1 // class template regex -*- C++ -*-
3 // Copyright (C) 2013-2015 Free Software Foundation, Inc.
5 // This file is part of the GNU ISO C++ Library. This library is free
6 // software; you can redistribute it and/or modify it under the
7 // terms of the GNU General Public License as published by the
8 // Free Software Foundation; either version 3, or (at your option)
11 // This library is distributed in the hope that it will be useful,
12 // but WITHOUT ANY WARRANTY; without even the implied warranty of
13 // MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 // GNU General Public License for more details.
16 // Under Section 7 of GPL version 3, you are granted additional
17 // permissions described in the GCC Runtime Library Exception, version
18 // 3.1, as published by the Free Software Foundation.
20 // You should have received a copy of the GNU General Public License and
21 // a copy of the GCC Runtime Library Exception along with this program;
22 // see the files COPYING3 and COPYING.RUNTIME respectively. If not, see
23 // <http://www.gnu.org/licenses/>.
26 * @file bits/regex_executor.tcc
27 * This is an internal header file, included by other library headers.
28 * Do not attempt to use it directly. @headername{regex}
31 namespace std _GLIBCXX_VISIBILITY(default)
35 _GLIBCXX_BEGIN_NAMESPACE_VERSION
37 template<typename _BiIter, typename _Alloc, typename _TraitsT,
39 bool _Executor<_BiIter, _Alloc, _TraitsT, __dfs_mode>::
42 if (_M_search_from_first())
44 if (_M_flags & regex_constants::match_continuous)
46 _M_flags |= regex_constants::match_prev_avail;
47 while (_M_begin != _M_end)
50 if (_M_search_from_first())
56 // The _M_main function operates in different modes, DFS mode or BFS mode,
57 // indicated by template parameter __dfs_mode, and dispatches to one of the
58 // _M_main_dispatch overloads.
60 // ------------------------------------------------------------
64 // It applies a Depth-First-Search (aka backtracking) on given NFA and input
66 // At the very beginning the executor stands in the start state, then it
67 // tries every possible state transition in current state recursively. Some
68 // state transitions consume input string, say, a single-char-matcher or a
69 // back-reference matcher; some don't, like assertion or other anchor nodes.
70 // When the input is exhausted and/or the current state is an accepting
71 // state, the whole executor returns true.
73 // TODO: This approach is exponentially slow for certain input.
74 // Try to compile the NFA to a DFA.
76 // Time complexity: \Omega(match_length), O(2^(_M_nfa.size()))
77 // Space complexity: \theta(match_results.size() + match_length)
79 template<typename _BiIter, typename _Alloc, typename _TraitsT,
81 bool _Executor<_BiIter, _Alloc, _TraitsT, __dfs_mode>::
82 _M_main_dispatch(_Match_mode __match_mode, __dfs)
85 *_M_states._M_get_sol_pos() = _BiIter();
86 _M_cur_results = _M_results;
87 _M_dfs(__match_mode, _M_states._M_start);
91 // ------------------------------------------------------------
95 // Russ Cox's article (http://swtch.com/~rsc/regexp/regexp1.html)
96 // explained this algorithm clearly.
98 // It first computes epsilon closure (states that can be achieved without
99 // consuming characters) for every state that's still matching,
100 // using the same DFS algorithm, but doesn't re-enter states (using
101 // _M_states._M_visited to check), nor follow _S_opcode_match.
103 // Then apply DFS using every _S_opcode_match (in _M_states._M_match_queue)
104 // as the start state.
106 // It significantly reduces potential duplicate states, so has a better
107 // upper bound; but it requires more overhead.
109 // Time complexity: \Omega(match_length * match_results.size())
110 // O(match_length * _M_nfa.size() * match_results.size())
111 // Space complexity: \Omega(_M_nfa.size() + match_results.size())
112 // O(_M_nfa.size() * match_results.size())
113 template<typename _BiIter, typename _Alloc, typename _TraitsT,
115 bool _Executor<_BiIter, _Alloc, _TraitsT, __dfs_mode>::
116 _M_main_dispatch(_Match_mode __match_mode, __bfs)
118 _M_states._M_queue(_M_states._M_start, _M_results);
123 if (_M_states._M_match_queue.empty())
125 std::fill_n(_M_states._M_visited_states.get(), _M_nfa.size(), false);
126 auto __old_queue = std::move(_M_states._M_match_queue);
127 for (auto& __task : __old_queue)
129 _M_cur_results = std::move(__task.second);
130 _M_dfs(__match_mode, __task.first);
132 if (__match_mode == _Match_mode::_Prefix)
134 if (_M_current == _M_end)
138 if (__match_mode == _Match_mode::_Exact)
140 _M_states._M_match_queue.clear();
144 // Return whether now match the given sub-NFA.
