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/functional/function_ref.h"
80 #include "absl/meta/type_traits.h"
81 #include "absl/strings/cord_analysis.h"
82 #include "absl/strings/cord_buffer.h"
83 #include "absl/strings/internal/cord_data_edge.h"
84 #include "absl/strings/internal/cord_internal.h"
85 #include "absl/strings/internal/cord_rep_btree.h"
86 #include "absl/strings/internal/cord_rep_btree_reader.h"
87 #include "absl/strings/internal/cord_rep_crc.h"
88 #include "absl/strings/internal/cord_rep_ring.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 kTotal,
113
114 // Counts the *approximate* number of bytes held in full or in part by this
115 // Cord weighted by the sharing ratio of that data. For example, if some data
116 // edge is shared by 4 different Cords, then each cord is attributed 1/4th of
117 // the total memory usage as a 'fair share' of the total memory usage.
118 kFairShare,
119 };
120
121 // Cord
122 //
123 // A Cord is a sequence of characters, designed to be more efficient than a
124 // `std::string` in certain circumstances: namely, large string data that needs
125 // to change over its lifetime or shared, especially when such data is shared
126 // across API boundaries.
127 //
128 // A Cord stores its character data in a structure that allows efficient prepend
129 // and append operations. This makes a Cord useful for large string data sent
130 // over in a wire format that may need to be prepended or appended at some point
131 // during the data exchange (e.g. HTTP, protocol buffers). For example, a
132 // Cord is useful for storing an HTTP request, and prepending an HTTP header to
133 // such a request.
134 //
135 // Cords should not be used for storing general string data, however. They
136 // require overhead to construct and are slower than strings for random access.
137 //
138 // The Cord API provides the following common API operations:
139 //
140 // * Create or assign Cords out of existing string data, memory, or other Cords
141 // * Append and prepend data to an existing Cord
142 // * Create new Sub-Cords from existing Cord data
143 // * Swap Cord data and compare Cord equality
144 // * Write out Cord data by constructing a `std::string`
145 //
146 // Additionally, the API provides iterator utilities to iterate through Cord
147 // data via chunks or character bytes.
148 //
149 class Cord {
150 private:
151 template <typename T>
152 using EnableIfString =
153 absl::enable_if_t<std::is_same<T, std::string>::value, int>;
154
155 public:
156 // Cord::Cord() Constructors.
157
158 // Creates an empty Cord.
159 constexpr Cord() noexcept;
160
161 // Creates a Cord from an existing Cord. Cord is copyable and efficiently
162 // movable. The moved-from state is valid but unspecified.
163 Cord(const Cord& src);
164 Cord(Cord&& src) noexcept;
165 Cord& operator=(const Cord& x);
166 Cord& operator=(Cord&& x) noexcept;
167
168 // Creates a Cord from a `src` string. This constructor is marked explicit to
169 // prevent implicit Cord constructions from arguments convertible to an
170 // `absl::string_view`.
171 explicit Cord(absl::string_view src);
172 Cord& operator=(absl::string_view src);
173
174 // Creates a Cord from a `std::string&&` rvalue. These constructors are
175 // templated to avoid ambiguities for types that are convertible to both
176 // `absl::string_view` and `std::string`, such as `const char*`.
177 template <typename T, EnableIfString<T> = 0>
178 explicit Cord(T&& src);
179 template <typename T, EnableIfString<T> = 0>
180 Cord& operator=(T&& src);
181
182 // Cord::~Cord()
183 //
184 // Destructs the Cord.
~Cord()185 ~Cord() {
186 if (contents_.is_tree()) DestroyCordSlow();
187 }
188
189 // MakeCordFromExternal()
190 //
191 // Creates a Cord that takes ownership of external string memory. The
192 // contents of `data` are not copied to the Cord; instead, the external
193 // memory is added to the Cord and reference-counted. This data may not be
194 // changed for the life of the Cord, though it may be prepended or appended
195 // to.
196 //
197 // `MakeCordFromExternal()` takes a callable "releaser" that is invoked when
198 // the reference count for `data` reaches zero. As noted above, this data must
199 // remain live until the releaser is invoked. The callable releaser also must:
200 //
201 // * be move constructible
202 // * support `void operator()(absl::string_view) const` or `void operator()`
203 //
204 // Example:
205 //
206 // Cord MakeCord(BlockPool* pool) {
207 // Block* block = pool->NewBlock();
208 // FillBlock(block);
209 // return absl::MakeCordFromExternal(
210 // block->ToStringView(),
211 // [pool, block](absl::string_view v) {
212 // pool->FreeBlock(block, v);
213 // });
214 // }
215 //
216 // WARNING: Because a Cord can be reference-counted, it's likely a bug if your
217 // releaser doesn't do anything. For example, consider the following:
218 //
219 // void Foo(const char* buffer, int len) {
220 // auto c = absl::MakeCordFromExternal(absl::string_view(buffer, len),
221 // [](absl::string_view) {});
222 //
223 // // BUG: If Bar() copies its cord for any reason, including keeping a
224 // // substring of it, the lifetime of buffer might be extended beyond
225 // // when Foo() returns.
226 // Bar(c);
227 // }
228 template <typename Releaser>
229 friend Cord MakeCordFromExternal(absl::string_view data, Releaser&& releaser);
230
231 // Cord::Clear()
232 //
233 // Releases the Cord data. Any nodes that share data with other Cords, if
234 // applicable, will have their reference counts reduced by 1.
235 ABSL_ATTRIBUTE_REINITIALIZES void Clear();
236
237 // Cord::Append()
238 //
239 // Appends data to the Cord, which may come from another Cord or other string
240 // data.
241 void Append(const Cord& src);
242 void Append(Cord&& src);
243 void Append(absl::string_view src);
244 template <typename T, EnableIfString<T> = 0>
245 void Append(T&& src);
246
247 // Appends `buffer` to this cord, unless `buffer` has a zero length in which
248 // case this method has no effect on this cord instance.
249 // This method is guaranteed to consume `buffer`.
250 void Append(CordBuffer buffer);
251
252 // Returns a CordBuffer, re-using potential existing capacity in this cord.
253 //
254 // Cord instances may have additional unused capacity in the last (or first)
255 // nodes of the underlying tree to facilitate amortized growth. This method
256 // allows applications to explicitly use this spare capacity if available,
257 // or create a new CordBuffer instance otherwise.
258 // If this cord has a final non-shared node with at least `min_capacity`
259 // available, then this method will return that buffer including its data
260 // contents. I.e.; the returned buffer will have a non-zero length, and
261 // a capacity of at least `buffer.length + min_capacity`. Otherwise, this
262 // method will return `CordBuffer::CreateWithDefaultLimit(capacity)`.
263 //
264 // Below an example of using GetAppendBuffer. Notice that in this example we
265 // use `GetAppendBuffer()` only on the first iteration. As we know nothing
266 // about any initial extra capacity in `cord`, we may be able to use the extra
267 // capacity. But as we add new buffers with fully utilized contents after that
268 // we avoid calling `GetAppendBuffer()` on subsequent iterations: while this
269 // works fine, it results in an unnecessary inspection of cord contents:
270 //
271 // void AppendRandomDataToCord(absl::Cord &cord, size_t n) {
272 // bool first = true;
273 // while (n > 0) {
274 // CordBuffer buffer = first ? cord.GetAppendBuffer(n)
275 // : CordBuffer::CreateWithDefaultLimit(n);
276 // absl::Span<char> data = buffer.available_up_to(n);
277 // FillRandomValues(data.data(), data.size());
278 // buffer.IncreaseLengthBy(data.size());
279 // cord.Append(std::move(buffer));
280 // n -= data.size();
281 // first = false;
282 // }
283 // }
284 CordBuffer GetAppendBuffer(size_t capacity, size_t min_capacity = 16);
285
286 // Returns a CordBuffer, re-using potential existing capacity in this cord.
287 //
288 // This function is identical to `GetAppendBuffer`, except that in the case
289 // where a new `CordBuffer` is allocated, it is allocated using the provided
290 // custom limit instead of the default limit. `GetAppendBuffer` will default
291 // to `CordBuffer::CreateWithDefaultLimit(capacity)` whereas this method
292 // will default to `CordBuffer::CreateWithCustomLimit(block_size, capacity)`.
293 // This method is equivalent to `GetAppendBuffer` if `block_size` is zero.
294 // See the documentation for `CreateWithCustomLimit` for more details on the
295 // restrictions and legal values for `block_size`.
