libstdc++
regex_executor.tcc
Go to the documentation of this file.
00001 // class template regex -*- C++ -*-
00002 
00003 // Copyright (C) 2013-2017 Free Software Foundation, Inc.
00004 //
00005 // This file is part of the GNU ISO C++ Library.  This library is free
00006 // software; you can redistribute it and/or modify it under the
00007 // terms of the GNU General Public License as published by the
00008 // Free Software Foundation; either version 3, or (at your option)
00009 // any later version.
00010 
00011 // This library is distributed in the hope that it will be useful,
00012 // but WITHOUT ANY WARRANTY; without even the implied warranty of
00013 // MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
00014 // GNU General Public License for more details.
00015 
00016 // Under Section 7 of GPL version 3, you are granted additional
00017 // permissions described in the GCC Runtime Library Exception, version
00018 // 3.1, as published by the Free Software Foundation.
00019 
00020 // You should have received a copy of the GNU General Public License and
00021 // a copy of the GCC Runtime Library Exception along with this program;
00022 // see the files COPYING3 and COPYING.RUNTIME respectively.  If not, see
00023 // <http://www.gnu.org/licenses/>.
00024 
00025 /**
00026  *  @file bits/regex_executor.tcc
00027  *  This is an internal header file, included by other library headers.
00028  *  Do not attempt to use it directly. @headername{regex}
00029  */
00030 
00031 namespace std _GLIBCXX_VISIBILITY(default)
00032 {
00033 namespace __detail
00034 {
00035 _GLIBCXX_BEGIN_NAMESPACE_VERSION
00036 
00037   template<typename _BiIter, typename _Alloc, typename _TraitsT,
00038            bool __dfs_mode>
00039     bool _Executor<_BiIter, _Alloc, _TraitsT, __dfs_mode>::
00040     _M_search()
00041     {
00042       if (_M_search_from_first())
00043         return true;
00044       if (_M_flags & regex_constants::match_continuous)
00045         return false;
00046       _M_flags |= regex_constants::match_prev_avail;
00047       while (_M_begin != _M_end)
00048         {
00049           ++_M_begin;
00050           if (_M_search_from_first())
00051             return true;
00052         }
00053       return false;
00054     }
00055 
00056   // The _M_main function operates in different modes, DFS mode or BFS mode,
00057   // indicated by template parameter __dfs_mode, and dispatches to one of the
00058   // _M_main_dispatch overloads.
00059   //
00060   // ------------------------------------------------------------
00061   //
00062   // DFS mode:
00063   //
00064   // It applies a Depth-First-Search (aka backtracking) on given NFA and input
00065   // string.
00066   // At the very beginning the executor stands in the start state, then it
00067   // tries every possible state transition in current state recursively. Some
00068   // state transitions consume input string, say, a single-char-matcher or a
00069   // back-reference matcher; some don't, like assertion or other anchor nodes.
00070   // When the input is exhausted and/or the current state is an accepting
00071   // state, the whole executor returns true.
00072   //
00073   // TODO: This approach is exponentially slow for certain input.
00074   //       Try to compile the NFA to a DFA.
00075   //
00076   // Time complexity: \Omega(match_length), O(2^(_M_nfa.size()))
00077   // Space complexity: \theta(match_results.size() + match_length)
00078   //
00079   template<typename _BiIter, typename _Alloc, typename _TraitsT,
00080            bool __dfs_mode>
00081     bool _Executor<_BiIter, _Alloc, _TraitsT, __dfs_mode>::
00082     _M_main_dispatch(_Match_mode __match_mode, __dfs)
00083     {
00084       _M_has_sol = false;
00085       *_M_states._M_get_sol_pos() = _BiIter();
00086       _M_cur_results = _M_results;
00087       _M_dfs(__match_mode, _M_states._M_start);
00088       return _M_has_sol;
00089     }
00090 
00091   // ------------------------------------------------------------
00092   //
00093   // BFS mode:
00094   //
00095   // Russ Cox's article (http://swtch.com/~rsc/regexp/regexp1.html)
00096   // explained this algorithm clearly.
00097   //
00098   // It first computes epsilon closure (states that can be achieved without
00099   // consuming characters) for every state that's still matching,
00100   // using the same DFS algorithm, but doesn't re-enter states (using
00101   // _M_states._M_visited to check), nor follow _S_opcode_match.
00102   //
00103   // Then apply DFS using every _S_opcode_match (in _M_states._M_match_queue)
00104   // as the start state.
