1 // class template regex -*- C++ -*-
3 // Copyright (C) 2013-2014 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_flags & regex_constants::match_continuous)
43 return _M_search_from_first();
44 auto __cur = _M_begin;
51 // Continue when __cur == _M_end
52 while (__cur++ != _M_end);
56 // This function operates in different modes, DFS mode or BFS mode, indicated
57 // by template parameter __dfs_mode. See _M_main for details.
59 // ------------------------------------------------------------
63 // It applies a Depth-First-Search (aka backtracking) on given NFA and input
65 // At the very beginning the executor stands in the start state, then it tries
66 // every possible state transition in current state recursively. Some state
67 // transitions consume input string, say, a single-char-matcher or a
68 // back-reference matcher; some don't, like assertion or other anchor nodes.
69 // When the input is exhausted and/or the current state is an accepting state,
70 // the whole executor returns true.
72 // TODO: This approach is exponentially slow for certain input.
73 // Try to compile the NFA to a DFA.
75 // Time complexity: \Omega(match_length), O(2^(_M_nfa.size()))
76 // Space complexity: \theta(match_results.size() + match_length)
78 // ------------------------------------------------------------
82 // Russ Cox's article (http://swtch.com/~rsc/regexp/regexp1.html)
83 // explained this algorithm clearly.
85 // It first computes epsilon closure (states that can be achieved without
86 // consuming characters) for every state that's still matching,
87 // using the same DFS algorithm, but doesn't re-enter states (find a true in
88 // _M_visited), nor follows _S_opcode_match.
90 // Then apply DFS using every _S_opcode_match (in _M_match_queue) as the start
93 // It significantly reduces potential duplicate states, so has a better
94 // upper bound; but it requires more overhead.
96 // Time complexity: \Omega(match_length * match_results.size())
97 // O(match_length * _M_nfa.size() * match_results.size())
98 // Space complexity: \Omega(_M_nfa.size() + match_results.size())
99 // O(_M_nfa.size() * match_results.size())
100 template<typename _BiIter, typename _Alloc, typename _TraitsT,
102 template<bool __match_mode>
103 bool _Executor<_BiIter, _Alloc, _TraitsT, __dfs_mode>::
109 _M_cur_results = _M_results;
110 _M_dfs<__match_mode>(_M_start_state);
115 _M_match_queue->push_back(make_pair(_M_start_state, _M_results));
120 if (_M_match_queue->empty())
122 _M_visited->assign(_M_visited->size(), false);
123 auto __old_queue = std::move(*_M_match_queue);
124 for (auto& __task : __old_queue)
126 _M_cur_results = std::move(__task.second);
127 _M_dfs<__match_mode>(__task.first);
131 if (_M_current == _M_end)
141 // Return whether now match the given sub-NFA.
142 template<typename _BiIter, typename _Alloc, typename _TraitsT,
144 bool _Executor<_BiIter, _Alloc, _TraitsT, __dfs_mode>::
145 _M_lookahead(_State<_TraitsT> __state)
147 // Backreferences may refer to captured content.
148 // We may want to make this faster by not copying,
149 // but let's not be clever prematurely.
150 _ResultsVec __what(_M_cur_results);
151 auto __sub = std::unique_ptr<_Executor>(new _Executor(_M_current,
156 __sub->_M_start_state = __state._M_alt;
157 if (__sub->_M_search_from_first())
159 for (size_t __i = 0; __i < __what.size(); __i++)
160 if (__what[__i].matched)
161 _M_cur_results[__i] = __what[__i];
167 // TODO: Use a function vector to dispatch, instead of using switch-case.
168 template<typename _BiIter, typename _Alloc, typename _TraitsT,
170 template<bool __match_mode>
171 void _Executor<_BiIter, _Alloc, _TraitsT, __dfs_mode>::
172 _M_dfs(_StateIdT __i)
176 if ((*_M_visited)[__i])
178 (*_M_visited)[__i] = true;
181 const auto& __state = _M_nfa[__i];
182 // Every change on _M_cur_results and _M_current will be rolled back after
183 // finishing the recursion step.
184 switch (__state._M_opcode)
186 // _M_alt branch is "match once more", while _M_next is "get me out
187 // of this quantifier". Executing _M_next first or _M_alt first don't
188 // mean the same thing, and we need to choose the correct order under
189 // given greedy mode.
190 case _S_opcode_alternative:
194 // "Once more" is preferred in greedy mode.
195 _M_dfs<__match_mode>(__state._M_alt);
196 // If it's DFS executor and already accepted, we're done.
197 if (!__dfs_mode || !_M_has_sol)
198 _M_dfs<__match_mode>(__state._M_next);
200 else // Non-greedy mode
205 _M_dfs<__match_mode>(__state._M_next);
207 _M_dfs<__match_mode>(__state._M_alt);
211 // DON'T attempt anything, because there's already another
212 // state with higher priority accepted. This state cannot be
213 // better by attempting its next node.
