/* * * Copyright (c) 2004 * John Maddock * * Use, modification and distribution are subject to the * Boost Software License, Version 1.0. (See accompanying file * LICENSE_1_0.txt or copy at http://www.boost.org/LICENSE_1_0.txt) * */ /* * LOCATION: see http://www.boost.org for most recent version. * FILE basic_regex_creator.cpp * VERSION see * DESCRIPTION: Declares template class basic_regex_creator which fills in * the data members of a regex_data object. */ #ifndef BOOST_REGEX_V4_BASIC_REGEX_CREATOR_HPP #define BOOST_REGEX_V4_BASIC_REGEX_CREATOR_HPP #ifdef BOOST_MSVC #pragma warning(push) #pragma warning(disable: 4103) #endif #ifdef BOOST_HAS_ABI_HEADERS # include BOOST_ABI_PREFIX #endif #ifdef BOOST_MSVC #pragma warning(pop) #endif #ifdef BOOST_MSVC # pragma warning(push) # pragma warning(disable: 4800) #endif namespace boost{ namespace re_detail{ template struct digraph : public std::pair { digraph() : std::pair(0, 0){} digraph(charT c1) : std::pair(c1, 0){} digraph(charT c1, charT c2) : std::pair(c1, c2) {} #if !BOOST_WORKAROUND(BOOST_MSVC, < 1300) digraph(const digraph& d) : std::pair(d.first, d.second){} #endif template digraph(const Seq& s) : std::pair() { BOOST_ASSERT(s.size() <= 2); BOOST_ASSERT(s.size()); this->first = s[0]; this->second = (s.size() > 1) ? s[1] : 0; } }; template class basic_char_set { public: typedef digraph digraph_type; typedef typename traits::string_type string_type; typedef typename traits::char_class_type m_type; basic_char_set() { m_negate = false; m_has_digraphs = false; m_classes = 0; m_negated_classes = 0; m_empty = true; } void add_single(const digraph_type& s) { m_singles.insert(m_singles.end(), s); if(s.second) m_has_digraphs = true; m_empty = false; } void add_range(const digraph_type& first, const digraph_type& end) { m_ranges.insert(m_ranges.end(), first); m_ranges.insert(m_ranges.end(), end); if(first.second) { m_has_digraphs = true; add_single(first); } if(end.second) { m_has_digraphs = true; add_single(end); } m_empty = false; } void add_class(m_type m) { m_classes |= m; m_empty = false; } void add_negated_class(m_type m) { m_negated_classes |= m; m_empty = false; } void add_equivalent(const digraph_type& s) { m_equivalents.insert(m_equivalents.end(), s); if(s.second) { m_has_digraphs = true; add_single(s); } m_empty = false; } void negate() { m_negate = true; //m_empty = false; } // // accessor functions: // bool has_digraphs()const { return m_has_digraphs; } bool is_negated()const { return m_negate; } typedef typename std::vector::const_iterator list_iterator; list_iterator singles_begin()const { return m_singles.begin(); } list_iterator singles_end()const { return m_singles.end(); } list_iterator ranges_begin()const { return m_ranges.begin(); } list_iterator ranges_end()const { return m_ranges.end(); } list_iterator equivalents_begin()const { return m_equivalents.begin(); } list_iterator equivalents_end()const { return m_equivalents.end(); } m_type classes()const { return m_classes; } m_type negated_classes()const { return m_negated_classes; } bool empty()const { return m_empty; } private: std::vector m_singles; // a list of single characters to match std::vector m_ranges; // a list of end points of our ranges bool m_negate; // true if the set is to be negated bool m_has_digraphs; // true if we have digraphs present m_type m_classes; // character classes to match m_type m_negated_classes; // negated character classes to match bool m_empty; // whether we've added anything yet std::vector m_equivalents; // a list of equivalence classes }; template class basic_regex_creator { public: basic_regex_creator(regex_data* data); std::ptrdiff_t getoffset(void* addr) { return getoffset(addr, m_pdata->m_data.data()); } std::ptrdiff_t getoffset(const void* addr, const void* base) { return static_cast(addr) - static_cast(base); } re_syntax_base* getaddress(std::ptrdiff_t off) { return getaddress(off, m_pdata->m_data.data()); } re_syntax_base* getaddress(std::ptrdiff_t off, void* base) { return static_cast(static_cast(static_cast(base) + off)); } void init(unsigned l_flags) { m_pdata->m_flags = l_flags; m_icase = l_flags & regex_constants::icase; } regbase::flag_type flags() { return m_pdata->m_flags; } void flags(regbase::flag_type f) { m_pdata->m_flags = f; if(m_icase != static_cast(f & regbase::icase)) { m_icase = static_cast(f & regbase::icase); } } re_syntax_base* append_state(syntax_element_type t, std::size_t s = sizeof(re_syntax_base)); re_syntax_base* insert_state(std::ptrdiff_t pos, syntax_element_type t, std::size_t s = sizeof(re_syntax_base)); re_literal* append_literal(charT c); re_syntax_base* append_set(const