libstdc++
regex_executor.tcc
Go to the documentation of this file.
00001 // class template regex -*- C++ -*-
00002 
00003 // Copyright (C) 2013-2015 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(_State<_TraitsT> __state)
00149     {
00150       _ResultsVec __what(_M_cur_results.size());
00151       _Executor __sub(_M_current, _M_end, __what, _M_re, _M_flags);
00152       __sub._M_states._M_start = __state._M_alt;
00153       if (__sub._M_search_from_first())
00154         {
00155           for (size_t __i = 0; __i < __what.size(); __i++)
00156             if (__what[__i].matched)
00157               _M_cur_results[__i] = __what[__i];
00158           return true;
00159         }
00160       return false;
00161     }
00162 
00163   // __rep_count records how many times (__rep_count.second)
00164   // this node is visited under certain input iterator
00165   // (__rep_count.first). This prevent the executor from entering
00166   // infinite loop by refusing to continue when it's already been
00167   // visited more than twice. It's `twice` instead of `once` because
00168   // we need to spare one more time for potential group capture.
00169   template<typename _BiIter, typename _Alloc, typename _TraitsT,
00170     bool __dfs_mode>
00171     void _Executor<_BiIter, _Alloc, _TraitsT, __dfs_mode>::
00172     _M_rep_once_more(_Match_mode __match_mode, _StateIdT __i)
00173     {
00174       const auto& __state = _M_nfa[__i];
00175       auto& __rep_count = _M_rep_count[__i];
00176       if (__rep_count.second == 0 || __rep_count.first != _M_current)
00177         {
00178           auto __back = __rep_count;
00179           __rep_count.first = _M_current;
00180           __rep_count.second = 1;
00181           _M_dfs(__match_mode, __state._M_alt);
00182           __rep_count = __back;
00183         }
00184       else
00185         {
00186           if (__rep_count.second < 2)
00187             {
00188               __rep_count.second++;
00189               _M_dfs(__match_mode, __state._M_alt);
00190               __rep_count.second--;
00191             }
00192         }
00193     };
00194 
00195   template<typename _BiIter, typename _Alloc, typename _TraitsT,
00196            bool __dfs_mode>
00197     void _Executor<_BiIter, _Alloc, _TraitsT, __dfs_mode>::
00198     _M_dfs(_Match_mode __match_mode, _StateIdT __i)
00199     {
00200       if (_M_states._M_visited(__i))
00201         return;
00202 
00203       const auto& __state = _M_nfa[__i];
00204       // Every change on _M_cur_results and _M_current will be rolled back after
00205       // finishing the recursion step.
00206       switch (__state._M_opcode)
00207         {
00208         // _M_alt branch is "match once more", while _M_next is "get me out
00209         // of this quantifier". Executing _M_next first or _M_alt first don't
00210         // mean the same thing, and we need to choose the correct order under
00211         // given greedy mode.
00212         case _S_opcode_repeat:
00213           {
00214             // Greedy.
00215             if (!__state._M_neg)
00216               {
00217                 _M_rep_once_more(__match_mode, __i);
00218                 // If it's DFS executor and already accepted, we're done.
00219                 if (!__dfs_mode || !_M_has_sol)
00220                   _M_dfs(__match_mode, __state._M_next);
00221               }
00222             else // Non-greedy mode
00223               {
00224                 if (__dfs_mode)
00225                   {
00226                     // vice-versa.
00227                     _M_dfs(__match_mode, __state._M_next);
00228                     if (!_M_has_sol)
00229                       _M_rep_once_more(__match_mode, __i);
00230                   }
00231                 else
00232                   {
00233                     // DON'T attempt anything, because there's already another
00234                     // state with higher priority accepted. This state cannot
00235                     // be better by attempting its next node.
00236                     if (!_M_has_sol)
00237                       {
00238                         _M_dfs(__match_mode, __state._M_next);
00239                         // DON'T attempt anything if it's already accepted. An
00240                         // accepted state *must* be better than a solution that
00241                         // matches a non-greedy quantifier one more time.
00242                         if (!_M_has_sol)
00243                           _M_rep_once_more(__match_mode, __i);
00244                       }
00245                   }
00246               }
00247             }
00248           break;
00249         case _S_opcode_subexpr_begin:
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           break;
00258         case _S_opcode_subexpr_end:
00259           {
00260             auto& __res = _M_cur_results[__state._M_subexpr];
00261             auto __back = __res;
00262             __res.second = _M_current;
00263             __res.matched = true;
00264             _M_dfs(__match_mode, __state._M_next);
00265             __res = __back;
00266           }
00267           break;
00268         case _S_opcode_line_begin_assertion:
00269           if (_M_at_begin())
00270             _M_dfs(__match_mode, __state._M_next);
00271           break;
00272         case _S_opcode_line_end_assertion:
00273           if (_M_at_end())
00274             _M_dfs(__match_mode, __state._M_next);
00275           break;
00276         case _S_opcode_word_boundary:
00277           if (_M_word_boundary() == !__state._M_neg)
00278             _M_dfs(__match_mode, __state._M_next);
00279           break;
00280         // Here __state._M_alt offers a single start node for a sub-NFA.
00281         // We recursively invoke our algorithm to match the sub-NFA.
00282         case _S_opcode_subexpr_lookahead:
00283           if (_M_lookahead(__state) == !__state._M_neg)
00284             _M_dfs(__match_mode, __state._M_next);
00285           break;
00286         case _S_opcode_match:
00287           if (_M_current == _M_end)
00288             break;
00289           if (__dfs_mode)
00290             {
00291               if (__state._M_matches(*_M_current))
00292                 {
00293                   ++_M_current;
00294                   _M_dfs(__match_mode, __state._M_next);
00295                   --_M_current;
00296                 }
00297             }
00298           else
00299             if (__state._M_matches(*_M_current))
00300               _M_states._M_queue(__state._M_next, _M_cur_results);
00301           break;
00302         // First fetch the matched result from _M_cur_results as __submatch;
00303         // then compare it with
00304         // (_M_current, _M_current + (__submatch.second - __submatch.first)).