00105   //
00106   // It significantly reduces potential duplicate states, so has a better
00107   // upper bound; but it requires more overhead.
00108   //
00109   // Time complexity: \Omega(match_length * match_results.size())
00110   //                  O(match_length * _M_nfa.size() * match_results.size())
00111   // Space complexity: \Omega(_M_nfa.size() + match_results.size())
00112   //                   O(_M_nfa.size() * match_results.size())
00113   template<typename _BiIter, typename _Alloc, typename _TraitsT,
00114            bool __dfs_mode>
00115     bool _Executor<_BiIter, _Alloc, _TraitsT, __dfs_mode>::
00116     _M_main_dispatch(_Match_mode __match_mode, __bfs)
00117     {
00118       _M_states._M_queue(_M_states._M_start, _M_results);
00119       bool __ret = false;
00120       while (1)
00121         {
00122           _M_has_sol = false;
00123           if (_M_states._M_match_queue.empty())
00124             break;
00125           std::fill_n(_M_states._M_visited_states.get(), _M_nfa.size(), false);
00126           auto __old_queue = std::move(_M_states._M_match_queue);
00127           for (auto& __task : __old_queue)
00128             {
00129               _M_cur_results = std::move(__task.second);
00130               _M_dfs(__match_mode, __task.first);
00131             }
00132           if (__match_mode == _Match_mode::_Prefix)
00133             __ret |= _M_has_sol;
00134           if (_M_current == _M_end)
00135             break;
00136           ++_M_current;
00137         }
00138       if (__match_mode == _Match_mode::_Exact)
00139         __ret = _M_has_sol;
00140       _M_states._M_match_queue.clear();
00141       return __ret;
00142     }
00143 
00144   // Return whether now match the given sub-NFA.
00145   template<typename _BiIter, typename _Alloc, typename _TraitsT,
00146            bool __dfs_mode>
00147     bool _Executor<_BiIter, _Alloc, _TraitsT, __dfs_mode>::
00148     _M_lookahead(_StateIdT __next)
00149     {
00150       // Backreferences may refer to captured content.
00151       // We may want to make this faster by not copying,
00152       // but let's not be clever prematurely.
00153       _ResultsVec __what(_M_cur_results);
00154       _Executor __sub(_M_current, _M_end, __what, _M_re, _M_flags);
00155       __sub._M_states._M_start = __next;
00156       if (__sub._M_search_from_first())
00157         {
00158           for (size_t __i = 0; __i < __what.size(); __i++)
00159             if (__what[__i].matched)
00160               _M_cur_results[__i] = __what[__i];
00161           return true;
00162         }
00163       return false;
00164     }
00165 
00166   // __rep_count records how many times (__rep_count.second)
00167   // this node is visited under certain input iterator
00168   // (__rep_count.first). This prevent the executor from entering
00169   // infinite loop by refusing to continue when it's already been
00170   // visited more than twice. It's `twice` instead of `once` because
00171   // we need to spare one more time for potential group capture.
00172   template<typename _BiIter, typename _Alloc, typename _TraitsT,
00173     bool __dfs_mode>
00174     void _Executor<_BiIter, _Alloc, _TraitsT, __dfs_mode>::
00175     _M_rep_once_more(_Match_mode __match_mode, _StateIdT __i)
00176     {
00177       const auto& __state = _M_nfa[__i];
00178       auto& __rep_count = _M_rep_count[__i];
00179       if (__rep_count.second == 0 || __rep_count.first != _M_current)
00180         {
00181           auto __back = __rep_count;
00182           __rep_count.first = _M_current;
00183           __rep_count.second = 1;
00184           _M_dfs(__match_mode, __state._M_alt);
00185           __rep_count = __back;
00186         }
00187       else
00188         {
00189           if (__rep_count.second < 2)
00190             {
00191               __rep_count.second++;
00192               _M_dfs(__match_mode, __state._M_alt);
00193               __rep_count.second--;
00194             }
00195         }
00196     };
00197 
00198   // _M_alt branch is "match once more", while _M_next is "get me out
00199   // of this quantifier". Executing _M_next first or _M_alt first don't
00200   // mean the same thing, and we need to choose the correct order under
00201   // given greedy mode.
00202   template<typename _BiIter, typename _Alloc, typename _TraitsT,
00203            bool __dfs_mode>
00204     void _Executor<_BiIter, _Alloc, _TraitsT, __dfs_mode>::
00205     _M_handle_repeat(_Match_mode __match_mode, _StateIdT __i)
00206     {
00207       const auto& __state = _M_nfa[__i];
00208 
00209       // Greedy.