basic_char_set& char_set); re_syntax_base* append_set(const basic_char_set& char_set, mpl::false_*); re_syntax_base* append_set(const basic_char_set& char_set, mpl::true_*); void finalize(const charT* p1, const charT* p2); protected: regex_data* m_pdata; // pointer to the basic_regex_data struct we are filling in const ::boost::regex_traits_wrapper& m_traits; // convenience reference to traits class re_syntax_base* m_last_state; // the last state we added bool m_icase; // true for case insensitive matches unsigned m_repeater_id; // the state_id of the next repeater bool m_has_backrefs; // true if there are actually any backrefs unsigned m_backrefs; // bitmask of permitted backrefs boost::uintmax_t m_bad_repeats; // bitmask of repeats we can't deduce a startmap for; bool m_has_recursions; // set when we have recursive expresisons to fixup std::vector m_recursion_checks; // notes which recursions we've followed while analysing this expression typename traits::char_class_type m_word_mask; // mask used to determine if a character is a word character typename traits::char_class_type m_mask_space; // mask used to determine if a character is a word character typename traits::char_class_type m_lower_mask; // mask used to determine if a character is a lowercase character typename traits::char_class_type m_upper_mask; // mask used to determine if a character is an uppercase character typename traits::char_class_type m_alpha_mask; // mask used to determine if a character is an alphabetic character private: basic_regex_creator& operator=(const basic_regex_creator&); basic_regex_creator(const basic_regex_creator&); void fixup_pointers(re_syntax_base* state); void fixup_recursions(re_syntax_base* state); void create_startmaps(re_syntax_base* state); int calculate_backstep(re_syntax_base* state); void create_startmap(re_syntax_base* state, unsigned char* l_map, unsigned int* pnull, unsigned char mask); unsigned get_restart_type(re_syntax_base* state); void set_all_masks(unsigned char* bits, unsigned char); bool is_bad_repeat(re_syntax_base* pt); void set_bad_repeat(re_syntax_base* pt); syntax_element_type get_repeat_type(re_syntax_base* state); void probe_leading_repeat(re_syntax_base* state); }; template basic_regex_creator::basic_regex_creator(regex_data* data) : m_pdata(data), m_traits(*(data->m_ptraits)), m_last_state(0), m_repeater_id(0), m_has_backrefs(false), m_backrefs(0), m_has_recursions(false) { m_pdata->m_data.clear(); m_pdata->m_status = ::boost::regex_constants::error_ok; static const charT w = 'w'; static const charT s = 's'; static const charT l[5] = { 'l', 'o', 'w', 'e', 'r', }; static const charT u[5] = { 'u', 'p', 'p', 'e', 'r', }; static const charT a[5] = { 'a', 'l', 'p', 'h', 'a', }; m_word_mask = m_traits.lookup_classname(&w, &w +1); m_mask_space = m_traits.lookup_classname(&s, &s +1); m_lower_mask = m_traits.lookup_classname(l, l + 5); m_upper_mask = m_traits.lookup_classname(u, u + 5); m_alpha_mask = m_traits.lookup_classname(a, a + 5); m_pdata->m_word_mask = m_word_mask; BOOST_ASSERT(m_word_mask != 0); BOOST_ASSERT(m_mask_space != 0); BOOST_ASSERT(m_lower_mask != 0); BOOST_ASSERT(m_upper_mask != 0); BOOST_ASSERT(m_alpha_mask != 0); } template re_syntax_base* basic_regex_creator::append_state(syntax_element_type t, std::size_t s) { // if the state is a backref then make a note of it: if(t == syntax_element_backref) this->m_has_backrefs = true; // append a new state, start by aligning our last one: m_pdata->m_data.align(); // set the offset to the next state in our last one: if(m_last_state) m_last_state->next.i = m_pdata->m_data.size() - getoffset(m_last_state); // now actually extent our data: m_last_state = static_cast(m_pdata->m_data.extend(s)); // fill in boilerplate options in the new state: m_last_state->next.i = 0; m_last_state->type = t; return m_last_state; } template re_syntax_base* basic_regex_creator::insert_state(std::ptrdiff_t pos, syntax_element_type t, std::size_t s) { // append a new state, start by aligning our last one: m_pdata->m_data.align(); // set the offset to the next state in our last one: if(m_last_state) m_last_state->next.i = m_pdata->m_data.size() - getoffset(m_last_state); // remember the last state position: std::ptrdiff_t off = getoffset(m_last_state) + s; // now actually insert our data: re_syntax_base* new_state = static_cast(m_pdata->m_data.insert(pos, s)); // fill in boilerplate options in the new state: new_state->next.i = s; new_state->type = t; m_last_state = getaddress(off); return new_state; } template re_literal* basic_regex_creator::append_literal(charT c) { re_literal* result; // start by seeing