| 1 |
/** |
| 2 |
* MIT License |
| 3 |
* |
| 4 |
* Copyright (c) 2017 Thibaut Goetghebuer-Planchon <[email protected]> |
| 5 |
* |
| 6 |
* Permission is hereby granted, free of charge, to any person obtaining a copy |
| 7 |
* of this software and associated documentation files (the "Software"), to deal |
| 8 |
* in the Software without restriction, including without limitation the rights |
| 9 |
* to use, copy, modify, merge, publish, distribute, sublicense, and/or sell |
| 10 |
* copies of the Software, and to permit persons to whom the Software is |
| 11 |
* furnished to do so, subject to the following conditions: |
| 12 |
* |
| 13 |
* The above copyright notice and this permission notice shall be included in all |
| 14 |
* copies or substantial portions of the Software. |
| 15 |
* |
| 16 |
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR |
| 17 |
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, |
| 18 |
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE |
| 19 |
* AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER |
| 20 |
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, |
| 21 |
* OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE |
| 22 |
* SOFTWARE. |
| 23 |
*/ |
| 24 |
#ifndef TSL_ARRAY_HASH_H |
| 25 |
#define TSL_ARRAY_HASH_H |
| 26 |
|
| 27 |
#include <algorithm> |
| 28 |
#include <cassert> |
| 29 |
#include <cmath> |
| 30 |
#include <cstddef> |
| 31 |
#include <cstdint> |
| 32 |
#include <cstdlib> |
| 33 |
#include <cstring> |
| 34 |
#include <iterator> |
| 35 |
#include <limits> |
| 36 |
#include <memory> |
| 37 |
#include <stdexcept> |
| 38 |
#include <type_traits> |
| 39 |
#include <utility> |
| 40 |
#include <vector> |
| 41 |
#include "array_growth_policy.h" |
| 42 |
|
| 43 |
|
| 44 |
/* |
| 45 |
* __has_include is a bit useless (https://gcc.gnu.org/bugzilla/show_bug.cgi?id=79433), |
| 46 |
* check also __cplusplus version. |
| 47 |
*/ |
| 48 |
#ifdef __has_include |
| 49 |
# if __has_include(<string_view>) && __cplusplus >= 201703L |
| 50 |
# define TSL_AH_HAS_STRING_VIEW |
| 51 |
# endif |
| 52 |
#endif |
| 53 |
|
| 54 |
|
| 55 |
#ifdef TSL_AH_HAS_STRING_VIEW |
| 56 |
# include <string_view> |
| 57 |
#endif |
| 58 |
|
| 59 |
|
| 60 |
#ifdef TSL_DEBUG |
| 61 |
# define tsl_ah_assert(expr) assert(expr) |
| 62 |
#else |
| 63 |
# define tsl_ah_assert(expr) (static_cast<void>(0)) |
| 64 |
#endif |
| 65 |
|
| 66 |
|
| 67 |
|
| 68 |
/** |
| 69 |
* Implementation of the array hash structure described in the |
| 70 |
* "Cache-conscious collision resolution in string hash tables." (Askitis Nikolas and Justin Zobel, 2005) paper. |
| 71 |
*/ |
| 72 |
namespace tsl { |
| 73 |
|
| 74 |
namespace ah { |
| 75 |
|
| 76 |
template<class CharT> |
| 77 |
struct str_hash { |
| 78 |
#ifdef TSL_AH_HAS_STRING_VIEW |
| 79 |
std::size_t operator()(const CharT* key, std::size_t key_size) const { |
| 80 |
return std::hash<std::basic_string_view<CharT>>()(std::basic_string_view<CharT>(key, key_size)); |
| 81 |
} |
| 82 |
#else |
| 83 |
/** |
| 84 |
* FNV-1a hash |
| 85 |
*/ |
| 86 |
std::size_t operator()(const CharT* key, std::size_t key_size) const { |
| 87 |
static const std::size_t init = std::size_t((sizeof(std::size_t) == 8)?0xcbf29ce484222325:0x811c9dc5); |
| 88 |
static const std::size_t multiplier = std::size_t((sizeof(std::size_t) == 8)?0x100000001b3:0x1000193); |
| 89 |
|
| 90 |
std::size_t hash = init; |
| 91 |
for (std::size_t i = 0; i < key_size; ++i) { |
| 92 |
hash ^= key[i]; |
| 93 |
hash *= multiplier; |
| 94 |
} |
| 95 |
|
| 96 |
return hash; |
| 97 |
} |
| 98 |
#endif |
| 99 |
}; |
| 100 |
|
| 101 |
template<class CharT> |
| 102 |
struct str_equal { |
| 103 |
bool operator()(const CharT* key_lhs, std::size_t key_size_lhs, |
| 104 |
const CharT* key_rhs, std::size_t key_size_rhs) const |
| 105 |
{ |
| 106 |
if(key_size_lhs != key_size_rhs) { |
| 107 |
return false; |
| 108 |
} |
| 109 |
else { |
| 110 |
return std::memcmp(key_lhs, key_rhs, key_size_lhs * sizeof(CharT)) == 0; |
| 111 |
} |
| 112 |
} |
| 113 |
}; |
| 114 |
} |
| 115 |
|
| 116 |
|
| 117 |
namespace detail_array_hash { |
| 118 |
|
| 119 |
template<typename T, typename = void> |
| 120 |
struct is_iterator: std::false_type { |
| 121 |
}; |
| 122 |
|
| 123 |
template<typename T> |
| 124 |
struct is_iterator<T, typename std::enable_if<!std::is_same<typename std::iterator_traits<T>::iterator_category, void>::value>::type>: std::true_type { |
| 125 |
}; |
| 126 |
|
| 127 |
static constexpr bool is_power_of_two(std::size_t value) { |
| 128 |
return value != 0 && (value & (value - 1)) == 0; |
| 129 |
} |
| 130 |
|
| 131 |
template<typename T, typename U> |
| 132 |
static T numeric_cast(U value, const char* error_message = "numeric_cast() failed.") { |
| 133 |
T ret = static_cast<T>(value); |
| 134 |
if(static_cast<U>(ret) != value) { |
| 135 |
THROW(std::runtime_error, error_message); |
| 136 |
} |
| 137 |
|
| 138 |
const bool is_same_signedness = (std::is_unsigned<T>::value && std::is_unsigned<U>::value) || |
| 139 |
(std::is_signed<T>::value && std::is_signed<U>::value); |
| 140 |
if(!is_same_signedness && (ret < T{}) != (value < U{})) { |
| 141 |
THROW(std::runtime_error, error_message); |
| 142 |
} |
| 143 |
|
| 144 |
return ret; |
| 145 |
} |
| 146 |
|
| 147 |
|
| 148 |
|
| 149 |
/** |
| 150 |
* Fixed size type used to represent size_type values on serialization. Need to be big enough |
| 151 |
* to represent a std::size_t on 32 and 64 bits platforms, and must be the same size on both platforms. |
| 152 |
*/ |
| 153 |
using slz_size_type = std::uint64_t; |
| 154 |
|
| 155 |
template<class T, class Deserializer> |
| 156 |
static T deserialize_value(Deserializer& deserializer) { |
| 157 |
// MSVC < 2017 is not conformant, circumvent the problem by removing the template keyword |
| 158 |
#if defined (_MSC_VER) && _MSC_VER < 1910 |
| 159 |
return deserializer.Deserializer::operator()<T>(); |
| 160 |
#else |
| 161 |
return deserializer.Deserializer::template operator()<T>(); |
| 162 |
#endif |
| 163 |
} |
| 164 |
|
| 165 |
/** |
| 166 |
* For each string in the bucket, store the size of the string, the chars of the string |
| 167 |
* and T, if it's not void. T should be either void or an unsigned type. |
| 168 |
* |
| 169 |
* End the buffer with END_OF_BUCKET flag. END_OF_BUCKET has the same type as the string size variable. |
| 170 |
* |
| 171 |
* m_buffer (CharT*): |
| 172 |
* | size of str1 (KeySizeT) | str1 (const CharT*) | value (T if T != void) | ... | |
| 173 |
* | size of strN (KeySizeT) | strN (const CharT*) | value (T if T != void) | END_OF_BUCKET (KeySizeT) | |
| 174 |
* |
| 175 |
* m_buffer is null if there is no string in the bucket. |
| 176 |
* |
| 177 |
* KeySizeT and T are extended to be a multiple of CharT when stored in the buffer. |
| 178 |
* |
| 179 |
* Use std::malloc and std::free instead of new and delete so we can have access to std::realloc. |
| 180 |
*/ |
| 181 |
template<class CharT, |
| 182 |
class T, |
| 183 |
class KeyEqual, |
| 184 |
class KeySizeT, |
| 185 |
bool StoreNullTerminator> |
| 186 |
class array_bucket { |
| 187 |
template<typename U> |
| 188 |
using has_mapped_type = typename std::integral_constant<bool, !std::is_same<U, void>::value>; |
| 189 |
|
| 190 |
static_assert(!has_mapped_type<T>::value || std::is_unsigned<T>::value, |
| 191 |
"T should be either void or an unsigned type."); |
| 192 |
|
| 193 |
static_assert(std::is_unsigned<KeySizeT>::value, "KeySizeT should be an unsigned type."); |
| 194 |
|
| 195 |
public: |
| 196 |
template<bool IsConst> |
| 197 |
class array_bucket_iterator; |
| 198 |
|
| 199 |
using char_type = CharT; |
| 200 |
using key_size_type = KeySizeT; |
| 201 |
using mapped_type = T; |
| 202 |
using size_type = std::size_t; |
| 203 |
using key_equal = KeyEqual; |
| 204 |
using iterator = array_bucket_iterator<false>; |
| 205 |
using const_iterator = array_bucket_iterator<true>; |
| 206 |
|
| 207 |
static_assert(sizeof(KeySizeT) <= sizeof(size_type), "sizeof(KeySizeT) should be <= sizeof(std::size_t;)"); |
| 208 |
static_assert(std::is_unsigned<size_type>::value, ""); |
| 209 |
|
| 210 |
private: |
| 211 |
/** |
| 212 |
* Return how much space in bytes the type U will take when stored in the buffer. |
| 213 |
* As the buffer is of type CharT, U may take more space than sizeof(U). |
| 214 |
* |
| 215 |
* Example: sizeof(CharT) = 4, sizeof(U) = 2 => U will take 4 bytes in the buffer instead of 2. |
| 216 |
*/ |
| 217 |
template<typename U> |
| 218 |
static constexpr size_type sizeof_in_buff() noexcept { |
| 219 |
static_assert(is_power_of_two(sizeof(U)), "sizeof(U) should be a power of two."); |