145 template<typename _BiIter, typename _Alloc, typename _TraitsT,
147 bool _Executor<_BiIter, _Alloc, _TraitsT, __dfs_mode>::
148 _M_lookahead(_StateIdT __next)
150 // Backreferences may refer to captured content.
151 // We may want to make this faster by not copying,
152 // but let's not be clever prematurely.
153 _ResultsVec __what(_M_cur_results);
154 _Executor __sub(_M_current, _M_end, __what, _M_re, _M_flags);
155 __sub._M_states._M_start = __next;
156 if (__sub._M_search_from_first())
158 for (size_t __i = 0; __i < __what.size(); __i++)
159 if (__what[__i].matched)
160 _M_cur_results[__i] = __what[__i];
166 // __rep_count records how many times (__rep_count.second)
167 // this node is visited under certain input iterator
168 // (__rep_count.first). This prevent the executor from entering
169 // infinite loop by refusing to continue when it's already been
170 // visited more than twice. It's `twice` instead of `once` because
171 // we need to spare one more time for potential group capture.
172 template<typename _BiIter, typename _Alloc, typename _TraitsT,
174 void _Executor<_BiIter, _Alloc, _TraitsT, __dfs_mode>::
175 _M_rep_once_more(_Match_mode __match_mode, _StateIdT __i)
177 const auto& __state = _M_nfa[__i];
178 auto& __rep_count = _M_rep_count[__i];
179 if (__rep_count.second == 0 || __rep_count.first != _M_current)
181 auto __back = __rep_count;
182 __rep_count.first = _M_current;
183 __rep_count.second = 1;
184 _M_dfs(__match_mode, __state._M_alt);
185 __rep_count = __back;
189 if (__rep_count.second < 2)
191 __rep_count.second++;
192 _M_dfs(__match_mode, __state._M_alt);
193 __rep_count.second--;
198 template<typename _BiIter, typename _Alloc, typename _TraitsT,
200 void _Executor<_BiIter, _Alloc, _TraitsT, __dfs_mode>::
201 _M_dfs(_Match_mode __match_mode, _StateIdT __i)
203 if (_M_states._M_visited(__i))
206 const auto& __state = _M_nfa[__i];
207 // Every change on _M_cur_results and _M_current will be rolled back after
208 // finishing the recursion step.
209 switch (__state._M_opcode())
211 // _M_alt branch is "match once more", while _M_next is "get me out
212 // of this quantifier". Executing _M_next first or _M_alt first don't
213 // mean the same thing, and we need to choose the correct order under
214 // given greedy mode.
215 case _S_opcode_repeat:
220 _M_rep_once_more(__match_mode, __i);
221 // If it's DFS executor and already accepted, we're done.
222 if (!__dfs_mode || !_M_has_sol)
223 _M_dfs(__match_mode, __state._M_next);
225 else // Non-greedy mode
230 _M_dfs(__match_mode, __state._M_next);
232 _M_rep_once_more(__match_mode, __i);
236 // DON'T attempt anything, because there's already another
237 // state with higher priority accepted. This state cannot
238 // be better by attempting its next node.
241 _M_dfs(__match_mode, __state._M_next);
242 // DON'T attempt anything if it's already accepted. An
243 // accepted state *must* be better than a solution that
244 // matches a non-greedy quantifier one more time.
246 _M_rep_once_more(__match_mode, __i);
252 case _S_opcode_subexpr_begin:
254 auto& __res = _M_cur_results[__state._M_subexpr];
255 auto __back = __res.first;
256 __res.first = _M_current;
257 _M_dfs(__match_mode, __state._M_next);
258 __res.first = __back;
261 case _S_opcode_subexpr_end:
263 auto& __res = _M_cur_results[__state._M_subexpr];
265 __res.second = _M_current;
266 __res.matched = true;
267 _M_dfs(__match_mode, __state._M_next);
271 case _S_opcode_line_begin_assertion:
273 _M_dfs(__match_mode, __state._M_next);
275 case _S_opcode_line_end_assertion:
277 _M_dfs(__match_mode, __state._M_next);
279 case _S_opcode_word_boundary:
280 if (_M_word_boundary() == !__state._M_neg)
281 _M_dfs(__match_mode, __state._M_next);
283 // Here __state._M_alt offers a single start node for a sub-NFA.