00210       if (!__state._M_neg)
00211         {
00212           _M_rep_once_more(__match_mode, __i);
00213           // If it's DFS executor and already accepted, we're done.
00214           if (!__dfs_mode || !_M_has_sol)
00215             _M_dfs(__match_mode, __state._M_next);
00216         }
00217       else // Non-greedy mode
00218         {
00219           if (__dfs_mode)
00220             {
00221               // vice-versa.
00222               _M_dfs(__match_mode, __state._M_next);
00223               if (!_M_has_sol)
00224                 _M_rep_once_more(__match_mode, __i);
00225             }
00226           else
00227             {
00228               // DON'T attempt anything, because there's already another
00229               // state with higher priority accepted. This state cannot
00230               // be better by attempting its next node.
00231               if (!_M_has_sol)
00232                 {
00233                   _M_dfs(__match_mode, __state._M_next);
00234                   // DON'T attempt anything if it's already accepted. An
00235                   // accepted state *must* be better than a solution that
00236                   // matches a non-greedy quantifier one more time.
00237                   if (!_M_has_sol)
00238                     _M_rep_once_more(__match_mode, __i);
00239                 }
00240             }
00241         }
00242     }
00243 
00244   template<typename _BiIter, typename _Alloc, typename _TraitsT,
00245            bool __dfs_mode>
00246     void _Executor<_BiIter, _Alloc, _TraitsT, __dfs_mode>::
00247     _M_handle_subexpr_begin(_Match_mode __match_mode, _StateIdT __i)
00248     {
00249       const auto& __state = _M_nfa[__i];
00250 
00251       auto& __res = _M_cur_results[__state._M_subexpr];
00252       auto __back = __res.first;
00253       __res.first = _M_current;
00254       _M_dfs(__match_mode, __state._M_next);
00255       __res.first = __back;
00256     }
00257 
00258   template<typename _BiIter, typename _Alloc, typename _TraitsT,
00259            bool __dfs_mode>
00260     void _Executor<_BiIter, _Alloc, _TraitsT, __dfs_mode>::
00261     _M_handle_subexpr_end(_Match_mode __match_mode, _StateIdT __i)
00262     {
00263       const auto& __state = _M_nfa[__i];
00264 
00265       auto& __res = _M_cur_results[__state._M_subexpr];
00266       auto __back = __res;
00267       __res.second = _M_current;
00268       __res.matched = true;
00269       _M_dfs(__match_mode, __state._M_next);
00270       __res = __back;
00271     }
00272 
00273   template<typename _BiIter, typename _Alloc, typename _TraitsT,
00274            bool __dfs_mode>
00275     inline void _Executor<_BiIter, _Alloc, _TraitsT, __dfs_mode>::
00276     _M_handle_line_begin_assertion(_Match_mode __match_mode, _StateIdT __i)
00277     {
00278       const auto& __state = _M_nfa[__i];
00279       if (_M_at_begin())
00280         _M_dfs(__match_mode, __state._M_next);
00281     }
00282 
00283   template<typename _BiIter, typename _Alloc, typename _TraitsT,
00284            bool __dfs_mode>
00285     inline void _Executor<_BiIter, _Alloc, _TraitsT, __dfs_mode>::
00286     _M_handle_line_end_assertion(_Match_mode __match_mode, _StateIdT __i)
00287     {
00288       const auto& __state = _M_nfa[__i];
00289       if (_M_at_end())
00290         _M_dfs(__match_mode, __state._M_next);
00291     }
00292 
00293   template<typename _BiIter, typename _Alloc, typename _TraitsT,
00294            bool __dfs_mode>
00295     inline void _Executor<_BiIter, _Alloc, _TraitsT, __dfs_mode>::
00296     _M_handle_word_boundary(_Match_mode __match_mode, _StateIdT __i)
00297     {
00298       const auto& __state = _M_nfa[__i];
00299       if (_M_word_boundary() == !__state._M_neg)
00300         _M_dfs(__match_mode, __state._M_next);
00301     }
00302 
00303   // Here __state._M_alt offers a single start node for a sub-NFA.
00304   // We recursively invoke our algorithm to match the sub-NFA.