| 220 |
static_assert(is_power_of_two(sizeof(CharT)), "sizeof(CharT) should be a power of two."); |
| 221 |
|
| 222 |
return std::max(sizeof(U), sizeof(CharT)); |
| 223 |
} |
| 224 |
|
| 225 |
/** |
| 226 |
* Same as sizeof_in_buff<U>, but instead of returning the size in bytes return it in term of sizeof(CharT). |
| 227 |
*/ |
| 228 |
template<typename U> |
| 229 |
static constexpr size_type size_as_char_t() noexcept { |
| 230 |
return sizeof_in_buff<U>() / sizeof(CharT); |
| 231 |
} |
| 232 |
|
| 233 |
static key_size_type read_key_size(const CharT* buffer) noexcept { |
| 234 |
key_size_type key_size; |
| 235 |
std::memcpy(&key_size, buffer, sizeof(key_size)); |
| 236 |
|
| 237 |
return key_size; |
| 238 |
} |
| 239 |
|
| 240 |
static mapped_type read_value(const CharT* buffer) noexcept { |
| 241 |
mapped_type value; |
| 242 |
std::memcpy(&value, buffer, sizeof(value)); |
| 243 |
|
| 244 |
return value; |
| 245 |
} |
| 246 |
|
| 247 |
static bool is_end_of_bucket(const CharT* buffer) noexcept { |
| 248 |
return read_key_size(buffer) == END_OF_BUCKET; |
| 249 |
} |
| 250 |
|
| 251 |
public: |
| 252 |
/** |
| 253 |
* Return the size required for an entry with a key of size 'key_size'. |
| 254 |
*/ |
| 255 |
template<class U = T, typename std::enable_if<!has_mapped_type<U>::value>::type* = nullptr> |
| 256 |
static size_type entry_required_bytes(size_type key_size) noexcept { |
| 257 |
return sizeof_in_buff<key_size_type>() + (key_size + KEY_EXTRA_SIZE) * sizeof(CharT); |
| 258 |
} |
| 259 |
|
| 260 |
template<class U = T, typename std::enable_if<has_mapped_type<U>::value>::type* = nullptr> |
| 261 |
static size_type entry_required_bytes(size_type key_size) noexcept { |
| 262 |
return sizeof_in_buff<key_size_type>() + (key_size + KEY_EXTRA_SIZE) * sizeof(CharT) + |
| 263 |
sizeof_in_buff<mapped_type>(); |
| 264 |
} |
| 265 |
|
| 266 |
private: |
| 267 |
/** |
| 268 |
* Return the size of the current entry in buffer. |
| 269 |
*/ |
| 270 |
static size_type entry_size_bytes(const CharT* buffer) noexcept { |
| 271 |
return entry_required_bytes(read_key_size(buffer)); |
| 272 |
} |
| 273 |
|
| 274 |
public: |
| 275 |
template<bool IsConst> |
| 276 |
class array_bucket_iterator { |
| 277 |
friend class array_bucket; |
| 278 |
|
| 279 |
using buffer_type = typename std::conditional<IsConst, const CharT, CharT>::type; |
| 280 |
|
| 281 |
explicit array_bucket_iterator(buffer_type* position) noexcept: m_position(position) { |
| 282 |
} |
| 283 |
|
| 284 |
public: |
| 285 |
using iterator_category = std::forward_iterator_tag; |
| 286 |
using value_type = void; |
| 287 |
using difference_type = std::ptrdiff_t; |
| 288 |
using reference = void; |
| 289 |
using pointer = void; |
| 290 |
|
| 291 |
public: |
| 292 |
array_bucket_iterator() noexcept: m_position(nullptr) { |
| 293 |
} |
| 294 |
|
| 295 |
const CharT* key() const { |
| 296 |
return m_position + size_as_char_t<key_size_type>(); |
| 297 |
} |
| 298 |
|
| 299 |
size_type key_size() const { |
| 300 |
return read_key_size(m_position); |
| 301 |
} |
| 302 |
|
| 303 |
template<class U = T, typename std::enable_if<has_mapped_type<U>::value>::type* = nullptr> |
| 304 |
U value() const { |
| 305 |
return read_value(m_position + size_as_char_t<key_size_type>() + key_size() + KEY_EXTRA_SIZE); |
| 306 |
} |
| 307 |
|
| 308 |
|
| 309 |
template<class U = T, typename std::enable_if<has_mapped_type<U>::value && !IsConst && std::is_same<U, T>::value>::type* = nullptr> |
| 310 |
void set_value(U value) noexcept { |
| 311 |
std::memcpy(m_position + size_as_char_t<key_size_type>() + key_size() + KEY_EXTRA_SIZE, |
| 312 |
&value, sizeof(value)); |
| 313 |
} |
| 314 |
|
| 315 |
array_bucket_iterator& operator++() { |
| 316 |
m_position += entry_size_bytes(m_position)/sizeof(CharT); |
| 317 |
if(is_end_of_bucket(m_position)) { |
| 318 |
m_position = nullptr; |
| 319 |
} |
| 320 |
|
| 321 |
return *this; |
| 322 |
} |
| 323 |
|
| 324 |
array_bucket_iterator operator++(int) { |
| 325 |
array_bucket_iterator tmp(*this); |
| 326 |
++*this; |
| 327 |
|
| 328 |
return tmp; |
| 329 |
} |
| 330 |
|
| 331 |
friend bool operator==(const array_bucket_iterator& lhs, const array_bucket_iterator& rhs) { |
| 332 |
return lhs.m_position == rhs.m_position; |
| 333 |
} |
| 334 |
|
| 335 |
friend bool operator!=(const array_bucket_iterator& lhs, const array_bucket_iterator& rhs) { |
| 336 |
return !(lhs == rhs); |
| 337 |
} |
| 338 |
|
| 339 |
private: |
| 340 |
buffer_type* m_position; |
| 341 |
}; |
| 342 |
|
| 343 |
|
| 344 |
|
| 345 |
static iterator end_it() noexcept { |
| 346 |
return iterator(nullptr); |
| 347 |
} |
| 348 |
|
| 349 |
static const_iterator cend_it() noexcept { |
| 350 |
return const_iterator(nullptr); |
| 351 |
} |
| 352 |
|
| 353 |
public: |
| 354 |
array_bucket(): m_buffer(nullptr) { |
| 355 |
} |
| 356 |
|
| 357 |
/** |
| 358 |
* Reserve 'size' in the buffer of the bucket. The created bucket is empty. |
| 359 |
*/ |
| 360 |
array_bucket(std::size_t size): m_buffer(nullptr) { |
| 361 |
if(size == 0) { |
| 362 |
return; |
| 363 |
} |
| 364 |
|
| 365 |
m_buffer = static_cast<CharT*>(std::malloc(size*sizeof(CharT) + sizeof_in_buff<decltype(END_OF_BUCKET)>())); |
| 366 |
if(m_buffer == nullptr) { |
| 367 |
THROW(std::runtime_error, "Out of memory"); |
| 368 |
} |
| 369 |
|
| 370 |
const auto end_of_bucket = END_OF_BUCKET; |
| 371 |
std::memcpy(m_buffer, &end_of_bucket, sizeof(end_of_bucket)); |
| 372 |
} |
| 373 |
|
| 374 |
~array_bucket() { |
| 375 |
clear(); |
| 376 |
} |
| 377 |
|
| 378 |
array_bucket(const array_bucket& other) { |
| 379 |
if(other.m_buffer == nullptr) { |
| 380 |
m_buffer = nullptr; |
| 381 |
return; |
| 382 |
} |
| 383 |
|
| 384 |
const size_type other_buffer_size = other.size(); |
| 385 |
m_buffer = static_cast<CharT*>(std::malloc(other_buffer_size*sizeof(CharT) + sizeof_in_buff<decltype(END_OF_BUCKET)>())); |
| 386 |
if(m_buffer == nullptr) { |
| 387 |
THROW(std::runtime_error, "Out of memory"); |
| 388 |
} |
| 389 |
|
| 390 |
std::memcpy(m_buffer, other.m_buffer, other_buffer_size*sizeof(CharT)); |
| 391 |
|
| 392 |
const auto end_of_bucket = END_OF_BUCKET; |
| 393 |
std::memcpy(m_buffer + other_buffer_size, &end_of_bucket, sizeof(end_of_bucket)); |
| 394 |
} |
| 395 |
|
| 396 |
array_bucket(array_bucket&& other) noexcept: m_buffer(other.m_buffer) { |
| 397 |
other.m_buffer = nullptr; |
| 398 |
} |
| 399 |
|
| 400 |
array_bucket& operator=(array_bucket other) noexcept { |
| 401 |
other.swap(*this); |
| 402 |
|
| 403 |
return *this; |
| 404 |
} |
| 405 |
|
| 406 |
void swap(array_bucket& other) noexcept { |
| 407 |
std::swap(m_buffer, other.m_buffer); |
| 408 |
} |
| 409 |
|
| 410 |
iterator begin() noexcept { return iterator(m_buffer); } |
| 411 |
iterator end() noexcept { return iterator(nullptr); } |
| 412 |
const_iterator begin() const noexcept { return cbegin(); } |
| 413 |
const_iterator end() const noexcept { return cend(); } |
| 414 |
const_iterator cbegin() const noexcept { return const_iterator(m_buffer); } |
| 415 |
const_iterator cend() const noexcept { return const_iterator(nullptr); } |
| 416 |
|
| 417 |
/** |
| 418 |
* Return an iterator pointing to the key entry if presents or, if not there, to the position |
| 419 |
* past the last element of the bucket. Return end() if the bucket has not be initialized yet. |
| 420 |
* |
| 421 |
* The boolean of the pair is set to true if the key is there, false otherwise. |
| 422 |
*/ |
| 423 |
std::pair<const_iterator, bool> find_or_end_of_bucket(const CharT* key, size_type key_size) const noexcept { |
| 424 |
if(m_buffer == nullptr) { |
| 425 |
return std::make_pair(cend(), false); |
| 426 |
} |
| 427 |
|
| 428 |
const CharT* buffer_ptr_in_out = m_buffer; |
| 429 |
const bool found = find_or_end_of_bucket_impl(key, key_size, buffer_ptr_in_out); |
| 430 |
|
| 431 |
return std::make_pair(const_iterator(buffer_ptr_in_out), found); |
| 432 |
} |
| 433 |
|
| 434 |
/** |
| 435 |
* Append the element 'key' with its potential value at the end of the bucket. |
| 436 |
* 'end_of_bucket' should point past the end of the last element in the bucket, end() if the bucket |
| 437 |
* was not initialized yet. You usually get this value from find_or_end_of_bucket. |
| 438 |
* |
| 439 |
* Return the position where the element was actually inserted. |
| 440 |
*/ |
| 441 |
template<class... ValueArgs> |
| 442 |
const_iterator append(const_iterator end_of_bucket, const CharT* key, size_type key_size, |
| 443 |
ValueArgs&&... value) |
| 444 |
{ |
| 445 |
const key_size_type key_sz = as_key_size_type(key_size); |
| 446 |
|
| 447 |