296 CordBuffer GetCustomAppendBuffer(size_t block_size, size_t capacity,
297 size_t min_capacity = 16);
298
299 // Cord::Prepend()
300 //
301 // Prepends data to the Cord, which may come from another Cord or other string
302 // data.
303 void Prepend(const Cord& src);
304 void Prepend(absl::string_view src);
305 template <typename T, EnableIfString<T> = 0>
306 void Prepend(T&& src);
307
308 // Prepends `buffer` to this cord, unless `buffer` has a zero length in which
309 // case this method has no effect on this cord instance.
310 // This method is guaranteed to consume `buffer`.
311 void Prepend(CordBuffer buffer);
312
313 // Cord::RemovePrefix()
314 //
315 // Removes the first `n` bytes of a Cord.
316 void RemovePrefix(size_t n);
317 void RemoveSuffix(size_t n);
318
319 // Cord::Subcord()
320 //
321 // Returns a new Cord representing the subrange [pos, pos + new_size) of
322 // *this. If pos >= size(), the result is empty(). If
323 // (pos + new_size) >= size(), the result is the subrange [pos, size()).
324 Cord Subcord(size_t pos, size_t new_size) const;
325
326 // Cord::swap()
327 //
328 // Swaps the contents of the Cord with `other`.
329 void swap(Cord& other) noexcept;
330
331 // swap()
332 //
333 // Swaps the contents of two Cords.
swap(Cord & x,Cord & y)334 friend void swap(Cord& x, Cord& y) noexcept { x.swap(y); }
335
336 // Cord::size()
337 //
338 // Returns the size of the Cord.
339 size_t size() const;
340
341 // Cord::empty()
342 //
343 // Determines whether the given Cord is empty, returning `true` is so.
344 bool empty() const;
345
346 // Cord::EstimatedMemoryUsage()
347 //
348 // Returns the *approximate* number of bytes held by this cord.
349 // See CordMemoryAccounting for more information on the accounting method.
350 size_t EstimatedMemoryUsage(CordMemoryAccounting accounting_method =
351 CordMemoryAccounting::kTotal) const;
352
353 // Cord::Compare()
354 //
355 // Compares 'this' Cord with rhs. This function and its relatives treat Cords
356 // as sequences of unsigned bytes. The comparison is a straightforward
357 // lexicographic comparison. `Cord::Compare()` returns values as follows:
358 //
359 // -1 'this' Cord is smaller
360 // 0 two Cords are equal
361 // 1 'this' Cord is larger
362 int Compare(absl::string_view rhs) const;
363 int Compare(const Cord& rhs) const;
364
365 // Cord::StartsWith()
366 //
367 // Determines whether the Cord starts with the passed string data `rhs`.
368 bool StartsWith(const Cord& rhs) const;
369 bool StartsWith(absl::string_view rhs) const;
370
371 // Cord::EndsWith()
372 //
373 // Determines whether the Cord ends with the passed string data `rhs`.
374 bool EndsWith(absl::string_view rhs) const;
375 bool EndsWith(const Cord& rhs) const;
376
377 // Cord::operator std::string()
378 //
379 // Converts a Cord into a `std::string()`. This operator is marked explicit to
380 // prevent unintended Cord usage in functions that take a string.
381 explicit operator std::string() const;
382
383 // CopyCordToString()
384 //
385 // Copies the contents of a `src` Cord into a `*dst` string.
386 //
387 // This function optimizes the case of reusing the destination string since it
388 // can reuse previously allocated capacity. However, this function does not
389 // guarantee that pointers previously returned by `dst->data()` remain valid
390 // even if `*dst` had enough capacity to hold `src`. If `*dst` is a new
391 // object, prefer to simply use the conversion operator to `std::string`.
392 friend void CopyCordToString(const Cord& src, std::string* dst);
393
394 class CharIterator;
395
396 //----------------------------------------------------------------------------
397 // Cord::ChunkIterator
398 //----------------------------------------------------------------------------
399 //
400 // A `Cord::ChunkIterator` allows iteration over the constituent chunks of its
401 // Cord. Such iteration allows you to perform non-const operations on the data
402 // of a Cord without modifying it.
403 //
404 // Generally, you do not instantiate a `Cord::ChunkIterator` directly;
405 // instead, you create one implicitly through use of the `Cord::Chunks()`
406 // member function.
407 //
408 // The `Cord::ChunkIterator` has the following properties:
409 //
410 // * The iterator is invalidated after any non-const operation on the
411 // Cord object over which it iterates.
412 // * The `string_view` returned by dereferencing a valid, non-`end()`
413 // iterator is guaranteed to be non-empty.
414 // * Two `ChunkIterator` objects can be compared equal if and only if they
415 // remain valid and iterate over the same Cord.
416 // * The iterator in this case is a proxy iterator; the `string_view`
417 // returned by the iterator does not live inside the Cord, and its
418 // lifetime is limited to the lifetime of the iterator itself. To help
419 // prevent lifetime issues, `ChunkIterator::reference` is not a true
420 // reference type and is equivalent to `value_type`.
421 // * The iterator keeps state that can grow for Cords that contain many
422 // nodes and are imbalanced due to sharing. Prefer to pass this type by
423 // const reference instead of by value.
424 class ChunkIterator {
425 public:
426 using iterator_category = std::input_iterator_tag;
427 using value_type = absl::string_view;
428 using difference_type = ptrdiff_t;
429 using pointer = const value_type*;
430 using reference = value_type;
431
432 ChunkIterator() = default;
433
434 ChunkIterator& operator++();
435 ChunkIterator operator++(int);
436 bool operator==(const ChunkIterator& other) const;
437 bool operator!=(const ChunkIterator& other) const;
438 reference operator*() const;
439 pointer operator->() const;
440
441 friend class Cord;
442 friend class CharIterator;
443
444 private:
445 using CordRep = absl::cord_internal::CordRep;
446 using CordRepBtree = absl::cord_internal::CordRepBtree;
447 using CordRepBtreeReader = absl::cord_internal::CordRepBtreeReader;
448
449 // Constructs a `begin()` iterator from `tree`. `tree` must not be null.
450 explicit ChunkIterator(cord_internal::CordRep* tree);
451
452 // Constructs a `begin()` iterator from `cord`.
453 explicit ChunkIterator(const Cord* cord);
454
455 // Initializes this instance from a tree. Invoked by constructors.
456 void InitTree(cord_internal::CordRep* tree);
457
458 // Removes `n` bytes from `current_chunk_`. Expects `n` to be smaller than
459 // `current_chunk_.size()`.
460 void RemoveChunkPrefix(size_t n);
461 Cord AdvanceAndReadBytes(size_t n);
462 void AdvanceBytes(size_t n);
463
464 // Btree specific operator++
465 ChunkIterator& AdvanceBtree();
466 void AdvanceBytesBtree(size_t n);
467
468 // A view into bytes of the current `CordRep`. It may only be a view to a
469 // suffix of bytes if this is being used by `CharIterator`.
470 absl::string_view current_chunk_;
471 // The current leaf, or `nullptr` if the iterator points to short data.
472 // If the current chunk is a substring node, current_leaf_ points to the
473 // underlying flat or external node.
474 absl::cord_internal::CordRep* current_leaf_ = nullptr;
475 // The number of bytes left in the `Cord` over which we are iterating.
476 size_t bytes_remaining_ = 0;
477
478 // Cord reader for cord btrees. Empty if not traversing a btree.
479 CordRepBtreeReader btree_reader_;
480 };
481
482 // Cord::chunk_begin()
483 //
484 // Returns an iterator to the first chunk of the `Cord`.
485 //
486 // Generally, prefer using `Cord::Chunks()` within a range-based for loop for
487 // iterating over the chunks of a Cord. This method may be useful for getting
488 // a `ChunkIterator` where range-based for-loops are not useful.
489 //
490 // Example:
491 //
492 // absl::Cord::ChunkIterator FindAsChunk(const absl::Cord& c,
493 // absl::string_view s) {
494 // return std::find(c.chunk_begin(), c.chunk_end(), s);
495 // }
496 ChunkIterator chunk_begin() const;
497
498 // Cord::chunk_end()
499 //
500 // Returns an iterator one increment past the last chunk of the `Cord`.
501 //
502 // Generally, prefer using `Cord::Chunks()` within a range-based for loop for
503 // iterating over the chunks of a Cord. This method may be useful for getting
504 // a `ChunkIterator` where range-based for-loops may not be available.
