OLD | NEW |
| (Empty) |
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/thread.h" | |
6 #include "util/test.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 static const bool UsingMallocCounter = false; | |
14 | |
15 DECLARE_bool(re2_dfa_bail_when_slow); | |
16 | |
17 DEFINE_int32(size, 8, "log2(number of DFA nodes)"); | |
18 DEFINE_int32(repeat, 2, "Repetition count."); | |
19 DEFINE_int32(threads, 4, "number of threads"); | |
20 | |
21 namespace re2 { | |
22 | |
23 // Check that multithreaded access to DFA class works. | |
24 | |
25 // Helper thread: builds entire DFA for prog. | |
26 class BuildThread : public Thread { | |
27 public: | |
28 BuildThread(Prog* prog) : prog_(prog) {} | |
29 virtual void Run() { | |
30 CHECK(prog_->BuildEntireDFA(Prog::kFirstMatch)); | |
31 } | |
32 | |
33 private: | |
34 Prog* prog_; | |
35 }; | |
36 | |
37 TEST(Multithreaded, BuildEntireDFA) { | |
38 // Create regexp with 2^FLAGS_size states in DFA. | |
39 string s = "a"; | |
40 for (int i = 0; i < FLAGS_size; i++) | |
41 s += "[ab]"; | |
42 s += "b"; | |
43 | |
44 // Check that single-threaded code works. | |
45 { | |
46 //LOG(INFO) << s; | |
47 Regexp* re = Regexp::Parse(s, Regexp::LikePerl, NULL); | |
48 CHECK(re); | |
49 Prog* prog = re->CompileToProg(0); | |
50 CHECK(prog); | |
51 BuildThread* t = new BuildThread(prog); | |
52 t->SetJoinable(true); | |
53 t->Start(); | |
54 t->Join(); | |
55 delete t; | |
56 delete prog; | |
57 re->Decref(); | |
58 } | |
59 | |
60 // Build the DFA simultaneously in a bunch of threads. | |
61 for (int i = 0; i < FLAGS_repeat; i++) { | |
62 Regexp* re = Regexp::Parse(s, Regexp::LikePerl, NULL); | |
63 CHECK(re); | |
64 Prog* prog = re->CompileToProg(0); | |
65 CHECK(prog); | |
66 | |
67 vector<BuildThread*> threads; | |
68 for (int j = 0; j < FLAGS_threads; j++) { | |
69 BuildThread *t = new BuildThread(prog); | |
70 t->SetJoinable(true); | |
71 threads.push_back(t); | |
72 } | |
73 for (int j = 0; j < FLAGS_threads; j++) | |
74 threads[j]->Start(); | |
75 for (int j = 0; j < FLAGS_threads; j++) { | |
76 threads[j]->Join(); | |
77 delete threads[j]; | |
78 } | |
79 | |
80 // One more compile, to make sure everything is okay. | |
81 prog->BuildEntireDFA(Prog::kFirstMatch); | |
82 delete prog; | |
83 re->Decref(); | |
84 } | |
85 } | |
86 | |
87 // Check that DFA size requirements are followed. | |
88 // BuildEntireDFA will, like SearchDFA, stop building out | |
89 // the DFA once the memory limits are reached. | |
90 TEST(SingleThreaded, BuildEntireDFA) { | |
91 // Create regexp with 2^30 states in DFA. | |
92 string s = "a"; | |
93 for (int i = 0; i < 30; i++) | |
94 s += "[ab]"; | |
95 s += "b"; | |
96 | |
97 Regexp* re = Regexp::Parse(s, Regexp::LikePerl, NULL); | |
98 CHECK(re); | |
99 int max = 24; | |
100 for (int i = 17; i < max; i++) { | |
101 int64 limit = 1<<i; | |
102 int64 usage; | |
103 //int64 progusage, dfamem; | |
104 { | |
105 testing::MallocCounter m(testing::MallocCounter::THIS_THREAD_ONLY); | |
106 Prog* prog = re->CompileToProg(limit); | |
107 CHECK(prog); | |
108 //progusage = m.HeapGrowth(); | |
109 //dfamem = prog->dfa_mem(); | |
110 prog->BuildEntireDFA(Prog::kFirstMatch); | |
111 prog->BuildEntireDFA(Prog::kLongestMatch); | |
112 usage = m.HeapGrowth(); | |
113 delete prog; | |
114 } | |
115 if (!UsingMallocCounter) | |
116 continue; | |
117 //LOG(INFO) << "limit " << limit << ", " | |
118 // << "prog usage " << progusage << ", " | |
119 // << "DFA budget " << dfamem << ", " | |
120 // << "total " << usage; | |
121 // Tolerate +/- 10%. | |
122 CHECK_GT(usage, limit*9/10); | |
123 CHECK_LT(usage, limit*11/10); | |
124 } | |
125 re->Decref(); | |
126 } | |
127 | |
128 // Generates and returns a string over binary alphabet {0,1} that contains | |
129 // all possible binary sequences of length n as subsequences. The obvious | |
130 // brute force method would generate a string of length n * 2^n, but this | |