if(end_of_bucket == cend()) { |
| 448 |
tsl_ah_assert(m_buffer == nullptr); |
| 449 |
|
| 450 |
const size_type buffer_size = entry_required_bytes(key_sz) + sizeof_in_buff<decltype(END_OF_BUCKET)>(); |
| 451 |
|
| 452 |
m_buffer = static_cast<CharT*>(std::malloc(buffer_size)); |
| 453 |
if(m_buffer == nullptr) { |
| 454 |
THROW(std::runtime_error, "Out of memory"); |
| 455 |
} |
| 456 |
|
| 457 |
append_impl(key, key_sz, m_buffer, std::forward<ValueArgs>(value)...); |
| 458 |
|
| 459 |
return const_iterator(m_buffer); |
| 460 |
} |
| 461 |
else { |
| 462 |
tsl_ah_assert(is_end_of_bucket(end_of_bucket.m_position)); |
| 463 |
|
| 464 |
const size_type current_size = ((end_of_bucket.m_position + size_as_char_t<decltype(END_OF_BUCKET)>()) - |
| 465 |
m_buffer) * sizeof(CharT); |
| 466 |
const size_type new_size = current_size + entry_required_bytes(key_sz); |
| 467 |
|
| 468 |
|
| 469 |
CharT* new_buffer = static_cast<CharT*>(std::realloc(m_buffer, new_size)); |
| 470 |
if(new_buffer == nullptr) { |
| 471 |
THROW(std::runtime_error, "Out of memory"); |
| 472 |
} |
| 473 |
m_buffer = new_buffer; |
| 474 |
|
| 475 |
|
| 476 |
CharT* buffer_append_pos = m_buffer + current_size / sizeof(CharT) - |
| 477 |
size_as_char_t<decltype(END_OF_BUCKET)>(); |
| 478 |
append_impl(key, key_sz, buffer_append_pos, std::forward<ValueArgs>(value)...); |
| 479 |
|
| 480 |
return const_iterator(buffer_append_pos); |
| 481 |
} |
| 482 |
|
| 483 |
} |
| 484 |
|
| 485 |
const_iterator erase(const_iterator position) noexcept { |
| 486 |
tsl_ah_assert(position.m_position != nullptr && !is_end_of_bucket(position.m_position)); |
| 487 |
|
| 488 |
// get mutable pointers |
| 489 |
CharT* start_entry = m_buffer + (position.m_position - m_buffer); |
| 490 |
CharT* start_next_entry = start_entry + entry_size_bytes(start_entry) / sizeof(CharT); |
| 491 |
|
| 492 |
|
| 493 |
CharT* end_buffer_ptr = start_next_entry; |
| 494 |
while(!is_end_of_bucket(end_buffer_ptr)) { |
| 495 |
end_buffer_ptr += entry_size_bytes(end_buffer_ptr) / sizeof(CharT); |
| 496 |
} |
| 497 |
end_buffer_ptr += size_as_char_t<decltype(END_OF_BUCKET)>(); |
| 498 |
|
| 499 |
|
| 500 |
const size_type size_to_move = (end_buffer_ptr - start_next_entry) * sizeof(CharT); |
| 501 |
std::memmove(start_entry, start_next_entry, size_to_move); |
| 502 |
|
| 503 |
|
| 504 |
if(is_end_of_bucket(m_buffer)) { |
| 505 |
clear(); |
| 506 |
return cend(); |
| 507 |
} |
| 508 |
else if(is_end_of_bucket(start_entry)) { |
| 509 |
return cend(); |
| 510 |
} |
| 511 |
else { |
| 512 |
return const_iterator(start_entry); |
| 513 |
} |
| 514 |
} |
| 515 |
|
| 516 |
/** |
| 517 |
* Return true if an element has been erased |
| 518 |
*/ |
| 519 |
bool erase(const CharT* key, size_type key_size) noexcept { |
| 520 |
if(m_buffer == nullptr) { |
| 521 |
return false; |
| 522 |
} |
| 523 |
|
| 524 |
const CharT* entry_buffer_ptr_in_out = m_buffer; |
| 525 |
bool found = find_or_end_of_bucket_impl(key, key_size, entry_buffer_ptr_in_out); |
| 526 |
if(found) { |
| 527 |
erase(const_iterator(entry_buffer_ptr_in_out)); |
| 528 |
|
| 529 |
return true; |
| 530 |
} |
| 531 |
else { |
| 532 |
return false; |
| 533 |
} |
| 534 |
} |
| 535 |
|
| 536 |
/** |
| 537 |
* Bucket should be big enough and there is no check to see if the key already exists. |
| 538 |
* No check on key_size. |
| 539 |
*/ |
| 540 |
template<class... ValueArgs> |
| 541 |
void append_in_reserved_bucket_no_check(const CharT* key, size_type key_size, ValueArgs&&... value) noexcept { |
| 542 |
CharT* buffer_ptr = m_buffer; |
| 543 |
while(!is_end_of_bucket(buffer_ptr)) { |
| 544 |
buffer_ptr += entry_size_bytes(buffer_ptr)/sizeof(CharT); |
| 545 |
} |
| 546 |
|
| 547 |
append_impl(key, key_size_type(key_size), buffer_ptr, std::forward<ValueArgs>(value)...); |
| 548 |
} |
| 549 |
|
| 550 |
bool empty() const noexcept { |
| 551 |
return m_buffer == nullptr || is_end_of_bucket(m_buffer); |
| 552 |
} |
| 553 |
|
| 554 |
void clear() noexcept { |
| 555 |
std::free(m_buffer); |
| 556 |
m_buffer = nullptr; |
| 557 |
} |
| 558 |
|
| 559 |
iterator mutable_iterator(const_iterator pos) noexcept { |
| 560 |
return iterator(m_buffer + (pos.m_position - m_buffer)); |
| 561 |
} |
| 562 |
|
| 563 |
template<class Serializer> |
| 564 |
void serialize(Serializer& serializer) const { |
| 565 |
const slz_size_type bucket_size = size(); |
| 566 |
tsl_ah_assert(m_buffer != nullptr || bucket_size == 0); |
| 567 |
|
| 568 |
serializer(bucket_size); |
| 569 |
serializer(m_buffer, bucket_size); |
| 570 |
} |
| 571 |
|
| 572 |
template<class Deserializer> |
| 573 |
static array_bucket deserialize(Deserializer& deserializer) { |
| 574 |
array_bucket bucket; |
| 575 |
const slz_size_type bucket_size_ds = deserialize_value<slz_size_type>(deserializer); |
| 576 |
|
| 577 |
if(bucket_size_ds == 0) { |
| 578 |
return bucket; |
| 579 |
} |
| 580 |
|
| 581 |
const std::size_t bucket_size = numeric_cast<std::size_t>(bucket_size_ds, "Deserialized bucket_size is too big."); |
| 582 |
bucket.m_buffer = static_cast<CharT*>(std::malloc(bucket_size*sizeof(CharT) + sizeof_in_buff<decltype(END_OF_BUCKET)>())); |
| 583 |
if(bucket.m_buffer == nullptr) { |
| 584 |
THROW(std::runtime_error, "Out of memory"); |
| 585 |
} |
| 586 |
|
| 587 |
|
| 588 |
deserializer(bucket.m_buffer, bucket_size); |
| 589 |
|
| 590 |
const auto end_of_bucket = END_OF_BUCKET; |
| 591 |
std::memcpy(bucket.m_buffer + bucket_size, &end_of_bucket, sizeof(end_of_bucket)); |
| 592 |
|
| 593 |
|
| 594 |
tsl_ah_assert(bucket.size() == bucket_size); |
| 595 |
return bucket; |
| 596 |
} |
| 597 |
|
| 598 |
private: |
| 599 |
key_size_type as_key_size_type(size_type key_size) const { |
| 600 |
if(key_size > MAX_KEY_SIZE) { |
| 601 |
THROW(std::length_error, "Key is too long."); |
| 602 |
} |
| 603 |
|
| 604 |
return key_size_type(key_size); |
| 605 |
} |
| 606 |
|
| 607 |
/* |
| 608 |
* Return true if found, false otherwise. |
| 609 |
* If true, buffer_ptr_in_out points to the start of the entry matching 'key'. |
| 610 |
* If false, buffer_ptr_in_out points to where the 'key' should be inserted. |
| 611 |
* |
| 612 |
* Start search from buffer_ptr_in_out. |
| 613 |
*/ |
| 614 |
bool find_or_end_of_bucket_impl(const CharT* key, size_type key_size, |
| 615 |
const CharT* & buffer_ptr_in_out) const noexcept |
| 616 |
{ |
| 617 |
while(!is_end_of_bucket(buffer_ptr_in_out)) { |
| 618 |
const key_size_type buffer_key_size = read_key_size(buffer_ptr_in_out); |
| 619 |
const CharT* buffer_str = buffer_ptr_in_out + size_as_char_t<key_size_type>(); |
| 620 |
if(KeyEqual()(buffer_str, buffer_key_size, key, key_size)) { |
| 621 |
return true; |
| 622 |
} |
| 623 |
|
| 624 |
buffer_ptr_in_out += entry_size_bytes(buffer_ptr_in_out)/sizeof(CharT); |
| 625 |
} |
| 626 |
|
| 627 |
return false; |
| 628 |
} |
| 629 |
|
| 630 |
template<typename U = T, typename std::enable_if<!has_mapped_type<U>::value>::type* = nullptr> |
| 631 |
void append_impl(const CharT* key, key_size_type key_size, CharT* buffer_append_pos) noexcept { |
| 632 |
std::memcpy(buffer_append_pos, &key_size, sizeof(key_size)); |
| 633 |
buffer_append_pos += size_as_char_t<key_size_type>(); |
| 634 |
|
| 635 |
std::memcpy(buffer_append_pos, key, key_size * sizeof(CharT)); |
| 636 |
buffer_append_pos += key_size; |
| 637 |
|
| 638 |
const CharT zero = 0; |
| 639 |
std::memcpy(buffer_append_pos, &zero, KEY_EXTRA_SIZE * sizeof(CharT)); |
| 640 |
buffer_append_pos += KEY_EXTRA_SIZE; |
| 641 |
|
| 642 |
const auto end_of_bucket = END_OF_BUCKET; |
| 643 |
std::memcpy(buffer_append_pos, &end_of_bucket, sizeof(end_of_bucket)); |
| 644 |
} |
| 645 |
|
| 646 |
template<typename U = T, typename std::enable_if<has_mapped_type<U>::value>::type* = nullptr> |
| 647 |
void append_impl(const CharT* key, key_size_type key_size, CharT* buffer_append_pos, |
| 648 |
typename array_bucket<CharT, U, KeyEqual, KeySizeT, StoreNullTerminator>::mapped_type value) noexcept |
| 649 |
{ |
| 650 |
std::memcpy(buffer_append_pos, &key_size, sizeof(key_size)); |
| 651 |
buffer_append_pos += size_as_char_t<key_size_type>(); |
| 652 |
|
| 653 |
std::memcpy(buffer_append_pos, key, key_size * sizeof(CharT)); |
| 654 |
buffer_append_pos += key_size; |
| 655 |
|
| 656 |
const CharT zero = 0; |
| 657 |
std::memcpy(buffer_append_pos, &zero, KEY_EXTRA_SIZE * sizeof(CharT)); |
| 658 |
buffer_append_pos += KEY_EXTRA_SIZE; |
| 659 |
|
| 660 |
std::memcpy(buffer_append_pos, &value, sizeof(value)); |
| 661 |
buffer_append_pos += size_as_char_t<mapped_type>(); |
| 662 |
|
| 663 |
const auto end_of_bucket = END_OF_BUCKET; |
| 664 |