505 ChunkIterator chunk_end() const;
506
507 //----------------------------------------------------------------------------
508 // Cord::ChunkRange
509 //----------------------------------------------------------------------------
510 //
511 // `ChunkRange` is a helper class for iterating over the chunks of the `Cord`,
512 // producing an iterator which can be used within a range-based for loop.
513 // Construction of a `ChunkRange` will return an iterator pointing to the
514 // first chunk of the Cord. Generally, do not construct a `ChunkRange`
515 // directly; instead, prefer to use the `Cord::Chunks()` method.
516 //
517 // Implementation note: `ChunkRange` is simply a convenience wrapper over
518 // `Cord::chunk_begin()` and `Cord::chunk_end()`.
519 class ChunkRange {
520 public:
521 // Fulfill minimum c++ container requirements [container.requirements]
522 // These (partial) container type definitions allow ChunkRange to be used
523 // in various utilities expecting a subset of [container.requirements].
524 // For example, the below enables using `::testing::ElementsAre(...)`
525 using value_type = absl::string_view;
526 using reference = value_type&;
527 using const_reference = const value_type&;
528 using iterator = ChunkIterator;
529 using const_iterator = ChunkIterator;
530
ChunkRange(const Cord * cord)531 explicit ChunkRange(const Cord* cord) : cord_(cord) {}
532
533 ChunkIterator begin() const;
534 ChunkIterator end() const;
535
536 private:
537 const Cord* cord_;
538 };
539
540 // Cord::Chunks()
541 //
542 // Returns a `Cord::ChunkRange` for iterating over the chunks of a `Cord` with
543 // a range-based for-loop. For most iteration tasks on a Cord, use
544 // `Cord::Chunks()` to retrieve this iterator.
545 //
546 // Example:
547 //
548 // void ProcessChunks(const Cord& cord) {
549 // for (absl::string_view chunk : cord.Chunks()) { ... }
550 // }
551 //
552 // Note that the ordinary caveats of temporary lifetime extension apply:
553 //
554 // void Process() {
555 // for (absl::string_view chunk : CordFactory().Chunks()) {
556 // // The temporary Cord returned by CordFactory has been destroyed!
557 // }
558 // }
559 ChunkRange Chunks() const;
560
561 //----------------------------------------------------------------------------
562 // Cord::CharIterator
563 //----------------------------------------------------------------------------
564 //
565 // A `Cord::CharIterator` allows iteration over the constituent characters of
566 // a `Cord`.
567 //
568 // Generally, you do not instantiate a `Cord::CharIterator` directly; instead,
569 // you create one implicitly through use of the `Cord::Chars()` member
570 // function.
571 //
572 // A `Cord::CharIterator` has the following properties:
573 //
574 // * The iterator is invalidated after any non-const operation on the
575 // Cord object over which it iterates.
576 // * Two `CharIterator` objects can be compared equal if and only if they
577 // remain valid and iterate over the same Cord.
578 // * The iterator keeps state that can grow for Cords that contain many
579 // nodes and are imbalanced due to sharing. Prefer to pass this type by
580 // const reference instead of by value.
581 // * This type cannot act as a forward iterator because a `Cord` can reuse
582 // sections of memory. This fact violates the requirement for forward
583 // iterators to compare equal if dereferencing them returns the same
584 // object.
585 class CharIterator {
586 public:
587 using iterator_category = std::input_iterator_tag;
588 using value_type = char;
589 using difference_type = ptrdiff_t;
590 using pointer = const char*;
591 using reference = const char&;
592
593 CharIterator() = default;
594
595 CharIterator& operator++();
596 CharIterator operator++(int);
597 bool operator==(const CharIterator& other) const;
598 bool operator!=(const CharIterator& other) const;
599 reference operator*() const;
600 pointer operator->() const;
601
602 friend Cord;
603
604 private:
CharIterator(const Cord * cord)605 explicit CharIterator(const Cord* cord) : chunk_iterator_(cord) {}
606
607 ChunkIterator chunk_iterator_;
608 };
609
610 // Cord::AdvanceAndRead()
611 //
612 // Advances the `Cord::CharIterator` by `n_bytes` and returns the bytes
613 // advanced as a separate `Cord`. `n_bytes` must be less than or equal to the
614 // number of bytes within the Cord; otherwise, behavior is undefined. It is
615 // valid to pass `char_end()` and `0`.
616 static Cord AdvanceAndRead(CharIterator* it, size_t n_bytes);
617
618 // Cord::Advance()
619 //
620 // Advances the `Cord::CharIterator` by `n_bytes`. `n_bytes` must be less than
621 // or equal to the number of bytes remaining within the Cord; otherwise,
622 // behavior is undefined. It is valid to pass `char_end()` and `0`.
623 static void Advance(CharIterator* it, size_t n_bytes);
624
625 // Cord::ChunkRemaining()
626 //
627 // Returns the longest contiguous view starting at the iterator's position.
628 //
629 // `it` must be dereferenceable.
630 static absl::string_view ChunkRemaining(const CharIterator& it);
631
632 // Cord::char_begin()
633 //
634 // Returns an iterator to the first character of the `Cord`.
635 //
636 // Generally, prefer using `Cord::Chars()` within a range-based for loop for
637 // iterating over the chunks of a Cord. This method may be useful for getting
638 // a `CharIterator` where range-based for-loops may not be available.
639 CharIterator char_begin() const;
640
641 // Cord::char_end()
642 //
643 // Returns an iterator to one past the last character of the `Cord`.
644 //
645 // Generally, prefer using `Cord::Chars()` within a range-based for loop for
646 // iterating over the chunks of a Cord. This method may be useful for getting
647 // a `CharIterator` where range-based for-loops are not useful.
648 CharIterator char_end() const;
649
650 // Cord::CharRange
651 //
652 // `CharRange` is a helper class for iterating over the characters of a
653 // producing an iterator which can be used within a range-based for loop.
654 // Construction of a `CharRange` will return an iterator pointing to the first
655 // character of the Cord. Generally, do not construct a `CharRange` directly;
656 // instead, prefer to use the `Cord::Chars()` method shown below.
657 //
658 // Implementation note: `CharRange` is simply a convenience wrapper over
659 // `Cord::char_begin()` and `Cord::char_end()`.
660 class CharRange {
661 public:
662 // Fulfill minimum c++ container requirements [container.requirements]
663 // Theses (partial) container type definitions allow CharRange to be used
664 // in various utilities expecting a subset of [container.requirements].
665 // For example, the below enables using `::testing::ElementsAre(...)`
666 using value_type = char;
667 using reference = value_type&;
668 using const_reference = const value_type&;
669 using iterator = CharIterator;
670 using const_iterator = CharIterator;
671
CharRange(const Cord * cord)672 explicit CharRange(const Cord* cord) : cord_(cord) {}
673
674 CharIterator begin() const;
675 CharIterator end() const;
676
677 private:
678 const Cord* cord_;
679 };
680
681 // Cord::Chars()
682 //
683 // Returns a `Cord::CharRange` for iterating over the characters of a `Cord`
684 // with a range-based for-loop. For most character-based iteration tasks on a
685 // Cord, use `Cord::Chars()` to retrieve this iterator.
686 //
687 // Example:
688 //
689 // void ProcessCord(const Cord& cord) {
690 // for (char c : cord.Chars()) { ... }
691 // }
692 //
693 // Note that the ordinary caveats of temporary lifetime extension apply:
694 //
695 // void Process() {
696 // for (char c : CordFactory().Chars()) {
697 // // The temporary Cord returned by CordFactory has been destroyed!
698 // }
699 // }
700 CharRange Chars() const;
701
702 // Cord::operator[]
703 //
704 // Gets the "i"th character of the Cord and returns it, provided that
705 // 0 <= i < Cord.size().
706 //
707 // NOTE: This routine is reasonably efficient. It is roughly
708 // logarithmic based on the number of chunks that make up the cord. Still,
709 // if you need to iterate over the contents of a cord, you should
710 // use a CharIterator/ChunkIterator rather than call operator[] or Get()
711 // repeatedly in a loop.
712 char operator[](size_t i) const;
713
714 // Cord::TryFlat()
715 //
716 // If this cord's representation is a single flat array, returns a
717 // string_view referencing that array. Otherwise returns nullopt.
718 absl::optional<absl::string_view> TryFlat() const;
719
720 // Cord::Flatten()
721 //
722 // Flattens the cord into a single array and returns a view of the data.
723 //
724 // If the cord was already flat, the contents are not modified.
