| 1 | /* |
| 2 | * Copyright (c) 2015-2017, Intel Corporation |
| 3 | * |
| 4 | * Redistribution and use in source and binary forms, with or without |
| 5 | * modification, are permitted provided that the following conditions are met: |
| 6 | * |
| 7 | * * Redistributions of source code must retain the above copyright notice, |
| 8 | * this list of conditions and the following disclaimer. |
| 9 | * * Redistributions in binary form must reproduce the above copyright |
| 10 | * notice, this list of conditions and the following disclaimer in the |
| 11 | * documentation and/or other materials provided with the distribution. |
| 12 | * * Neither the name of Intel Corporation nor the names of its contributors |
| 13 | * may be used to endorse or promote products derived from this software |
| 14 | * without specific prior written permission. |
| 15 | * |
| 16 | * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" |
| 17 | * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE |
| 18 | * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE |
| 19 | * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE |
| 20 | * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR |
| 21 | * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF |
| 22 | * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS |
| 23 | * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN |
| 24 | * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) |
| 25 | * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE |
| 26 | * POSSIBILITY OF SUCH DAMAGE. |
| 27 | */ |
| 28 | |
| 29 | /** \file |
| 30 | * \brief Corpus Generation tool. |
| 31 | */ |
| 32 | |
| 33 | #include "config.h" |
| 34 | |
| 35 | #include "ng_corpus_generator.h" |
| 36 | |
| 37 | #include "ng_corpus_editor.h" |
| 38 | #include "compiler/compiler.h" |
| 39 | #include "nfagraph/ng.h" |
| 40 | #include "nfagraph/ng_util.h" |
| 41 | #include "ue2common.h" |
| 42 | #include "util/container.h" |
| 43 | #include "util/graph_range.h" |
| 44 | #include "util/make_unique.h" |
| 45 | #include "util/ue2string.h" |
| 46 | #include "util/unicode_def.h" |
| 47 | #include "util/unicode_set.h" |
| 48 | |
| 49 | #include <algorithm> |
| 50 | #include <deque> |
| 51 | #include <memory> |
| 52 | #include <set> |
| 53 | #include <sstream> |
| 54 | #include <unordered_set> |
| 55 | #include <vector> |
| 56 | |
| 57 | #include <boost/utility.hpp> |
| 58 | |
| 59 | using namespace std; |
| 60 | using namespace ue2; |
| 61 | |
| 62 | typedef vector<NFAVertex> VertexPath; |
| 63 | |
| 64 | #if defined(DEBUG) |
| 65 | // For debugging output |
| 66 | static |
| 67 | string pathToString(const NGHolder &g, const VertexPath &p) { |
| 68 | ostringstream oss; |
| 69 | oss << '['; |
| 70 | for (auto i = p.begin(); i != p.end(); ++i) { |
| 71 | if (i != p.begin()) { |
| 72 | oss << ','; |
| 73 | } |
| 74 | oss << g[*i].index; |
| 75 | } |
| 76 | oss << ']'; |
| 77 | return oss.str(); |
| 78 | } |
| 79 | #endif |
| 80 | |
| 81 | /** True if this graph has no non-special successors of start or startDs. */ |
| 82 | static |
| 83 | bool graph_is_empty(const NGHolder &g) { |
| 84 | for (const auto &v : adjacent_vertices_range(g.start, g)) { |
| 85 | if (!is_special(v, g)) { |
| 86 | return false; |
| 87 | } |
| 88 | } |