std::memcpy(buffer_append_pos, &end_of_bucket, sizeof(end_of_bucket)); |
| 665 |
} |
| 666 |
|
| 667 |
/** |
| 668 |
* Return the number of CharT in m_buffer. As the size of the buffer is not stored to gain some space, |
| 669 |
* the method need to find the EOF marker and is thus in O(n). |
| 670 |
*/ |
| 671 |
size_type size() const noexcept { |
| 672 |
if(m_buffer == nullptr) { |
| 673 |
return 0; |
| 674 |
} |
| 675 |
|
| 676 |
CharT* buffer_ptr = m_buffer; |
| 677 |
while(!is_end_of_bucket(buffer_ptr)) { |
| 678 |
buffer_ptr += entry_size_bytes(buffer_ptr)/sizeof(CharT); |
| 679 |
} |
| 680 |
|
| 681 |
return buffer_ptr - m_buffer; |
| 682 |
} |
| 683 |
|
| 684 |
private: |
| 685 |
static const key_size_type END_OF_BUCKET = std::numeric_limits<key_size_type>::max(); |
| 686 |
static const key_size_type KEY_EXTRA_SIZE = StoreNullTerminator?1:0; |
| 687 |
|
| 688 |
CharT* m_buffer; |
| 689 |
|
| 690 |
public: |
| 691 |
static const key_size_type MAX_KEY_SIZE = |
| 692 |
// -1 for END_OF_BUCKET |
| 693 |
key_size_type(std::numeric_limits<key_size_type>::max() - KEY_EXTRA_SIZE - 1); |
| 694 |
}; |
| 695 |
|
| 696 |
|
| 697 |
template<class T> |
| 698 |
class value_container { |
| 699 |
public: |
| 700 |
void clear() noexcept { |
| 701 |
m_values.clear(); |
| 702 |
} |
| 703 |
|
| 704 |
void reserve(std::size_t new_cap) { |
| 705 |
m_values.reserve(new_cap); |
| 706 |
} |
| 707 |
|
| 708 |
void shrink_to_fit() { |
| 709 |
m_values.shrink_to_fit(); |
| 710 |
} |
| 711 |
|
| 712 |
friend void swap(value_container& lhs, value_container& rhs) { |
| 713 |
lhs.m_values.swap(rhs.m_values); |
| 714 |
} |
| 715 |
|
| 716 |
protected: |
| 717 |
static constexpr float VECTOR_GROWTH_RATE = 1.5f; |
| 718 |
|
| 719 |
// TODO use a sparse array? or a std::deque |
| 720 |
std::vector<T> m_values; |
| 721 |
}; |
| 722 |
|
| 723 |
template<> |
| 724 |
class value_container<void> { |
| 725 |
public: |
| 726 |
void clear() noexcept { |
| 727 |
} |
| 728 |
|
| 729 |
void shrink_to_fit() { |
| 730 |
} |
| 731 |
|
| 732 |
void reserve(std::size_t /*new_cap*/) { |
| 733 |
} |
| 734 |
}; |
| 735 |
|
| 736 |
|
| 737 |
|
| 738 |
/** |
| 739 |
* If there is no value in the array_hash (in the case of a set for example), T should be void. |
| 740 |
* |
| 741 |
* The size of a key string is limited to std::numeric_limits<KeySizeT>::max() - 1. |
| 742 |
* |
| 743 |
* The number of elements in the map is limited to std::numeric_limits<IndexSizeT>::max(). |
| 744 |
*/ |
| 745 |
template<class CharT, |
| 746 |
class T, |
| 747 |
class Hash, |
| 748 |
class KeyEqual, |
| 749 |
bool StoreNullTerminator, |
| 750 |
class KeySizeT, |
| 751 |
class IndexSizeT, |
| 752 |
class GrowthPolicy> |
| 753 |
class array_hash: private value_container<T>, private Hash, private GrowthPolicy { |
| 754 |
private: |
| 755 |
template<typename U> |
| 756 |
using has_mapped_type = typename std::integral_constant<bool, !std::is_same<U, void>::value>; |
| 757 |
|
| 758 |
/** |
| 759 |
* If there is a mapped type in array_hash, we store the values in m_values of value_container class |
| 760 |
* and we store an index to m_values in the bucket. The index is of type IndexSizeT. |
| 761 |
*/ |
| 762 |
using array_bucket = tsl::detail_array_hash::array_bucket<CharT, |
| 763 |
typename std::conditional<has_mapped_type<T>::value, |
| 764 |
IndexSizeT, |
| 765 |
void>::type, |
| 766 |
KeyEqual, KeySizeT, StoreNullTerminator>; |
| 767 |
|
| 768 |
public: |
| 769 |
template<bool IsConst> |
| 770 |
class array_hash_iterator; |
| 771 |
|
| 772 |
using char_type = CharT; |
| 773 |
using key_size_type = KeySizeT; |
| 774 |
using index_size_type = IndexSizeT; |
| 775 |
using size_type = std::size_t; |
| 776 |
using hasher = Hash; |
| 777 |
using key_equal = KeyEqual; |
| 778 |
using iterator = array_hash_iterator<false>; |
| 779 |
using const_iterator = array_hash_iterator<true>; |
| 780 |
|
| 781 |
|
| 782 |
/* |
| 783 |
* Iterator classes |
| 784 |
*/ |
| 785 |
public: |
| 786 |
template<bool IsConst> |
| 787 |
class array_hash_iterator { |
| 788 |
friend class array_hash; |
| 789 |
|
| 790 |
private: |
| 791 |
using iterator_array_bucket = typename array_bucket::const_iterator; |
| 792 |
|
| 793 |
using iterator_buckets = typename std::conditional<IsConst, |
| 794 |
typename std::vector<array_bucket>::const_iterator, |
| 795 |
typename std::vector<array_bucket>::iterator>::type; |
| 796 |
|
| 797 |
using array_hash_ptr = typename std::conditional<IsConst, |
| 798 |
const array_hash*, |
| 799 |
array_hash*>::type; |
| 800 |
|
| 801 |
public: |
| 802 |
using iterator_category = std::forward_iterator_tag; |
| 803 |
using value_type = typename std::conditional<has_mapped_type<T>::value, T, void>::type; |
| 804 |
using difference_type = std::ptrdiff_t; |
| 805 |
using reference = typename std::conditional<has_mapped_type<T>::value, |
| 806 |
typename std::conditional< |
| 807 |
IsConst, |
| 808 |
typename std::add_lvalue_reference<const T>::type, |
| 809 |
typename std::add_lvalue_reference<T>::type>::type, |
| 810 |
void>::type; |
| 811 |
using pointer = typename std::conditional<has_mapped_type<T>::value, |
| 812 |
typename std::conditional<IsConst, const T*, T*>::type, |
| 813 |
void>::type; |
| 814 |
|
| 815 |
|
| 816 |
private: |
| 817 |
array_hash_iterator(iterator_buckets buckets_iterator, iterator_array_bucket array_bucket_iterator, |
| 818 |
array_hash_ptr array_hash_p) noexcept: |
| 819 |
m_buckets_iterator(buckets_iterator), |
| 820 |
m_array_bucket_iterator(array_bucket_iterator), |
| 821 |
m_array_hash(array_hash_p) |
| 822 |
{ |
| 823 |
tsl_ah_assert(m_array_hash != nullptr); |
| 824 |
} |
| 825 |
|
| 826 |
public: |
| 827 |
array_hash_iterator() noexcept: m_array_hash(nullptr) { |
| 828 |
} |
| 829 |
|
| 830 |
template<bool TIsConst = IsConst, typename std::enable_if<TIsConst>::type* = nullptr> |
| 831 |
array_hash_iterator(const array_hash_iterator<!TIsConst>& other) noexcept : |
| 832 |
m_buckets_iterator(other.m_buckets_iterator), |
| 833 |
m_array_bucket_iterator(other.m_array_bucket_iterator), |
| 834 |
m_array_hash(other.m_array_hash) |
| 835 |
{ |
| 836 |
} |
| 837 |
|
| 838 |
array_hash_iterator(const array_hash_iterator& other) = default; |
| 839 |
array_hash_iterator(array_hash_iterator&& other) = default; |
| 840 |
array_hash_iterator& operator=(const array_hash_iterator& other) = default; |
| 841 |
array_hash_iterator& operator=(array_hash_iterator&& other) = default; |
| 842 |
|
| 843 |
const CharT* key() const { |
| 844 |
return m_array_bucket_iterator.key(); |
| 845 |
} |
| 846 |
|
| 847 |
size_type key_size() const { |
| 848 |
return m_array_bucket_iterator.key_size(); |
| 849 |
} |
| 850 |
|
| 851 |
#ifdef TSL_AH_HAS_STRING_VIEW |
| 852 |
std::basic_string_view<CharT> key_sv() const { |
| 853 |
return std::basic_string_view<CharT>(key(), key_size()); |
| 854 |
} |
| 855 |
#endif |
| 856 |
|
| 857 |
template<class U = T, typename std::enable_if<has_mapped_type<U>::value>::type* = nullptr> |
| 858 |
reference value() const { |
| 859 |
return this->m_array_hash->m_values[value_position()]; |
| 860 |
} |
| 861 |
|
| 862 |
template<class U = T, typename std::enable_if<has_mapped_type<U>::value>::type* = nullptr> |
| 863 |
reference operator*() const { |
| 864 |
return value(); |
| 865 |
} |
| 866 |
|
| 867 |
template<class U = T, typename std::enable_if<has_mapped_type<U>::value>::type* = nullptr> |
| 868 |
pointer operator->() const { |
| 869 |
return std::addressof(value()); |
| 870 |
} |
| 871 |
|
| 872 |
array_hash_iterator& operator++() { |
| 873 |
tsl_ah_assert(m_buckets_iterator != m_array_hash->m_buckets_data.end()); |
| 874 |
tsl_ah_assert(m_array_bucket_iterator != m_buckets_iterator->cend()); |
| 875 |
|
| 876 |
++m_array_bucket_iterator; |
| 877 |
if(m_array_bucket_iterator == m_buckets_iterator->cend()) { |
| 878 |
do { |
| 879 |
++m_buckets_iterator; |
| 880 |
} while(m_buckets_iterator != m_array_hash->m_buckets_data.end() && |
| 881 |
m_buckets_iterator->empty()); |
| 882 |
|
| 883 |
if(m_buckets_iterator != m_array_hash->m_buckets_data.end()) { |
| 884 |
m_array_bucket_iterator = m_buckets_iterator->cbegin(); |
| 885 |
} |
| 886 |
} |
| 887 |
|
| 888 |
return *this; |
| 889 |
} |
| 890 |
|
| 891 |
array_hash_iterator operator++(int) { |
| 892 |
array_hash_iterator tmp(*this); |
| 893 |
++*this; |
| 894 |
|
| 895 |
return tmp; |
| 896 |
} |
| 897 |
|
| 898 |
friend bool operator==(const array_hash_iterator& lhs, const array_hash_iterator& rhs) { |
| 899 |
return lhs.m_buckets_iterator == rhs.m_buckets_iterator && |
| 900 |
lhs.m_array_bucket_iterator == rhs.m_array_bucket_iterator && |
| 901 |