if we have an existing re_literal we can extend: if((0 == m_last_state) || (m_last_state->type != syntax_element_literal)) { // no existing re_literal, create a new one: result = static_cast(append_state(syntax_element_literal, sizeof(re_literal) + sizeof(charT))); result->length = 1; *static_cast(static_cast(result+1)) = m_traits.translate(c, m_icase); } else { // we have an existing re_literal, extend it: std::ptrdiff_t off = getoffset(m_last_state); m_pdata->m_data.extend(sizeof(charT)); m_last_state = result = static_cast(getaddress(off)); charT* characters = static_cast(static_cast(result+1)); characters[result->length] = m_traits.translate(c, m_icase); ++(result->length); } return result; } template inline re_syntax_base* basic_regex_creator::append_set( const basic_char_set& char_set) { typedef mpl::bool_< (sizeof(charT) == 1) > truth_type; return char_set.has_digraphs() ? append_set(char_set, static_cast(0)) : append_set(char_set, static_cast(0)); } template re_syntax_base* basic_regex_creator::append_set( const basic_char_set& char_set, mpl::false_*) { typedef typename traits::string_type string_type; typedef typename basic_char_set::list_iterator item_iterator; typedef typename traits::char_class_type m_type; re_set_long* result = static_cast*>(append_state(syntax_element_long_set, sizeof(re_set_long))); // // fill in the basics: // result->csingles = static_cast(::boost::re_detail::distance(char_set.singles_begin(), char_set.singles_end())); result->cranges = static_cast(::boost::re_detail::distance(char_set.ranges_begin(), char_set.ranges_end())) / 2; result->cequivalents = static_cast(::boost::re_detail::distance(char_set.equivalents_begin(), char_set.equivalents_end())); result->cclasses = char_set.classes(); result->cnclasses = char_set.negated_classes(); if(flags() & regbase::icase) { // adjust classes as needed: if(((result->cclasses & m_lower_mask) == m_lower_mask) || ((result->cclasses & m_upper_mask) == m_upper_mask)) result->cclasses |= m_alpha_mask; if(((result->cnclasses & m_lower_mask) == m_lower_mask) || ((result->cnclasses & m_upper_mask) == m_upper_mask)) result->cnclasses |= m_alpha_mask; } result->isnot = char_set.is_negated(); result->singleton = !char_set.has_digraphs(); // // remember where the state is for later: // std::ptrdiff_t offset = getoffset(result); // // now extend with all the singles: // item_iterator first, last; first = char_set.singles_begin(); last = char_set.singles_end(); while(first != last) { charT* p = static_cast(this->m_pdata->m_data.extend(sizeof(charT) * (first->second ? 3 : 2))); p[0] = m_traits.translate(first->first, m_icase); if(first->second) { p[1] = m_traits.translate(first->second, m_icase); p[2] = 0; } else p[1] = 0; ++first; } // // now extend with all the ranges: // first = char_set.ranges_begin(); last = char_set.ranges_end(); while(first != last) { // first grab the endpoints of the range: digraph c1 = *first; c1.first = this->m_traits.translate(c1.first, this->m_icase); c1.second = this->m_traits.translate(c1.second, this->m_icase); ++first; digraph c2 = *first; c2.first = this->m_traits.translate(c2.first, this->m_icase); c2.second = this->m_traits.translate(c2.second, this->m_icase); ++first; string_type s1, s2; // different actions now depending upon whether collation is turned on: if(flags() & regex_constants::collate) { // we need to transform our range into sort keys: #if BOOST_WORKAROUND(__GNUC__, < 3) string_type in(3, charT(0)); in[0] = c1.first; in[1] = c1.second; s1 = this->m_traits.transform(in.c_str(), (in[1] ? in.c_str()+2 : in.c_str()+1)); in[0] = c2.first; in[1] = c2.second; s2 = this->m_traits.transform(in.c_str(), (in[1] ? in.c_str()+2 : in.c_str()+1)); #else charT a1[3] = { c1.first, c1.second, charT(0), }; charT a2[3] = { c2.first, c2.second, charT(0), }; s1 = this->m_traits.transform(a1, (a1[1] ? a1+2 : a1+1)); s2 = this->m_traits.transform(a2, (a2[1] ? a2+2 : a2+1)); #endif if(s1.size() == 0) s1 = string_type(1, charT(0)); if(s2.size() == 0) s2 = string_type(1, charT(0)); } else { if(c1.second) { s1.insert(s1.end(), c1.first); s1.insert(s1.end(), c1.second); } else s1 = string_type(1, c1.first); if(c2.second) { s2.insert(s2.end(), c2.first); s2.insert(s2.end(), c2.second); } else s2.insert(s2.end(), c2.first); } if(s1 > s2) { // Oops error: return 0; } charT* p = static_cast(this->m_pdata->m_data.extend(sizeof(charT) * (s1.size() + s2.size() + 2) ) ); re_detail::copy(s1.begin(), s1.end(), p); p[s1.size()] = charT(0); p += s1.size() + 1; re_detail::copy(s2.begin(), s2.end(), p); p[s2.size()] = charT(0); } // // now process the equivalence classes: // first = char_set.equivalents_begin(); last = char_set.equivalents_end(); while(first != last) { string_type s; if(first->second) { #if BOOST_WORKAROUND(__GNUC__, < 3) string_type in(3, charT(0)); in[0] = first->first; in[1] = first->second; s = m_traits.transform_primary(in.c_str(), in.c_str()+2); #else charT cs[3] = { first->first, first->second, charT(0), }; s = m_traits.transform_primary(cs, cs+2); #endif } else s = m_traits.transform_primary(&first->first, &first->first+1); if(s.empty()) return 0; // invalid or unsupported equivalence class charT* p = static_cast(this->m_pdata->m_data.extend(sizeof(charT) * (s.size()+1) ) ); re_detail::copy(s.begin(), s.end(), p); p[s.size()] = charT(0); ++first; } // // finally reset the address of our last state: // m_last_state = result = static_cast*>(getaddress(offset)); return result; } template inline bool char_less(T t1, T t2) { return t1 < t2; } inline bool char_less(char t1, char t2) { return static_cast(t1) < static_cast(t2); } inline bool char_less(signed char t1, signed char t2) { return static_cast(t1) < static_cast(t2); } template re_syntax_base* basic_regex_creator::append_set( const basic_char_set& char_set, mpl::true_*) { typedef typename traits::string_type string_type; typedef typename basic_char_set::list_iterator item_iterator; re_set* result = static_cast(append_state(syntax_element_set, sizeof(re_set))); bool negate = char_set.is_negated(); std::memset(result->_map, 0, sizeof(result->_map)); // // handle singles first: // item_iterator first, last; first = char_set.singles_begin(); last = char_set.singles_end(); while(first != last) { for(unsigned int i = 0; i < (1 << CHAR_BIT); ++i) { if(this->m_traits.translate(static_cast(i), this->m_icase) == this->m_traits.translate(first->first, this->m_icase)) result->_map[i] = true; } ++first; } // // OK now handle ranges: // first = char_set.ranges_begin(); last = char_set.ranges_end(); while(first != last) { // first grab the endpoints of the range: charT c1 = this->m_traits.translate(first->first, this->m_icase); ++first; charT c2 = this->m_traits.translate(first->first, this->m_icase); ++first; // different actions now depending upon whether collation is turned on: if(flags() & regex_constants::collate) { // we need to transform our range into sort keys: charT c3[2] = { c1, charT(0), }; string_type s1 = this->m_traits.transform(c3, c3+1); c3[0] = c2; string_type s2 = this->m_traits.transform(c3, c3+1); if(s1 > s2) { // Oops error: return 0; } BOOST_ASSERT(c3[1] == charT(0)); for(unsigned i = 0; i < (1u << CHAR_BIT); ++i) { c3[0] = static_cast(i); string_type s3 = this->m_traits.transform(c3, c3 +1); if((s1 <= s3) && (s3 <= s2)) result->_map[i] = true; } } else { if(char_less(c2, c1)) { // Oops error: return 0; } // everything in range matches: std::memset(result->_map + static_cast(c1), true, 1 + static_cast(c2) - static_cast(c1)); } } // // and now the classes: // typedef typename traits::char_class_type m_type; m_type m = char_set.classes(); if(flags() & regbase::icase) { // adjust m as needed: if(((m & m_lower_mask) == m_lower_mask) || ((m & m_upper_mask) == m_upper_mask)) m |= m_alpha_mask; } if(m != 0) { for(unsigned i = 0; i < (1u << CHAR_BIT); ++i) { if(this->m_traits.isctype(static_cast(i), m)) result->_map[i] = true; } } // // and now the negated classes: // m = char_set.negated_classes(); if(flags() & regbase::icase) { // adjust m as needed: if(((m & m_lower_mask) == m_lower_mask) || ((m & m_upper_mask) == m_upper_mask)) m |= m_alpha_mask; } if(m != 0) { for(unsigned i = 0; i < (1u << CHAR_BIT); ++i) { if(0 == this->m_traits.isctype(static_cast(i), m)) result->_map[i] = true; } } // // now process the equivalence classes: // first = char_set.equivalents_begin(); last = char_set.equivalents_end(); while(first != last) { string_type s; BOOST_ASSERT(static_cast(0) == first->second); s = m_traits.transform_primary(&first->first, &first->first+1); if(s.empty()) return 0; // invalid or unsupported equivalence class for(unsigned i = 0; i < (1u << CHAR_BIT); ++i) { charT c[2] = { (static_cast(i)), charT(0), }; string_type s2 = this->m_traits.transform_primary(c, c+1); if(s == s2) result->_map[i] = true; } ++first; } if(negate) { for(unsigned i = 0; i < (1u << CHAR_BIT); ++i) { result->_map[i] = !