725 absl::string_view Flatten();
726
727 // Supports absl::Cord as a sink object for absl::Format().
AbslFormatFlush(absl::Cord * cord,absl::string_view part)728 friend void AbslFormatFlush(absl::Cord* cord, absl::string_view part) {
729 cord->Append(part);
730 }
731
732 // Cord::SetExpectedChecksum()
733 //
734 // Stores a checksum value with this non-empty cord instance, for later
735 // retrieval.
736 //
737 // The expected checksum is a number stored out-of-band, alongside the data.
738 // It is preserved across copies and assignments, but any mutations to a cord
739 // will cause it to lose its expected checksum.
740 //
741 // The expected checksum is not part of a Cord's value, and does not affect
742 // operations such as equality or hashing.
743 //
744 // This field is intended to store a CRC32C checksum for later validation, to
745 // help support end-to-end checksum workflows. However, the Cord API itself
746 // does no CRC validation, and assigns no meaning to this number.
747 //
748 // This call has no effect if this cord is empty.
749 void SetExpectedChecksum(uint32_t crc);
750
751 // Returns this cord's expected checksum, if it has one. Otherwise, returns
752 // nullopt.
753 absl::optional<uint32_t> ExpectedChecksum() const;
754
755 template <typename H>
AbslHashValue(H hash_state,const absl::Cord & c)756 friend H AbslHashValue(H hash_state, const absl::Cord& c) {
757 absl::optional<absl::string_view> maybe_flat = c.TryFlat();
758 if (maybe_flat.has_value()) {
759 return H::combine(std::move(hash_state), *maybe_flat);
760 }
761 return c.HashFragmented(std::move(hash_state));
762 }
763
764 // Create a Cord with the contents of StringConstant<T>::value.
765 // No allocations will be done and no data will be copied.
766 // This is an INTERNAL API and subject to change or removal. This API can only
767 // be used by spelling absl::strings_internal::MakeStringConstant, which is
768 // also an internal API.
769 template <typename T>
770 // NOLINTNEXTLINE(google-explicit-constructor)
771 constexpr Cord(strings_internal::StringConstant<T>);
772
773 private:
774 using CordRep = absl::cord_internal::CordRep;
775 using CordRepFlat = absl::cord_internal::CordRepFlat;
776 using CordzInfo = cord_internal::CordzInfo;
777 using CordzUpdateScope = cord_internal::CordzUpdateScope;
778 using CordzUpdateTracker = cord_internal::CordzUpdateTracker;
779 using InlineData = cord_internal::InlineData;
780 using MethodIdentifier = CordzUpdateTracker::MethodIdentifier;
781
782 // Creates a cord instance with `method` representing the originating
783 // public API call causing the cord to be created.
784 explicit Cord(absl::string_view src, MethodIdentifier method);
785
786 friend class CordTestPeer;
787 friend bool operator==(const Cord& lhs, const Cord& rhs);
788 friend bool operator==(const Cord& lhs, absl::string_view rhs);
789
790 friend const CordzInfo* GetCordzInfoForTesting(const Cord& cord);
791
792 // Calls the provided function once for each cord chunk, in order. Unlike
793 // Chunks(), this API will not allocate memory.
794 void ForEachChunk(absl::FunctionRef<void(absl::string_view)>) const;
795
796 // Allocates new contiguous storage for the contents of the cord. This is
797 // called by Flatten() when the cord was not already flat.
798 absl::string_view FlattenSlowPath();
799
800 // Actual cord contents are hidden inside the following simple
801 // class so that we can isolate the bulk of cord.cc from changes
802 // to the representation.
803 //
804 // InlineRep holds either a tree pointer, or an array of kMaxInline bytes.
805 class InlineRep {
806 public:
807 static constexpr unsigned char kMaxInline = cord_internal::kMaxInline;
808 static_assert(kMaxInline >= sizeof(absl::cord_internal::CordRep*), "");
809
InlineRep()810 constexpr InlineRep() : data_() {}
InlineRep(InlineData::DefaultInitType init)811 explicit InlineRep(InlineData::DefaultInitType init) : data_(init) {}
812 InlineRep(const InlineRep& src);
813 InlineRep(InlineRep&& src);
814 InlineRep& operator=(const InlineRep& src);
815 InlineRep& operator=(InlineRep&& src) noexcept;
816
817 explicit constexpr InlineRep(cord_internal::InlineData data);
818
819 void Swap(InlineRep* rhs);
820 bool empty() const;
821 size_t size() const;
822 const char* data() const; // Returns nullptr if holding pointer
823 void set_data(const char* data, size_t n); // Discards pointer, if any
824 char* set_data(size_t n); // Write data to the result
825 // Returns nullptr if holding bytes
826 absl::cord_internal::CordRep* tree() const;
827 absl::cord_internal::CordRep* as_tree() const;
828 const char* as_chars() const;
829 // Returns non-null iff was holding a pointer
830 absl::cord_internal::CordRep* clear();
831 // Converts to pointer if necessary.
832 void reduce_size(size_t n); // REQUIRES: holding data
833 void remove_prefix(size_t n); // REQUIRES: holding data
834 void AppendArray(absl::string_view src, MethodIdentifier method);
835 absl::string_view FindFlatStartPiece() const;
836
837 // Creates a CordRepFlat instance from the current inlined data with `extra'
838 // bytes of desired additional capacity.
839 CordRepFlat* MakeFlatWithExtraCapacity(size_t extra);
840
841 // Sets the tree value for this instance. `rep` must not be null.
842 // Requires the current instance to hold a tree, and a lock to be held on
843 // any CordzInfo referenced by this instance. The latter is enforced through
844 // the CordzUpdateScope argument. If the current instance is sampled, then
845 // the CordzInfo instance is updated to reference the new `rep` value.
846 void SetTree(CordRep* rep, const CordzUpdateScope& scope);
847
848 // Identical to SetTree(), except that `rep` is allowed to be null, in
849 // which case the current instance is reset to an empty value.
850 void SetTreeOrEmpty(CordRep* rep, const CordzUpdateScope& scope);
851
852 // Sets the tree value for this instance, and randomly samples this cord.
853 // This function disregards existing contents in `data_`, and should be
854 // called when a Cord is 'promoted' from an 'uninitialized' or 'inlined'
855 // value to a non-inlined (tree / ring) value.
856 void EmplaceTree(CordRep* rep, MethodIdentifier method);
857
858 // Identical to EmplaceTree, except that it copies the parent stack from
859 // the provided `parent` data if the parent is sampled.
860 void EmplaceTree(CordRep* rep, const InlineData& parent,
861 MethodIdentifier method);
862
863 // Commits the change of a newly created, or updated `rep` root value into
864 // this cord. `old_rep` indicates the old (inlined or tree) value of the
865 // cord, and determines if the commit invokes SetTree() or EmplaceTree().
866 void CommitTree(const CordRep* old_rep, CordRep* rep,
867 const CordzUpdateScope& scope, MethodIdentifier method);
868
869 void AppendTreeToInlined(CordRep* tree, MethodIdentifier method);
870 void AppendTreeToTree(CordRep* tree, MethodIdentifier method);
871 void AppendTree(CordRep* tree, MethodIdentifier method);
872 void PrependTreeToInlined(CordRep* tree, MethodIdentifier method);
873 void PrependTreeToTree(CordRep* tree, MethodIdentifier method);
874 void PrependTree(CordRep* tree, MethodIdentifier method);
875
IsSame(const InlineRep & other)876 bool IsSame(const InlineRep& other) const {
877 return memcmp(&data_, &other.data_, sizeof(data_)) == 0;
878 }
CopyTo(std::string * dst)879 void CopyTo(std::string* dst) const {
880 // memcpy is much faster when operating on a known size. On most supported
881 // platforms, the small string optimization is large enough that resizing
882 // to 15 bytes does not cause a memory allocation.
883 absl::strings_internal::STLStringResizeUninitialized(dst, kMaxInline);
884 memcpy(&(*dst)[0], data_.as_chars(), kMaxInline);
885 // erase is faster than resize because the logic for memory allocation is
886 // not needed.
887 dst->erase(inline_size());
888 }
889
890 // Copies the inline contents into `dst`. Assumes the cord is not empty.
