Alexander Gutkin | 0d4c523 | 2013-02-28 13:47:27 +0000 | [diff] [blame] | 1 | // Copyright 2006-2008 The RE2 Authors. All Rights Reserved. |
| 2 | // Use of this source code is governed by a BSD-style |
| 3 | // license that can be found in the LICENSE file. |
| 4 | |
| 5 | #include "util/test.h" |
| 6 | #include "util/thread.h" |
| 7 | #include "re2/prog.h" |
| 8 | #include "re2/re2.h" |
| 9 | #include "re2/regexp.h" |
| 10 | #include "re2/testing/regexp_generator.h" |
| 11 | #include "re2/testing/string_generator.h" |
| 12 | |
| 13 | DECLARE_bool(re2_dfa_bail_when_slow); |
| 14 | |
| 15 | DEFINE_int32(size, 8, "log2(number of DFA nodes)"); |
| 16 | DEFINE_int32(repeat, 2, "Repetition count."); |
| 17 | DEFINE_int32(threads, 4, "number of threads"); |
| 18 | |
| 19 | namespace re2 { |
| 20 | |
| 21 | // Check that multithreaded access to DFA class works. |
| 22 | |
| 23 | // Helper thread: builds entire DFA for prog. |
| 24 | class BuildThread : public Thread { |
| 25 | public: |
| 26 | BuildThread(Prog* prog) : prog_(prog) {} |
| 27 | virtual void Run() { |
| 28 | CHECK(prog_->BuildEntireDFA(Prog::kFirstMatch)); |
| 29 | } |
| 30 | |
| 31 | private: |
| 32 | Prog* prog_; |
| 33 | }; |
| 34 | |
| 35 | TEST(Multithreaded, BuildEntireDFA) { |
| 36 | // Create regexp with 2^FLAGS_size states in DFA. |
| 37 | string s = "a"; |
| 38 | for (int i = 0; i < FLAGS_size; i++) |
| 39 | s += "[ab]"; |
| 40 | s += "b"; |
| 41 | |
| 42 | // Check that single-threaded code works. |
| 43 | { |
| 44 | //LOG(INFO) << s; |
| 45 | Regexp* re = Regexp::Parse(s.c_str(), Regexp::LikePerl, NULL); |
| 46 | CHECK(re); |
| 47 | Prog* prog = re->CompileToProg(0); |
| 48 | CHECK(prog); |
| 49 | BuildThread* t = new BuildThread(prog); |
| 50 | t->SetJoinable(true); |
| 51 | t->Start(); |
| 52 | t->Join(); |
| 53 | delete t; |
| 54 | delete prog; |
| 55 | re->Decref(); |
| 56 | } |
| 57 | |
| 58 | // Build the DFA simultaneously in a bunch of threads. |
| 59 | for (int i = 0; i < FLAGS_repeat; i++) { |
| 60 | Regexp* re = Regexp::Parse(s.c_str(), Regexp::LikePerl, NULL); |
| 61 | CHECK(re); |
| 62 | Prog* prog = re->CompileToProg(0); |
| 63 | CHECK(prog); |
| 64 | |
| 65 | vector<BuildThread*> threads; |
| 66 | for (int j = 0; j < FLAGS_threads; j++) { |
| 67 | BuildThread *t = new BuildThread(prog); |
| 68 | t->SetJoinable(true); |
| 69 | threads.push_back(t); |
| 70 | } |
| 71 | for (int j = 0; j < FLAGS_threads; j++) |
| 72 | threads[j]->Start(); |
| 73 | for (int j = 0; j < FLAGS_threads; j++) { |
| 74 | threads[j]->Join(); |
| 75 | delete threads[j]; |
| 76 | } |
| 77 | |
| 78 | // One more compile, to make sure everything is okay. |
| 79 | prog->BuildEntireDFA(Prog::kFirstMatch); |
| 80 | delete prog; |
| 81 | re->Decref(); |
| 82 | } |
| 83 | } |
| 84 | |
| 85 | // Check that DFA size requirements are followed. |
| 86 | // BuildEntireDFA will, like SearchDFA, stop building out |
| 87 | // the DFA once the memory limits are reached. |
| 88 | TEST(SingleThreaded, BuildEntireDFA) { |
| 89 | // Create regexp with 2^30 states in DFA. |
| 90 | string s = "a"; |
| 91 | for (int i = 0; i < 30; i++) |
| 92 | s += "[ab]"; |
| 93 | s += "b"; |
| 94 | |
| 95 | //LOG(INFO) << s; |