(result->_map[i]); } } return result; } template void basic_regex_creator::finalize(const charT* p1, const charT* p2) { if(this->m_pdata->m_status) return; // we've added all the states we need, now finish things off. // start by adding a terminating state: append_state(syntax_element_match); // extend storage to store original expression: std::ptrdiff_t len = p2 - p1; m_pdata->m_expression_len = len; charT* ps = static_cast(m_pdata->m_data.extend(sizeof(charT) * (1 + (p2 - p1)))); m_pdata->m_expression = ps; re_detail::copy(p1, p2, ps); ps[p2 - p1] = 0; // fill in our other data... // successful parsing implies a zero status: m_pdata->m_status = 0; // get the first state of the machine: m_pdata->m_first_state = static_cast(m_pdata->m_data.data()); // fixup pointers in the machine: fixup_pointers(m_pdata->m_first_state); if(m_has_recursions) { m_pdata->m_has_recursions = true; fixup_recursions(m_pdata->m_first_state); if(this->m_pdata->m_status) return; } else m_pdata->m_has_recursions = false; // create nested startmaps: create_startmaps(m_pdata->m_first_state); // create main startmap: std::memset(m_pdata->m_startmap, 0, sizeof(m_pdata->m_startmap)); m_pdata->m_can_be_null = 0; m_bad_repeats = 0; if(m_has_recursions) m_recursion_checks.assign(1 + m_pdata->m_mark_count, false); create_startmap(m_pdata->m_first_state, m_pdata->m_startmap, &(m_pdata->m_can_be_null), mask_all); // get the restart type: m_pdata->m_restart_type = get_restart_type(m_pdata->m_first_state); // optimise a leading repeat if there is one: probe_leading_repeat(m_pdata->m_first_state); } template void basic_regex_creator::fixup_pointers(re_syntax_base* state) { while(state) { switch(state->type) { case syntax_element_recurse: m_has_recursions = true; if(state->next.i) state->next.p = getaddress(state->next.i, state); else state->next.p = 0; break; case syntax_element_rep: case syntax_element_dot_rep: case syntax_element_char_rep: case syntax_element_short_set_rep: case syntax_element_long_set_rep: // set the state_id of this repeat: static_cast(state)->state_id = m_repeater_id++; BOOST_FALLTHROUGH; case syntax_element_alt: std::memset(static_cast(state)->_map, 0, sizeof(static_cast(state)->_map)); static_cast(state)->can_be_null = 0; BOOST_FALLTHROUGH; case syntax_element_jump: static_cast(state)->alt.p = getaddress(static_cast(state)->alt.i, state); BOOST_FALLTHROUGH; default: if(state->next.i) state->next.p = getaddress(state->next.i, state); else state->next.p = 0; } state = state->next.p; } } template void basic_regex_creator::fixup_recursions(re_syntax_base* state) { re_syntax_base* base = state; while(state) { switch(state->type) { case syntax_element_assert_backref: { // just check that the index is valid: int idx = static_cast(state)->index; if(idx < 0) { idx = -idx-1; if(idx >= 10000) { idx = m_pdata->get_id(idx); if(idx <= 0) { // check of sub-expression that doesn't exist: if(0 == this->m_pdata->m_status) // update the error code if not already set this->m_pdata->m_status = boost::regex_constants::error_bad_pattern; // // clear the expression, we should be empty: // this->m_pdata->m_expression = 0; this->m_pdata->m_expression_len = 0; // // and throw if required: // if(0 == (this->flags() & regex_constants::no_except)) { std::string message = "Encountered a forward reference to a marked sub-expression that does not exist."; boost::regex_error e(message, boost::regex_constants::error_bad_pattern, 0); e.raise(); } } } } } break; case syntax_element_recurse: { bool ok = false; re_syntax_base* p = base; std::ptrdiff_t idx = static_cast(state)->alt.i; if(idx > 10000) { // // There may be more than one capture group with this hash, just do what Perl // does and recurse to the leftmost: // idx = m_pdata->get_id(static_cast(idx)); } while(p) { if((p->type == syntax_element_startmark) && (static_cast(p)->index == idx)) { // // We've found the target of the recursion, set the jump target: // static_cast(state)->alt.p = p; ok = true; // // Now scan the target for nested repeats: // p = p->next.p; int next_rep_id = 0; while(p) { switch(p->type) { case syntax_element_rep: case syntax_element_dot_rep: case syntax_element_char_rep: case syntax_element_short_set_rep: case syntax_element_long_set_rep: next_rep_id = static_cast(p)->state_id; break; case syntax_element_endmark: if(static_cast(p)->index == idx) next_rep_id = -1; break; default: break; } if(next_rep_id) break; p = p->next.p; } if(next_rep_id > 0) { static_cast(state)->state_id = next_rep_id - 1; } break; } p = p->next.p; } if(!ok) { // recursion to sub-expression that doesn't exist: if(0 == this->m_pdata->m_status) // update the error code if not already set this->m_pdata->m_status = boost::regex_constants::error_bad_pattern; // // clear the expression, we should be empty: // this->m_pdata->m_expression = 0; this->m_pdata->m_expression_len = 0; // // and throw if required: // if(0 == (this->flags() & regex_constants::no_except)) { std::string message = "Encountered a forward reference to