00305   template<typename _BiIter, typename _Alloc, typename _TraitsT,
00306            bool __dfs_mode>
00307     void _Executor<_BiIter, _Alloc, _TraitsT, __dfs_mode>::
00308     _M_handle_subexpr_lookahead(_Match_mode __match_mode, _StateIdT __i)
00309     {
00310       const auto& __state = _M_nfa[__i];
00311       if (_M_lookahead(__state._M_alt) == !__state._M_neg)
00312         _M_dfs(__match_mode, __state._M_next);
00313     }
00314 
00315   template<typename _BiIter, typename _Alloc, typename _TraitsT,
00316            bool __dfs_mode>
00317     void _Executor<_BiIter, _Alloc, _TraitsT, __dfs_mode>::
00318     _M_handle_match(_Match_mode __match_mode, _StateIdT __i)
00319     {
00320       const auto& __state = _M_nfa[__i];
00321 
00322       if (_M_current == _M_end)
00323         return;
00324       if (__dfs_mode)
00325         {
00326           if (__state._M_matches(*_M_current))
00327             {
00328               ++_M_current;
00329               _M_dfs(__match_mode, __state._M_next);
00330               --_M_current;
00331             }
00332         }
00333       else
00334         if (__state._M_matches(*_M_current))
00335           _M_states._M_queue(__state._M_next, _M_cur_results);
00336     }
00337 
00338   // First fetch the matched result from _M_cur_results as __submatch;
00339   // then compare it with
00340   // (_M_current, _M_current + (__submatch.second - __submatch.first)).
00341   // If matched, keep going; else just return and try another state.
00342   template<typename _BiIter, typename _Alloc, typename _TraitsT,
00343            bool __dfs_mode>
00344     void _Executor<_BiIter, _Alloc, _TraitsT, __dfs_mode>::
00345     _M_handle_backref(_Match_mode __match_mode, _StateIdT __i)
00346     {
00347       __glibcxx_assert(__dfs_mode);
00348 
00349       const auto& __state = _M_nfa[__i];
00350       auto& __submatch = _M_cur_results[__state._M_backref_index];
00351       if (!__submatch.matched)
00352         return;
00353       auto __last = _M_current;
00354       for (auto __tmp = __submatch.first;
00355            __last != _M_end && __tmp != __submatch.second;
00356            ++__tmp)
00357         ++__last;
00358       if (_M_re._M_automaton->_M_traits.transform(__submatch.first,
00359                                                   __submatch.second)
00360           == _M_re._M_automaton->_M_traits.transform(_M_current, __last))
00361         {
00362           if (__last != _M_current)
00363             {
00364               auto __backup = _M_current;
00365               _M_current = __last;
00366               _M_dfs(__match_mode, __state._M_next);
00367               _M_current = __backup;
00368             }
00369           else
00370             _M_dfs(__match_mode, __state._M_next);
00371         }
00372     }
00373 
00374   template<typename _BiIter, typename _Alloc, typename _TraitsT,
00375            bool __dfs_mode>
00376     void _Executor<_BiIter, _Alloc, _TraitsT, __dfs_mode>::
00377     _M_handle_accept(_Match_mode __match_mode, _StateIdT __i)
00378     {
00379       if (__dfs_mode)
00380         {
00381           __glibcxx_assert(!_M_has_sol);
00382           if (__match_mode == _Match_mode::_Exact)
00383             _M_has_sol = _M_current == _M_end;
00384           else
00385             _M_has_sol = true;
00386           if (_M_current == _M_begin
00387               && (_M_flags & regex_constants::match_not_null))
00388             _M_has_sol = false;
00389           if (_M_has_sol)
00390             {
00391               if (_M_nfa._M_flags & regex_constants::ECMAScript)
00392                 _M_results = _M_cur_results;
00393               else // POSIX
00394                 {
00395                   __glibcxx_assert(_M_states._M_get_sol_pos());
00396                   // Here's POSIX's logic: match the longest one. However
00397                   // we never know which one (lhs or rhs of "|") is longer
00398                   // unless we try both of them and compare the results.
00399                   // The member variable _M_sol_pos records the end
00400                   // position of the last successful match. It's better
00401                   // to be larger, because POSIX regex is always greedy.
00402                   // TODO: This could be slow.
