| 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 | #include "rdfa_merge.h" |
| 30 | |
| 31 | #include "grey.h" |
| 32 | #include "dfa_min.h" |
| 33 | #include "mcclellancompile_util.h" |
| 34 | #include "rdfa.h" |
| 35 | #include "ue2common.h" |
| 36 | #include "nfagraph/ng_mcclellan_internal.h" |
| 37 | #include "util/container.h" |
| 38 | #include "util/determinise.h" |
| 39 | #include "util/flat_containers.h" |
| 40 | #include "util/make_unique.h" |
| 41 | #include "util/report_manager.h" |
| 42 | #include "util/unordered.h" |
| 43 | |
| 44 | #include <algorithm> |
| 45 | #include <queue> |
| 46 | |
| 47 | using namespace std; |
| 48 | |
| 49 | namespace ue2 { |
| 50 | |
| 51 | #define MAX_DFA_STATES 16383 |
| 52 | |
| 53 | namespace { |
| 54 | |
| 55 | class Automaton_Merge { |
| 56 | public: |
| 57 | using StateSet = vector<u16>; |
| 58 | using StateMap = ue2_unordered_map<StateSet, dstate_id_t>; |
| 59 | |
| 60 | Automaton_Merge(const raw_dfa *rdfa1, const raw_dfa *rdfa2, |
| 61 | const ReportManager *rm_in, const Grey &grey_in) |
| 62 | : rm(rm_in), grey(grey_in), nfas{rdfa1, rdfa2}, dead(2) { |
| 63 | calculateAlphabet(); |
| 64 | populateAsFs(); |
| 65 | prunable = isPrunable(); |
| 66 | } |
| 67 | |
| 68 | Automaton_Merge(const vector<const raw_dfa *> &dfas, |
| 69 | const ReportManager *rm_in, const Grey &grey_in) |
| 70 | : rm(rm_in), grey(grey_in), nfas(dfas), dead(nfas.size()) { |
| 71 | calculateAlphabet(); |
| 72 | populateAsFs(); |
| 73 | prunable = isPrunable(); |
| 74 | } |
| 75 | |
| 76 | void populateAsFs(void) { |
| 77 | bool fs_same = true; |
| 78 | bool fs_dead = true; |
| 79 | |
| 80 | as.resize(nfas.size()); |
| 81 | fs.resize(nfas.size()); |
| 82 | for (size_t i = 0, end = nfas.size(); i < end; i++) { |
| 83 | as[i] = nfas[i]->start_anchored; |
| 84 | fs[i] = nfas[i]->start_floating; |
| 85 | |
| 86 | if (fs[i]) { |
| 87 | fs_dead = false; |
| 88 | } |
| 89 | |
| 90 | if (as[i] != fs[i]) { |
| 91 | fs_same = false; |
| 92 | } |
| 93 | } |
| 94 | |
| 95 | start_anchored = DEAD_STATE + 1; |
| 96 | if (fs_same) { |
| 97 | start_floating = start_anchored; |
| 98 | } else if (fs_dead) { |
| 99 | start_floating = DEAD_STATE; |
| 100 | } else { |
| 101 | start_floating = start_anchored + 1; |
| 102 | } |
| 103 | } |
| 104 | |
| 105 | void calculateAlphabet(void) { |
| 106 | DEBUG_PRINTF("calculating alphabet\n" ); |
| 107 | vector<CharReach> esets = {CharReach::dot()}; |
| 108 | |
| 109 | for (const auto &rdfa : nfas) { |
| 110 | DEBUG_PRINTF("...next dfa alphabet\n" ); |
| 111 | assert(rdfa); |
| 112 | const auto &alpha_remap = rdfa->alpha_remap; |
| 113 | |
| 114 | for (size_t i = 0; i < esets.size(); i++) { |
| 115 | assert(esets[i].count()); |
| 116 | if (esets[i].count() == 1) { |
| 117 | DEBUG_PRINTF("skipping singleton eq set\n" ); |
| 118 | continue; |
| 119 | } |
| 120 | |
| 121 | CharReach t; |
| 122 | u8 leader_s = alpha_remap[esets[i].find_first()]; |
| 123 | |
| 124 | DEBUG_PRINTF("checking eq set, leader %02hhx \n" , leader_s); |
| 125 | |