a recursive sub-expression that does not exist."; boost::regex_error e(message, boost::regex_constants::error_bad_pattern, 0); e.raise(); } } } break; default: break; } state = state->next.p; } } template void basic_regex_creator::create_startmaps(re_syntax_base* state) { // non-recursive implementation: // create the last map in the machine first, so that earlier maps // can make use of the result... // // This was originally a recursive implementation, but that caused stack // overflows with complex expressions on small stacks (think COM+). // start by saving the case setting: bool l_icase = m_icase; std::vector > v; while(state) { switch(state->type) { case syntax_element_toggle_case: // we need to track case changes here: m_icase = static_cast(state)->icase; state = state->next.p; continue; case syntax_element_alt: case syntax_element_rep: case syntax_element_dot_rep: case syntax_element_char_rep: case syntax_element_short_set_rep: case syntax_element_long_set_rep: // just push the state onto our stack for now: v.push_back(std::pair(m_icase, state)); state = state->next.p; break; case syntax_element_backstep: // we need to calculate how big the backstep is: static_cast(state)->index = this->calculate_backstep(state->next.p); if(static_cast(state)->index < 0) { // Oops error: if(0 == this->m_pdata->m_status) // update the error code if not already set this->m_pdata->m_status = boost::regex_constants::error_bad_pattern; // // clear the expression, we should be empty: // this->m_pdata->m_expression = 0; this->m_pdata->m_expression_len = 0; // // and throw if required: // if(0 == (this->flags() & regex_constants::no_except)) { std::string message = "Invalid lookbehind assertion encountered in the regular expression."; boost::regex_error e(message, boost::regex_constants::error_bad_pattern, 0); e.raise(); } } BOOST_FALLTHROUGH; default: state = state->next.p; } } // now work through our list, building all the maps as we go: while(v.size()) { // Initialize m_recursion_checks if we need it: if(m_has_recursions) m_recursion_checks.assign(1 + m_pdata->m_mark_count, false); const std::pair& p = v.back(); m_icase = p.first; state = p.second; v.pop_back(); // Build maps: m_bad_repeats = 0; create_startmap(state->next.p, static_cast(state)->_map, &static_cast(state)->can_be_null, mask_take); m_bad_repeats = 0; if(m_has_recursions) m_recursion_checks.assign(1 + m_pdata->m_mark_count, false); create_startmap(static_cast(state)->alt.p, static_cast(state)->_map, &static_cast(state)->can_be_null, mask_skip); // adjust the type of the state to allow for faster matching: state->type = this->get_repeat_type(state); } // restore case sensitivity: m_icase = l_icase; } template int basic_regex_creator::calculate_backstep(re_syntax_base* state) { typedef typename traits::char_class_type m_type; int result = 0; while(state) { switch(state->type) { case syntax_element_startmark: if((static_cast(state)->index == -1) || (static_cast(state)->index == -2)) { state = static_cast(state->next.p)->alt.p->next.p; continue; } else if(static_cast(state)->index == -3) { state = state->next.p->next.p; continue; } break; case syntax_element_endmark: if((static_cast(state)->index == -1) || (static_cast(state)->index == -2)) return result; break; case syntax_element_literal: result += static_cast(state)->length; break; case syntax_element_wild: case syntax_element_set: result += 1; break; case syntax_element_dot_rep: case syntax_element_char_rep: case syntax_element_short_set_rep: case syntax_element_backref: case syntax_element_rep: case syntax_element_combining: case syntax_element_long_set_rep: case syntax_element_backstep: { re_repeat* rep = static_cast(state); // adjust the type of the state to allow for faster matching: state->type = this->get_repeat_type(state); if((state->type == syntax_element_dot_rep) || (state->type == syntax_element_char_rep) || (state->type == syntax_element_short_set_rep)) { if(rep->max != rep->min) return -1; result += static_cast(rep->min); state = rep->alt.p; continue; } else if(state->type == syntax_element_long_set_rep) { BOOST_ASSERT(rep->next.p->type == syntax_element_long_set); if(static_cast*>(rep->next.p)->singleton == 0) return -1; if(rep->max != rep->min) return -1; result += static_cast(rep->min); state = rep->alt.p; continue; } } return -1; case syntax_element_long_set: if(static_cast*>(state)->singleton == 0) return -1; result += 1; break; case syntax_element_jump: state = static_cast(state)->alt.p; continue; case syntax_element_alt: { int r1 = calculate_backstep(state->next.p); int r2 = calculate_backstep(static_cast(state)->alt.p); if((r1 < 