lhs.m_array_hash == rhs.m_array_hash; |
| 902 |
} |
| 903 |
|
| 904 |
friend bool operator!=(const array_hash_iterator& lhs, const array_hash_iterator& rhs) { |
| 905 |
return !(lhs == rhs); |
| 906 |
} |
| 907 |
|
| 908 |
private: |
| 909 |
template<class U = T, typename std::enable_if<has_mapped_type<U>::value>::type* = nullptr> |
| 910 |
IndexSizeT value_position() const { |
| 911 |
return this->m_array_bucket_iterator.value(); |
| 912 |
} |
| 913 |
|
| 914 |
private: |
| 915 |
iterator_buckets m_buckets_iterator; |
| 916 |
iterator_array_bucket m_array_bucket_iterator; |
| 917 |
|
| 918 |
array_hash_ptr m_array_hash; |
| 919 |
}; |
| 920 |
|
| 921 |
|
| 922 |
|
| 923 |
public: |
| 924 |
array_hash(size_type bucket_count, |
| 925 |
const Hash& hash, |
| 926 |
float max_load_factor): value_container<T>(), |
| 927 |
Hash(hash), |
| 928 |
GrowthPolicy(bucket_count), |
| 929 |
m_buckets_data(bucket_count > max_bucket_count()? |
| 930 |
max_bucket_count(): |
| 931 |
bucket_count), |
| 932 |
m_buckets(m_buckets_data.empty()?static_empty_bucket_ptr():m_buckets_data.data()), |
| 933 |
m_nb_elements(0) |
| 934 |
{ |
| 935 |
this->max_load_factor(max_load_factor); |
| 936 |
} |
| 937 |
|
| 938 |
array_hash(const array_hash& other): value_container<T>(other), |
| 939 |
Hash(other), |
| 940 |
GrowthPolicy(other), |
| 941 |
m_buckets_data(other.m_buckets_data), |
| 942 |
m_buckets(m_buckets_data.empty()?static_empty_bucket_ptr():m_buckets_data.data()), |
| 943 |
m_nb_elements(other.m_nb_elements), |
| 944 |
m_max_load_factor(other.m_max_load_factor), |
| 945 |
m_load_threshold(other.m_load_threshold) |
| 946 |
{ |
| 947 |
} |
| 948 |
|
| 949 |
array_hash(array_hash&& other) noexcept(std::is_nothrow_move_constructible<value_container<T>>::value && |
| 950 |
std::is_nothrow_move_constructible<Hash>::value && |
| 951 |
std::is_nothrow_move_constructible<GrowthPolicy>::value && |
| 952 |
std::is_nothrow_move_constructible<std::vector<array_bucket>>::value) |
| 953 |
: value_container<T>(std::move(other)), |
| 954 |
Hash(std::move(other)), |
| 955 |
GrowthPolicy(std::move(other)), |
| 956 |
m_buckets_data(std::move(other.m_buckets_data)), |
| 957 |
m_buckets(m_buckets_data.empty()?static_empty_bucket_ptr():m_buckets_data.data()), |
| 958 |
m_nb_elements(other.m_nb_elements), |
| 959 |
m_max_load_factor(other.m_max_load_factor), |
| 960 |
m_load_threshold(other.m_load_threshold) |
| 961 |
{ |
| 962 |
other.value_container<T>::clear(); |
| 963 |
other.GrowthPolicy::clear(); |
| 964 |
other.m_buckets_data.clear(); |
| 965 |
other.m_buckets = static_empty_bucket_ptr(); |
| 966 |
other.m_nb_elements = 0; |
| 967 |
other.m_load_threshold = 0; |
| 968 |
} |
| 969 |
|
| 970 |
array_hash& operator=(const array_hash& other) { |
| 971 |
if(&other != this) { |
| 972 |
value_container<T>::operator=(other); |
| 973 |
Hash::operator=(other); |
| 974 |
GrowthPolicy::operator=(other); |
| 975 |
|
| 976 |
m_buckets_data = other.m_buckets_data; |
| 977 |
m_buckets = m_buckets_data.empty()?static_empty_bucket_ptr(): |
| 978 |
m_buckets_data.data(); |
| 979 |
m_nb_elements = other.m_nb_elements; |
| 980 |
m_max_load_factor = other.m_max_load_factor; |
| 981 |
m_load_threshold = other.m_load_threshold; |
| 982 |
} |
| 983 |
|
| 984 |
return *this; |
| 985 |
} |
| 986 |
|
| 987 |
array_hash& operator=(array_hash&& other) { |
| 988 |
other.swap(*this); |
| 989 |
other.clear(); |
| 990 |
|
| 991 |
return *this; |
| 992 |
} |
| 993 |
|
| 994 |
|
| 995 |
/* |
| 996 |
* Iterators |
| 997 |
*/ |
| 998 |
iterator begin() noexcept { |
| 999 |
auto begin = m_buckets_data.begin(); |
| 1000 |
while(begin != m_buckets_data.end() && begin->empty()) { |
| 1001 |
++begin; |
| 1002 |
} |
| 1003 |
|
| 1004 |
return (begin != m_buckets_data.end())?iterator(begin, begin->cbegin(), this):end(); |
| 1005 |
} |
| 1006 |
|
| 1007 |
const_iterator begin() const noexcept { |
| 1008 |
return cbegin(); |
| 1009 |
} |
| 1010 |
|
| 1011 |
const_iterator cbegin() const noexcept { |
| 1012 |
auto begin = m_buckets_data.cbegin(); |
| 1013 |
while(begin != m_buckets_data.cend() && begin->empty()) { |
| 1014 |
++begin; |
| 1015 |
} |
| 1016 |
|
| 1017 |
return (begin != m_buckets_data.cend())?const_iterator(begin, begin->cbegin(), this):cend(); |
| 1018 |
} |
| 1019 |
|
| 1020 |
iterator end() noexcept { |
| 1021 |
return iterator(m_buckets_data.end(), array_bucket::cend_it(), this); |
| 1022 |
} |
| 1023 |
|
| 1024 |
const_iterator end() const noexcept { |
| 1025 |
return cend(); |
| 1026 |
} |
| 1027 |
|
| 1028 |
const_iterator cend() const noexcept { |
| 1029 |
return const_iterator(m_buckets_data.end(), array_bucket::cend_it(), this); |
| 1030 |
} |
| 1031 |
|
| 1032 |
|
| 1033 |
/* |
| 1034 |
* Capacity |
| 1035 |
*/ |
| 1036 |
bool empty() const noexcept { |
| 1037 |
return m_nb_elements == 0; |
| 1038 |
} |
| 1039 |
|
| 1040 |
size_type size() const noexcept { |
| 1041 |
return m_nb_elements; |
| 1042 |
} |
| 1043 |
|
| 1044 |
size_type max_size() const noexcept { |
| 1045 |
return std::numeric_limits<IndexSizeT>::max(); |
| 1046 |
} |
| 1047 |
|
| 1048 |
size_type max_key_size() const noexcept { |
| 1049 |
return MAX_KEY_SIZE; |
| 1050 |
} |
| 1051 |
|
| 1052 |
void shrink_to_fit() { |
| 1053 |
clear_old_erased_values(); |
| 1054 |
value_container<T>::shrink_to_fit(); |
| 1055 |
|
| 1056 |
rehash_impl(size_type(std::ceil(float(size())/max_load_factor()))); |
| 1057 |
} |
| 1058 |
|
| 1059 |
/* |
| 1060 |
* Modifiers |
| 1061 |
*/ |
| 1062 |
void clear() noexcept { |
| 1063 |
value_container<T>::clear(); |
| 1064 |
|
| 1065 |
for(auto& bucket: m_buckets_data) { |
| 1066 |
bucket.clear(); |
| 1067 |
} |
| 1068 |
|
| 1069 |
m_nb_elements = 0; |
| 1070 |
} |
| 1071 |
|
| 1072 |
|
| 1073 |
|
| 1074 |
template<class... ValueArgs> |
| 1075 |
std::pair<iterator, bool> emplace(const CharT* key, size_type key_size, ValueArgs&&... value_args) { |
| 1076 |
const std::size_t hash = hash_key(key, key_size); |
| 1077 |
std::size_t ibucket = bucket_for_hash(hash); |
| 1078 |
|
| 1079 |
auto it_find = m_buckets[ibucket].find_or_end_of_bucket(key, key_size); |
| 1080 |
if(it_find.second) { |
| 1081 |
return std::make_pair(iterator(m_buckets_data.begin() + ibucket, it_find.first, this), false); |
| 1082 |
} |
| 1083 |
|
| 1084 |
if(grow_on_high_load()) { |
| 1085 |
ibucket = bucket_for_hash(hash); |
| 1086 |
it_find = m_buckets[ibucket].find_or_end_of_bucket(key, key_size); |
| 1087 |
} |
| 1088 |
|
| 1089 |
return emplace_impl(ibucket, it_find.first, key, key_size, std::forward<ValueArgs>(value_args)...); |
| 1090 |
} |
| 1091 |
|
| 1092 |
template<class M> |
| 1093 |
std::pair<iterator, bool> insert_or_assign(const CharT* key, size_type key_size, M&& obj) { |
| 1094 |
auto it = emplace(key, key_size, std::forward<M>(obj)); |
| 1095 |
if(!it.second) { |
| 1096 |
it.first.value() = std::forward<M>(obj); |
| 1097 |
} |
| 1098 |
|
| 1099 |
return it; |
| 1100 |
} |
| 1101 |
|
| 1102 |
|
| 1103 |
|
| 1104 |
iterator erase(const_iterator pos) { |
| 1105 |
if(should_clear_old_erased_values()) { |
| 1106 |
clear_old_erased_values(); |
| 1107 |
} |
| 1108 |
|
| 1109 |
return erase_from_bucket(mutable_iterator(pos)); |
| 1110 |
} |
| 1111 |
|
| 1112 |
iterator erase(const_iterator first, const_iterator last) { |
| 1113 |
if(first == last) { |
| 1114 |
return mutable_iterator(first); |
| 1115 |
} |
| 1116 |
|
| 1117 |
/** |
| 1118 |
* When erasing an element from a bucket with erase_from_bucket, it invalidates all the iterators |
| 1119 |
* in the array bucket of the element (m_array_bucket_iterator) but not the iterators of the buckets |
| 1120 |
* itself (m_buckets_iterator). |
| 1121 |
* |
| 1122 |
* So first erase all the values between first and last which are not part of the bucket of last, |
| 1123 |
* and then erase carefully the values in last's bucket. |
| 1124 |
*/ |
| 1125 |
auto to_delete = mutable_iterator(first); |
| 1126 |
while(to_delete.m_buckets_iterator != last.m_buckets_iterator) { |
| 1127 |
to_delete = erase_from_bucket(to_delete); |
| 1128 |
} |
| 1129 |
|
| 1130 |
std::size_t nb_elements_until_last = std::distance(to_delete.m_array_bucket_iterator, |
| 1131 |
last.m_array_bucket_iterator); |
| 1132 |
while(nb_elements_until_last > 0) { |
| 1133 |
to_delete = erase_from_bucket(to_delete); |
| 1134 |
nb_elements_until_last--; |
| 1135 |
} |
| 1136 |
|
| 1137 |
if(should_clear_old_erased_values()) { |
| 1138 |
clear_old_erased_values(); |
| 1139 |
} |
| 1140 |
|
| 1141 |
return to_delete; |
| 1142 |
} |
| 1143 |
|
| 1144 |
|
| 1145 |
|
| 1146 |
size_type erase(const CharT* key, size_type key_size) { |