216 _M_dfs<__match_mode>(__state._M_next);
217 // DON'T attempt anything if it's already accepted. An
218 // accepted state *must* be better than a solution that
219 // matches a non-greedy quantifier one more time.
221 _M_dfs<__match_mode>(__state._M_alt);
226 case _S_opcode_subexpr_begin:
227 // If there's nothing changed since last visit, do NOT continue.
228 // This prevents the executor from get into infinite loop when using
229 // "()*" to match "".
230 if (!_M_cur_results[__state._M_subexpr].matched
231 || _M_cur_results[__state._M_subexpr].first != _M_current)
233 auto& __res = _M_cur_results[__state._M_subexpr];
234 auto __back = __res.first;
235 __res.first = _M_current;
236 _M_dfs<__match_mode>(__state._M_next);
237 __res.first = __back;
240 case _S_opcode_subexpr_end:
241 if (_M_cur_results[__state._M_subexpr].second != _M_current
242 || _M_cur_results[__state._M_subexpr].matched != true)
244 auto& __res = _M_cur_results[__state._M_subexpr];
246 __res.second = _M_current;
247 __res.matched = true;
248 _M_dfs<__match_mode>(__state._M_next);
252 _M_dfs<__match_mode>(__state._M_next);
254 case _S_opcode_line_begin_assertion:
256 _M_dfs<__match_mode>(__state._M_next);
258 case _S_opcode_line_end_assertion:
260 _M_dfs<__match_mode>(__state._M_next);
262 case _S_opcode_word_boundary:
263 if (_M_word_boundary(__state) == !__state._M_neg)
264 _M_dfs<__match_mode>(__state._M_next);
266 // Here __state._M_alt offers a single start node for a sub-NFA.
267 // We recursively invoke our algorithm to match the sub-NFA.
268 case _S_opcode_subexpr_lookahead:
269 if (_M_lookahead(__state) == !__state._M_neg)
270 _M_dfs<__match_mode>(__state._M_next);
272 case _S_opcode_match:
273 if (_M_current == _M_end)
277 if (__state._M_matches(*_M_current))
280 _M_dfs<__match_mode>(__state._M_next);
285 if (__state._M_matches(*_M_current))
286 _M_match_queue->push_back(make_pair(__state._M_next,
289 // First fetch the matched result from _M_cur_results as __submatch;
290 // then compare it with
291 // (_M_current, _M_current + (__submatch.second - __submatch.first)).
292 // If matched, keep going; else just return and try another state.
293 case _S_opcode_backref:
295 _GLIBCXX_DEBUG_ASSERT(__dfs_mode);
296 auto& __submatch = _M_cur_results[__state._M_backref_index];
297 if (!__submatch.matched)
299 auto __last = _M_current;
300 for (auto __tmp = __submatch.first;
301 __last != _M_end && __tmp != __submatch.second;
304 if (_M_re._M_traits.transform(__submatch.first,
306 == _M_re._M_traits.transform(_M_current, __last))
308 if (__last != _M_current)
310 auto __backup = _M_current;
312 _M_dfs<__match_mode>(__state._M_next);
313 _M_current = __backup;
316 _M_dfs<__match_mode>(__state._M_next);
320 case _S_opcode_accept:
323 _GLIBCXX_DEBUG_ASSERT(!_M_has_sol);
325 _M_has_sol = _M_current == _M_end;
328 if (_M_current == _M_begin
329 && (_M_flags & regex_constants::match_not_null))
332 _M_results = _M_cur_results;
336 if (_M_current == _M_begin
337 && (_M_flags & regex_constants::match_not_null))
339 if (!__match_mode || _M_current == _M_end)
343 _M_results = _M_cur_results;
348 _GLIBCXX_DEBUG_ASSERT(false);
352 // Return whether now is at some word boundary.
353 template<typename _BiIter, typename _Alloc, typename _TraitsT,
355 bool _Executor<_BiIter, _Alloc, _TraitsT, __dfs_mode>::
356 _M_word_boundary(_State<_TraitsT> __state) const
358 bool __left_is_word = false;
359 if (_M_current != _M_begin
360 || (_M_flags & regex_constants::match_prev_avail))
362 auto __prev = _M_current;
363 if (_M_is_word(*std::prev(__prev)))
364 __left_is_word = true;
366 bool __right_is_word =
367 _M_current != _M_end && _M_is_word(*_M_current);
369 if (__left_is_word == __right_is_word)
371 if (__left_is_word && !(_M_flags & regex_constants::match_not_eow))
373 if (__right_is_word && !(_M_flags & regex_constants::match_not_bow))
378 _GLIBCXX_END_NAMESPACE_VERSION
379 } // namespace __detail