| 126 | for (size_t s = esets[i].find_first(); s != CharReach::npos; |
| 127 | s = esets[i].find_next(s)) { |
| 128 | if (alpha_remap[s] != leader_s) { |
| 129 | t.set(s); |
| 130 | } |
| 131 | } |
| 132 | |
| 133 | if (t.any() && t != esets[i]) { |
| 134 | esets[i] &= ~t; |
| 135 | esets.push_back(t); |
| 136 | } |
| 137 | } |
| 138 | } |
| 139 | |
| 140 | // Sort so that our alphabet mapping isn't dependent on the order of |
| 141 | // rdfas passed in. |
| 142 | sort(esets.begin(), esets.end()); |
| 143 | |
| 144 | alphasize = buildAlphabetFromEquivSets(esets, alpha, unalpha); |
| 145 | } |
| 146 | |
| 147 | bool isPrunable() const { |
| 148 | if (!grey.highlanderPruneDFA || !rm) { |
| 149 | DEBUG_PRINTF("disabled, or not managed reports\n" ); |
| 150 | return false; |
| 151 | } |
| 152 | |
| 153 | assert(!nfas.empty()); |
| 154 | if (!generates_callbacks(nfas.front()->kind)) { |
| 155 | DEBUG_PRINTF("doesn't generate callbacks\n" ); |
| 156 | return false; |
| 157 | } |
| 158 | |
| 159 | // Collect all reports from all merge candidates. |
| 160 | flat_set<ReportID> merge_reports; |
| 161 | for (const auto &rdfa : nfas) { |
| 162 | insert(&merge_reports, all_reports(*rdfa)); |
| 163 | } |
| 164 | |
| 165 | DEBUG_PRINTF("all reports: %s\n" , as_string_list(merge_reports).c_str()); |
| 166 | |
| 167 | // Return true if they're all exhaustible with the same exhaustion key. |
| 168 | u32 ekey = INVALID_EKEY; |
| 169 | for (const auto &report_id : merge_reports) { |
| 170 | const Report &r = rm->getReport(report_id); |
| 171 | if (!isSimpleExhaustible(r)) { |
| 172 | DEBUG_PRINTF("report %u not simple exhaustible\n" , report_id); |
| 173 | return false; |
| 174 | } |
| 175 | assert(r.ekey != INVALID_EKEY); |
| 176 | if (ekey == INVALID_EKEY) { |
| 177 | ekey = r.ekey; |
| 178 | } else if (ekey != r.ekey) { |
| 179 | DEBUG_PRINTF("two different ekeys, %u and %u\n" , ekey, r.ekey); |
| 180 | return false; |
| 181 | } |
| 182 | } |
| 183 | |
| 184 | DEBUG_PRINTF("is prunable\n" ); |
| 185 | return true; |
| 186 | } |
| 187 | |
| 188 | |
| 189 | void transition(const StateSet &in, StateSet *next) { |
| 190 | u16 t[ALPHABET_SIZE]; |
| 191 | |
| 192 | for (u32 i = 0; i < alphasize; i++) { |
| 193 | next[i].resize(nfas.size()); |
| 194 | } |
| 195 | |
| 196 | for (size_t j = 0, j_end = nfas.size(); j < j_end; j++) { |
| 197 | getFullTransitionFromState(*nfas[j], in[j], t); |
| 198 | for (u32 i = 0; i < alphasize; i++) { |
| 199 | next[i][j] = t[unalpha[i]]; |
| 200 | } |
| 201 | } |
| 202 | } |
| 203 | |
| 204 | const vector<StateSet> initial() { |
| 205 | vector<StateSet> rv = {as}; |
| 206 | if (start_floating != DEAD_STATE && start_floating != start_anchored) { |
| 207 | rv.push_back(fs); |
| 208 | } |
| 209 | return rv; |
| 210 | } |
| 211 | |
| 212 | private: |
| 213 | void reports_i(const StateSet &in, flat_set<ReportID> dstate::*r_set, |
| 214 | flat_set<ReportID> &r) const { |
| 215 | for (size_t i = 0, end = nfas.size(); i < end; i++) { |
| 216 | const auto &rs = nfas[i]->states[in[i]].*r_set; |