| 1147 |
return erase(key, key_size, hash_key(key, key_size)); |
| 1148 |
} |
| 1149 |
|
| 1150 |
size_type erase(const CharT* key, size_type key_size, std::size_t hash) { |
| 1151 |
if(should_clear_old_erased_values()) { |
| 1152 |
clear_old_erased_values(); |
| 1153 |
} |
| 1154 |
|
| 1155 |
const std::size_t ibucket = bucket_for_hash(hash); |
| 1156 |
if(m_buckets[ibucket].erase(key, key_size)) { |
| 1157 |
m_nb_elements--; |
| 1158 |
return 1; |
| 1159 |
} |
| 1160 |
else { |
| 1161 |
return 0; |
| 1162 |
} |
| 1163 |
} |
| 1164 |
|
| 1165 |
|
| 1166 |
|
| 1167 |
void swap(array_hash& other) { |
| 1168 |
using std::swap; |
| 1169 |
|
| 1170 |
swap(static_cast<value_container<T>&>(*this), static_cast<value_container<T>&>(other)); |
| 1171 |
swap(static_cast<Hash&>(*this), static_cast<Hash&>(other)); |
| 1172 |
swap(static_cast<GrowthPolicy&>(*this), static_cast<GrowthPolicy&>(other)); |
| 1173 |
swap(m_buckets_data, other.m_buckets_data); |
| 1174 |
swap(m_buckets, other.m_buckets); |
| 1175 |
swap(m_nb_elements, other.m_nb_elements); |
| 1176 |
swap(m_max_load_factor, other.m_max_load_factor); |
| 1177 |
swap(m_load_threshold, other.m_load_threshold); |
| 1178 |
} |
| 1179 |
|
| 1180 |
/* |
| 1181 |
* Lookup |
| 1182 |
*/ |
| 1183 |
template<class U = T, typename std::enable_if<has_mapped_type<U>::value>::type* = nullptr> |
| 1184 |
U& at(const CharT* key, size_type key_size) { |
| 1185 |
return at(key, key_size, hash_key(key, key_size)); |
| 1186 |
} |
| 1187 |
|
| 1188 |
template<class U = T, typename std::enable_if<has_mapped_type<U>::value>::type* = nullptr> |
| 1189 |
const U& at(const CharT* key, size_type key_size) const { |
| 1190 |
return at(key, key_size, hash_key(key, key_size)); |
| 1191 |
} |
| 1192 |
|
| 1193 |
template<class U = T, typename std::enable_if<has_mapped_type<U>::value>::type* = nullptr> |
| 1194 |
U& at(const CharT* key, size_type key_size, std::size_t hash) { |
| 1195 |
return const_cast<U&>(static_cast<const array_hash*>(this)->at(key, key_size, hash)); |
| 1196 |
} |
| 1197 |
|
| 1198 |
template<class U = T, typename std::enable_if<has_mapped_type<U>::value>::type* = nullptr> |
| 1199 |
const U& at(const CharT* key, size_type key_size, std::size_t hash) const { |
| 1200 |
const std::size_t ibucket = bucket_for_hash(hash); |
| 1201 |
|
| 1202 |
auto it_find = m_buckets[ibucket].find_or_end_of_bucket(key, key_size); |
| 1203 |
if(it_find.second) { |
| 1204 |
return this->m_values[it_find.first.value()]; |
| 1205 |
} |
| 1206 |
else { |
| 1207 |
THROW(std::out_of_range, "Couldn't find key."); |
| 1208 |
} |
| 1209 |
} |
| 1210 |
|
| 1211 |
|
| 1212 |
|
| 1213 |
template<class U = T, typename std::enable_if<has_mapped_type<U>::value>::type* = nullptr> |
| 1214 |
U& access_operator(const CharT* key, size_type key_size) { |
| 1215 |
const std::size_t hash = hash_key(key, key_size); |
| 1216 |
std::size_t ibucket = bucket_for_hash(hash); |
| 1217 |
|
| 1218 |
auto it_find = m_buckets[ibucket].find_or_end_of_bucket(key, key_size); |
| 1219 |
if(it_find.second) { |
| 1220 |
return this->m_values[it_find.first.value()]; |
| 1221 |
} |
| 1222 |
else { |
| 1223 |
if(grow_on_high_load()) { |
| 1224 |
ibucket = bucket_for_hash(hash); |
| 1225 |
it_find = m_buckets[ibucket].find_or_end_of_bucket(key, key_size); |
| 1226 |
} |
| 1227 |
|
| 1228 |
return emplace_impl(ibucket, it_find.first, key, key_size, U{}).first.value(); |
| 1229 |
} |
| 1230 |
} |
| 1231 |
|
| 1232 |
|
| 1233 |
|
| 1234 |
size_type count(const CharT* key, size_type key_size) const { |
| 1235 |
return count(key, key_size, hash_key(key, key_size)); |
| 1236 |
} |
| 1237 |
|
| 1238 |
size_type count(const CharT* key, size_type key_size, std::size_t hash) const { |
| 1239 |
const std::size_t ibucket = bucket_for_hash(hash); |
| 1240 |
|
| 1241 |
auto it_find = m_buckets[ibucket].find_or_end_of_bucket(key, key_size); |
| 1242 |
if(it_find.second) { |
| 1243 |
return 1; |
| 1244 |
} |
| 1245 |
else { |
| 1246 |
return 0; |
| 1247 |
} |
| 1248 |
} |
| 1249 |
|
| 1250 |
|
| 1251 |
|
| 1252 |
iterator find(const CharT* key, size_type key_size) { |
| 1253 |
return find(key, key_size, hash_key(key, key_size)); |
| 1254 |
} |
| 1255 |
|
| 1256 |
const_iterator find(const CharT* key, size_type key_size) const { |
| 1257 |
return find(key, key_size, hash_key(key, key_size)); |
| 1258 |
} |
| 1259 |
|
| 1260 |
iterator find(const CharT* key, size_type key_size, std::size_t hash) { |
| 1261 |
const std::size_t ibucket = bucket_for_hash(hash); |
| 1262 |
|
| 1263 |
auto it_find = m_buckets[ibucket].find_or_end_of_bucket(key, key_size); |
| 1264 |
if(it_find.second) { |
| 1265 |
return iterator(m_buckets_data.begin() + ibucket, it_find.first, this); |
| 1266 |
} |
| 1267 |
else { |
| 1268 |
return end(); |
| 1269 |
} |
| 1270 |
} |
| 1271 |
|
| 1272 |
const_iterator find(const CharT* key, size_type key_size, std::size_t hash) const { |
| 1273 |
const std::size_t ibucket = bucket_for_hash(hash); |
| 1274 |
|
| 1275 |
auto it_find = m_buckets[ibucket].find_or_end_of_bucket(key, key_size); |
| 1276 |
if(it_find.second) { |
| 1277 |
return const_iterator(m_buckets_data.cbegin() + ibucket, it_find.first, this); |
| 1278 |
} |
| 1279 |
else { |
| 1280 |
return cend(); |
| 1281 |
} |
| 1282 |
} |
| 1283 |
|
| 1284 |
|
| 1285 |
|
| 1286 |
std::pair<iterator, iterator> equal_range(const CharT* key, size_type key_size) { |
| 1287 |
return equal_range(key, key_size, hash_key(key, key_size)); |
| 1288 |
} |
| 1289 |
|
| 1290 |
std::pair<const_iterator, const_iterator> equal_range(const CharT* key, size_type key_size) const { |
| 1291 |
return equal_range(key, key_size, hash_key(key, key_size)); |
| 1292 |
} |
| 1293 |
|
| 1294 |
std::pair<iterator, iterator> equal_range(const CharT* key, size_type key_size, std::size_t hash) { |
| 1295 |
iterator it = find(key, key_size, hash); |
| 1296 |
return std::make_pair(it, (it == end())?it:std::next(it)); |
| 1297 |
} |
| 1298 |
|
| 1299 |
std::pair<const_iterator, const_iterator> equal_range(const CharT* key, size_type key_size, |
| 1300 |
std::size_t hash) const |
| 1301 |
{ |
| 1302 |
const_iterator it = find(key, key_size, hash); |
| 1303 |
return std::make_pair(it, (it == cend())?it:std::next(it)); |
| 1304 |
} |
| 1305 |
|
| 1306 |
/* |
| 1307 |
* Bucket interface |
| 1308 |
*/ |
| 1309 |
size_type bucket_count() const { |
| 1310 |
return m_buckets_data.size(); |
| 1311 |
} |
| 1312 |
|
| 1313 |
size_type max_bucket_count() const { |
| 1314 |
return std::min(GrowthPolicy::max_bucket_count(), m_buckets_data.max_size()); |
| 1315 |
} |
| 1316 |
|
| 1317 |
|
| 1318 |
/* |
| 1319 |
* Hash policy |
| 1320 |
*/ |
| 1321 |
float load_factor() const { |
| 1322 |
if(bucket_count() == 0) { |
| 1323 |
return 0; |
| 1324 |
} |
| 1325 |
|
| 1326 |
return float(m_nb_elements) / float(bucket_count()); |
| 1327 |
} |
| 1328 |
|
| 1329 |
float max_load_factor() const { |
| 1330 |
return m_max_load_factor; |
| 1331 |
} |
| 1332 |
|
| 1333 |
void max_load_factor(float ml) { |
| 1334 |
m_max_load_factor = std::max(0.1f, ml); |
| 1335 |
m_load_threshold = size_type(float(bucket_count())*m_max_load_factor); |
| 1336 |
} |
| 1337 |
|
| 1338 |
void rehash(size_type count) { |
| 1339 |
count = std::max(count, size_type(std::ceil(float(size())/max_load_factor()))); |
| 1340 |
rehash_impl(count); |
| 1341 |
} |
| 1342 |
|
| 1343 |
void reserve(size_type count) { |
| 1344 |
rehash(size_type(std::ceil(float(count)/max_load_factor()))); |
| 1345 |
} |
| 1346 |
|
| 1347 |
/* |
| 1348 |
* Observers |
| 1349 |
*/ |
| 1350 |
hasher hash_function() const { |
| 1351 |
return static_cast<const hasher&>(*this); |
| 1352 |
} |
| 1353 |
|
| 1354 |
// TODO add support for statefull KeyEqual |
| 1355 |
key_equal key_eq() const { |
| 1356 |
return KeyEqual(); |
| 1357 |
} |
| 1358 |
|
| 1359 |
/* |
| 1360 |
* Other |
| 1361 |
*/ |
| 1362 |
iterator mutable_iterator(const_iterator it) noexcept { |
| 1363 |
auto it_bucket = m_buckets_data.begin() + std::distance(m_buckets_data.cbegin(), it.m_buckets_iterator); |
| 1364 |
return iterator(it_bucket, it.m_array_bucket_iterator, this); |
| 1365 |
} |
| 1366 |
|
| 1367 |
template<class Serializer> |
| 1368 |
void serialize(Serializer& serializer) const { |
| 1369 |
serialize_impl(serializer); |
| 1370 |
} |
| 1371 |
|
| 1372 |
template<class Deserializer> |
| 1373 |
void deserialize(Deserializer& deserializer, bool hash_compatible) { |
| 1374 |
deserialize_impl(deserializer, hash_compatible); |
| 1375 |
} |
| 1376 |
|
| 1377 |
private: |
| 1378 |
std::size_t hash_key(const CharT* key, size_type key_size) const { |
| 1379 |