284 // We recursively invoke our algorithm to match the sub-NFA.
285 case _S_opcode_subexpr_lookahead:
286 if (_M_lookahead(__state._M_alt) == !__state._M_neg)
287 _M_dfs(__match_mode, __state._M_next);
289 case _S_opcode_match:
290 if (_M_current == _M_end)
294 if (__state._M_matches(*_M_current))
297 _M_dfs(__match_mode, __state._M_next);
302 if (__state._M_matches(*_M_current))
303 _M_states._M_queue(__state._M_next, _M_cur_results);
305 // First fetch the matched result from _M_cur_results as __submatch;
306 // then compare it with
307 // (_M_current, _M_current + (__submatch.second - __submatch.first)).
308 // If matched, keep going; else just return and try another state.
309 case _S_opcode_backref:
311 __glibcxx_assert(__dfs_mode);
312 auto& __submatch = _M_cur_results[__state._M_backref_index];
313 if (!__submatch.matched)
315 auto __last = _M_current;
316 for (auto __tmp = __submatch.first;
317 __last != _M_end && __tmp != __submatch.second;
320 if (_M_re._M_automaton->_M_traits.transform(__submatch.first,
322 == _M_re._M_automaton->_M_traits.transform(_M_current, __last))
324 if (__last != _M_current)
326 auto __backup = _M_current;
328 _M_dfs(__match_mode, __state._M_next);
329 _M_current = __backup;
332 _M_dfs(__match_mode, __state._M_next);
336 case _S_opcode_accept:
339 __glibcxx_assert(!_M_has_sol);
340 if (__match_mode == _Match_mode::_Exact)
341 _M_has_sol = _M_current == _M_end;
344 if (_M_current == _M_begin
345 && (_M_flags & regex_constants::match_not_null))
349 if (_M_nfa._M_flags & regex_constants::ECMAScript)
350 _M_results = _M_cur_results;
353 __glibcxx_assert(_M_states._M_get_sol_pos());
354 // Here's POSIX's logic: match the longest one. However
355 // we never know which one (lhs or rhs of "|") is longer
356 // unless we try both of them and compare the results.
357 // The member variable _M_sol_pos records the end
358 // position of the last successful match. It's better
359 // to be larger, because POSIX regex is always greedy.
360 // TODO: This could be slow.
361 if (*_M_states._M_get_sol_pos() == _BiIter()
362 || std::distance(_M_begin,
363 *_M_states._M_get_sol_pos())
364 < std::distance(_M_begin, _M_current))
366 *_M_states._M_get_sol_pos() = _M_current;
367 _M_results = _M_cur_results;
374 if (_M_current == _M_begin
375 && (_M_flags & regex_constants::match_not_null))
377 if (__match_mode == _Match_mode::_Prefix || _M_current == _M_end)
381 _M_results = _M_cur_results;
385 case _S_opcode_alternative:
386 if (_M_nfa._M_flags & regex_constants::ECMAScript)
388 // TODO: Fix BFS support. It is wrong.
389 _M_dfs(__match_mode, __state._M_alt);
390 // Pick lhs if it matches. Only try rhs if it doesn't.
392 _M_dfs(__match_mode, __state._M_next);
396 // Try both and compare the result.
397 // See "case _S_opcode_accept:" handling above.
398 _M_dfs(__match_mode, __state._M_alt);
399 auto __has_sol = _M_has_sol;
401 _M_dfs(__match_mode, __state._M_next);
402 _M_has_sol |= __has_sol;
406 __glibcxx_assert(false);
410 // Return whether now is at some word boundary.
411 template<typename _BiIter, typename _Alloc, typename _TraitsT,
413 bool _Executor<_BiIter, _Alloc, _TraitsT, __dfs_mode>::
414 _M_word_boundary() const
416 bool __left_is_word = false;
417 if (_M_current != _M_begin
418 || (_M_flags & regex_constants::match_prev_avail))
420 auto __prev = _M_current;
421 if (_M_is_word(*std::prev(__prev)))
422 __left_is_word = true;
424 bool __right_is_word =
425 _M_current != _M_end && _M_is_word(*_M_current);
427 if (__left_is_word == __right_is_word)
429 if (__left_is_word && !(_M_flags & regex_constants::match_not_eow))
431 if (__right_is_word && !(_M_flags & regex_constants::match_not_bow))
436 _GLIBCXX_END_NAMESPACE_VERSION
437 } // namespace __detail