891 void CopyToArray(char* dst) const;
892
is_tree()893 bool is_tree() const { return data_.is_tree(); }
894
895 // Returns true if the Cord is being profiled by cordz.
is_profiled()896 bool is_profiled() const { return data_.is_tree() && data_.is_profiled(); }
897
898 // Returns the available inlined capacity, or 0 if is_tree() == true.
remaining_inline_capacity()899 size_t remaining_inline_capacity() const {
900 return data_.is_tree() ? 0 : kMaxInline - data_.inline_size();
901 }
902
903 // Returns the profiled CordzInfo, or nullptr if not sampled.
cordz_info()904 absl::cord_internal::CordzInfo* cordz_info() const {
905 return data_.cordz_info();
906 }
907
908 // Sets the profiled CordzInfo. `cordz_info` must not be null.
set_cordz_info(cord_internal::CordzInfo * cordz_info)909 void set_cordz_info(cord_internal::CordzInfo* cordz_info) {
910 assert(cordz_info != nullptr);
911 data_.set_cordz_info(cordz_info);
912 }
913
914 // Resets the current cordz_info to null / empty.
clear_cordz_info()915 void clear_cordz_info() { data_.clear_cordz_info(); }
916
917 private:
918 friend class Cord;
919
920 void AssignSlow(const InlineRep& src);
921 // Unrefs the tree and stops profiling.
922 void UnrefTree();
923
ResetToEmpty()924 void ResetToEmpty() { data_ = {}; }
925
set_inline_size(size_t size)926 void set_inline_size(size_t size) { data_.set_inline_size(size); }
inline_size()927 size_t inline_size() const { return data_.inline_size(); }
928
929 // Empty cords that carry a checksum have a CordRepCrc node with a null
930 // child node. The code can avoid lots of special cases where it would
931 // otherwise transition from tree to inline storage if we just remove the
932 // CordRepCrc node before mutations. Must never be called inside a
933 // CordzUpdateScope since it untracks the cordz info.
934 void MaybeRemoveEmptyCrcNode();
935
936 cord_internal::InlineData data_;
937 };
938 InlineRep contents_;
939
940 // Helper for GetFlat() and TryFlat().
941 static bool GetFlatAux(absl::cord_internal::CordRep* rep,
942 absl::string_view* fragment);
943
944 // Helper for ForEachChunk().
945 static void ForEachChunkAux(
946 absl::cord_internal::CordRep* rep,
947 absl::FunctionRef<void(absl::string_view)> callback);
948
949 // The destructor for non-empty Cords.
950 void DestroyCordSlow();
951
952 // Out-of-line implementation of slower parts of logic.
953 void CopyToArraySlowPath(char* dst) const;
954 int CompareSlowPath(absl::string_view rhs, size_t compared_size,
955 size_t size_to_compare) const;
956 int CompareSlowPath(const Cord& rhs, size_t compared_size,
957 size_t size_to_compare) const;
958 bool EqualsImpl(absl::string_view rhs, size_t size_to_compare) const;
959 bool EqualsImpl(const Cord& rhs, size_t size_to_compare) const;
960 int CompareImpl(const Cord& rhs) const;
961
962 template <typename ResultType, typename RHS>
963 friend ResultType GenericCompare(const Cord& lhs, const RHS& rhs,
964 size_t size_to_compare);
965 static absl::string_view GetFirstChunk(const Cord& c);
966 static absl::string_view GetFirstChunk(absl::string_view sv);
967
968 // Returns a new reference to contents_.tree(), or steals an existing
969 // reference if called on an rvalue.
970 absl::cord_internal::CordRep* TakeRep() const&;
971 absl::cord_internal::CordRep* TakeRep() &&;
972
973 // Helper for Append().
974 template <typename C>
975 void AppendImpl(C&& src);
976
977 // Appends / Prepends `src` to this instance, using precise sizing.
978 // This method does explicitly not attempt to use any spare capacity
979 // in any pending last added private owned flat.
980 // Requires `src` to be <= kMaxFlatLength.
981 void AppendPrecise(absl::string_view src, MethodIdentifier method);
982 void PrependPrecise(absl::string_view src, MethodIdentifier method);
983
984 CordBuffer GetAppendBufferSlowPath(size_t block_size, size_t capacity,
985 size_t min_capacity);
986
987 // Prepends the provided data to this instance. `method` contains the public
988 // API method for this action which is tracked for Cordz sampling purposes.
989 void PrependArray(absl::string_view src, MethodIdentifier method);
990
991 // Assigns the value in 'src' to this instance, 'stealing' its contents.
992 // Requires src.length() > kMaxBytesToCopy.
993 Cord& AssignLargeString(std::string&& src);
994
995 // Helper for AbslHashValue().
996 template <typename H>
HashFragmented(H hash_state)997 H HashFragmented(H hash_state) const {
998 typename H::AbslInternalPiecewiseCombiner combiner;
999 ForEachChunk([&combiner, &hash_state](absl::string_view chunk) {
1000 hash_state = combiner.add_buffer(std::move(hash_state), chunk.data(),
1001 chunk.size());
1002 });
1003 return H::combine(combiner.finalize(std::move(hash_state)), size());
1004 }
1005 };
1006
1007 ABSL_NAMESPACE_END
1008 } // namespace absl
1009
1010 namespace absl {
1011 ABSL_NAMESPACE_BEGIN
1012
1013 // allow a Cord to be logged
1014 extern std::ostream& operator<<(std::ostream& out, const Cord& cord);
1015
1016 // ------------------------------------------------------------------
1017 // Internal details follow. Clients should ignore.
1018
1019 namespace cord_internal {
1020
1021 // Fast implementation of memmove for up to 15 bytes. This implementation is
1022 // safe for overlapping regions. If nullify_tail is true, the destination is
1023 // padded with '\0' up to 15 bytes.
1024 template <bool nullify_tail = false>
SmallMemmove(char * dst,const char * src,size_t n)1025 inline void SmallMemmove(char* dst, const char* src, size_t n) {
1026 if (n >= 8) {
1027 assert(n <= 15);
1028 uint64_t buf1;
1029 uint64_t buf2;
1030 memcpy(&buf1, src, 8);
1031 memcpy(&buf2, src + n - 8, 8);
1032 if (nullify_tail) {
1033 memset(dst + 7, 0, 8);
1034 }
1035 memcpy(dst, &buf1, 8);
1036 memcpy(dst + n - 8, &buf2, 8);
1037 } else if (n >= 4) {
1038 uint32_t buf1;
1039 uint32_t buf2;
1040 memcpy(&buf1, src, 4);
1041 memcpy(&buf2, src + n - 4, 4);
1042 if (nullify_tail) {
1043 memset(dst + 4, 0, 4);
1044 memset(dst + 7, 0, 8);
1045 }
1046 memcpy(dst, &buf1, 4);
1047 memcpy(dst + n - 4, &buf2, 4);
1048 } else {
1049 if (n != 0) {
1050 dst[0] = src[0];
1051 dst[n / 2] = src[n / 2];
1052 dst[n - 1] = src[n - 1];
1053 }
1054 if (nullify_tail) {
1055 memset(dst + 7, 0, 8);
1056 memset(dst + n, 0, 8);
1057 }
1058 }
1059 }
1060
1061 // Does non-template-specific `CordRepExternal` initialization.
1062 // Requires `data` to be non-empty.
1063 void InitializeCordRepExternal(absl::string_view data, CordRepExternal* rep);
1064
1065 // Creates a new `CordRep` that owns `data` and `releaser` and returns a pointer
1066 // to it. Requires `data` to be non-empty.
1067 template <typename Releaser>
1068 // NOLINTNEXTLINE - suppress clang-tidy raw pointer return.
NewExternalRep(absl::string_view data,Releaser && releaser)1069 CordRep* NewExternalRep(absl::string_view data, Releaser&& releaser) {
1070 assert(!data.empty());
1071 using ReleaserType = absl::decay_t<Releaser>;
1072 CordRepExternal* rep = new CordRepExternalImpl<ReleaserType>(
1073 std::forward<Releaser>(releaser), 0);
1074 InitializeCordRepExternal(data, rep);
1075 return rep;
1076 }
1077
1078 // Overload for function reference types that dispatches using a function
1079 // pointer because there are no `alignof()` or `sizeof()` a function reference.
1080 // NOLINTNEXTLINE - suppress clang-tidy raw pointer return.