return Hash::operator()(key, key_size); |
| 1380 |
} |
| 1381 |
|
| 1382 |
std::size_t bucket_for_hash(std::size_t hash) const { |
| 1383 |
return GrowthPolicy::bucket_for_hash(hash); |
| 1384 |
} |
| 1385 |
|
| 1386 |
/** |
| 1387 |
* If there is a mapped_type, the mapped value in m_values is not erased now. |
| 1388 |
* It will be erased when the ratio between the size of the map and |
| 1389 |
* the size of the map + the number of deleted values still stored is low enough (see clear_old_erased_values). |
| 1390 |
*/ |
| 1391 |
iterator erase_from_bucket(iterator pos) noexcept { |
| 1392 |
auto array_bucket_next_it = pos.m_buckets_iterator->erase(pos.m_array_bucket_iterator); |
| 1393 |
m_nb_elements--; |
| 1394 |
|
| 1395 |
if(array_bucket_next_it != pos.m_buckets_iterator->cend()) { |
| 1396 |
return iterator(pos.m_buckets_iterator, array_bucket_next_it, this); |
| 1397 |
} |
| 1398 |
else { |
| 1399 |
do { |
| 1400 |
++pos.m_buckets_iterator; |
| 1401 |
} while(pos.m_buckets_iterator != m_buckets_data.end() && pos.m_buckets_iterator->empty()); |
| 1402 |
|
| 1403 |
if(pos.m_buckets_iterator != m_buckets_data.end()) { |
| 1404 |
return iterator(pos.m_buckets_iterator, pos.m_buckets_iterator->cbegin(), this); |
| 1405 |
} |
| 1406 |
else { |
| 1407 |
return end(); |
| 1408 |
} |
| 1409 |
} |
| 1410 |
} |
| 1411 |
|
| 1412 |
|
| 1413 |
template<class U = T, typename std::enable_if<!has_mapped_type<U>::value>::type* = nullptr> |
| 1414 |
bool should_clear_old_erased_values(float /*threshold*/ = DEFAULT_CLEAR_OLD_ERASED_VALUE_THRESHOLD) const { |
| 1415 |
return false; |
| 1416 |
} |
| 1417 |
|
| 1418 |
template<class U = T, typename std::enable_if<has_mapped_type<U>::value>::type* = nullptr> |
| 1419 |
bool should_clear_old_erased_values(float threshold = DEFAULT_CLEAR_OLD_ERASED_VALUE_THRESHOLD) const { |
| 1420 |
if(this->m_values.size() == 0) { |
| 1421 |
return false; |
| 1422 |
} |
| 1423 |
|
| 1424 |
return float(m_nb_elements)/float(this->m_values.size()) < threshold; |
| 1425 |
} |
| 1426 |
|
| 1427 |
template<class U = T, typename std::enable_if<!has_mapped_type<U>::value>::type* = nullptr> |
| 1428 |
void clear_old_erased_values() { |
| 1429 |
} |
| 1430 |
|
| 1431 |
template<class U = T, typename std::enable_if<has_mapped_type<U>::value>::type* = nullptr> |
| 1432 |
void clear_old_erased_values() { |
| 1433 |
static_assert(std::is_nothrow_move_constructible<U>::value || |
| 1434 |
std::is_copy_constructible<U>::value, |
| 1435 |
"mapped_value must be either copy constructible or nothrow move constructible."); |
| 1436 |
|
| 1437 |
if(m_nb_elements == this->m_values.size()) { |
| 1438 |
return; |
| 1439 |
} |
| 1440 |
|
| 1441 |
std::vector<T> new_values; |
| 1442 |
new_values.reserve(size()); |
| 1443 |
|
| 1444 |
for(auto it = begin(); it != end(); ++it) { |
| 1445 |
new_values.push_back(std::move_if_noexcept(it.value())); |
| 1446 |
} |
| 1447 |
|
| 1448 |
|
| 1449 |
IndexSizeT ivalue = 0; |
| 1450 |
for(auto it = begin(); it != end(); ++it) { |
| 1451 |
auto it_array_bucket = it.m_buckets_iterator->mutable_iterator(it.m_array_bucket_iterator); |
| 1452 |
it_array_bucket.set_value(ivalue); |
| 1453 |
ivalue++; |
| 1454 |
} |
| 1455 |
|
| 1456 |
new_values.swap(this->m_values); |
| 1457 |
tsl_ah_assert(m_nb_elements == this->m_values.size()); |
| 1458 |
} |
| 1459 |
|
| 1460 |
/** |
| 1461 |
* Return true if a rehash occurred. |
| 1462 |
*/ |
| 1463 |
bool grow_on_high_load() { |
| 1464 |
if(size() >= m_load_threshold) { |
| 1465 |
rehash_impl(GrowthPolicy::next_bucket_count()); |
| 1466 |
return true; |
| 1467 |
} |
| 1468 |
|
| 1469 |
return false; |
| 1470 |
} |
| 1471 |
|
| 1472 |
template<class... ValueArgs, class U = T, typename std::enable_if<has_mapped_type<U>::value>::type* = nullptr> |
| 1473 |
std::pair<iterator, bool> emplace_impl(std::size_t ibucket, typename array_bucket::const_iterator end_of_bucket, |
| 1474 |
const CharT* key, size_type key_size, ValueArgs&&... value_args) |
| 1475 |
{ |
| 1476 |
if(this->m_values.size() >= max_size()) { |
| 1477 |
// Try to clear old erased values lingering in m_values. Throw if it doesn't change anything. |
| 1478 |
clear_old_erased_values(); |
| 1479 |
if(this->m_values.size() >= max_size()) { |
| 1480 |
THROW(std::length_error, "Can't insert value, too much values in the map."); |
| 1481 |
} |
| 1482 |
} |
| 1483 |
|
| 1484 |
if(this->m_values.size() == this->m_values.capacity()) { |
| 1485 |
this->m_values.reserve(std::size_t(float(this->m_values.size()) * value_container<T>::VECTOR_GROWTH_RATE)); |
| 1486 |
} |
| 1487 |
|
| 1488 |
|
| 1489 |
this->m_values.emplace_back(std::forward<ValueArgs>(value_args)...); |
| 1490 |
|
| 1491 |
auto it = m_buckets[ibucket].append(end_of_bucket, key, key_size, IndexSizeT(this->m_values.size() - 1)); |
| 1492 |
m_nb_elements++; |
| 1493 |
|
| 1494 |
return std::make_pair(iterator(m_buckets_data.begin() + ibucket, it, this), true); |
| 1495 |
} |
| 1496 |
|
| 1497 |
template<class U = T, typename std::enable_if<!has_mapped_type<U>::value>::type* = nullptr> |
| 1498 |
std::pair<iterator, bool> emplace_impl(std::size_t ibucket, typename array_bucket::const_iterator end_of_bucket, |
| 1499 |
const CharT* key, size_type key_size) |
| 1500 |
{ |
| 1501 |
if(m_nb_elements >= max_size()) { |
| 1502 |
THROW(std::length_error, "Can't insert value, too much values in the map."); |
| 1503 |
} |
| 1504 |
|
| 1505 |
auto it = m_buckets[ibucket].append(end_of_bucket, key, key_size); |
| 1506 |
m_nb_elements++; |
| 1507 |
|
| 1508 |
return std::make_pair(iterator(m_buckets_data.begin() + ibucket, it, this), true); |
| 1509 |
} |
| 1510 |
|
| 1511 |
void rehash_impl(size_type bucket_count) { |
| 1512 |
GrowthPolicy new_growth_policy(bucket_count); |
| 1513 |
if(bucket_count == this->bucket_count()) { |
| 1514 |
return; |
| 1515 |
} |
| 1516 |
|
| 1517 |
|
| 1518 |
if(should_clear_old_erased_values(REHASH_CLEAR_OLD_ERASED_VALUE_THRESHOLD)) { |
| 1519 |
clear_old_erased_values(); |
| 1520 |
} |
| 1521 |
|
| 1522 |
|
| 1523 |
std::vector<std::size_t> required_size_for_bucket(bucket_count, 0); |
| 1524 |
std::vector<std::size_t> bucket_for_ivalue(size(), 0); |
| 1525 |
|
| 1526 |
std::size_t ivalue = 0; |
| 1527 |
for(auto it = begin(); it != end(); ++it) { |
| 1528 |
const std::size_t hash = hash_key(it.key(), it.key_size()); |
| 1529 |
const std::size_t ibucket = new_growth_policy.bucket_for_hash(hash); |
| 1530 |
|
| 1531 |
bucket_for_ivalue[ivalue] = ibucket; |
| 1532 |
required_size_for_bucket[ibucket] += array_bucket::entry_required_bytes(it.key_size()); |
| 1533 |
ivalue++; |
| 1534 |
} |
| 1535 |
|
| 1536 |
|
| 1537 |
|
| 1538 |
|
| 1539 |
std::vector<array_bucket> new_buckets; |
| 1540 |
new_buckets.reserve(bucket_count); |
| 1541 |
for(std::size_t ibucket = 0; ibucket < bucket_count; ibucket++) { |
| 1542 |
new_buckets.emplace_back(required_size_for_bucket[ibucket]); |
| 1543 |
} |
| 1544 |
|
| 1545 |
|
| 1546 |
ivalue = 0; |
| 1547 |
for(auto it = begin(); it != end(); ++it) { |
| 1548 |
const std::size_t ibucket = bucket_for_ivalue[ivalue]; |
| 1549 |
append_iterator_in_reserved_bucket_no_check(new_buckets[ibucket], it); |
| 1550 |
|
| 1551 |
ivalue++; |
| 1552 |
} |
| 1553 |
|
| 1554 |
|
| 1555 |
using std::swap; |
| 1556 |
swap(static_cast<GrowthPolicy&>(*this), new_growth_policy); |
| 1557 |
|
| 1558 |
m_buckets_data.swap(new_buckets); |
| 1559 |
m_buckets = !m_buckets_data.empty()?m_buckets_data.data(): |
| 1560 |
static_empty_bucket_ptr(); |
| 1561 |
|
| 1562 |
// Call max_load_factor to change m_load_threshold |
| 1563 |
max_load_factor(m_max_load_factor); |
| 1564 |
} |
| 1565 |
|
| 1566 |
template<class U = T, typename std::enable_if<!has_mapped_type<U>::value>::type* = nullptr> |
| 1567 |
void append_iterator_in_reserved_bucket_no_check(array_bucket& bucket, iterator it) { |
| 1568 |
bucket.append_in_reserved_bucket_no_check(it.key(), it.key_size()); |
| 1569 |
} |
| 1570 |
|
| 1571 |
template<class U = T, typename std::enable_if<has_mapped_type<U>::value>::type* = nullptr> |
| 1572 |
void append_iterator_in_reserved_bucket_no_check(array_bucket& bucket, iterator it) { |
| 1573 |
bucket.append_in_reserved_bucket_no_check(it.key(), it.key_size(), it.value_position()); |
| 1574 |
} |
| 1575 |
|
| 1576 |
|
| 1577 |
|
| 1578 |
/** |
| 1579 |
* On serialization the values of each bucket (if has_mapped_type is true) are serialized |
| 1580 |
* next to the bucket. The potential old erased values in value_container are thus not serialized. |
| 1581 |
* |
| 1582 |
* On deserialization, when hash_compatible is true, we reaffect the value index (IndexSizeT) of each |