| 89 | for (const auto &v : adjacent_vertices_range(g.start, g)) { |
| 90 | if (!is_special(v, g)) { |
| 91 | return false; |
| 92 | } |
| 93 | } |
| 94 | |
| 95 | return true; |
| 96 | } |
| 97 | |
| 98 | static |
| 99 | string encodeUtf8(const vector<unichar> &v) { |
| 100 | string rv; |
| 101 | for (const unichar &cp : v) { |
| 102 | if (cp < UTF_2CHAR_MIN) { |
| 103 | rv.push_back(cp); |
| 104 | } else if (cp < UTF_3CHAR_MIN) { |
| 105 | rv.push_back(UTF_TWO_BYTE_HEADER | (cp >> UTF_CONT_SHIFT)); |
| 106 | rv.push_back(makeContByte(cp)); |
| 107 | } else if (cp < UTF_4CHAR_MIN) { |
| 108 | rv.push_back(UTF_THREE_BYTE_HEADER | (cp >> (2 * UTF_CONT_SHIFT))); |
| 109 | rv.push_back(makeContByte(cp >> UTF_CONT_SHIFT)); |
| 110 | rv.push_back(makeContByte(cp)); |
| 111 | } else { |
| 112 | rv.push_back(UTF_FOUR_BYTE_HEADER | (cp >> (3 * UTF_CONT_SHIFT))); |
| 113 | rv.push_back(makeContByte(cp >> (2 * UTF_CONT_SHIFT))); |
| 114 | rv.push_back(makeContByte(cp >> UTF_CONT_SHIFT)); |
| 115 | rv.push_back(makeContByte(cp)); |
| 116 | } |
| 117 | } |
| 118 | return rv; |
| 119 | } |
| 120 | |
| 121 | template<class Iter, class Val> |
| 122 | static |
| 123 | bool has_greater_than(Iter it, Iter end, const Val &v, size_t limit) { |
| 124 | for (; it != end; ++it) { |
| 125 | if (*it == v) { |
| 126 | if (limit == 0) { |
| 127 | return true; |
| 128 | } |
| 129 | --limit; |
| 130 | } |
| 131 | } |
| 132 | return false; |
| 133 | } |
| 134 | |
| 135 | static |
| 136 | void findPaths(const NGHolder &g, CorpusProperties &cProps, |
| 137 | vector<VertexPath> &allPaths, size_t cycleLimit, |
| 138 | size_t corpusLimit) { |
| 139 | // The maximum number of open (in progress) paths. New paths beyond this |
| 140 | // limit will evict a random existing one. |
| 141 | const size_t MAX_OPEN = min((size_t)1000, corpusLimit * 10); |
| 142 | |
| 143 | vector<unique_ptr<VertexPath>> open; |
| 144 | open.push_back(ue2::make_unique<VertexPath>(1, g.start)); |
| 145 | |
| 146 | unordered_set<NFAVertex> one_way_in; |
| 147 | for (const auto &v : vertices_range(g)) { |
| 148 | if (in_degree(v, g) <= 1) { |
| 149 | one_way_in.insert(v); |
| 150 | } |
| 151 | } |
| 152 | |
| 153 | while (!open.empty()) { |
| 154 | u32 slot = cProps.rand(0, open.size() - 1); |
| 155 | swap(open.at(slot), open.back()); |
| 156 | auto p = std::move(open.back()); |
| 157 | open.pop_back(); |
| 158 | NFAVertex u = p->back(); |
| 159 | |
| 160 | DEBUG_PRINTF("dequeuing path %s, back %zu\n" , |
| 161 | pathToString(g, *p).c_str(), g[u].index); |
| 162 | |
| 163 | NGHolder::adjacency_iterator ai, ae; |
| 164 | for (tie(ai, ae) = adjacent_vertices(u, g); ai != ae; ++ai) { |
| 165 | NFAVertex v = *ai; |
| 166 | |
| 167 | if (u == g.startDs && v == g.startDs) { |
| 168 | // explicitly avoid following startDs self-loop, as we have |
| 169 | // other mechanisms for adding prefixes to our corpora. |
| 170 | continue; |
| 171 | } |
| 172 | |
| 173 | // Accept vertices generate completed paths. |
| 174 | if (v == g.accept || v == g.acceptEod) { |
| 175 | DEBUG_PRINTF("path complete: %s\n" , |
| 176 | pathToString(g, *p).c_str()); |
| 177 | allPaths.push_back(*p); |
| 178 | if (allPaths.size() >= corpusLimit) { |
| 179 | DEBUG_PRINTF("full, going home\n" ); |
| 180 | return; |
| 181 | } |
| 182 | |