| 217 | insert(&r, rs); |
| 218 | } |
| 219 | } |
| 220 | |
| 221 | public: |
| 222 | void reports(const StateSet &in, flat_set<ReportID> &rv) const { |
| 223 | reports_i(in, &dstate::reports, rv); |
| 224 | } |
| 225 | void reportsEod(const StateSet &in, flat_set<ReportID> &rv) const { |
| 226 | reports_i(in, &dstate::reports_eod, rv); |
| 227 | } |
| 228 | |
| 229 | bool canPrune(const flat_set<ReportID> &test_reports) const { |
| 230 | if (!grey.highlanderPruneDFA || !prunable) { |
| 231 | return false; |
| 232 | } |
| 233 | |
| 234 | // Must all be external reports. |
| 235 | assert(rm); |
| 236 | for (const auto &report_id : test_reports) { |
| 237 | if (!isExternalReport(rm->getReport(report_id))) { |
| 238 | return false; |
| 239 | } |
| 240 | } |
| 241 | |
| 242 | return true; |
| 243 | } |
| 244 | |
| 245 | /** True if the minimization algorithm should be run after merging. */ |
| 246 | bool shouldMinimize() const { |
| 247 | // We only need to run minimization if our merged DFAs shared a report. |
| 248 | flat_set<ReportID> seen_reports; |
| 249 | for (const auto &rdfa : nfas) { |
| 250 | for (const auto &report_id : all_reports(*rdfa)) { |
| 251 | if (!seen_reports.insert(report_id).second) { |
| 252 | DEBUG_PRINTF("report %u in several dfas\n" , report_id); |
| 253 | return true; |
| 254 | } |
| 255 | } |
| 256 | } |
| 257 | |
| 258 | return false; |
| 259 | } |
| 260 | |
| 261 | private: |
| 262 | const ReportManager *rm; |
| 263 | const Grey &grey; |
| 264 | |
| 265 | vector<const raw_dfa *> nfas; |
| 266 | vector<dstate_id_t> as; |
| 267 | vector<dstate_id_t> fs; |
| 268 | |
| 269 | bool prunable = false; |
| 270 | |
| 271 | public: |
| 272 | std::array<u16, ALPHABET_SIZE> alpha; |
| 273 | std::array<u16, ALPHABET_SIZE> unalpha; |
| 274 | u16 alphasize; |
| 275 | StateSet dead; |
| 276 | |
| 277 | u16 start_anchored; |
| 278 | u16 start_floating; |
| 279 | }; |
| 280 | |
| 281 | } // namespace |
| 282 | |
| 283 | unique_ptr<raw_dfa> mergeTwoDfas(const raw_dfa *d1, const raw_dfa *d2, |
| 284 | size_t max_states, const ReportManager *rm, |
| 285 | const Grey &grey) { |
| 286 | assert(d1 && d2); |
| 287 | assert(d1->kind == d2->kind); |
| 288 | assert(max_states <= MAX_DFA_STATES); |
| 289 | |
| 290 | auto rdfa = ue2::make_unique<raw_dfa>(d1->kind); |
| 291 | |
| 292 | Automaton_Merge autom(d1, d2, rm, grey); |
| 293 | if (determinise(autom, rdfa->states, max_states)) { |
| 294 | rdfa->start_anchored = autom.start_anchored; |
| 295 | rdfa->start_floating = autom.start_floating; |
| 296 | rdfa->alpha_size = autom.alphasize; |
| 297 | rdfa->alpha_remap = autom.alpha; |
| 298 | DEBUG_PRINTF("merge succeeded, %zu states\n" , rdfa->states.size()); |
| 299 | |
| 300 | if (autom.shouldMinimize()) { |
| 301 | minimize_hopcroft(*rdfa, grey); |
| 302 | DEBUG_PRINTF("minimized, %zu states\n" , rdfa->states.size()); |
| 303 | } |
| 304 | |
| 305 | return rdfa; |
| 306 | } |
| 307 | |
| 308 | return nullptr; |
| 309 | } |
| 310 | |
| 311 | void mergeDfas(vector<unique_ptr<raw_dfa>> &dfas, size_t max_states, |