0) || (r1 != r2)) return -1; return result + r1; } default: break; } state = state->next.p; } return -1; } template void basic_regex_creator::create_startmap(re_syntax_base* state, unsigned char* l_map, unsigned int* pnull, unsigned char mask) { int not_last_jump = 1; re_syntax_base* recursion_start = 0; int recursion_sub = 0; re_syntax_base* recursion_restart = 0; // track case sensitivity: bool l_icase = m_icase; while(state) { switch(state->type) { case syntax_element_toggle_case: l_icase = static_cast(state)->icase; state = state->next.p; break; case syntax_element_literal: { // don't set anything in *pnull, set each element in l_map // that could match the first character in the literal: if(l_map) { l_map[0] |= mask_init; charT first_char = *static_cast(static_cast(static_cast(state) + 1)); for(unsigned int i = 0; i < (1u << CHAR_BIT); ++i) { if(m_traits.translate(static_cast(i), l_icase) == first_char) l_map[i] |= mask; } } return; } case syntax_element_end_line: { // next character must be a line separator (if there is one): if(l_map) { l_map[0] |= mask_init; l_map[static_cast('\n')] |= mask; l_map[static_cast('\r')] |= mask; l_map[static_cast('\f')] |= mask; l_map[0x85] |= mask; } // now figure out if we can match a NULL string at this point: if(pnull) create_startmap(state->next.p, 0, pnull, mask); return; } case syntax_element_recurse: { if(state->type == syntax_element_startmark) recursion_sub = static_cast(state)->index; else recursion_sub = 0; if(m_recursion_checks[recursion_sub]) { // Infinite recursion!! if(0 == this->m_pdata->m_status) // update the error code if not already set this->m_pdata->m_status = boost::regex_constants::error_bad_pattern; // // clear the expression, we should be empty: // this->m_pdata->m_expression = 0; this->m_pdata->m_expression_len = 0; // // and throw if required: // if(0 == (this->flags() & regex_constants::no_except)) { std::string message = "Encountered an infinite recursion."; boost::regex_error e(message, boost::regex_constants::error_bad_pattern, 0); e.raise(); } } else if(recursion_start == 0) { recursion_start = state; recursion_restart = state->next.p; state = static_cast(state)->alt.p; m_recursion_checks[recursion_sub] = true; break; } m_recursion_checks[recursion_sub] = true; // can't handle nested recursion here... BOOST_FALLTHROUGH; } case syntax_element_backref: // can be null, and any character can match: if(pnull) *pnull |= mask; BOOST_FALLTHROUGH; case syntax_element_wild: { // can't be null, any character can match: set_all_masks(l_map, mask); return; } case syntax_element_match: { // must be null, any character can match: set_all_masks(l_map, mask); if(pnull) *pnull |= mask; return; } case syntax_element_word_start: { // recurse, then AND with all the word characters: create_startmap(state->next.p, l_map, pnull, mask); if(l_map) { l_map[0] |= mask_init; for(unsigned int i = 0; i < (1u << CHAR_BIT); ++i) { if(!m_traits.isctype(static_cast(i), m_word_mask)) l_map[i] &= static_cast(~mask); } } return; } case syntax_element_word_end: { // recurse, then AND with all the word characters: create_startmap(state->next.p, l_map, pnull, mask); if(l_map) { l_map[0] |= mask_init; for(unsigned int i = 0; i < (1u << CHAR_BIT); ++i) { if(m_traits.isctype(static_cast(i), m_word_mask)) l_map[i] &= static_cast(~mask); } } return; } case syntax_element_buffer_end: { // we *must be null* : if(pnull) *pnull |= mask; return; } case syntax_element_long_set: if(l_map) { typedef typename traits::char_class_type m_type; if(static_cast*>(state)->singleton) { l_map[0] |= mask_init; for(unsigned int i = 0; i < (1u << CHAR_BIT); ++i) { charT c = static_cast(i); if(&c != re_is_set_member(&c, &c + 1, static_cast*>(state), *m_pdata, l_icase)) l_map[i] |= mask; } } else set_all_masks(l_map, mask); } return; case syntax_element_set: if(l_map) { l_map[0] |= mask_init; for(unsigned int i = 0; i < (1u << CHAR_BIT); ++i) { if(static_cast(state)->_map[ static_cast(m_traits.translate(static_cast(i), l_icase))]) l_map[i] |= mask; } } return; case syntax_element_jump: // take the jump: state = static_cast(state)->alt.p; not_last_jump = -1; break; case syntax_element_alt: case syntax_element_rep: case syntax_element_dot_rep: case syntax_element_char_rep: case syntax_element_short_set_rep: case syntax_element_long_set_rep: { re_alt* rep = static_cast(state); if(rep->_map[0] & mask_init) { if(l_map) { // copy previous results: l_map[0] |= mask_init; for(unsigned int i = 0; i <= UCHAR_MAX; ++i) { if(rep->_map[i] & mask_any) l_map[i] |= mask; } } if(pnull) { if(rep->can_be_null & mask_any) *pnull |= mask; } } else { // we haven't created a