| 183 | // No meaningful edges out of accept or acceptEod. |
| 184 | continue; |
| 185 | } |
| 186 | |
| 187 | if (!contains(one_way_in, v) && |
| 188 | has_greater_than(p->begin(), p->end(), v, cycleLimit)) { |
| 189 | // Note that vertices that only have one predecessor don't need |
| 190 | // their cycle limit checked, as their predecessors will have |
| 191 | // the same count. |
| 192 | DEBUG_PRINTF("exceeded cycle limit for v=%zu, pruning path\n" , |
| 193 | g[v].index); |
| 194 | continue; |
| 195 | } |
| 196 | |
| 197 | // If we've got no further adjacent vertices, re-use p rather than |
| 198 | // copying it for the next path. |
| 199 | unique_ptr<VertexPath> new_path; |
| 200 | if (boost::next(ai) == ae) { |
| 201 | new_path = std::move(p); |
| 202 | } else { |
| 203 | new_path = make_unique<VertexPath>(*p); |
| 204 | } |
| 205 | |
| 206 | new_path->push_back(v); |
| 207 | if (open.size() < MAX_OPEN) { |
| 208 | open.push_back(std::move(new_path)); |
| 209 | } else { |
| 210 | u32 victim = cProps.rand(0, open.size() - 1); |
| 211 | open[victim] = std::move(new_path); |
| 212 | } |
| 213 | } |
| 214 | } |
| 215 | DEBUG_PRINTF("bored, going home\n" ); |
| 216 | } |
| 217 | |
| 218 | namespace { |
| 219 | |
| 220 | /** \brief Concrete implementation */ |
| 221 | class CorpusGeneratorImpl : public CorpusGenerator { |
| 222 | public: |
| 223 | CorpusGeneratorImpl(const NGHolder &graph_in, const ExpressionInfo &expr_in, |
| 224 | CorpusProperties &props); |
| 225 | ~CorpusGeneratorImpl() = default; |
| 226 | |
| 227 | void generateCorpus(vector<string> &data); |
| 228 | |
| 229 | private: |
| 230 | unsigned char getRandomChar(); |
| 231 | unsigned char getMatchChar(const CharReach &cr); |
| 232 | unsigned char getUnmatchChar(const CharReach &cr); |
| 233 | |
| 234 | unsigned char getChar(NFAVertex v); |
| 235 | void newGenerator(vector<string> &data); |
| 236 | string pathToCorpus(const VertexPath &path); |
| 237 | |
| 238 | /** \brief Generate a string of random bytes between minLen and maxLen |
| 239 | * bytes in length. */ |
| 240 | void addRandom(const min_max &mm, string *out); |
| 241 | |
| 242 | /** \brief Info about this expression. */ |
| 243 | const ExpressionInfo &expr; |
| 244 | |
| 245 | /** \brief The NFA graph we operate over. */ |
| 246 | const NGHolder &graph; |
| 247 | |
| 248 | /** \brief Reference to our corpus generator properties object (stores some |
| 249 | * state) */ |
| 250 | CorpusProperties &cProps; |
| 251 | }; |
| 252 | |
| 253 | CorpusGeneratorImpl::CorpusGeneratorImpl(const NGHolder &graph_in, |
| 254 | const ExpressionInfo &expr_in, |
| 255 | CorpusProperties &props) |
| 256 | : expr(expr_in), graph(graph_in), cProps(props) { |
| 257 | // if this pattern is to be matched approximately |
| 258 | if ((expr.edit_distance || expr.hamm_distance) && !props.editDistance) { |
| 259 | props.editDistance = |
| 260 | props.rand(0, expr.hamm_distance + expr.edit_distance + 1); |
| 261 | } |
| 262 | } |
| 263 | |
| 264 | void CorpusGeneratorImpl::generateCorpus(vector<string> &data) { |
| 265 | newGenerator(data); |
| 266 | |
| 267 | if (cProps.editDistance && !data.empty() && |
| 268 | data.size() < cProps.corpusLimit) { |
| 269 | // Create more entries by copying the corpora and applying edits |
