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_compiler.tcc
27 * This is an internal header file, included by other library headers.
28 * Do not attempt to use it directly. @headername{regex}
31 // FIXME make comments doxygen format.
33 // This compiler refers to "Regular Expression Matching Can Be Simple And Fast"
34 // (http://swtch.com/~rsc/regexp/regexp1.html"),
35 // but doesn't strictly follow it.
37 // When compiling, states are *chained* instead of tree- or graph-constructed.
38 // It's more like structured programs: there's if statement and loop statement.
40 // For alternative structure (say "a|b"), aka "if statement", two branches
41 // should be constructed. However, these two shall merge to an "end_tag" at
42 // the end of this operator:
46 // => begin_tag end_tag =>
50 // This is the difference between this implementation and that in Russ's
53 // That's why we introduced dummy node here ------ "end_tag" is a dummy node.
54 // All dummy node will be eliminated at the end of compiling process.
56 namespace std _GLIBCXX_VISIBILITY(default)
60 _GLIBCXX_BEGIN_NAMESPACE_VERSION
62 template<typename _TraitsT>
64 _Compiler(_IterT __b, _IterT __e,
65 const typename _TraitsT::locale_type& __loc, _FlagT __flags)
67 & (regex_constants::ECMAScript
68 | regex_constants::basic
69 | regex_constants::extended
70 | regex_constants::grep
71 | regex_constants::egrep
72 | regex_constants::awk))
74 : __flags | regex_constants::ECMAScript),
75 _M_scanner(__b, __e, _M_flags, __loc),
76 _M_nfa(make_shared<_RegexT>(__loc, _M_flags)),
77 _M_traits(_M_nfa->_M_traits),
78 _M_ctype(std::use_facet<_CtypeT>(__loc))
80 _StateSeqT __r(*_M_nfa, _M_nfa->_M_start());
81 __r._M_append(_M_nfa->_M_insert_subexpr_begin());
82 this->_M_disjunction();
83 if (!_M_match_token(_ScannerT::_S_token_eof))
84 __throw_regex_error(regex_constants::error_paren);
85 __r._M_append(_M_pop());
86 _GLIBCXX_DEBUG_ASSERT(_M_stack.empty());
87 __r._M_append(_M_nfa->_M_insert_subexpr_end());
88 __r._M_append(_M_nfa->_M_insert_accept());
89 _M_nfa->_M_eliminate_dummy();
92 template<typename _TraitsT>
97 this->_M_alternative();
98 while (_M_match_token(_ScannerT::_S_token_or))
100 _StateSeqT __alt1 = _M_pop();
101 this->_M_alternative();
102 _StateSeqT __alt2 = _M_pop();
103 auto __end = _M_nfa->_M_insert_dummy();
104 __alt1._M_append(__end);
105 __alt2._M_append(__end);
106 // __alt2 is state._M_next, __alt1 is state._M_alt. The executor
107 // executes _M_alt before _M_next, as well as executing left
108 // alternative before right one.
109 _M_stack.push(_StateSeqT(*_M_nfa,
110 _M_nfa->_M_insert_alt(
111 __alt2._M_start, __alt1._M_start, false),
116 template<typename _TraitsT>
118 _Compiler<_TraitsT>::
123 _StateSeqT __re = _M_pop();
124 this->_M_alternative();
125 __re._M_append(_M_pop());
129 _M_stack.push(_StateSeqT(*_M_nfa, _M_nfa->_M_insert_dummy()));
132 template<typename _TraitsT>
134 _Compiler<_TraitsT>::
137 if (this->_M_assertion())
141 while (this->_M_quantifier());
147 template<typename _TraitsT>
149 _Compiler<_TraitsT>::
152 if (_M_match_token(_ScannerT::_S_token_line_begin))
153 _M_stack.push(_StateSeqT(*_M_nfa, _M_nfa->_M_insert_line_begin()));
154 else if (_M_match_token(_ScannerT::_S_token_line_end))
155 _M_stack.push(_StateSeqT(*_M_nfa, _M_nfa->_M_insert_line_end()));
156 else if (_M_match_token(_ScannerT::_S_token_word_bound))
157 // _M_value[0] == 'n' means it's negative, say "not word boundary".