| 312 | const ReportManager *rm, const Grey &grey) { |
| 313 | assert(max_states <= MAX_DFA_STATES); |
| 314 | |
| 315 | if (dfas.size() <= 1) { |
| 316 | return; |
| 317 | } |
| 318 | |
| 319 | DEBUG_PRINTF("before merging, we have %zu dfas\n" , dfas.size()); |
| 320 | |
| 321 | queue<unique_ptr<raw_dfa>> q; |
| 322 | for (auto &dfa : dfas) { |
| 323 | q.push(move(dfa)); |
| 324 | } |
| 325 | |
| 326 | // All DFAs are now on the queue, so we'll clear the vector and use it for |
| 327 | // output from here. |
| 328 | dfas.clear(); |
| 329 | |
| 330 | while (q.size() > 1) { |
| 331 | // Attempt to merge the two front elements of the queue. |
| 332 | unique_ptr<raw_dfa> d1 = move(q.front()); |
| 333 | q.pop(); |
| 334 | unique_ptr<raw_dfa> d2 = move(q.front()); |
| 335 | q.pop(); |
| 336 | |
| 337 | auto rdfa = mergeTwoDfas(d1.get(), d2.get(), max_states, rm, grey); |
| 338 | if (rdfa) { |
| 339 | q.push(move(rdfa)); |
| 340 | } else { |
| 341 | DEBUG_PRINTF("failed to merge\n" ); |
| 342 | // Put the larger of the two DFAs on the output list, retain the |
| 343 | // smaller one on the queue for further merge attempts. |
| 344 | if (d2->states.size() > d1->states.size()) { |
| 345 | dfas.push_back(move(d2)); |
| 346 | q.push(move(d1)); |
| 347 | } else { |
| 348 | dfas.push_back(move(d1)); |
| 349 | q.push(move(d2)); |
| 350 | } |
| 351 | } |
| 352 | } |
| 353 | |
| 354 | while (!q.empty()) { |
| 355 | dfas.push_back(move(q.front())); |
| 356 | q.pop(); |
| 357 | } |
| 358 | |
| 359 | DEBUG_PRINTF("after merging, we have %zu dfas\n" , dfas.size()); |
| 360 | } |
| 361 | |
| 362 | unique_ptr<raw_dfa> mergeAllDfas(const vector<const raw_dfa *> &dfas, |
| 363 | size_t max_states, const ReportManager *rm, |
| 364 | const Grey &grey) { |
| 365 | assert(max_states <= MAX_DFA_STATES); |
| 366 | assert(!dfas.empty()); |
| 367 | |
| 368 | // All the DFAs should be of the same kind. |
| 369 | const auto kind = dfas.front()->kind; |
| 370 | assert(all_of(begin(dfas), end(dfas), |
| 371 | [&kind](const raw_dfa *rdfa) { return rdfa->kind == kind; })); |
| 372 | |
| 373 | auto rdfa = ue2::make_unique<raw_dfa>(kind); |
| 374 | Automaton_Merge n(dfas, rm, grey); |
| 375 | |
| 376 | DEBUG_PRINTF("merging dfa\n" ); |
| 377 | |
| 378 | if (!determinise(n, rdfa->states, max_states)) { |
| 379 | DEBUG_PRINTF("state limit (%zu) exceeded\n" , max_states); |
| 380 | return nullptr; /* over state limit */ |
| 381 | } |
| 382 | |
| 383 | rdfa->start_anchored = n.start_anchored; |
| 384 | rdfa->start_floating = n.start_floating; |
| 385 | rdfa->alpha_size = n.alphasize; |
| 386 | rdfa->alpha_remap = n.alpha; |
| 387 | |
| 388 | DEBUG_PRINTF("merged, building impl dfa (a,f) = (%hu,%hu)\n" , |
| 389 | rdfa->start_anchored, rdfa->start_floating); |
| 390 | |
| 391 | if (n.shouldMinimize()) { |
| 392 | minimize_hopcroft(*rdfa, grey); |
| 393 | DEBUG_PRINTF("minimized, %zu states\n" , rdfa->states.size()); |
| 394 | } |
| 395 | |
| 396 | return rdfa; |
| 397 | } |
| 398 | |
| 399 | } // namespace ue2 |
| 400 | |