| 270 | size_t diff = cProps.corpusLimit - data.size(); |
| 271 | size_t repeats = diff / data.size(); |
| 272 | size_t remains = diff % data.size(); |
| 273 | vector<string> newdata; |
| 274 | for (size_t i = 0; i < repeats; i++) { |
| 275 | std::copy(data.begin(), data.end(), std::back_inserter(newdata)); |
| 276 | } |
| 277 | if (remains) { |
| 278 | std::copy_n(data.begin(), remains, std::back_inserter(newdata)); |
| 279 | } |
| 280 | for (auto &s : newdata) { |
| 281 | editCorpus(&s, cProps); |
| 282 | } |
| 283 | std::move(newdata.begin(), newdata.end(), back_inserter(data)); |
| 284 | } else if (cProps.editDistance) { |
| 285 | // If the caller has asked us, apply edit distance to corpora |
| 286 | for (auto &s : data) { |
| 287 | editCorpus(&s, cProps); |
| 288 | } |
| 289 | } |
| 290 | } |
| 291 | |
| 292 | /** \brief Generate a random character, taking care to stick to the alphabet |
| 293 | * that we've been asked for. */ |
| 294 | u8 CorpusGeneratorImpl::getRandomChar() { |
| 295 | return 'a' + cProps.rand(0, min(cProps.alphabetSize, (u32)CharReach::npos)); |
| 296 | } |
| 297 | |
| 298 | /** \brief Select a random character from the given string of valid match |
| 299 | * characters. */ |
| 300 | unsigned char CorpusGeneratorImpl::getMatchChar(const CharReach &cr) { |
| 301 | unsigned int num = cr.count(); |
| 302 | if (num == 0) { |
| 303 | return 0; |
| 304 | } else if (num == 1) { |
| 305 | return (unsigned char)cr.find_first(); |
| 306 | } else if (num == 256) { |
| 307 | // Dot class, any character is OK! |
| 308 | return (unsigned char)cProps.rand(0, 255); |
| 309 | } |
| 310 | else { |
| 311 | unsigned idx = cProps.rand(0, num - 1); |
| 312 | return (unsigned char)cr.find_nth(idx); |
| 313 | } |
| 314 | } |
| 315 | |
| 316 | /** \brief Select a character that does not belong to the given bitset. This |
| 317 | * makes no guarantees on unmatchability if the bitset is full. */ |
| 318 | unsigned char CorpusGeneratorImpl::getUnmatchChar(const CharReach &cr) { |
| 319 | return getMatchChar(~cr); |
| 320 | } |
| 321 | |
| 322 | void CorpusGeneratorImpl::addRandom(const min_max &mm, string *out) { |
| 323 | assert(mm.min <= mm.max); |
| 324 | u32 range = mm.max - mm.min; |
| 325 | u32 len = mm.min + (range ? cProps.rand(0, range - 1) : 0); |
| 326 | for (u32 i = 0; i < len; ++i) { |
| 327 | out->push_back(getRandomChar()); |
| 328 | } |
| 329 | } |
| 330 | |
| 331 | unsigned char CorpusGeneratorImpl::getChar(NFAVertex v) { |
| 332 | const CharReach &cr = graph[v].char_reach; |
| 333 | |
| 334 | switch (cProps.throwDice()) { |
| 335 | case CorpusProperties::ROLLED_MATCH: |
| 336 | return getMatchChar(cr); |
| 337 | case CorpusProperties::ROLLED_UNMATCH: |
| 338 | return getUnmatchChar(cr); |
| 339 | case CorpusProperties::ROLLED_RANDOM: /* character pulled from hat */ |
| 340 | return getRandomChar(); |
| 341 | } |
| 342 | assert(0); |
| 343 | return 0; |
| 344 | } |
| 345 | |
| 346 | /** \brief Convert a path through the graph to a corpus string. */ |
| 347 | string CorpusGeneratorImpl::pathToCorpus(const VertexPath &path) { |
| 348 | string s; |
| 349 | |
| 350 | // Add random prefix |
| 351 | if (cProps.prefixRange.max) { |
| 352 | addRandom(cProps.prefixRange, &s); |
| 353 | } |
| 354 | |
| 355 | // Generate a corpus from our path |