| 96 | Regexp* re = Regexp::Parse(s.c_str(), Regexp::LikePerl, NULL); |
| 97 | CHECK(re); |
| 98 | int max = 24; |
| 99 | for (int i = 17; i < max; i++) { |
| 100 | int limit = 1<<i; |
| 101 | int usage; |
| 102 | //int progusage, dfamem; |
| 103 | { |
| 104 | testing::MallocCounter m(testing::MallocCounter::THIS_THREAD_ONLY); |
| 105 | Prog* prog = re->CompileToProg(limit); |
| 106 | CHECK(prog); |
| 107 | //progusage = m.HeapGrowth(); |
| 108 | //dfamem = prog->dfa_mem(); |
| 109 | prog->BuildEntireDFA(Prog::kFirstMatch); |
| 110 | prog->BuildEntireDFA(Prog::kLongestMatch); |
| 111 | usage = m.HeapGrowth(); |
| 112 | delete prog; |
| 113 | } |
| 114 | if (!UsingMallocCounter) |
| 115 | continue; |
| 116 | //LOG(INFO) << StringPrintf("Limit %d: prog used %d, DFA budget %d, total %d\n", |
| 117 | // limit, progusage, dfamem, usage); |
| 118 | CHECK_GT(usage, limit*9/10); |
| 119 | CHECK_LT(usage, limit + (16<<10)); // 16kB of slop okay |
| 120 | } |
| 121 | re->Decref(); |
| 122 | } |
| 123 | |
| 124 | // Generates and returns a string over binary alphabet {0,1} that contains |
| 125 | // all possible binary sequences of length n as subsequences. The obvious |
| 126 | // brute force method would generate a string of length n * 2^n, but this |
| 127 | // generates a string of length n + 2^n - 1 called a De Bruijn cycle. |
| 128 | // See Knuth, The Art of Computer Programming, Vol 2, Exercise 3.2.2 #17. |
| 129 | // Such a string is useful for testing a DFA. If you have a DFA |
| 130 | // where distinct last n bytes implies distinct states, then running on a |
| 131 | // DeBruijn string causes the DFA to need to create a new state at every |
| 132 | // position in the input, never reusing any states until it gets to the |
| 133 | // end of the string. This is the worst possible case for DFA execution. |
| 134 | static string DeBruijnString(int n) { |
| 135 | CHECK_LT(n, 8*sizeof(int)); |
| 136 | CHECK_GT(n, 0); |
| 137 | |
| 138 | vector<bool> did(1<<n); |
| 139 | for (int i = 0; i < 1<<n; i++) |
| 140 | did[i] = false; |
| 141 | |
| 142 | string s; |
| 143 | for (int i = 0; i < n-1; i++) |
| 144 | s.append("0"); |
| 145 | int bits = 0; |
| 146 | int mask = (1<<n) - 1; |
| 147 | for (int i = 0; i < (1<<n); i++) { |
| 148 | bits <<= 1; |
| 149 | bits &= mask; |
| 150 | if (!did[bits|1]) { |
| 151 | bits |= 1; |
| 152 | s.append("1"); |
| 153 | } else { |
| 154 | s.append("0"); |
| 155 | } |
| 156 | CHECK(!did[bits]); |
| 157 | did[bits] = true; |
| 158 | } |
| 159 | return s; |
| 160 | } |
| 161 | |
| 162 | // Test that the DFA gets the right result even if it runs |
| 163 | // out of memory during a search. The regular expression |
| 164 | // 0[01]{n}$ matches a binary string of 0s and 1s only if |
| 165 | // the (n+1)th-to-last character is a 0. Matching this in |
| 166 | // a single forward pass (as done by the DFA) requires |
| 167 | // keeping one bit for each of the last n+1 characters |
| 168 | // (whether each was a 0), or 2^(n+1) possible states. |
| 169 | // If we run this regexp to search in a string that contains |
| 170 | // every possible n-character binary string as a substring, |
| 171 | // then it will have to run through at least 2^n states. |
