chunk_.size());
1462 bytes_remaining_ -= n;
1463 if (bytes_remaining_) {
1464 if (n == current_chunk_.size()) {
1465 current_chunk_ = btree_reader_.Next();
1466 } else {
1467 size_t offset = btree_reader_.length() - bytes_remaining_;
1468 current_chunk_ = btree_reader_.Seek(offset);
1469 }
1470 } else {
1471 current_chunk_ = {};
1472 }
1473 }
1474
1475 inline Cord::ChunkIterator& Cord::ChunkIterator::operator++() {
1476 ABSL_HARDENING_ASSERT(bytes_remaining_ > 0 &&
1477 "Attempted to iterate past `end()`");
1478 assert(bytes_remaining_ >= current_chunk_.size());
1479 bytes_remaining_ -= current_chunk_.size();
1480 if (bytes_remaining_ > 0) {
1481 if (btree_reader_) {
1482 return AdvanceBtree();
1483 } else {
1484 assert(!current_chunk_.empty()); // Called on invalid iterator.
1485 }
1486 current_chunk_ = {};
1487 }
1488 return *this;
1489 }
1490
1491 inline Cord::ChunkIterator Cord::ChunkIterator::operator++(int) {
1492 ChunkIterator tmp(*this);
1493 operator++();
1494 return tmp;
1495 }
1496
1497 inline bool Cord::ChunkIterator::operator==(const ChunkIterator& other) const {
1498 return bytes_remaining_ == other.bytes_remaining_;
1499 }
1500
1501 inline bool Cord::ChunkIterator::operator!=(const ChunkIterator& other) const {
1502 return !(*this == other);
1503 }
1504
1505 inline Cord::ChunkIterator::reference Cord::ChunkIterator::operator*() const {
1506 ABSL_HARDENING_ASSERT(bytes_remaining_ != 0);
1507 return current_chunk_;
1508 }
1509
1510 inline Cord::ChunkIterator::pointer Cord::ChunkIterator::operator->() const {
1511 ABSL_HARDENING_ASSERT(bytes_remaining_ != 0);
1512 return ¤t_chunk_;
1513 }
1514
RemoveChunkPrefix(size_t n)1515 inline void Cord::ChunkIterator::RemoveChunkPrefix(size_t n) {
1516 assert(n < current_chunk_.size());
1517 current_chunk_.remove_prefix(n);
1518 bytes_remaining_ -= n;
1519 }
1520
AdvanceBytes(size_t n)1521 inline void Cord::ChunkIterator::AdvanceBytes(size_t n) {
1522 assert(bytes_remaining_ >= n);
1523 if (ABSL_PREDICT_TRUE(n < current_chunk_.size())) {
1524 RemoveChunkPrefix(n);
1525 } else if (n != 0) {
1526 if (btree_reader_) {
1527 AdvanceBytesBtree(n);
1528 } else {
1529 bytes_remaining_ = 0;
1530 }
1531 }
1532 }
1533
chunk_begin()1534 inline Cord::ChunkIterator Cord::chunk_begin() const {
1535 return ChunkIterator(this);
1536 }
1537
chunk_end()1538 inline Cord::ChunkIterator Cord::chunk_end() const { return ChunkIterator(); }
1539
begin()1540 inline Cord::ChunkIterator Cord::ChunkRange::begin() const {
1541 return cord_->chunk_begin();
1542 }
1543
end()1544 inline Cord::ChunkIterator Cord::ChunkRange::end() const {
1545 return cord_->chunk_end();
1546 }
1547
Chunks()1548 inline Cord::ChunkRange Cord::Chunks() const { return ChunkRange(this); }
1549
1550 inline Cord::CharIterator& Cord::CharIterator::operator++() {
1551 if (ABSL_PREDICT_TRUE(chunk_iterator_->size() > 1)) {
1552 chunk_iterator_.RemoveChunkPrefix(1);
1553 } else {
1554 ++chunk_iterator_;
1555 }
1556 return *this;
1557 }
1558
1559 inline Cord::CharIterator Cord::CharIterator::operator++(int) {
1560 CharIterator tmp(*this);
1561 operator++();
1562 return tmp;
1563 }
1564
1565 inline bool Cord::CharIterator::operator==(const CharIterator& other) const {
1566 return chunk_iterator_ == other.chunk_iterator_;
1567 }
1568
1569 inline bool Cord::CharIterator::operator!=(const CharIterator& other) const {
1570 return !(*this == other);
1571 }
1572
1573 inline Cord::CharIterator::reference Cord::CharIterator::operator*() const {
1574 return *chunk_iterator_->data();
1575 }
1576
1577 inline Cord::CharIterator::pointer Cord::CharIterator::operator->() const {
1578 return chunk_iterator_->data();
1579 }
1580
AdvanceAndRead(CharIterator * it,size_t n_bytes)1581 inline Cord Cord::AdvanceAndRead(CharIterator* it, size_t n_bytes) {
1582 assert(it != nullptr);
1583 return it->chunk_iterator_.AdvanceAndReadBytes(n_bytes);