| 356 | for (const auto &e : path) { |
| 357 | if (!is_special(e, graph)) { |
| 358 | s += getChar(e); |
| 359 | } |
| 360 | } |
| 361 | |
| 362 | // Add random suffix |
| 363 | if (cProps.suffixRange.max) { |
| 364 | addRandom(cProps.suffixRange, &s); |
| 365 | } |
| 366 | |
| 367 | return s; |
| 368 | } |
| 369 | |
| 370 | void CorpusGeneratorImpl::newGenerator(vector<string> &outdata) { |
| 371 | const unsigned int maxCycles = cProps.getCycleLimit().second; |
| 372 | DEBUG_PRINTF("generating up to %u corpora, cycle limit of %u\n" , |
| 373 | cProps.corpusLimit, maxCycles); |
| 374 | |
| 375 | vector<VertexPath> allPaths; |
| 376 | |
| 377 | // Special case: if the graph has ONLY special vertices, then this is |
| 378 | // likely to be an odd vacuous pattern or a pattern that can never match. |
| 379 | // In these cases, an empty corpus is useful. |
| 380 | if (graph_is_empty(graph)) { |
| 381 | VertexPath empty(1, graph.start); |
| 382 | allPaths.push_back(empty); |
| 383 | } |
| 384 | |
| 385 | // build a set of unique paths |
| 386 | findPaths(graph, cProps, allPaths, maxCycles, cProps.corpusLimit); |
| 387 | |
| 388 | // transform paths into corpora: we do this repeatedly until we (a) hit our |
| 389 | // limit, or (b) don't generate any new corpora for any of our paths. |
| 390 | set<string> data; |
| 391 | while (data.size() < cProps.corpusLimit) { |
| 392 | size_t count = data.size(); |
| 393 | for (const auto &path : allPaths) { |
| 394 | string s = pathToCorpus(path); |
| 395 | if (data.insert(s).second) { |
| 396 | DEBUG_PRINTF("corpus %zu (%zu bytes): '%s'\n" , data.size(), |
| 397 | s.size(), escapeString(s).c_str()); |
| 398 | if (data.size() == cProps.corpusLimit) { |
| 399 | goto hit_limit; |
| 400 | } |
| 401 | } |
| 402 | } |
| 403 | if (data.size() == count) { |
| 404 | break; // we're finding it hard to generate more corpora |
| 405 | } |
| 406 | } |
| 407 | |
| 408 | hit_limit: |
| 409 | DEBUG_PRINTF("%zu corpora built\n" , data.size()); |
| 410 | |
| 411 | // populate the output vector from the set we built. |
| 412 | outdata.reserve(data.size()); |
| 413 | copy(data.begin(), data.end(), back_inserter(outdata)); |
| 414 | } |
| 415 | |
| 416 | /** \brief Concrete implementation for UTF-8 */ |
| 417 | class CorpusGeneratorUtf8 : public CorpusGenerator { |
| 418 | public: |
| 419 | CorpusGeneratorUtf8(const NGHolder &graph_in, const ExpressionInfo &expr_in, |
| 420 | CorpusProperties &props); |
| 421 | ~CorpusGeneratorUtf8() = default; |
| 422 | |
| 423 | void generateCorpus(vector<string> &data); |
| 424 | |
| 425 | private: |
| 426 | unichar getRandomChar(); |
| 427 | unichar getMatchChar(CodePointSet cps); |
| 428 | unichar getUnmatchChar(const CodePointSet &cps); |
| 429 | |
| 430 | unichar getChar(const CodePointSet &cps); |
| 431 | void newGenerator(vector<vector<unichar> > &data); |
| 432 | vector<unichar> pathToCorpus(const vector<CodePointSet> &path); |
| 433 | |
| 434 | /** \brief Generate a random string between min and max codepoints in |
| 435 | * length. */ |
| 436 | void addRandom(const min_max &mm, vector<unichar> *out); |
| 437 | |
| 438 | /** \brief Info about this expression. */ |
| 439 | const ExpressionInfo &expr; |
| 440 | |
| 441 | /** \brief The NFA graph we operate over. */ |
| 442 | const NGHolder &graph; |