00403                   if (*_M_states._M_get_sol_pos() == _BiIter()
00404                       || std::distance(_M_begin,
00405                                        *_M_states._M_get_sol_pos())
00406                          < std::distance(_M_begin, _M_current))
00407                     {
00408                       *_M_states._M_get_sol_pos() = _M_current;
00409                       _M_results = _M_cur_results;
00410                     }
00411                 }
00412             }
00413         }
00414       else
00415         {
00416           if (_M_current == _M_begin
00417               && (_M_flags & regex_constants::match_not_null))
00418             return;
00419           if (__match_mode == _Match_mode::_Prefix || _M_current == _M_end)
00420             if (!_M_has_sol)
00421               {
00422                 _M_has_sol = true;
00423                 _M_results = _M_cur_results;
00424               }
00425         }
00426     }
00427 
00428   template<typename _BiIter, typename _Alloc, typename _TraitsT,
00429            bool __dfs_mode>
00430     void _Executor<_BiIter, _Alloc, _TraitsT, __dfs_mode>::
00431     _M_handle_alternative(_Match_mode __match_mode, _StateIdT __i)
00432     {
00433       const auto& __state = _M_nfa[__i];
00434 
00435       if (_M_nfa._M_flags & regex_constants::ECMAScript)
00436         {
00437           // TODO: Fix BFS support. It is wrong.
00438           _M_dfs(__match_mode, __state._M_alt);
00439           // Pick lhs if it matches. Only try rhs if it doesn't.
00440           if (!_M_has_sol)
00441             _M_dfs(__match_mode, __state._M_next);
00442         }
00443       else
00444         {
00445           // Try both and compare the result.
00446           // See "case _S_opcode_accept:" handling above.
00447           _M_dfs(__match_mode, __state._M_alt);
00448           auto __has_sol = _M_has_sol;
00449           _M_has_sol = false;
00450           _M_dfs(__match_mode, __state._M_next);
00451           _M_has_sol |= __has_sol;
00452         }
00453     }
00454 
00455   template<typename _BiIter, typename _Alloc, typename _TraitsT,
00456            bool __dfs_mode>
00457     void _Executor<_BiIter, _Alloc, _TraitsT, __dfs_mode>::
00458     _M_dfs(_Match_mode __match_mode, _StateIdT __i)
00459     {
00460       if (_M_states._M_visited(__i))
00461         return;
00462 
00463       switch (_M_nfa[__i]._M_opcode())
00464         {
00465         case _S_opcode_repeat:
00466           _M_handle_repeat(__match_mode, __i); break;
00467         case _S_opcode_subexpr_begin:
00468           _M_handle_subexpr_begin(__match_mode, __i); break;
00469         case _S_opcode_subexpr_end:
00470           _M_handle_subexpr_end(__match_mode, __i); break;
00471         case _S_opcode_line_begin_assertion:
00472           _M_handle_line_begin_assertion(__match_mode, __i); break;
00473         case _S_opcode_line_end_assertion:
00474           _M_handle_line_end_assertion(__match_mode, __i); break;
00475         case _S_opcode_word_boundary:
00476           _M_handle_word_boundary(__match_mode, __i); break;
00477         case _S_opcode_subexpr_lookahead:
00478           _M_handle_subexpr_lookahead(__match_mode, __i); break;
00479         case _S_opcode_match:
00480           _M_handle_match(__match_mode, __i); break;
00481         case _S_opcode_backref:
00482           _M_handle_backref(__match_mode, __i); break;
00483         case _S_opcode_accept:
00484           _M_handle_accept(__match_mode, __i); break;
00485         case _S_opcode_alternative:
00486           _M_handle_alternative(__match_mode, __i); break;
00487         default:
00488           __glibcxx_assert(false);
00489         }
00490     }
00491 
00492   // Return whether now is at some word boundary.
00493   template<typename _BiIter, typename _Alloc, typename _TraitsT,
00494            bool __dfs_mode>
00495     bool _Executor<_BiIter, _Alloc, _TraitsT, __dfs_mode>::
00496     _M_word_boundary() const
00497     {
00498       if (_M_current == _M_begin && (_M_flags & regex_constants::match_not_bow))
00499         return false;
00500       if (_M_current == _M_end && (_M_flags & regex_constants::match_not_eow))
00501         return false;
00502 
00503       bool __left_is_word = false;
00504       if (_M_current != _M_begin
00505           || (_M_flags & regex_constants::match_prev_avail))
00506         {
00507           auto __prev = _M_current;
00508           if (_M_is_word(*std::prev(__prev)))
00509             __left_is_word = true;
00510         }
00511       bool __right_is_word =
00512         _M_current != _M_end && _M_is_word(*_M_current);
00513 
00514       return __left_is_word != __right_is_word;
00515     }
00516 
00517 _GLIBCXX_END_NAMESPACE_VERSION
00518 } // namespace __detail
00519 } // namespace