158 _M_stack.push(_StateSeqT(*_M_nfa, _M_nfa->
159 _M_insert_word_bound(_M_value[0] == 'n')));
160 else if (_M_match_token(_ScannerT::_S_token_subexpr_lookahead_begin))
162 auto __neg = _M_value[0] == 'n';
163 this->_M_disjunction();
164 if (!_M_match_token(_ScannerT::_S_token_subexpr_end))
165 __throw_regex_error(regex_constants::error_paren);
166 auto __tmp = _M_pop();
167 __tmp._M_append(_M_nfa->_M_insert_accept());
171 _M_nfa->_M_insert_lookahead(__tmp._M_start, __neg)));
178 template<typename _TraitsT>
180 _Compiler<_TraitsT>::
183 bool __neg = (_M_flags & regex_constants::ECMAScript);
184 auto __init = [this, &__neg]()
186 if (_M_stack.empty())
187 __throw_regex_error(regex_constants::error_badrepeat);
188 __neg = __neg && _M_match_token(_ScannerT::_S_token_opt);
190 if (_M_match_token(_ScannerT::_S_token_closure0))
194 _StateSeqT __r(*_M_nfa,
195 _M_nfa->_M_insert_repeat(_S_invalid_state_id,
196 __e._M_start, __neg));
200 else if (_M_match_token(_ScannerT::_S_token_closure1))
204 __e._M_append(_M_nfa->_M_insert_repeat(_S_invalid_state_id,
205 __e._M_start, __neg));
208 else if (_M_match_token(_ScannerT::_S_token_opt))
212 auto __end = _M_nfa->_M_insert_dummy();
213 _StateSeqT __r(*_M_nfa,
214 _M_nfa->_M_insert_repeat(_S_invalid_state_id,
215 __e._M_start, __neg));
216 __e._M_append(__end);
217 __r._M_append(__end);
220 else if (_M_match_token(_ScannerT::_S_token_interval_begin))
222 if (_M_stack.empty())
223 __throw_regex_error(regex_constants::error_badrepeat);
224 if (!_M_match_token(_ScannerT::_S_token_dup_count))
225 __throw_regex_error(regex_constants::error_badbrace);
226 _StateSeqT __r(_M_pop());
227 _StateSeqT __e(*_M_nfa, _M_nfa->_M_insert_dummy());
228 long __min_rep = _M_cur_int_value(10);
233 if (_M_match_token(_ScannerT::_S_token_comma))
234 if (_M_match_token(_ScannerT::_S_token_dup_count)) // {3,7}
235 __n = _M_cur_int_value(10) - __min_rep;
240 if (!_M_match_token(_ScannerT::_S_token_interval_end))
241 __throw_regex_error(regex_constants::error_brace);
243 __neg = __neg && _M_match_token(_ScannerT::_S_token_opt);
245 for (long __i = 0; __i < __min_rep; ++__i)
246 __e._M_append(__r._M_clone());
250 auto __tmp = __r._M_clone();
251 _StateSeqT __s(*_M_nfa,
252 _M_nfa->_M_insert_repeat(_S_invalid_state_id,
253 __tmp._M_start, __neg));
254 __tmp._M_append(__s);
260 __throw_regex_error(regex_constants::error_badbrace);
261 auto __end = _M_nfa->_M_insert_dummy();
262 // _M_alt is the "match more" branch, and _M_next is the
263 // "match less" one. Switch _M_alt and _M_next of all created
264 // nodes. This is a hack but IMO works well.
