| 1 | #ifndef SIMDJSON_INLINE_DOCUMENT_STREAM_H |
| 2 | #define SIMDJSON_INLINE_DOCUMENT_STREAM_H |
| 3 | |
| 4 | #include "simdjson/dom/document_stream.h" |
| 5 | #include <algorithm> |
| 6 | #include <limits> |
| 7 | #include <stdexcept> |
| 8 | namespace simdjson { |
| 9 | namespace dom { |
| 10 | |
| 11 | #ifdef SIMDJSON_THREADS_ENABLED |
| 12 | inline void stage1_worker::finish() { |
| 13 | // After calling "run" someone would call finish() to wait |
| 14 | // for the end of the processing. |
| 15 | // This function will wait until either the thread has done |
| 16 | // the processing or, else, the destructor has been called. |
| 17 | std::unique_lock<std::mutex> lock(locking_mutex); |
| 18 | cond_var.wait(lock&: lock, p: [this]{return has_work == false;}); |
| 19 | } |
| 20 | |
| 21 | inline stage1_worker::~stage1_worker() { |
| 22 | // The thread may never outlive the stage1_worker instance |
| 23 | // and will always be stopped/joined before the stage1_worker |
| 24 | // instance is gone. |
| 25 | stop_thread(); |
| 26 | } |
| 27 | |
| 28 | inline void stage1_worker::start_thread() { |
| 29 | std::unique_lock<std::mutex> lock(locking_mutex); |
| 30 | if(thread.joinable()) { |
| 31 | return; // This should never happen but we never want to create more than one thread. |
| 32 | } |
| 33 | thread = std::thread([this]{ |
| 34 | while(true) { |
| 35 | std::unique_lock<std::mutex> thread_lock(locking_mutex); |
| 36 | // We wait for either "run" or "stop_thread" to be called. |
| 37 | cond_var.wait(lock&: thread_lock, p: [this]{return has_work || !can_work;}); |
| 38 | // If, for some reason, the stop_thread() method was called (i.e., the |
| 39 | // destructor of stage1_worker is called, then we want to immediately destroy |
| 40 | // the thread (and not do any more processing). |
| 41 | if(!can_work) { |
| 42 | break; |
| 43 | } |
| 44 | this->owner->stage1_thread_error = this->owner->run_stage1(p&: *this->stage1_thread_parser, |
| 45 | batch_start: this->_next_batch_start); |
| 46 | this->has_work = false; |
| 47 | // The condition variable call should be moved after thread_lock.unlock() for performance |
| 48 | // reasons but thread sanitizers may report it as a data race if we do. |
| 49 | // See https://stackoverflow.com/questions/35775501/c-should-condition-variable-be-notified-under-lock |
| 50 | cond_var.notify_one(); // will notify "finish" |
| 51 | thread_lock.unlock(); |
| 52 | } |
| 53 | } |
| 54 | ); |
| 55 | } |
| 56 | |
| 57 | |
| 58 | inline void stage1_worker::stop_thread() { |
| 59 | std::unique_lock<std::mutex> lock(locking_mutex); |
| 60 | // We have to make sure that all locks can be released. |
| 61 | can_work = false; |
| 62 | has_work = false; |
| 63 | cond_var.notify_all(); |
| 64 | lock.unlock(); |
| 65 | if(thread.joinable()) { |
| 66 | thread.join(); |
| 67 | } |
| 68 | } |
| 69 | |
| 70 | inline void stage1_worker::run(document_stream * ds, dom::parser * stage1, size_t next_batch_start) { |
| 71 | std::unique_lock<std::mutex> lock(locking_mutex); |
| 72 | owner = ds; |
| 73 | _next_batch_start = next_batch_start; |
| 74 | stage1_thread_parser = stage1; |
| 75 | has_work = true; |
| 76 | // The condition variable call should be moved after thread_lock.unlock() for performance |
| 77 | // reasons but thread sanitizers may report it as a data race if we do. |
| 78 | // See https://stackoverflow.com/questions/35775501/c-should-condition-variable-be-notified-under-lock |
| 79 | cond_var.notify_one(); // will notify the thread lock that we have work |
| 80 | lock.unlock(); |
| 81 | } |
| 82 | #endif |
| 83 | |
| 84 | simdjson_inline document_stream::document_stream( |
| 85 | dom::parser &_parser, |
| 86 | const uint8_t *_buf, |
| 87 | size_t _len, |
| 88 | size_t _batch_size |
| 89 | ) noexcept |
| 90 | : parser{&_parser}, |