| 443 | |
| 444 | /** \brief Reference to our corpus generator properties object (stores some |
| 445 | * state) */ |
| 446 | CorpusProperties &cProps; |
| 447 | }; |
| 448 | |
| 449 | CorpusGeneratorUtf8::CorpusGeneratorUtf8(const NGHolder &graph_in, |
| 450 | const ExpressionInfo &expr_in, |
| 451 | CorpusProperties &props) |
| 452 | : expr(expr_in), graph(graph_in), cProps(props) { |
| 453 | // we do not support Utf8 for approximate matching |
| 454 | if (expr.edit_distance) { |
| 455 | throw CorpusGenerationFailure("UTF-8 for edited patterns is not " |
| 456 | "supported." ); |
| 457 | } |
| 458 | } |
| 459 | |
| 460 | void CorpusGeneratorUtf8::generateCorpus(vector<string> &data) { |
| 461 | vector<vector<unichar>> raw; |
| 462 | newGenerator(raw); |
| 463 | |
| 464 | // If the caller has asked us, apply edit distance to corpora |
| 465 | if (cProps.editDistance) { |
| 466 | for (auto &e : raw) { |
| 467 | editCorpus(&e, cProps); |
| 468 | } |
| 469 | } |
| 470 | |
| 471 | for (const auto &e : raw) { |
| 472 | data.push_back(encodeUtf8(e)); |
| 473 | } |
| 474 | } |
| 475 | |
| 476 | /** \brief Generate a random character, taking care to stick to the alphabet |
| 477 | * that we've been asked for. */ |
| 478 | unichar CorpusGeneratorUtf8::getRandomChar() { |
| 479 | u32 range = MAX_UNICODE + 1 |
| 480 | - (UNICODE_SURROGATE_MAX + UNICODE_SURROGATE_MIN + 1); |
| 481 | range = min(cProps.alphabetSize, range); |
| 482 | assert(range); |
| 483 | |
| 484 | unichar c = 'a' + cProps.rand(0, range - 1); |
| 485 | |
| 486 | if (c >= UNICODE_SURROGATE_MIN) { |
| 487 | c =+ UNICODE_SURROGATE_MAX + 1; |
| 488 | } |
| 489 | |
| 490 | return c % (MAX_UNICODE + 1); |
| 491 | } |
| 492 | |
| 493 | /** \brief Select a random character from the given string of valid match |
| 494 | * characters. */ |
| 495 | unichar CorpusGeneratorUtf8::getMatchChar(CodePointSet cps) { |
| 496 | cps.unsetRange(UNICODE_SURROGATE_MIN, UNICODE_SURROGATE_MAX); |
| 497 | u32 num = cps.count(); |
| 498 | if (num == 0) { |
| 499 | return 0; |
| 500 | } else if (num == 1) { |
| 501 | return lower(*cps.begin()); |
| 502 | } else { |
| 503 | unichar rv = cps.at(cProps.rand(0, num - 1)); |
| 504 | assert(rv != INVALID_UNICODE); |
| 505 | return rv; |
| 506 | } |
| 507 | } |
| 508 | |
| 509 | /** \brief Select a character that does not belong to the given bitset. This |
| 510 | * makes no guarantees on unmatchability if the bitset is full. */ |
| 511 | unichar CorpusGeneratorUtf8::getUnmatchChar(const CodePointSet &cps) { |
| 512 | return getMatchChar(~cps); |
| 513 | } |
| 514 | |
| 515 | void CorpusGeneratorUtf8::addRandom(const min_max &mm, vector<unichar> *out) { |
| 516 | assert(mm.min <= mm.max); |
| 517 | u32 range = mm.max - mm.min; |
| 518 | u32 len = mm.min + (range ? cProps.rand(0, range - 1) : 0); |
| 519 | for (u32 i = 0; i < len; ++i) { |
| 520 | out->push_back(getRandomChar()); |
| 521 | } |
| 522 | } |
| 523 | |
| 524 | unichar CorpusGeneratorUtf8::getChar(const CodePointSet &cps) { |
| 525 | switch (cProps.throwDice()) { |
| 526 | case CorpusProperties::ROLLED_MATCH: |
| 527 | return getMatchChar(cps); |
| 528 | case CorpusProperties::ROLLED_UNMATCH: |
| 529 | return getUnmatchChar(cps); |
| 530 | case CorpusProperties::ROLLED_RANDOM: /* character pulled from hat */ |