| 172 | // States are big data structures -- certainly more than 1 byte -- |
| 173 | // so if the DFA can search correctly while staying within a |
| 174 | // 2^n byte limit, it must be handling out-of-memory conditions |
| 175 | // gracefully. |
| 176 | TEST(SingleThreaded, SearchDFA) { |
| 177 | // Choice of n is mostly arbitrary, except that: |
| 178 | // * making n too big makes the test run for too long. |
| 179 | // * making n too small makes the DFA refuse to run, |
| 180 | // because it has so little memory compared to the program size. |
| 181 | // Empirically, n = 18 is a good compromise between the two. |
| 182 | const int n = 18; |
| 183 | |
| 184 | Regexp* re = Regexp::Parse(StringPrintf("0[01]{%d}$", n), |
| 185 | Regexp::LikePerl, NULL); |
| 186 | CHECK(re); |
| 187 | |
| 188 | // The De Bruijn string for n ends with a 1 followed by n 0s in a row, |
| 189 | // which is not a match for 0[01]{n}$. Adding one more 0 is a match. |
| 190 | string no_match = DeBruijnString(n); |
| 191 | string match = no_match + "0"; |
| 192 | |
| 193 | // The De Bruijn string is the worst case input for this regexp. |
| 194 | // By default, the DFA will notice that it is flushing its cache |
| 195 | // too frequently and will bail out early, so that RE2 can use the |
| 196 | // NFA implementation instead. (The DFA loses its speed advantage |
| 197 | // if it can't get a good cache hit rate.) |
| 198 | // Tell the DFA to trudge along instead. |
| 199 | FLAGS_re2_dfa_bail_when_slow = false; |
| 200 | |
| 201 | int64 usage; |
| 202 | int64 peak_usage; |
| 203 | { |
| 204 | testing::MallocCounter m(testing::MallocCounter::THIS_THREAD_ONLY); |
| 205 | Prog* prog = re->CompileToProg(1<<n); |
| 206 | CHECK(prog); |
| 207 | for (int i = 0; i < 10; i++) { |
| 208 | bool matched, failed = false; |
| 209 | matched = prog->SearchDFA(match, NULL, |
| 210 | Prog::kUnanchored, Prog::kFirstMatch, |
| 211 | NULL, &failed, NULL); |
| 212 | CHECK(!failed); |
| 213 | CHECK(matched); |
| 214 | matched = prog->SearchDFA(no_match, NULL, |
| 215 | Prog::kUnanchored, Prog::kFirstMatch, |
| 216 | NULL, &failed, NULL); |
| 217 | CHECK(!failed); |
| 218 | CHECK(!matched); |
| 219 | } |
| 220 | usage = m.HeapGrowth(); |
| 221 | peak_usage = m.PeakHeapGrowth(); |
| 222 | delete prog; |
| 223 | } |
| 224 | re->Decref(); |
| 225 | |
| 226 | if (!UsingMallocCounter) |
| 227 | return; |
| 228 | //LOG(INFO) << "usage " << usage << " " << peak_usage; |
| 229 | CHECK_LT(usage, 1<<n); |
| 230 | CHECK_LT(peak_usage, 1<<n); |
| 231 | } |
| 232 | |
| 233 | // Helper thread: searches for match, which should match, |
| 234 | // and no_match, which should not. |
| 235 | class SearchThread : public Thread { |
| 236 | public: |
| 237 | SearchThread(Prog* prog, const StringPiece& match, |
| 238 | const StringPiece& no_match) |
| 239 | : prog_(prog), match_(match), no_match_(no_match) {} |
| 240 | |
| 241 | virtual void Run() { |
| 242 | for (int i = 0; i < 2; i++) { |
| 243 | bool matched, failed = false; |
| 244 | matched = prog_->SearchDFA(match_, NULL, |
| 245 | Prog::kUnanchored, Prog::kFirstMatch, |
| 246 | NULL, &failed, NULL); |
| 247 | CHECK(!failed); |