| 91 | buf{_buf}, |
| 92 | len{_len}, |
| 93 | batch_size{_batch_size <= MINIMAL_BATCH_SIZE ? MINIMAL_BATCH_SIZE : _batch_size}, |
| 94 | error{SUCCESS} |
| 95 | #ifdef SIMDJSON_THREADS_ENABLED |
| 96 | , use_thread(_parser.threaded) // we need to make a copy because _parser.threaded can change |
| 97 | #endif |
| 98 | { |
| 99 | #ifdef SIMDJSON_THREADS_ENABLED |
| 100 | if(worker.get() == nullptr) { |
| 101 | error = MEMALLOC; |
| 102 | } |
| 103 | #endif |
| 104 | } |
| 105 | |
| 106 | simdjson_inline document_stream::document_stream() noexcept |
| 107 | : parser{nullptr}, |
| 108 | buf{nullptr}, |
| 109 | len{0}, |
| 110 | batch_size{0}, |
| 111 | error{UNINITIALIZED} |
| 112 | #ifdef SIMDJSON_THREADS_ENABLED |
| 113 | , use_thread(false) |
| 114 | #endif |
| 115 | { |
| 116 | } |
| 117 | |
| 118 | simdjson_inline document_stream::~document_stream() noexcept { |
| 119 | #ifdef SIMDJSON_THREADS_ENABLED |
| 120 | worker.reset(); |
| 121 | #endif |
| 122 | } |
| 123 | |
| 124 | simdjson_inline document_stream::iterator::iterator() noexcept |
| 125 | : stream{nullptr}, finished{true} { |
| 126 | } |
| 127 | |
| 128 | simdjson_inline document_stream::iterator document_stream::begin() noexcept { |
| 129 | start(); |
| 130 | // If there are no documents, we're finished. |
| 131 | return iterator(this, error == EMPTY); |
| 132 | } |
| 133 | |
| 134 | simdjson_inline document_stream::iterator document_stream::end() noexcept { |
| 135 | return iterator(this, true); |
| 136 | } |
| 137 | |
| 138 | simdjson_inline document_stream::iterator::iterator(document_stream* _stream, bool is_end) noexcept |
| 139 | : stream{_stream}, finished{is_end} { |
| 140 | } |
| 141 | |
| 142 | simdjson_inline document_stream::iterator::reference document_stream::iterator::operator*() noexcept { |
| 143 | // Note that in case of error, we do not yet mark |
| 144 | // the iterator as "finished": this detection is done |
| 145 | // in the operator++ function since it is possible |
| 146 | // to call operator++ repeatedly while omitting |
| 147 | // calls to operator*. |
| 148 | if (stream->error) { return stream->error; } |
| 149 | return stream->parser->doc.root(); |
| 150 | } |
| 151 | |
| 152 | simdjson_inline document_stream::iterator& document_stream::iterator::operator++() noexcept { |
| 153 | // If there is an error, then we want the iterator |
| 154 | // to be finished, no matter what. (E.g., we do not |
| 155 | // keep generating documents with errors, or go beyond |
| 156 | // a document with errors.) |
| 157 | // |
| 158 | // Users do not have to call "operator*()" when they use operator++, |
| 159 | // so we need to end the stream in the operator++ function. |
| 160 | // |
| 161 | // Note that setting finished = true is essential otherwise |
| 162 | // we would enter an infinite loop. |
| 163 | if (stream->error) { finished = true; } |
| 164 | // Note that stream->error() is guarded against error conditions |
| 165 | // (it will immediately return if stream->error casts to false). |
| 166 | // In effect, this next function does nothing when (stream->error) |
| 167 | // is true (hence the risk of an infinite loop). |
| 168 | stream->next(); |
| 169 | // If that was the last document, we're finished. |
| 170 | // It is the only type of error we do not want to appear |
| 171 | // in operator*. |
| 172 | if (stream->error == EMPTY) { finished = true; } |
| 173 | // If we had any other kind of error (not EMPTY) then we want |
| 174 | // to pass it along to the operator* and we cannot mark the result |
| 175 | // as "finished" just yet. |
| 176 | return *this; |
| 177 | } |
| 178 | |
| 179 | simdjson_inline bool document_stream::iterator::operator!=(const document_stream::iterator &other) const noexcept { |
| 180 | return finished != other.finished; |
| 181 | } |