265 std::stack<_StateIdT> __stack;
266 for (long __i = 0; __i < __n; ++__i)
268 auto __tmp = __r._M_clone();
269 auto __alt = _M_nfa->_M_insert_repeat(__tmp._M_start,
272 __e._M_append(_StateSeqT(*_M_nfa, __alt, __tmp._M_end));
274 __e._M_append(__end);
275 while (!__stack.empty())
277 auto& __tmp = (*_M_nfa)[__stack.top()];
279 std::swap(__tmp._M_next, __tmp._M_alt);
289 #define __INSERT_REGEX_MATCHER(__func, args...)\
291 if (!(_M_flags & regex_constants::icase))\
292 if (!(_M_flags & regex_constants::collate))\
293 __func<false, false>(args);\
295 __func<false, true>(args);\
297 if (!(_M_flags & regex_constants::collate))\
298 __func<true, false>(args);\
300 __func<true, true>(args);\
303 template<typename _TraitsT>
305 _Compiler<_TraitsT>::
308 if (_M_match_token(_ScannerT::_S_token_anychar))
310 if (!(_M_flags & regex_constants::ECMAScript))
311 __INSERT_REGEX_MATCHER(_M_insert_any_matcher_posix);
313 __INSERT_REGEX_MATCHER(_M_insert_any_matcher_ecma);
315 else if (_M_try_char())
316 __INSERT_REGEX_MATCHER(_M_insert_char_matcher);
317 else if (_M_match_token(_ScannerT::_S_token_backref))
318 _M_stack.push(_StateSeqT(*_M_nfa, _M_nfa->
319 _M_insert_backref(_M_cur_int_value(10))));
320 else if (_M_match_token(_ScannerT::_S_token_quoted_class))
321 __INSERT_REGEX_MATCHER(_M_insert_character_class_matcher);
322 else if (_M_match_token(_ScannerT::_S_token_subexpr_no_group_begin))
324 _StateSeqT __r(*_M_nfa, _M_nfa->_M_insert_dummy());
325 this->_M_disjunction();
326 if (!_M_match_token(_ScannerT::_S_token_subexpr_end))
327 __throw_regex_error(regex_constants::error_paren);
328 __r._M_append(_M_pop());
331 else if (_M_match_token(_ScannerT::_S_token_subexpr_begin))
333 _StateSeqT __r(*_M_nfa, _M_nfa->_M_insert_subexpr_begin());
334 this->_M_disjunction();
335 if (!_M_match_token(_ScannerT::_S_token_subexpr_end))
336 __throw_regex_error(regex_constants::error_paren);
337 __r._M_append(_M_pop());
338 __r._M_append(_M_nfa->_M_insert_subexpr_end());
341 else if (!_M_bracket_expression())
346 template<typename _TraitsT>
348 _Compiler<_TraitsT>::
349 _M_bracket_expression()
352 _M_match_token(_ScannerT::_S_token_bracket_neg_begin);
353 if (!(__neg || _M_match_token(_ScannerT::_S_token_bracket_begin)))
355 __INSERT_REGEX_MATCHER(_M_insert_bracket_matcher, __neg);
358 #undef __INSERT_REGEX_MATCHER
360 template<typename _TraitsT>
361 template<bool __icase, bool __collate>
363 _Compiler<_TraitsT>::
364 _M_insert_any_matcher_ecma()
366 _M_stack.push(_StateSeqT(*_M_nfa,
367 _M_nfa->_M_insert_matcher
368 (_AnyMatcher<_TraitsT, true, __icase, __collate>
372 template<typename _TraitsT>
373 template<bool __icase, bool __collate>
375 _Compiler<_TraitsT>::
376 _M_insert_any_matcher_posix()
378 _M_stack.push(_StateSeqT(*_M_nfa,
379 _M_nfa->_M_insert_matcher
380 (_AnyMatcher<_TraitsT, false, __icase, __collate>
384 template<typename _TraitsT>
385 template<bool __icase, bool __collate>
387 _Compiler<_TraitsT>::
388 _M_insert_char_matcher()
390 _M_stack.push(_StateSeqT(*_M_nfa,
391 _M_nfa->_M_insert_matcher
392 (_CharMatcher<_TraitsT, __icase, __collate>
393 (_M_value[0], _M_traits))));
396 template<typename _TraitsT>
397 template<bool __icase, bool __collate>
399 _Compiler<_TraitsT>::
400 _M_insert_character_class_matcher()
402 _GLIBCXX_DEBUG_ASSERT(_M_value.size() == 1);
403 _BracketMatcher<_TraitsT, __icase, __collate> __matcher
404 (_M_ctype.is(_CtypeT::upper, _M_value[0]), _M_traits);
405 __matcher._M_add_character_class(_M_value, false);
406 __matcher._M_ready();
407 _M_stack.push(_StateSeqT(*_M_nfa,
408 _M_nfa->_M_insert_matcher(std::move(__matcher))));
411 template<typename _TraitsT>
412 template<bool __icase, bool __collate>
414 _Compiler<_TraitsT>::
415 _M_insert_bracket_matcher(bool __neg)
417 _BracketMatcher<_TraitsT, __icase, __collate> __matcher(__neg, _M_traits);
418 pair<bool, _CharT> __last_char; // Optional<_CharT>
419 __last_char.first = false;
420 if (!(_M_flags & regex_constants::ECMAScript))
423 __matcher._M_add_char(_M_value[0]);
424 __last_char.first = true;
425 __last_char.second = _M_value[0];
427 while (!_M_match_token(_ScannerT::_S_token_bracket_end))
428 _M_expression_term(__last_char, __matcher);
429 __matcher._M_ready();
430 _M_stack.push(_StateSeqT(
432 _M_nfa->_M_insert_matcher(std::move(__matcher))));
435 template<typename _TraitsT>
436 template<bool __icase, bool __collate>
438 _Compiler<_TraitsT>::
439 _M_expression_term(pair<bool, _CharT>& __last_char,
440 _BracketMatcher<_TraitsT, __icase, __collate>& __matcher)
442 if (_M_match_token(_ScannerT::_S_token_collsymbol))
443 __matcher._M_add_collating_element(_M_value);
444 else if (_M_match_token(_ScannerT::_S_token_equiv_class_name))
445 __matcher._M_add_equivalence_class(_M_value);
446 else if (_M_match_token(_ScannerT::_S_token_char_class_name))
447 __matcher._M_add_character_class(_M_value, false);
448 // POSIX doesn't permit '-' as a start-range char (say [a-z--0]),
449 // except when the '-' is the first character in the bracket expression
450 // ([--0]). ECMAScript treats all '-' after a range as a normal character.