00305         // If matched, keep going; else just return and try another state.
00306         case _S_opcode_backref:
00307           {
00308             _GLIBCXX_DEBUG_ASSERT(__dfs_mode);
00309             auto& __submatch = _M_cur_results[__state._M_backref_index];
00310             if (!__submatch.matched)
00311               break;
00312             auto __last = _M_current;
00313             for (auto __tmp = __submatch.first;
00314                  __last != _M_end && __tmp != __submatch.second;
00315                  ++__tmp)
00316               ++__last;
00317             if (_M_re._M_automaton->_M_traits.transform(__submatch.first,
00318                                                         __submatch.second)
00319                 == _M_re._M_automaton->_M_traits.transform(_M_current, __last))
00320               {
00321                 if (__last != _M_current)
00322                   {
00323                     auto __backup = _M_current;
00324                     _M_current = __last;
00325                     _M_dfs(__match_mode, __state._M_next);
00326                     _M_current = __backup;
00327                   }
00328                 else
00329                   _M_dfs(__match_mode, __state._M_next);
00330               }
00331           }
00332           break;
00333         case _S_opcode_accept:
00334           if (__dfs_mode)
00335             {
00336               _GLIBCXX_DEBUG_ASSERT(!_M_has_sol);
00337               if (__match_mode == _Match_mode::_Exact)
00338                 _M_has_sol = _M_current == _M_end;
00339               else
00340                 _M_has_sol = true;
00341               if (_M_current == _M_begin
00342                   && (_M_flags & regex_constants::match_not_null))
00343                 _M_has_sol = false;
00344               if (_M_has_sol)
00345                 {
00346                   if (_M_nfa._M_flags & regex_constants::ECMAScript)
00347                     _M_results = _M_cur_results;
00348                   else // POSIX
00349                     {
00350                       _GLIBCXX_DEBUG_ASSERT(_M_states._M_get_sol_pos());
00351                       // Here's POSIX's logic: match the longest one. However
00352                       // we never know which one (lhs or rhs of "|") is longer
00353                       // unless we try both of them and compare the results.
00354                       // The member variable _M_sol_pos records the end
00355                       // position of the last successful match. It's better
00356                       // to be larger, because POSIX regex is always greedy.
00357                       // TODO: This could be slow.
00358                       if (*_M_states._M_get_sol_pos() == _BiIter()
00359                           || std::distance(_M_begin,
00360                                            *_M_states._M_get_sol_pos())
00361                              < std::distance(_M_begin, _M_current))
00362                         {
00363                           *_M_states._M_get_sol_pos() = _M_current;
00364                           _M_results = _M_cur_results;
00365                         }
00366                     }
00367                 }
00368             }
00369           else
00370             {
00371               if (_M_current == _M_begin
00372                   && (_M_flags & regex_constants::match_not_null))
00373                 break;
00374               if (__match_mode == _Match_mode::_Prefix || _M_current == _M_end)
00375                 if (!_M_has_sol)
00376                   {
00377                     _M_has_sol = true;
00378                     _M_results = _M_cur_results;
00379                   }
00380             }
00381           break;
00382         case _S_opcode_alternative:
00383           if (_M_nfa._M_flags & regex_constants::ECMAScript)
00384             {
00385               // TODO: Let BFS support ECMAScript's alternative operation.
00386               _GLIBCXX_DEBUG_ASSERT(__dfs_mode);
00387               _M_dfs(__match_mode, __state._M_alt);
00388               // Pick lhs if it matches. Only try rhs if it doesn't.
00389               if (!_M_has_sol)
00390                 _M_dfs(__match_mode, __state._M_next);
00391             }
00392           else
00393             {
00394               // Try both and compare the result.
00395               // See "case _S_opcode_accept:" handling above.
00396               _M_dfs(__match_mode, __state._M_alt);
00397               auto __has_sol = _M_has_sol;
00398               _M_has_sol = false;
00399               _M_dfs(__match_mode, __state._M_next);
00400               _M_has_sol |= __has_sol;
00401             }
00402           break;
00403         default:
00404           _GLIBCXX_DEBUG_ASSERT(false);
00405         }
00406     }
00407 
00408   // Return whether now is at some word boundary.
00409   template<typename _BiIter, typename _Alloc, typename _TraitsT,
00410            bool __dfs_mode>
00411     bool _Executor<_BiIter, _Alloc, _TraitsT, __dfs_mode>::
00412     _M_word_boundary() const
00413     {
00414       bool __left_is_word = false;
00415       if (_M_current != _M_begin
00416           || (_M_flags & regex_constants::match_prev_avail))
00417         {
00418           auto __prev = _M_current;
00419           if (_M_is_word(*std::prev(__prev)))
00420             __left_is_word = true;
00421         }
00422       bool __right_is_word =
00423         _M_current != _M_end && _M_is_word(*_M_current);
00424 
00425       if (__left_is_word == __right_is_word)
00426         return false;
00427       if (__left_is_word && !(_M_flags & regex_constants::match_not_eow))
00428         return true;
00429       if (__right_is_word && !(_M_flags & regex_constants::match_not_bow))
00430         return true;
00431       return false;
00432     }
00433 
00434 _GLIBCXX_END_NAMESPACE_VERSION
00435 } // namespace __detail
00436 } // namespace