| 182 | |
| 183 | inline void document_stream::start() noexcept { |
| 184 | if (error) { return; } |
| 185 | error = parser->ensure_capacity(desired_capacity: batch_size); |
| 186 | if (error) { return; } |
| 187 | // Always run the first stage 1 parse immediately |
| 188 | batch_start = 0; |
| 189 | error = run_stage1(p&: *parser, batch_start); |
| 190 | while(error == EMPTY) { |
| 191 | // In exceptional cases, we may start with an empty block |
| 192 | batch_start = next_batch_start(); |
| 193 | if (batch_start >= len) { return; } |
| 194 | error = run_stage1(p&: *parser, batch_start); |
| 195 | } |
| 196 | if (error) { return; } |
| 197 | #ifdef SIMDJSON_THREADS_ENABLED |
| 198 | if (use_thread && next_batch_start() < len) { |
| 199 | // Kick off the first thread if needed |
| 200 | error = stage1_thread_parser.ensure_capacity(desired_capacity: batch_size); |
| 201 | if (error) { return; } |
| 202 | worker->start_thread(); |
| 203 | start_stage1_thread(); |
| 204 | if (error) { return; } |
| 205 | } |
| 206 | #endif // SIMDJSON_THREADS_ENABLED |
| 207 | next(); |
| 208 | } |
| 209 | |
| 210 | simdjson_inline size_t document_stream::iterator::current_index() const noexcept { |
| 211 | return stream->doc_index; |
| 212 | } |
| 213 | |
| 214 | simdjson_inline std::string_view document_stream::iterator::source() const noexcept { |
| 215 | const char* start = reinterpret_cast<const char*>(stream->buf) + current_index(); |
| 216 | bool object_or_array = ((*start == '[') || (*start == '{')); |
| 217 | if(object_or_array) { |
| 218 | size_t next_doc_index = stream->batch_start + stream->parser->implementation->structural_indexes[stream->parser->implementation->next_structural_index - 1]; |
| 219 | return std::string_view(start, next_doc_index - current_index() + 1); |
| 220 | } else { |
| 221 | size_t next_doc_index = stream->batch_start + stream->parser->implementation->structural_indexes[stream->parser->implementation->next_structural_index]; |
| 222 | return std::string_view(reinterpret_cast<const char*>(stream->buf) + current_index(), next_doc_index - current_index() - 1); |
| 223 | } |
| 224 | } |
| 225 | |
| 226 | |
| 227 | inline void document_stream::next() noexcept { |
| 228 | // We always exit at once, once in an error condition. |
| 229 | if (error) { return; } |
| 230 | |
| 231 | // Load the next document from the batch |
| 232 | doc_index = batch_start + parser->implementation->structural_indexes[parser->implementation->next_structural_index]; |
| 233 | error = parser->implementation->stage2_next(doc&: parser->doc); |
| 234 | // If that was the last document in the batch, load another batch (if available) |
| 235 | while (error == EMPTY) { |
| 236 | batch_start = next_batch_start(); |
| 237 | if (batch_start >= len) { break; } |
| 238 | |
| 239 | #ifdef SIMDJSON_THREADS_ENABLED |
| 240 | if(use_thread) { |
| 241 | load_from_stage1_thread(); |
| 242 | } else { |
| 243 | error = run_stage1(p&: *parser, batch_start); |
| 244 | } |
| 245 | #else |
| 246 | error = run_stage1(*parser, batch_start); |
| 247 | #endif |
| 248 | if (error) { continue; } // If the error was EMPTY, we may want to load another batch. |
| 249 | // Run stage 2 on the first document in the batch |
| 250 | doc_index = batch_start + parser->implementation->structural_indexes[parser->implementation->next_structural_index]; |
| 251 | error = parser->implementation->stage2_next(doc&: parser->doc); |
| 252 | } |
| 253 | } |
| 254 | inline size_t document_stream::size_in_bytes() const noexcept { |
| 255 | return len; |
| 256 | } |
| 257 | |
| 258 | inline size_t document_stream::truncated_bytes() const noexcept { |
| 259 | if(error == CAPACITY) { return len - batch_start; } |
| 260 | return parser->implementation->structural_indexes[parser->implementation->n_structural_indexes] - parser->implementation->structural_indexes[parser->implementation->n_structural_indexes + 1]; |