451 // Also see above, where _M_expression_term gets called.
453 // As a result, POSIX rejects [-----], but ECMAScript doesn't.
454 // Boost (1.57.0) always uses POSIX style even in its ECMAScript syntax.
455 // Clang (3.5) always uses ECMAScript style even in its POSIX syntax.
457 // It turns out that no one reads BNFs ;)
458 else if (_M_try_char())
460 if (!__last_char.first)
462 if (_M_value[0] == '-'
463 && !(_M_flags & regex_constants::ECMAScript))
464 __throw_regex_error(regex_constants::error_range);
465 __matcher._M_add_char(_M_value[0]);
466 __last_char.first = true;
467 __last_char.second = _M_value[0];
471 if (_M_value[0] == '-')
475 __matcher._M_make_range(__last_char.second , _M_value[0]);
476 __last_char.first = false;
480 if (_M_scanner._M_get_token()
481 != _ScannerT::_S_token_bracket_end)
482 __throw_regex_error(regex_constants::error_range);
483 __matcher._M_add_char(_M_value[0]);
488 __matcher._M_add_char(_M_value[0]);
489 __last_char.second = _M_value[0];
493 else if (_M_match_token(_ScannerT::_S_token_quoted_class))
494 __matcher._M_add_character_class(_M_value,
495 _M_ctype.is(_CtypeT::upper,
498 __throw_regex_error(regex_constants::error_brack);
501 template<typename _TraitsT>
503 _Compiler<_TraitsT>::
506 bool __is_char = false;
507 if (_M_match_token(_ScannerT::_S_token_oct_num))
510 _M_value.assign(1, _M_cur_int_value(8));
512 else if (_M_match_token(_ScannerT::_S_token_hex_num))
515 _M_value.assign(1, _M_cur_int_value(16));
517 else if (_M_match_token(_ScannerT::_S_token_ord_char))
522 template<typename _TraitsT>
524 _Compiler<_TraitsT>::
525 _M_match_token(_TokenT token)
527 if (token == _M_scanner._M_get_token())
529 _M_value = _M_scanner._M_get_value();
530 _M_scanner._M_advance();
536 template<typename _TraitsT>
538 _Compiler<_TraitsT>::
539 _M_cur_int_value(int __radix)
542 for (typename _StringT::size_type __i = 0;
543 __i < _M_value.length(); ++__i)
544 __v =__v * __radix + _M_traits.value(_M_value[__i], __radix);
548 template<typename _TraitsT, bool __icase, bool __collate>
550 _BracketMatcher<_TraitsT, __icase, __collate>::
551 _M_apply(_CharT __ch, false_type) const
553 bool __ret = std::binary_search(_M_char_set.begin(), _M_char_set.end(),
554 _M_translator._M_translate(__ch));
557 auto __s = _M_translator._M_transform(__ch);
558 for (auto& __it : _M_range_set)
559 if (__it.first <= __s && __s <= __it.second)
564 if (_M_traits.isctype(__ch, _M_class_set))
566 else if (std::find(_M_equiv_set.begin(), _M_equiv_set.end(),
567 _M_traits.transform_primary(&__ch, &__ch+1))
568 != _M_equiv_set.end())
572 for (auto& __it : _M_neg_class_set)
573 if (!_M_traits.isctype(__ch, __it))
580 if (_M_is_non_matching)
586 _GLIBCXX_END_NAMESPACE_VERSION
587 } // namespace __detail