NewExternalRep(absl::string_view data,void (& releaser)(absl::string_view))1081 inline CordRep* NewExternalRep(absl::string_view data,
1082 void (&releaser)(absl::string_view)) {
1083 return NewExternalRep(data, &releaser);
1084 }
1085
1086 } // namespace cord_internal
1087
1088 template <typename Releaser>
MakeCordFromExternal(absl::string_view data,Releaser && releaser)1089 Cord MakeCordFromExternal(absl::string_view data, Releaser&& releaser) {
1090 Cord cord;
1091 if (ABSL_PREDICT_TRUE(!data.empty())) {
1092 cord.contents_.EmplaceTree(::absl::cord_internal::NewExternalRep(
1093 data, std::forward<Releaser>(releaser)),
1094 Cord::MethodIdentifier::kMakeCordFromExternal);
1095 } else {
1096 using ReleaserType = absl::decay_t<Releaser>;
1097 cord_internal::InvokeReleaser(
1098 cord_internal::Rank0{}, ReleaserType(std::forward<Releaser>(releaser)),
1099 data);
1100 }
1101 return cord;
1102 }
1103
InlineRep(cord_internal::InlineData data)1104 constexpr Cord::InlineRep::InlineRep(cord_internal::InlineData data)
1105 : data_(data) {}
1106
InlineRep(const Cord::InlineRep & src)1107 inline Cord::InlineRep::InlineRep(const Cord::InlineRep& src)
1108 : data_(InlineData::kDefaultInit) {
1109 if (CordRep* tree = src.tree()) {
1110 EmplaceTree(CordRep::Ref(tree), src.data_,
1111 CordzUpdateTracker::kConstructorCord);
1112 } else {
1113 data_ = src.data_;
1114 }
1115 }
1116
InlineRep(Cord::InlineRep && src)1117 inline Cord::InlineRep::InlineRep(Cord::InlineRep&& src) : data_(src.data_) {
1118 src.ResetToEmpty();
1119 }
1120
1121 inline Cord::InlineRep& Cord::InlineRep::operator=(const Cord::InlineRep& src) {
1122 if (this == &src) {
1123 return *this;
1124 }
1125 if (!is_tree() && !src.is_tree()) {
1126 data_ = src.data_;
1127 return *this;
1128 }
1129 AssignSlow(src);
1130 return *this;
1131 }
1132
1133 inline Cord::InlineRep& Cord::InlineRep::operator=(
1134 Cord::InlineRep&& src) noexcept {
1135 if (is_tree()) {
1136 UnrefTree();
1137 }
1138 data_ = src.data_;
1139 src.ResetToEmpty();
1140 return *this;
1141 }
1142
Swap(Cord::InlineRep * rhs)1143 inline void Cord::InlineRep::Swap(Cord::InlineRep* rhs) {
1144 if (rhs == this) {
1145 return;
1146 }
1147 std::swap(data_, rhs->data_);
1148 }
1149
data()1150 inline const char* Cord::InlineRep::data() const {
1151 return is_tree() ? nullptr : data_.as_chars();
1152 }
1153
as_chars()1154 inline const char* Cord::InlineRep::as_chars() const {
1155 assert(!data_.is_tree());
1156 return data_.as_chars();
1157 }
1158
as_tree()1159 inline absl::cord_internal::CordRep* Cord::InlineRep::as_tree() const {
1160 assert(data_.is_tree());
1161 return data_.as_tree();
1162 }
1163
tree()1164 inline absl::cord_internal::CordRep* Cord::InlineRep::tree() const {
1165 if (is_tree()) {
1166 return as_tree();
1167 } else {
1168 return nullptr;
1169 }
1170 }
1171
empty()1172 inline bool Cord::InlineRep::empty() const { return data_.is_empty(); }
1173
size()1174 inline size_t Cord::InlineRep::size() const {
1175 return is_tree() ? as_tree()->length : inline_size();
1176 }
1177
MakeFlatWithExtraCapacity(size_t extra)1178 inline cord_internal::CordRepFlat* Cord::InlineRep::MakeFlatWithExtraCapacity(
1179 size_t extra) {
1180 static_assert(cord_internal::kMinFlatLength >= sizeof(data_), "");
1181 size_t len = data_.inline_size();
1182 auto* result = CordRepFlat::New(len + extra);
1183 result->length = len;
1184 memcpy(result->Data(), data_.as_chars(), InlineRep::kMaxInline);
1185 return result;
1186 }
1187
EmplaceTree(CordRep * rep,MethodIdentifier method)1188 inline void Cord::InlineRep::EmplaceTree(CordRep* rep,
1189 MethodIdentifier method) {
1190 assert(rep);
1191 data_.make_tree(rep);
1192 CordzInfo::MaybeTrackCord(data_, method);
1193 }
1194
EmplaceTree(CordRep * rep,const InlineData & parent,MethodIdentifier method)1195 inline void Cord::InlineRep::EmplaceTree(CordRep* rep, const InlineData& parent,
1196 MethodIdentifier method) {
1197 data_.make_tree(rep);
1198 CordzInfo::MaybeTrackCord(data_, parent, method);
1199 }
1200
SetTree(CordRep * rep,const CordzUpdateScope & scope)1201 inline void Cord::InlineRep::SetTree(CordRep* rep,
1202 const CordzUpdateScope& scope) {
1203 assert(rep);
1204 assert(data_.is_tree());
1205 data_.set_tree(rep);
1206 scope.SetCordRep(rep);
1207 }
1208
SetTreeOrEmpty(CordRep * rep,const CordzUpdateScope & scope)1209 inline void Cord::InlineRep::SetTreeOrEmpty(CordRep* rep,
1210 const CordzUpdateScope& scope) {
1211 assert(data_.is_tree());
1212 if (rep) {
1213 data_.set_tree(rep);
1214 } else {
1215 data_ = {};
1216 }
1217 scope.SetCordRep(rep);
1218 }
1219
CommitTree(const CordRep * old_rep,CordRep * rep,const CordzUpdateScope & scope,MethodIdentifier method)1220 inline void Cord::InlineRep::CommitTree(const CordRep* old_rep, CordRep* rep,
1221 const CordzUpdateScope& scope,
1222 MethodIdentifier method) {
1223 if (old_rep) {
1224 SetTree(rep, scope);
1225 } else {
1226 EmplaceTree(rep, method);
1227 }
1228 }
1229
clear()1230 inline absl::cord_internal::CordRep* Cord::InlineRep::clear() {
1231 if (is_tree()) {
1232 CordzInfo::MaybeUntrackCord(cordz_info());
1233 }
1234 absl::cord_internal::CordRep* result = tree();
1235 ResetToEmpty();
1236 return result;
1237 }
1238
CopyToArray(char * dst)1239 inline void Cord::InlineRep::CopyToArray(char* dst) const {
1240 assert(!is_tree());
1241 size_t n = inline_size();
1242 assert(n != 0);
1243 cord_internal::SmallMemmove(dst, data_.as_chars(), n);
1244 }
1245
MaybeRemoveEmptyCrcNode()1246 inline void Cord::InlineRep::MaybeRemoveEmptyCrcNode() {
1247 CordRep* rep = tree();
1248 if (rep == nullptr || ABSL_PREDICT_TRUE(rep->length > 0)) {
1249 return;
1250 }
1251 assert(rep->IsCrc());
1252 assert(rep->crc()->child == nullptr);
1253 CordzInfo::MaybeUntrackCord(cordz_info());
1254 CordRep::Unref(rep);
1255 ResetToEmpty();
1256 }
1257
Cord()1258 constexpr inline Cord::Cord() noexcept {}
1259
Cord(absl::string_view src)1260 inline Cord::Cord(absl::string_view src)
1261 : Cord(src, CordzUpdateTracker::kConstructorString) {}
1262
1263 template <typename T>
Cord(strings_internal::StringConstant<T>)1264 constexpr Cord::Cord(strings_internal::StringConstant<T>)
1265 : contents_(strings_internal::StringConstant<T>::value.size() <=
1266 cord_internal::kMaxInline
1267 ? cord_internal::InlineData(
1268 strings_internal::StringConstant<T>::value)
1269 : cord_internal::InlineData(
1270 &cord_internal::ConstInitExternalStorage<
1271 strings_internal::StringConstant<T>>::value)) {}
1272
1273 inline Cord& Cord::operator=(const Cord& x) {
1274 contents_ = x.contents_;
1275 return *this;
1276 }
1277
1278 template <typename T, Cord::EnableIfString<T>>
1279 Cord& Cord::operator=(T&& src) {
1280 if (src.size() <= cord_internal::kMaxBytesToCopy) {
1281 return operator=(absl::string_view(src));
1282 } else {
1283 return AssignLargeString(std::forward<T>(src));