| 531 | return getRandomChar(); |
| 532 | } |
| 533 | assert(0); |
| 534 | return 0; |
| 535 | } |
| 536 | |
| 537 | /** \brief Convert a path through the graph to a corpus string. */ |
| 538 | vector<unichar> |
| 539 | CorpusGeneratorUtf8::pathToCorpus(const vector<CodePointSet> &path) { |
| 540 | vector<unichar> s; |
| 541 | |
| 542 | // Add random prefix |
| 543 | if (cProps.prefixRange.max) { |
| 544 | addRandom(cProps.prefixRange, &s); |
| 545 | } |
| 546 | |
| 547 | // Generate a corpus from our path |
| 548 | for (const auto &e : path) { |
| 549 | s.push_back(getChar(e)); |
| 550 | } |
| 551 | |
| 552 | // Add random suffix |
| 553 | if (cProps.suffixRange.max) { |
| 554 | addRandom(cProps.suffixRange, &s); |
| 555 | } |
| 556 | |
| 557 | return s; |
| 558 | } |
| 559 | |
| 560 | static |
| 561 | u32 classify_vertex(const NGHolder &g, NFAVertex v) { |
| 562 | const CharReach &cr = g[v].char_reach; |
| 563 | if (cr.isSubsetOf(UTF_ASCII_CR)) { |
| 564 | return 1; |
| 565 | } else if (cr.isSubsetOf(UTF_TWO_START_CR)) { |
| 566 | return 2; |
| 567 | } else if (cr.isSubsetOf(UTF_THREE_START_CR)) { |
| 568 | return 3; |
| 569 | } else if (cr.isSubsetOf(UTF_FOUR_START_CR)) { |
| 570 | return 4; |
| 571 | } |
| 572 | |
| 573 | /* this can happen due to dummy vertices from zwa */ |
| 574 | return 1; |
| 575 | } |
| 576 | |
| 577 | static |
| 578 | void fillCodePointSet(const CharReach &cr, CodePointSet *out, u8 mask = 0xff) { |
| 579 | for (u32 i = cr.find_first(); i != CharReach::npos; i = cr.find_next(i)) { |
| 580 | out->set(i & mask); |
| 581 | } |
| 582 | } |
| 583 | |
| 584 | static |
| 585 | void expandCodePointSet(const CharReach &cr, CodePointSet *out, u32 mask, |
| 586 | u32 n) { |
| 587 | CodePointSet base; |
| 588 | base.swap(*out); |
| 589 | for (u32 i = cr.find_first(); i != CharReach::npos; i = cr.find_next(i)) { |
| 590 | u32 val = (i & mask) << (n * UTF_CONT_SHIFT); |
| 591 | for (const auto &cp : base) { |
| 592 | unichar ll = lower(cp); |
| 593 | unichar uu = upper(cp); |
| 594 | out->setRange(val + ll, MIN(val + uu, MAX_UNICODE)); |
| 595 | } |
| 596 | } |
| 597 | } |
| 598 | |
| 599 | static |
| 600 | void decodePath(const NGHolder &g, const VertexPath &in, |
| 601 | vector<CodePointSet> &out) { |
| 602 | VertexPath::const_iterator it = in.begin(); |
| 603 | while (it != in.end()) { |
| 604 | if (is_special(*it, g)) { |
| 605 | ++it; |
| 606 | continue; |
| 607 | } |
| 608 | |
| 609 | out.push_back(CodePointSet()); |
| 610 | CodePointSet &cps = out.back(); |
| 611 | |
| 612 | switch (classify_vertex(g, *it)) { |
| 613 | case 1: |
| 614 | fillCodePointSet(g[*it].char_reach, &cps); |
| 615 | ++it; |
| 616 | break; |
| 617 | case 2: |
| 618 | fillCodePointSet(g[*(it + 1)].char_reach, &cps, |
| 619 | UTF_CONT_BYTE_VALUE_MASK); |
| 620 | expandCodePointSet(g[*it].char_reach, &cps, |
| 621 | ~UTF_TWO_BYTE_HEADER, 1); |
| 622 | it += 2; |
| 623 | break; |
| 624 | case 3: |
| 625 | fillCodePointSet(g[*(it + 2)].char_reach, &cps, |
| 626 | UTF_CONT_BYTE_VALUE_MASK); |
| 627 | expandCodePointSet(g[*(it + 1)].char_reach, &cps, |
| 628 | UTF_CONT_BYTE_VALUE_MASK, 1); |
| 629 | expandCodePointSet(g[*it].char_reach, &cps, |
| 630 | ~UTF_THREE_BYTE_HEADER, 2); |
| 631 | it += 3; |
| 632 | break; |
| 633 | case 4: |