| 248 | CHECK(matched); |
| 249 | matched = prog_->SearchDFA(no_match_, NULL, |
| 250 | Prog::kUnanchored, Prog::kFirstMatch, |
| 251 | NULL, &failed, NULL); |
| 252 | CHECK(!failed); |
| 253 | CHECK(!matched); |
| 254 | } |
| 255 | } |
| 256 | |
| 257 | private: |
| 258 | Prog* prog_; |
| 259 | StringPiece match_; |
| 260 | StringPiece no_match_; |
| 261 | }; |
| 262 | |
| 263 | TEST(Multithreaded, SearchDFA) { |
| 264 | // Same as single-threaded test above. |
| 265 | const int n = 18; |
| 266 | Regexp* re = Regexp::Parse(StringPrintf("0[01]{%d}$", n), |
| 267 | Regexp::LikePerl, NULL); |
| 268 | CHECK(re); |
| 269 | string no_match = DeBruijnString(n); |
| 270 | string match = no_match + "0"; |
| 271 | FLAGS_re2_dfa_bail_when_slow = false; |
| 272 | |
| 273 | // Check that single-threaded code works. |
| 274 | { |
| 275 | Prog* prog = re->CompileToProg(1<<n); |
| 276 | CHECK(prog); |
| 277 | SearchThread* t = new SearchThread(prog, match, no_match); |
| 278 | t->SetJoinable(true); |
| 279 | t->Start(); |
| 280 | t->Join(); |
| 281 | delete t; |
| 282 | delete prog; |
| 283 | } |
| 284 | |
| 285 | // Run the search simultaneously in a bunch of threads. |
| 286 | // Reuse same flags for Multithreaded.BuildDFA above. |
| 287 | for (int i = 0; i < FLAGS_repeat; i++) { |
| 288 | //LOG(INFO) << "Search " << i; |
| 289 | Prog* prog = re->CompileToProg(1<<n); |
| 290 | CHECK(prog); |
| 291 | |
| 292 | vector<SearchThread*> threads; |
| 293 | for (int j = 0; j < FLAGS_threads; j++) { |
| 294 | SearchThread *t = new SearchThread(prog, match, no_match); |
| 295 | t->SetJoinable(true); |
| 296 | threads.push_back(t); |
| 297 | } |
| 298 | for (int j = 0; j < FLAGS_threads; j++) |
| 299 | threads[j]->Start(); |
| 300 | for (int j = 0; j < FLAGS_threads; j++) { |
| 301 | threads[j]->Join(); |
| 302 | delete threads[j]; |
| 303 | } |
| 304 | delete prog; |
| 305 | } |
| 306 | re->Decref(); |
| 307 | } |
| 308 | |
| 309 | struct ReverseTest { |
| 310 | const char *regexp; |
| 311 | const char *text; |
| 312 | bool match; |
| 313 | }; |
| 314 | |
| 315 | // Test that reverse DFA handles anchored/unanchored correctly. |
| 316 | // It's in the DFA interface but not used by RE2. |
| 317 | ReverseTest reverse_tests[] = { |
| 318 | { "\\A(a|b)", "abc", true }, |
| 319 | { "(a|b)\\z", "cba", true }, |
| 320 | { "\\A(a|b)", "cba", false }, |
| 321 | { "(a|b)\\z", "abc", false }, |
| 322 | }; |
| 323 | |
| 324 | TEST(DFA, ReverseMatch) { |
| 325 | int nfail = 0; |
| 326 | for (int i = 0; i < arraysize(reverse_tests); i++) { |
| 327 | const ReverseTest& t = reverse_tests[i]; |
| 328 | Regexp* re = Regexp::Parse(t.regexp, Regexp::LikePerl, NULL); |
| 329 | CHECK(re); |
| 330 | Prog *prog = re->CompileToReverseProg(0); |
| 331 | CHECK(prog); |
| 332 | bool failed = false; |
| 333 | bool matched = prog->SearchDFA(t.text, NULL, Prog::kUnanchored, Prog::kFirstMatch, NULL, &failed, NULL); |
| 334 | if (matched != t.match) { |
| 335 | LOG(ERROR) << t.regexp << " on " << t.text << ": want " << t.match; |
| 336 | nfail++; |
| 337 | } |
| 338 | delete prog; |
| 339 | re->Decref(); |
| 340 | } |
| 341 | EXPECT_EQ(nfail, 0); |
| 342 | } |
| 343 | |
| 344 | } // namespace re2 |