1284 }
1285 }
1286
Cord(const Cord & src)1287 inline Cord::Cord(const Cord& src) : contents_(src.contents_) {}
1288
Cord(Cord && src)1289 inline Cord::Cord(Cord&& src) noexcept : contents_(std::move(src.contents_)) {}
1290
swap(Cord & other)1291 inline void Cord::swap(Cord& other) noexcept {
1292 contents_.Swap(&other.contents_);
1293 }
1294
1295 inline Cord& Cord::operator=(Cord&& x) noexcept {
1296 contents_ = std::move(x.contents_);
1297 return *this;
1298 }
1299
1300 extern template Cord::Cord(std::string&& src);
1301
size()1302 inline size_t Cord::size() const {
1303 // Length is 1st field in str.rep_
1304 return contents_.size();
1305 }
1306
empty()1307 inline bool Cord::empty() const { return size() == 0; }
1308
EstimatedMemoryUsage(CordMemoryAccounting accounting_method)1309 inline size_t Cord::EstimatedMemoryUsage(
1310 CordMemoryAccounting accounting_method) const {
1311 size_t result = sizeof(Cord);
1312 if (const absl::cord_internal::CordRep* rep = contents_.tree()) {
1313 if (accounting_method == CordMemoryAccounting::kFairShare) {
1314 result += cord_internal::GetEstimatedFairShareMemoryUsage(rep);
1315 } else {
1316 result += cord_internal::GetEstimatedMemoryUsage(rep);
1317 }
1318 }
1319 return result;
1320 }
1321
TryFlat()1322 inline absl::optional<absl::string_view> Cord::TryFlat() const {
1323 absl::cord_internal::CordRep* rep = contents_.tree();
1324 if (rep == nullptr) {
1325 return absl::string_view(contents_.data(), contents_.size());
1326 }
1327 absl::string_view fragment;
1328 if (GetFlatAux(rep, &fragment)) {
1329 return fragment;
1330 }
1331 return absl::nullopt;
1332 }
1333
Flatten()1334 inline absl::string_view Cord::Flatten() {
1335 absl::cord_internal::CordRep* rep = contents_.tree();
1336 if (rep == nullptr) {
1337 return absl::string_view(contents_.data(), contents_.size());
1338 } else {
1339 absl::string_view already_flat_contents;
1340 if (GetFlatAux(rep, &already_flat_contents)) {
1341 return already_flat_contents;
1342 }
1343 }
1344 return FlattenSlowPath();
1345 }
1346
Append(absl::string_view src)1347 inline void Cord::Append(absl::string_view src) {
1348 contents_.AppendArray(src, CordzUpdateTracker::kAppendString);
1349 }
1350
Prepend(absl::string_view src)1351 inline void Cord::Prepend(absl::string_view src) {
1352 PrependArray(src, CordzUpdateTracker::kPrependString);
1353 }
1354
Append(CordBuffer buffer)1355 inline void Cord::Append(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_.AppendTree(rep, CordzUpdateTracker::kAppendCordBuffer);
1360 } else {
1361 AppendPrecise(short_value, CordzUpdateTracker::kAppendCordBuffer);
1362 }
1363 }
1364
Prepend(CordBuffer buffer)1365 inline void Cord::Prepend(CordBuffer buffer) {
1366 if (ABSL_PREDICT_FALSE(buffer.length() == 0)) return;
1367 absl::string_view short_value;
1368 if (CordRep* rep = buffer.ConsumeValue(short_value)) {
1369 contents_.PrependTree(rep, CordzUpdateTracker::kPrependCordBuffer);
1370 } else {
1371 PrependPrecise(short_value, CordzUpdateTracker::kPrependCordBuffer);
1372 }
1373 }
1374
GetAppendBuffer(size_t capacity,size_t min_capacity)1375 inline CordBuffer Cord::GetAppendBuffer(size_t capacity, size_t min_capacity) {
1376 if (empty()) return CordBuffer::CreateWithDefaultLimit(capacity);
1377 return GetAppendBufferSlowPath(0, capacity, min_capacity);
1378 }
1379
GetCustomAppendBuffer(size_t block_size,size_t capacity,size_t min_capacity)1380 inline CordBuffer Cord::GetCustomAppendBuffer(size_t block_size,
1381 size_t capacity,
1382 size_t min_capacity) {
1383 if (empty()) {
1384 return block_size ? CordBuffer::CreateWithCustomLimit(block_size, capacity)
1385 : CordBuffer::CreateWithDefaultLimit(capacity);
1386 }
1387 return GetAppendBufferSlowPath(block_size, capacity, min_capacity);
1388 }
1389
1390 extern template void Cord::Append(std::string&& src);
1391 extern template void Cord::Prepend(std::string&& src);
1392
Compare(const Cord & rhs)1393 inline int Cord::Compare(const Cord& rhs) const {
1394 if (!contents_.is_tree() && !rhs.contents_.is_tree()) {
1395 return contents_.data_.Compare(rhs.contents_.data_);
1396 }
1397
1398 return CompareImpl(rhs);
1399 }
1400
1401 // Does 'this' cord start/end with rhs
StartsWith(const Cord & rhs)1402 inline bool Cord::StartsWith(const Cord& rhs) const {
1403 if (contents_.IsSame(rhs.contents_)) return true;
1404 size_t rhs_size = rhs.size();
1405 if (size() < rhs_size) return false;
1406 return EqualsImpl(rhs, rhs_size);
1407 }
1408
StartsWith(absl::string_view rhs)1409 inline bool Cord::StartsWith(absl::string_view rhs) const {
1410 size_t rhs_size = rhs.size();
1411 if (size() < rhs_size) return false;
1412 return EqualsImpl(rhs, rhs_size);
1413 }
1414
InitTree(cord_internal::CordRep * tree)1415 inline void Cord::ChunkIterator::InitTree(cord_internal::CordRep* tree) {
1416 tree = cord_internal::SkipCrcNode(tree);
1417 if (tree->tag == cord_internal::BTREE) {
1418 current_chunk_ = btree_reader_.Init(tree->btree());
1419 } else {
1420 current_leaf_ = tree;
1421 current_chunk_ = cord_internal::EdgeData(tree);
1422 }
1423 }
1424
ChunkIterator(cord_internal::CordRep * tree)1425 inline Cord::ChunkIterator::ChunkIterator(cord_internal::CordRep* tree) {
1426 bytes_remaining_ = tree->length;
1427 InitTree(tree);
1428 }
1429
ChunkIterator(const Cord * cord)1430 inline Cord::ChunkIterator::ChunkIterator(const Cord* cord) {
1431 if (CordRep* tree = cord->contents_.tree()) {
1432 bytes_remaining_ = tree->length;
1433 if (ABSL_PREDICT_TRUE(bytes_remaining_ != 0)) {
1434 InitTree(tree);
1435 } else {
1436 current_chunk_ = {};
1437 }
1438 } else {
1439 bytes_remaining_ = cord->contents_.inline_size();
1440 current_chunk_ = {cord->contents_.data(), bytes_remaining_};
1441 }
1442 }
1443
AdvanceBtree()1444 inline Cord::ChunkIterator& Cord::ChunkIterator::AdvanceBtree() {
1445 current_chunk_ = btree_reader_.Next();
1446 return *this;
1447 }
1448
AdvanceBytesBtree(size_t n)1449 inline void Cord::ChunkIterator::AdvanceBytesBtree(size_t n) {
1450 assert(n >= current_chunk_.size());
1451 bytes_remaining_ -= n;
1452 if (bytes_remaining_) {
1453 if (n == current_chunk_.size()) {
1454 current_chunk_ = btree_reader_.Next();
1455 } else {
1456 size_t offset = btree_reader_.length() - bytes_remaining_;
1457 current_chunk_ = btree_reader_.Seek(offset);
1458 }
1459 } else {
1460 current_chunk_ = {};
1461 }
1462 }
1463
1464 inline Cord::ChunkIterator& Cord::ChunkIterator::operator++() {
1465 ABSL_HARDENING_ASSERT(bytes_remaining_ > 0 &&
1466 "Attempted to iterate past `end()`");
1467 assert(bytes_remaining_ >= current_chunk_.size());
1468 bytes_remaining_ -= current_chunk_.size();
1469 if (bytes_remaining_ > 0) {
1470 if (btree_reader_) {
1471 return AdvanceBtree();
1472 } else {
1473 assert(!current_chunk_.empty()); // Called on invalid iterator.