| 634 | fillCodePointSet(g[*(it + 3)].char_reach, &cps, |
| 635 | UTF_CONT_BYTE_VALUE_MASK); |
| 636 | expandCodePointSet(g[*(it + 2)].char_reach, &cps, |
| 637 | UTF_CONT_BYTE_VALUE_MASK, 1); |
| 638 | expandCodePointSet(g[*(it + 1)].char_reach, &cps, |
| 639 | UTF_CONT_BYTE_VALUE_MASK, 2); |
| 640 | expandCodePointSet(g[*it].char_reach, &cps, |
| 641 | ~UTF_FOUR_BYTE_HEADER, 3); |
| 642 | it += 4; |
| 643 | break; |
| 644 | default:; |
| 645 | assert(0); |
| 646 | ++it; |
| 647 | } |
| 648 | } |
| 649 | } |
| 650 | |
| 651 | static |
| 652 | void translatePaths(const NGHolder &graph, |
| 653 | const vector<VertexPath> &allPathsTemp, |
| 654 | vector<vector<CodePointSet>> *out) { |
| 655 | assert(out); |
| 656 | for (const auto &path : allPathsTemp) { |
| 657 | out->push_back(vector<CodePointSet>()); |
| 658 | decodePath(graph, path, out->back()); |
| 659 | } |
| 660 | } |
| 661 | |
| 662 | void CorpusGeneratorUtf8::newGenerator(vector<vector<unichar>> &outdata) { |
| 663 | const u32 maxCycles = cProps.getCycleLimit().second; |
| 664 | DEBUG_PRINTF("generating up to %u corpora, cycle limit of %u\n" , |
| 665 | cProps.corpusLimit, maxCycles); |
| 666 | |
| 667 | vector<vector<CodePointSet>> allPaths; |
| 668 | |
| 669 | // Special case: if the graph has ONLY special vertices, then this is |
| 670 | // likely to be an odd vacuous pattern or a pattern that can never match. |
| 671 | // In these cases, an empty corpus is useful. |
| 672 | if (graph_is_empty(graph)) { |
| 673 | allPaths.push_back(vector<CodePointSet>()); |
| 674 | } else { |
| 675 | // build a set of unique paths |
| 676 | vector<VertexPath> allPathsTemp; |
| 677 | findPaths(graph, cProps, allPathsTemp, maxCycles, cProps.corpusLimit); |
| 678 | translatePaths(graph, allPathsTemp, &allPaths); |
| 679 | } |
| 680 | |
| 681 | // transform paths into corpora: we do this repeatedly until we (a) hit our |
| 682 | // limit, or (b) don't generate any new corpora for any of our paths. |
| 683 | set<vector<unichar> > data; |
| 684 | while (data.size() < cProps.corpusLimit) { |
| 685 | size_t count = data.size(); |
| 686 | for (const auto &path : allPaths) { |
| 687 | vector<unichar> vu = pathToCorpus(path); |
| 688 | if (data.insert(vu).second) { |
| 689 | if (data.size() == cProps.corpusLimit) { |
| 690 | goto hit_limit; |
| 691 | } |
| 692 | } |
| 693 | } |
| 694 | if (data.size() == count) { |
| 695 | break; // we're finding it hard to generate more corpora |
| 696 | } |
| 697 | } |
| 698 | |
| 699 | hit_limit: |
| 700 | DEBUG_PRINTF("%zu corpora built\n" , data.size()); |
| 701 | |
| 702 | // populate the output vector from the set we built. |
| 703 | outdata.reserve(data.size()); |
| 704 | copy(data.begin(), data.end(), back_inserter(outdata)); |
| 705 | } |
| 706 | |
| 707 | } // namespace |
| 708 | |
| 709 | CorpusGenerator::~CorpusGenerator() { } |
| 710 | |
| 711 | // External entry point |
| 712 | |
| 713 | unique_ptr<CorpusGenerator> makeCorpusGenerator(const NGHolder &graph, |
| 714 | const ExpressionInfo &expr, |
| 715 | CorpusProperties &props) { |
| 716 | if (expr.utf8) { |
| 717 | return ue2::make_unique<CorpusGeneratorUtf8>(graph, expr, props); |
| 718 | } else { |
| 719 | return ue2::make_unique<CorpusGeneratorImpl>(graph, expr, props); |
| 720 | } |
| 721 | } |
| 722 | |