| 1583 |
* bucket with set_value as the position of each value is no more the same in value_container compared |
| 1584 |
* to when they were serialized. |
| 1585 |
* |
| 1586 |
* It's done this way as we can't call clear_old_erased_values() because we want the serialize |
| 1587 |
* method to remain const and we don't want to serialize/deserialize old erased values. As we may |
| 1588 |
* not serialize all the values in value_container, the values we keep can change of index. |
| 1589 |
* We thus have to modify the value indexes in the buckets. |
| 1590 |
*/ |
| 1591 |
template<class Serializer> |
| 1592 |
void serialize_impl(Serializer& serializer) const { |
| 1593 |
const slz_size_type version = SERIALIZATION_PROTOCOL_VERSION; |
| 1594 |
serializer(version); |
| 1595 |
|
| 1596 |
const slz_size_type bucket_count = m_buckets_data.size(); |
| 1597 |
serializer(bucket_count); |
| 1598 |
|
| 1599 |
const slz_size_type nb_elements = m_nb_elements; |
| 1600 |
serializer(nb_elements); |
| 1601 |
|
| 1602 |
const float max_load_factor = m_max_load_factor; |
| 1603 |
serializer(max_load_factor); |
| 1604 |
|
| 1605 |
for(const array_bucket& bucket: m_buckets_data) { |
| 1606 |
bucket.serialize(serializer); |
| 1607 |
serialize_bucket_values(serializer, bucket); |
| 1608 |
} |
| 1609 |
} |
| 1610 |
|
| 1611 |
template<class Serializer, class U = T, |
| 1612 |
typename std::enable_if<!has_mapped_type<U>::value>::type* = nullptr> |
| 1613 |
void serialize_bucket_values(Serializer& /*serializer*/, const array_bucket& /*bucket*/) const { |
| 1614 |
} |
| 1615 |
|
| 1616 |
template<class Serializer, class U = T, |
| 1617 |
typename std::enable_if<has_mapped_type<U>::value>::type* = nullptr> |
| 1618 |
void serialize_bucket_values(Serializer& serializer, const array_bucket& bucket) const { |
| 1619 |
for(auto it = bucket.begin(); it != bucket.end(); ++it) { |
| 1620 |
serializer(this->m_values[it.value()]); |
| 1621 |
} |
| 1622 |
} |
| 1623 |
|
| 1624 |
template<class Deserializer> |
| 1625 |
void deserialize_impl(Deserializer& deserializer, bool hash_compatible) { |
| 1626 |
tsl_ah_assert(m_buckets_data.empty()); // Current hash table must be empty |
| 1627 |
|
| 1628 |
const slz_size_type version = deserialize_value<slz_size_type>(deserializer); |
| 1629 |
// For now we only have one version of the serialization protocol. |
| 1630 |
// If it doesn't match there is a problem with the file. |
| 1631 |
if(version != SERIALIZATION_PROTOCOL_VERSION) { |
| 1632 |
THROW(std::runtime_error, "Can't deserialize the array_map/set. The protocol version header is invalid."); |
| 1633 |
} |
| 1634 |
|
| 1635 |
const slz_size_type bucket_count_ds = deserialize_value<slz_size_type>(deserializer); |
| 1636 |
const slz_size_type nb_elements = deserialize_value<slz_size_type>(deserializer); |
| 1637 |
const float max_load_factor = deserialize_value<float>(deserializer); |
| 1638 |
|
| 1639 |
|
| 1640 |
m_nb_elements = numeric_cast<IndexSizeT>(nb_elements, "Deserialized nb_elements is too big."); |
| 1641 |
|
| 1642 |
size_type bucket_count = numeric_cast<size_type>(bucket_count_ds, "Deserialized bucket_count is too big."); |
| 1643 |
GrowthPolicy::operator=(GrowthPolicy(bucket_count)); |
| 1644 |
|
| 1645 |
|
| 1646 |
this->max_load_factor(max_load_factor); |
| 1647 |
value_container<T>::reserve(m_nb_elements); |
| 1648 |
|
| 1649 |
|
| 1650 |
if(hash_compatible) { |
| 1651 |
if(bucket_count != bucket_count_ds) { |
| 1652 |
THROW(std::runtime_error, "The GrowthPolicy is not the same even though hash_compatible is true."); |
| 1653 |
} |
| 1654 |
|
| 1655 |
m_buckets_data.reserve(bucket_count); |
| 1656 |
for(size_type i = 0; i < bucket_count; i++) { |
| 1657 |
m_buckets_data.push_back(array_bucket::deserialize(deserializer)); |
| 1658 |
deserialize_bucket_values(deserializer, m_buckets_data.back()); |
| 1659 |
} |
| 1660 |
} |
| 1661 |
else { |
| 1662 |
m_buckets_data.resize(bucket_count); |
| 1663 |
for(size_type i = 0; i < bucket_count; i++) { |
| 1664 |
// TODO use buffer to avoid reallocation on each deserialization. |
| 1665 |
array_bucket bucket = array_bucket::deserialize(deserializer); |
| 1666 |
deserialize_bucket_values(deserializer, bucket); |
| 1667 |
|
| 1668 |
for(auto it_val = bucket.cbegin(); it_val != bucket.cend(); ++it_val) { |
| 1669 |
const std::size_t ibucket = bucket_for_hash(hash_key(it_val.key(), it_val.key_size())); |
| 1670 |
|
| 1671 |
auto it_find = m_buckets_data[ibucket].find_or_end_of_bucket(it_val.key(), it_val.key_size()); |
| 1672 |
if(it_find.second) { |
| 1673 |
THROW(std::runtime_error, "Error on deserialization, the same key is presents multiple times."); |
| 1674 |
} |
| 1675 |
|
| 1676 |
append_array_bucket_iterator_in_bucket(m_buckets_data[ibucket], it_find.first, it_val); |
| 1677 |
} |
| 1678 |
} |
| 1679 |
} |
| 1680 |
|
| 1681 |
m_buckets = m_buckets_data.data(); |
| 1682 |
|
| 1683 |
|
| 1684 |
if(load_factor() > this->max_load_factor()) { |
| 1685 |
THROW(std::runtime_error, "Invalid max_load_factor. Check that the serializer and deserializer support " |
| 1686 |
"floats correctly as they can be converted implicitely to ints."); |
| 1687 |
} |
| 1688 |
} |
| 1689 |
|
| 1690 |
template<class Deserializer, class U = T, |
| 1691 |
typename std::enable_if<!has_mapped_type<U>::value>::type* = nullptr> |
| 1692 |
void deserialize_bucket_values(Deserializer& /*deserializer*/, array_bucket& /*bucket*/) { |
| 1693 |
} |
| 1694 |
|
| 1695 |
template<class Deserializer, class U = T, |
| 1696 |
typename std::enable_if<has_mapped_type<U>::value>::type* = nullptr> |
| 1697 |
void deserialize_bucket_values(Deserializer& deserializer, array_bucket& bucket) { |
| 1698 |
for(auto it = bucket.begin(); it != bucket.end(); ++it) { |
| 1699 |
this->m_values.emplace_back(deserialize_value<U>(deserializer)); |
| 1700 |
|
| 1701 |
tsl_ah_assert(this->m_values.size() - 1 <= std::numeric_limits<IndexSizeT>::max()); |
| 1702 |
it.set_value(static_cast<IndexSizeT>(this->m_values.size() - 1)); |
| 1703 |
} |
| 1704 |
} |
| 1705 |
|
| 1706 |
template<class U = T, typename std::enable_if<!has_mapped_type<U>::value>::type* = nullptr> |
| 1707 |
void append_array_bucket_iterator_in_bucket(array_bucket& bucket, |
| 1708 |
typename array_bucket::const_iterator end_of_bucket, |
| 1709 |
typename array_bucket::const_iterator it_val) |
| 1710 |
{ |
| 1711 |
bucket.append(end_of_bucket, it_val.key(), it_val.key_size()); |
| 1712 |
} |
| 1713 |
|
| 1714 |
template<class U = T, typename std::enable_if<has_mapped_type<U>::value>::type* = nullptr> |
| 1715 |
void append_array_bucket_iterator_in_bucket(array_bucket& bucket, |
| 1716 |
typename array_bucket::const_iterator end_of_bucket, |
| 1717 |
typename array_bucket::const_iterator it_val) |
| 1718 |
{ |
| 1719 |
bucket.append(end_of_bucket, it_val.key(), it_val.key_size(), it_val.value()); |
| 1720 |
} |
| 1721 |
|
| 1722 |
public: |
| 1723 |
static const size_type DEFAULT_INIT_BUCKET_COUNT = 0; |
| 1724 |
static constexpr float DEFAULT_MAX_LOAD_FACTOR = 2.0f; |
| 1725 |
static const size_type MAX_KEY_SIZE = array_bucket::MAX_KEY_SIZE; |
| 1726 |
|
| 1727 |
private: |
| 1728 |
/** |
| 1729 |
* Protocol version currenlty used for serialization. |
| 1730 |
*/ |
| 1731 |
static const slz_size_type SERIALIZATION_PROTOCOL_VERSION = 1; |
| 1732 |
|
| 1733 |
|
| 1734 |
static constexpr float DEFAULT_CLEAR_OLD_ERASED_VALUE_THRESHOLD = 0.6f; |
| 1735 |
static constexpr float REHASH_CLEAR_OLD_ERASED_VALUE_THRESHOLD = 0.9f; |
| 1736 |
|
| 1737 |
|
| 1738 |
/** |
| 1739 |
* Return an always valid pointer to a static empty array_bucket. |
| 1740 |
*/ |
| 1741 |
array_bucket* static_empty_bucket_ptr() { |
| 1742 |
static array_bucket empty_bucket; |
| 1743 |
return &empty_bucket; |
| 1744 |
} |
| 1745 |
|
| 1746 |
private: |
| 1747 |
std::vector<array_bucket> m_buckets_data; |
| 1748 |
|
| 1749 |
/** |
| 1750 |
* Points to m_buckets_data.data() if !m_buckets_data.empty() otherwise points to static_empty_bucket_ptr. |
| 1751 |
* This variable is useful to avoid the cost of checking if m_buckets_data is empty when trying |
| 1752 |
* to find an element. |
| 1753 |
* |
| 1754 |
* TODO Remove m_buckets_data and only use a pointer+size instead of a pointer+vector to save some space in the array_hash object. |
| 1755 |
*/ |
| 1756 |
array_bucket* m_buckets; |
| 1757 |
|
| 1758 |
IndexSizeT m_nb_elements; |
| 1759 |
float m_max_load_factor; |
| 1760 |
size_type m_load_threshold; |
| 1761 |
}; |
| 1762 |
|
| 1763 |
} // end namespace detail_array_hash |
| 1764 |
} //end namespace tsl |
| 1765 |
|
| 1766 |
#endif |