1474 }
1475 current_chunk_ = {};
1476 }
1477 return *this;
1478 }
1479
1480 inline Cord::ChunkIterator Cord::ChunkIterator::operator++(int) {
1481 ChunkIterator tmp(*this);
1482 operator++();
1483 return tmp;
1484 }
1485
1486 inline bool Cord::ChunkIterator::operator==(const ChunkIterator& other) const {
1487 return bytes_remaining_ == other.bytes_remaining_;
1488 }
1489
1490 inline bool Cord::ChunkIterator::operator!=(const ChunkIterator& other) const {
1491 return !(*this == other);
1492 }
1493
1494 inline Cord::ChunkIterator::reference Cord::ChunkIterator::operator*() const {
1495 ABSL_HARDENING_ASSERT(bytes_remaining_ != 0);
1496 return current_chunk_;
1497 }
1498
1499 inline Cord::ChunkIterator::pointer Cord::ChunkIterator::operator->() const {
1500 ABSL_HARDENING_ASSERT(bytes_remaining_ != 0);
1501 return ¤t_chunk_;
1502 }
1503
RemoveChunkPrefix(size_t n)1504 inline void Cord::ChunkIterator::RemoveChunkPrefix(size_t n) {
1505 assert(n < current_chunk_.size());
1506 current_chunk_.remove_prefix(n);
1507 bytes_remaining_ -= n;
1508 }
1509
AdvanceBytes(size_t n)1510 inline void Cord::ChunkIterator::AdvanceBytes(size_t n) {
1511 assert(bytes_remaining_ >= n);
1512 if (ABSL_PREDICT_TRUE(n < current_chunk_.size())) {
1513 RemoveChunkPrefix(n);
1514 } else if (n != 0) {
1515 if (btree_reader_) {
1516 AdvanceBytesBtree(n);
1517 } else {
1518 bytes_remaining_ = 0;
1519 }
1520 }
1521 }
1522
chunk_begin()1523 inline Cord::ChunkIterator Cord::chunk_begin() const {
1524 return ChunkIterator(this);
1525 }
1526
chunk_end()1527 inline Cord::ChunkIterator Cord::chunk_end() const { return ChunkIterator(); }
1528
begin()1529 inline Cord::ChunkIterator Cord::ChunkRange::begin() const {
1530 return cord_->chunk_begin();
1531 }
1532
end()1533 inline Cord::ChunkIterator Cord::ChunkRange::end() const {
1534 return cord_->chunk_end();
1535 }
1536
Chunks()1537 inline Cord::ChunkRange Cord::Chunks() const { return ChunkRange(this); }
1538
1539 inline Cord::CharIterator& Cord::CharIterator::operator++() {
1540 if (ABSL_PREDICT_TRUE(chunk_iterator_->size() > 1)) {
1541 chunk_iterator_.RemoveChunkPrefix(1);
1542 } else {
1543 ++chunk_iterator_;
1544 }
1545 return *this;
1546 }
1547
1548 inline Cord::CharIterator Cord::CharIterator::operator++(int) {
1549 CharIterator tmp(*this);
1550 operator++();
1551 return tmp;
1552 }
1553
1554 inline bool Cord::CharIterator::operator==(const CharIterator& other) const {
1555 return chunk_iterator_ == other.chunk_iterator_;
1556 }
1557
1558 inline bool Cord::CharIterator::operator!=(const CharIterator& other) const {
1559 return !(*this == other);
1560 }
1561
1562 inline Cord::CharIterator::reference Cord::CharIterator::operator*() const {
1563 return *chunk_iterator_->data();
1564 }
1565
1566 inline Cord::CharIterator::pointer Cord::CharIterator::operator->() const {
1567 return chunk_iterator_->data();
1568 }
1569
AdvanceAndRead(CharIterator * it,size_t n_bytes)1570 inline Cord Cord::AdvanceAndRead(CharIterator* it, size_t n_bytes) {
1571 assert(it != nullptr);
1572 return it->chunk_iterator_.AdvanceAndReadBytes(n_bytes);
1573 }
1574
Advance(CharIterator * it,size_t n_bytes)1575 inline void Cord::Advance(CharIterator* it, size_t n_bytes) {
1576 assert(it != nullptr);
1577 it->chunk_iterator_.AdvanceBytes(n_bytes);
1578 }
1579
ChunkRemaining(const CharIterator & it)1580 inline absl::string_view Cord::ChunkRemaining(const CharIterator& it) {
1581 return *it.chunk_iterator_;
1582 }
1583
char_begin()1584 inline Cord::CharIterator Cord::char_begin() const {
1585 return CharIterator(this);
1586 }
1587
char_end()1588 inline Cord::CharIterator Cord::char_end() const { return CharIterator(); }
1589
begin()1590 inline Cord::CharIterator Cord::CharRange::begin() const {
1591 return cord_->char_begin();
1592 }
1593
end()1594 inline Cord::CharIterator Cord::CharRange::end() const {
1595 return cord_->char_end();
1596 }
1597
Chars()1598 inline Cord::CharRange Cord::Chars() const { return CharRange(this); }
1599
ForEachChunk(absl::FunctionRef<void (absl::string_view)> callback)1600 inline void Cord::ForEachChunk(
1601 absl::FunctionRef<void(absl::string_view)> callback) const {
1602 absl::cord_internal::CordRep* rep = contents_.tree();
1603 if (rep == nullptr) {
1604 callback(absl::string_view(contents_.data(), contents_.size()));
1605 } else {
1606 ForEachChunkAux(rep, callback);
1607 }
1608 }
1609
1610 // Nonmember Cord-to-Cord relational operators.
1611 inline bool operator==(const Cord& lhs, const Cord& rhs) {
1612 if (lhs.contents_.IsSame(rhs.contents_)) return true;
1613 size_t rhs_size = rhs.size();
1614 if (lhs.size() != rhs_size) return false;
1615 return lhs.EqualsImpl(rhs, rhs_size);
1616 }
1617
1618 inline bool operator!=(const Cord& x, const Cord& y) { return !(x == y); }
1619 inline bool operator<(const Cord& x, const Cord& y) { return x.Compare(y) < 0; }
1620 inline bool operator>(const Cord& x, const Cord& y) { return x.Compare(y) > 0; }
1621 inline bool operator<=(const Cord& x, const Cord& y) {
1622 return x.Compare(y) <= 0;
1623 }
1624 inline bool operator>=(const Cord& x, const Cord& y) {
1625 return x.Compare(y) >= 0;
1626 }
1627
1628 // Nonmember Cord-to-absl::string_view relational operators.
1629 //
1630 // Due to implicit conversions, these also enable comparisons of Cord with
1631 // with std::string, ::string, and const char*.
1632 inline bool operator==(const Cord& lhs, absl::string_view rhs) {
1633 size_t lhs_size = lhs.size();
1634 size_t rhs_size = rhs.size();
1635 if (lhs_size != rhs_size) return false;
1636 return lhs.EqualsImpl(rhs, rhs_size);
1637 }
1638
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 !(x == y); }
1641 inline bool operator!=(absl::string_view x, const Cord& y) { return !(x == y); }
1642 inline bool operator<(const Cord& x, absl::string_view y) {
1643 return x.Compare(y) < 0;
1644 }
1645 inline bool operator<(absl::string_view x, const Cord& y) {
1646 return y.Compare(x) > 0;
1647 }
1648 inline bool operator>(const Cord& x, absl::string_view y) { return y < x; }
1649 inline bool operator>(absl::string_view x, const Cord& y) { return y < x; }
1650 inline bool operator<=(const Cord& x, absl::string_view y) { return !(y < x); }
1651 inline bool operator<=(absl::string_view x, const Cord& y) { return !(y < x); }
1652 inline bool operator>=(const Cord& x, absl::string_view y) { return !(x < y); }
1653 inline bool operator>=(absl::string_view x, const Cord& y) { return !(x < y); }
1654
1655 // Some internals exposed to test code.
1656 namespace strings_internal {
1657 class CordTestAccess {
1658 public:
1659 static size_t FlatOverhead();
1660 static size_t MaxFlatLength();
1661 static size_t SizeofCordRepExternal();
1662 static size_t SizeofCordRepSubstring();
1663 static size_t FlatTagToLength(uint8_t tag);
1664 static uint8_t LengthToTag(size_t s);
1665 };
1666 } // namespace strings_internal
1667 ABSL_NAMESPACE_END
1668 } // namespace absl
1669
1670 #endif // ABSL_STRINGS_CORD_H_
1671