131 // generates a string of length n + 2^n - 1 called a De Bruijn cycle. | |
132 // See Knuth, The Art of Computer Programming, Vol 2, Exercise 3.2.2 #17. | |
133 // Such a string is useful for testing a DFA. If you have a DFA | |
134 // where distinct last n bytes implies distinct states, then running on a | |
135 // DeBruijn string causes the DFA to need to create a new state at every | |
136 // position in the input, never reusing any states until it gets to the | |
137 // end of the string. This is the worst possible case for DFA execution. | |
138 static string DeBruijnString(int n) { | |
139 CHECK_LT(n, static_cast<int>(8*sizeof(int))); | |
140 CHECK_GT(n, 0); | |
141 | |
142 vector<bool> did(1<<n); | |
143 for (int i = 0; i < 1<<n; i++) | |
144 did[i] = false; | |
145 | |
146 string s; | |
147 for (int i = 0; i < n-1; i++) | |
148 s.append("0"); | |
149 int bits = 0; | |
150 int mask = (1<<n) - 1; | |
151 for (int i = 0; i < (1<<n); i++) { | |
152 bits <<= 1; | |
153 bits &= mask; | |
154 if (!did[bits|1]) { | |
155 bits |= 1; | |
156 s.append("1"); | |
157 } else { | |
158 s.append("0"); | |
159 } | |
160 CHECK(!did[bits]); | |
161 did[bits] = true; | |
162 } | |
163 return s; | |
164 } | |
165 | |
166 // Test that the DFA gets the right result even if it runs | |
167 // out of memory during a search. The regular expression | |
168 // 0[01]{n}$ matches a binary string of 0s and 1s only if | |
169 // the (n+1)th-to-last character is a 0. Matching this in | |
170 // a single forward pass (as done by the DFA) requires | |
171 // keeping one bit for each of the last n+1 characters | |
172 // (whether each was a 0), or 2^(n+1) possible states. | |
173 // If we run this regexp to search in a string that contains | |
174 // every possible n-character binary string as a substring, | |
175 // then it will have to run through at least 2^n states. | |
176 // States are big data structures -- certainly more than 1 byte -- | |
177 // so if the DFA can search correctly while staying within a | |
178 // 2^n byte limit, it must be handling out-of-memory conditions | |
179 // gracefully. | |
180 TEST(SingleThreaded, SearchDFA) { | |
181 // Choice of n is mostly arbitrary, except that: | |
182 // * making n too big makes the test run for too long. | |
183 // * making n too small makes the DFA refuse to run, | |
184 // because it has so little memory compared to the program size. | |
185 // Empirically, n = 18 is a good compromise between the two. | |
186 const int n = 18; | |
187 | |
188 Regexp* re = Regexp::Parse(StringPrintf("0[01]{%d}$", n), | |
189 Regexp::LikePerl, NULL); | |
190 CHECK(re); | |
191 | |
192 // The De Bruijn string for n ends with a 1 followed by n 0s in a row, | |
193 // which is not a match for 0[01]{n}$. Adding one more 0 is a match. | |
194 string no_match = DeBruijnString(n); | |
195 string match = no_match + "0"; | |
196 | |
197 // The De Bruijn string is the worst case input for this regexp. | |
198 // By default, the DFA will notice that it is flushing its cache | |
199 // too frequently and will bail out early, so that RE2 can use the | |
200 // NFA implementation instead. (The DFA loses its speed advantage | |
201 // if it can't get a good cache hit rate.) | |
202 // Tell the DFA to trudge along instead. | |
203 FLAGS_re2_dfa_bail_when_slow = false; | |
204 | |
205 int64 usage; | |
206 int64 peak_usage; | |
207 { | |
208 testing::MallocCounter m(testing::MallocCounter::THIS_THREAD_ONLY); | |
209 Prog* prog = re->CompileToProg(1<<n); | |
210 CHECK(prog); | |
211 for (int i = 0; i < 10; i++) { | |
212 bool matched, failed = false; | |
213 matched = prog->SearchDFA(match, NULL, | |
214 Prog::kUnanchored, Prog::kFirstMatch, | |
215 NULL, &failed, NULL); | |
216 CHECK(!failed); | |
217 CHECK(matched); | |
218 matched = prog->SearchDFA(no_match, NULL, | |
219 Prog::kUnanchored, Prog::kFirstMatch, | |
220 NULL, &failed, NULL); | |
221 CHECK(!failed); | |
222 CHECK(!matched); | |
223 } | |