| 261 | } |
| 262 | |
| 263 | inline size_t document_stream::next_batch_start() const noexcept { |
| 264 | return batch_start + parser->implementation->structural_indexes[parser->implementation->n_structural_indexes]; |
| 265 | } |
| 266 | |
| 267 | inline error_code document_stream::run_stage1(dom::parser &p, size_t _batch_start) noexcept { |
| 268 | size_t remaining = len - _batch_start; |
| 269 | if (remaining <= batch_size) { |
| 270 | return p.implementation->stage1(buf: &buf[_batch_start], len: remaining, streaming: stage1_mode::streaming_final); |
| 271 | } else { |
| 272 | return p.implementation->stage1(buf: &buf[_batch_start], len: batch_size, streaming: stage1_mode::streaming_partial); |
| 273 | } |
| 274 | } |
| 275 | |
| 276 | #ifdef SIMDJSON_THREADS_ENABLED |
| 277 | |
| 278 | inline void document_stream::load_from_stage1_thread() noexcept { |
| 279 | worker->finish(); |
| 280 | // Swap to the parser that was loaded up in the thread. Make sure the parser has |
| 281 | // enough memory to swap to, as well. |
| 282 | std::swap(a&: *parser, b&: stage1_thread_parser); |
| 283 | error = stage1_thread_error; |
| 284 | if (error) { return; } |
| 285 | |
| 286 | // If there's anything left, start the stage 1 thread! |
| 287 | if (next_batch_start() < len) { |
| 288 | start_stage1_thread(); |
| 289 | } |
| 290 | } |
| 291 | |
| 292 | inline void document_stream::start_stage1_thread() noexcept { |
| 293 | // we call the thread on a lambda that will update |
| 294 | // this->stage1_thread_error |
| 295 | // there is only one thread that may write to this value |
| 296 | // TODO this is NOT exception-safe. |
| 297 | this->stage1_thread_error = UNINITIALIZED; // In case something goes wrong, make sure it's an error |
| 298 | size_t _next_batch_start = this->next_batch_start(); |
| 299 | |
| 300 | worker->run(ds: this, stage1: & this->stage1_thread_parser, next_batch_start: _next_batch_start); |
| 301 | } |
| 302 | |
| 303 | #endif // SIMDJSON_THREADS_ENABLED |
| 304 | |
| 305 | } // namespace dom |
| 306 | |
| 307 | simdjson_inline simdjson_result<dom::document_stream>::simdjson_result() noexcept |
| 308 | : simdjson_result_base() { |
| 309 | } |
| 310 | simdjson_inline simdjson_result<dom::document_stream>::simdjson_result(error_code error) noexcept |
| 311 | : simdjson_result_base(error) { |
| 312 | } |
| 313 | simdjson_inline simdjson_result<dom::document_stream>::simdjson_result(dom::document_stream &&value) noexcept |
| 314 | : simdjson_result_base(std::forward<dom::document_stream>(t&: value)) { |
| 315 | } |
| 316 | |
| 317 | #if SIMDJSON_EXCEPTIONS |
| 318 | simdjson_inline dom::document_stream::iterator simdjson_result<dom::document_stream>::begin() noexcept(false) { |
| 319 | if (error()) { throw simdjson_error(error()); } |
| 320 | return first.begin(); |
| 321 | } |
| 322 | simdjson_inline dom::document_stream::iterator simdjson_result<dom::document_stream>::end() noexcept(false) { |
| 323 | if (error()) { throw simdjson_error(error()); } |
| 324 | return first.end(); |
| 325 | } |
| 326 | #else // SIMDJSON_EXCEPTIONS |
| 327 | #ifndef SIMDJSON_DISABLE_DEPRECATED_API |
| 328 | simdjson_inline dom::document_stream::iterator simdjson_result<dom::document_stream>::begin() noexcept { |
| 329 | first.error = error(); |
| 330 | return first.begin(); |
| 331 | } |
| 332 | simdjson_inline dom::document_stream::iterator simdjson_result<dom::document_stream>::end() noexcept { |
| 333 | first.error = error(); |
| 334 | return first.end(); |
| 335 | } |
| 336 | #endif // SIMDJSON_DISABLE_DEPRECATED_API |
| 337 | #endif // SIMDJSON_EXCEPTIONS |
| 338 | |
| 339 | } // namespace simdjson |
| 340 | #endif // SIMDJSON_INLINE_DOCUMENT_STREAM_H |
| 341 | |