1584 }
1585
Advance(CharIterator * it,size_t n_bytes)1586 inline void Cord::Advance(CharIterator* it, size_t n_bytes) {
1587 assert(it != nullptr);
1588 it->chunk_iterator_.AdvanceBytes(n_bytes);
1589 }
1590
ChunkRemaining(const CharIterator & it)1591 inline absl::string_view Cord::ChunkRemaining(const CharIterator& it) {
1592 return *it.chunk_iterator_;
1593 }
1594
char_begin()1595 inline Cord::CharIterator Cord::char_begin() const {
1596 return CharIterator(this);
1597 }
1598
char_end()1599 inline Cord::CharIterator Cord::char_end() const { return CharIterator(); }
1600
begin()1601 inline Cord::CharIterator Cord::CharRange::begin() const {
1602 return cord_->char_begin();
1603 }
1604
end()1605 inline Cord::CharIterator Cord::CharRange::end() const {
1606 return cord_->char_end();
1607 }
1608
Chars()1609 inline Cord::CharRange Cord::Chars() const { return CharRange(this); }
1610
ForEachChunk(absl::FunctionRef<void (absl::string_view)> callback)1611 inline void Cord::ForEachChunk(
1612 absl::FunctionRef<void(absl::string_view)> callback) const {
1613 absl::cord_internal::CordRep* rep = contents_.tree();
1614 if (rep == nullptr) {
1615 callback(absl::string_view(contents_.data(), contents_.size()));
1616 } else {
1617 ForEachChunkAux(rep, callback);
1618 }
1619 }
1620
1621 // Nonmember Cord-to-Cord relational operators.
1622 inline bool operator==(const Cord& lhs, const Cord& rhs) {
1623 if (lhs.contents_.IsSame(rhs.contents_)) return true;
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!=(const Cord& x, const Cord& y) { return !(x == y); }
1630 inline bool operator<(const Cord& x, const Cord& y) { return x.Compare(y) < 0; }
1631 inline bool operator>(const Cord& x, const Cord& y) { return x.Compare(y) > 0; }
1632 inline bool operator<=(const Cord& x, const Cord& y) {
1633 return x.Compare(y) <= 0;
1634 }
1635 inline bool operator>=(const Cord& x, const Cord& y) {
1636 return x.Compare(y) >= 0;
1637 }
1638
1639 // Nonmember Cord-to-absl::string_view relational operators.
1640 //
1641 // Due to implicit conversions, these also enable comparisons of Cord with
1642 // std::string and const char*.
1643 inline bool operator==(const Cord& lhs, absl::string_view rhs) {
1644 size_t lhs_size = lhs.size();
1645 size_t rhs_size = rhs.size();
1646 if (lhs_size != rhs_size) return false;
1647 return lhs.EqualsImpl(rhs, rhs_size);
1648 }
1649
1650 inline bool operator==(absl::string_view x, const Cord& y) { return y == x; }
1651 inline bool operator!=(const Cord& x, absl::string_view y) { return !(x == y); }
1652 inline bool operator!=(absl::string_view x, const Cord& y) { return !(x == y); }
1653 inline bool operator<(const Cord& x, absl::string_view y) {
1654 return x.Compare(y) < 0;
1655 }
1656 inline bool operator<(absl::string_view x, const Cord& y) {
1657 return y.Compare(x) > 0;
1658 }
1659 inline bool operator>(const Cord& x, absl::string_view y) { return y < x; }
1660 inline bool operator>(absl::string_view x, const Cord& y) { return y < x; }
1661 inline bool operator<=(const Cord& x, absl::string_view y) { return !(y < x); }
1662 inline bool operator<=(absl::string_view x, const Cord& y) { return !(y < x); }
1663 inline bool operator>=(const Cord& x, absl::string_view y) { return !(x < y); }
1664 inline bool operator>=(absl::string_view x, const Cord& y) { return !(x < y); }
1665
1666 // Some internals exposed to test code.
1667 namespace strings_internal {
1668 class CordTestAccess {
1669 public:
1670 static size_t FlatOverhead();
1671 static size_t MaxFlatLength();
1672 static size_t SizeofCordRepExternal();
1673 static size_t SizeofCordRepSubstring();
1674 static size_t FlatTagToLength(uint8_t tag);
1675 static uint8_t LengthToTag(size_t s);
1676 };
1677 } // namespace strings_internal
1678 ABSL_NAMESPACE_END
1679 } // namespace absl
1680
1681 #endif // ABSL_STRINGS_CORD_H_
1682