224 usage = m.HeapGrowth(); | |
225 peak_usage = m.PeakHeapGrowth(); | |
226 delete prog; | |
227 } | |
228 if (!UsingMallocCounter) | |
229 return; | |
230 //LOG(INFO) << "usage " << usage << ", " | |
231 // << "peak usage " << peak_usage; | |
232 CHECK_LT(usage, 1<<n); | |
233 CHECK_LT(peak_usage, 1<<n); | |
234 re->Decref(); | |
235 } | |
236 | |
237 // Helper thread: searches for match, which should match, | |
238 // and no_match, which should not. | |
239 class SearchThread : public Thread { | |
240 public: | |
241 SearchThread(Prog* prog, const StringPiece& match, | |
242 const StringPiece& no_match) | |
243 : prog_(prog), match_(match), no_match_(no_match) {} | |
244 | |
245 virtual void Run() { | |
246 for (int i = 0; i < 2; i++) { | |
247 bool matched, failed = false; | |
248 matched = prog_->SearchDFA(match_, NULL, | |
249 Prog::kUnanchored, Prog::kFirstMatch, | |
250 NULL, &failed, NULL); | |
251 CHECK(!failed); | |
252 CHECK(matched); | |
253 matched = prog_->SearchDFA(no_match_, NULL, | |
254 Prog::kUnanchored, Prog::kFirstMatch, | |
255 NULL, &failed, NULL); | |
256 CHECK(!failed); | |
257 CHECK(!matched); | |
258 } | |
259 } | |
260 | |
261 private: | |
262 Prog* prog_; | |
263 StringPiece match_; | |
264 StringPiece no_match_; | |
265 }; | |
266 | |
267 TEST(Multithreaded, SearchDFA) { | |
268 // Same as single-threaded test above. | |
269 const int n = 18; | |
270 Regexp* re = Regexp::Parse(StringPrintf("0[01]{%d}$", n), | |
271 Regexp::LikePerl, NULL); | |
272 CHECK(re); | |
273 string no_match = DeBruijnString(n); | |
274 string match = no_match + "0"; | |
275 FLAGS_re2_dfa_bail_when_slow = false; | |
276 | |
277 // Check that single-threaded code works. | |
278 { | |
279 Prog* prog = re->CompileToProg(1<<n); | |
280 CHECK(prog); | |
281 SearchThread* t = new SearchThread(prog, match, no_match); | |
282 t->SetJoinable(true); | |
283 t->Start(); | |
284 t->Join(); | |
285 delete t; | |
286 delete prog; | |
287 } | |
288 | |
289 // Run the search simultaneously in a bunch of threads. | |
290 // Reuse same flags for Multithreaded.BuildDFA above. | |
291 for (int i = 0; i < FLAGS_repeat; i++) { | |
292 //LOG(INFO) << "Search " << i; | |
293 Prog* prog = re->CompileToProg(1<<n); | |
294 CHECK(prog); | |
295 | |
296 vector<SearchThread*> threads; | |
297 for (int j = 0; j < FLAGS_threads; j++) { | |
298 SearchThread *t = new SearchThread(prog, match, no_match); | |
299 t->SetJoinable(true); | |
300 threads.push_back(t); | |
301 } | |
302 for (int j = 0; j < FLAGS_threads; j++) | |
303 threads[j]->Start(); | |
304 for (int j = 0; j < FLAGS_threads; j++) { | |
305 threads[j]->Join(); | |
306 delete threads[j]; | |
307 } | |
308 delete prog; | |
309 } | |
310 re->Decref(); | |
311 } | |
312 | |
313 struct ReverseTest { | |
314 const char *regexp; | |
315 const char *text; | |
316 bool match; | |
317 }; | |
318 | |
319 // Test that reverse DFA handles anchored/unanchored correctly. | |
320 // It's in the DFA interface but not used by RE2. | |
321 ReverseTest reverse_tests[] = { | |
322 { "\\A(a|b)", "abc", true }, | |
323 { "(a|b)\\z", "cba", true }, | |
324 { "\\A(a|b)", "cba", false }, | |
325 { "(a|b)\\z", "abc", false }, | |
326 }; | |
327 | |
328 TEST(DFA, ReverseMatch) { | |
329 int nfail = 0; | |
330 for (int i = 0; i < arraysize(reverse_tests); i++) { | |
331 const ReverseTest& t = reverse_tests[i]; | |
332 Regexp* re = Regexp::Parse(t.regexp, Regexp::LikePerl, NULL); | |
333 CHECK(re); | |
334 Prog *prog = re->CompileToReverseProg(0); | |
335 CHECK(prog); | |
336 bool failed = false; | |
337 bool matched = prog->SearchDFA(t.text, NULL, Prog::kUnanchored, Prog::kFirst
Match, NULL, &failed, NULL); | |
338 if (matched != t.match) { | |
339 LOG(ERROR) << t.regexp << " on " << t.text << ": want " << t.match; | |
340 nfail++; | |
341 } | |
342 delete prog; | |
343 re->Decref(); | |
344 } | |
345 EXPECT_EQ(nfail, 0); | |
346 } | |
347 | |
348 } // namespace re2 | |
OLD | NEW |