startmap for this alternative yet // so take the union of the two options: if(is_bad_repeat(state)) { set_all_masks(l_map, mask); if(pnull) *pnull |= mask; return; } set_bad_repeat(state); create_startmap(state->next.p, l_map, pnull, mask); if((state->type == syntax_element_alt) || (static_cast(state)->min == 0) || (not_last_jump == 0)) create_startmap(rep->alt.p, l_map, pnull, mask); } } return; case syntax_element_soft_buffer_end: // match newline or null: if(l_map) { l_map[0] |= mask_init; l_map[static_cast('\n')] |= mask; l_map[static_cast('\r')] |= mask; } if(pnull) *pnull |= mask; return; case syntax_element_endmark: // need to handle independent subs as a special case: if(static_cast(state)->index < 0) { // can be null, any character can match: set_all_masks(l_map, mask); if(pnull) *pnull |= mask; return; } else if(recursion_start && (recursion_sub != 0) && (recursion_sub == static_cast(state)->index)) { // recursion termination: recursion_start = 0; state = recursion_restart; break; } // // Normally we just go to the next state... but if this sub-expression is // the target of a recursion, then we might be ending a recursion, in which // case we should check whatever follows that recursion, as well as whatever // follows this state: // if(m_pdata->m_has_recursions && static_cast(state)->index) { bool ok = false; re_syntax_base* p = m_pdata->m_first_state; while(p) { if(p->type == syntax_element_recurse) { re_brace* p2 = static_cast(static_cast(p)->alt.p); if((p2->type == syntax_element_startmark) && (p2->index == static_cast(state)->index)) { ok = true; break; } } p = p->next.p; } if(ok) { create_startmap(p->next.p, l_map, pnull, mask); } } state = state->next.p; break; case syntax_element_startmark: // need to handle independent subs as a special case: if(static_cast(state)->index == -3) { state = state->next.p->next.p; break; } BOOST_FALLTHROUGH; default: state = state->next.p; } ++not_last_jump; } } template unsigned basic_regex_creator::get_restart_type(re_syntax_base* state) { // // find out how the machine starts, so we can optimise the search: // while(state) { switch(state->type) { case syntax_element_startmark: case syntax_element_endmark: state = state->next.p; continue; case syntax_element_start_line: return regbase::restart_line; case syntax_element_word_start: return regbase::restart_word; case syntax_element_buffer_start: return regbase::restart_buf; case syntax_element_restart_continue: return regbase::restart_continue; default: state = 0; continue; } } return regbase::restart_any; } template void basic_regex_creator::set_all_masks(unsigned char* bits, unsigned char mask) { // // set mask in all of bits elements, // if bits[0] has mask_init not set then we can // optimise this to a call to memset: // if(bits) { if(bits[0] == 0) (std::memset)(bits, mask, 1u << CHAR_BIT); else { for(unsigned i = 0; i < (1u << CHAR_BIT); ++i) bits[i] |= mask; } bits[0] |= mask_init; } } template bool basic_regex_creator::is_bad_repeat(re_syntax_base* pt) { switch(pt->type) { case syntax_element_rep: case syntax_element_dot_rep: case syntax_element_char_rep: case syntax_element_short_set_rep: case syntax_element_long_set_rep: { unsigned state_id = static_cast(pt)->state_id; if(state_id > sizeof(m_bad_repeats) * CHAR_BIT) return true; // run out of bits, assume we can't traverse this one. static const boost::uintmax_t one = 1uL; return m_bad_repeats & (one << state_id); } default: return false; } } template void basic_regex_creator::set_bad_repeat(re_syntax_base* pt) { switch(pt->type) { case syntax_element_rep: case syntax_element_dot_rep: case syntax_element_char_rep: case syntax_element_short_set_rep: case syntax_element_long_set_rep: { unsigned state_id = static_cast(pt)->state_id; static const boost::uintmax_t one = 1uL; if(state_id <= sizeof(m_bad_repeats) * CHAR_BIT) m_bad_repeats |= (one << state_id); } break; default: break; } } template syntax_element_type basic_regex_creator::get_repeat_type(re_syntax_base* state) { typedef typename traits::char_class_type m_type; if(state->type == syntax_element_rep) { // check to see if we are repeating a single state: if(state->next.p->next.p->next.p == static_cast(state)->alt.p) { switch(state->next.p->type) { case re_detail::syntax_element_wild: return re_detail::syntax_element_dot_rep; case re_detail::syntax_element_literal: return re_detail::syntax_element_char_rep; case re_detail::syntax_element_set: return re_detail::syntax_element_short_set_rep; case re_detail::syntax_element_long_set: if(static_cast*>(state->next.p)->singleton) return re_detail::syntax_element_long_set_rep; break; default: break; } } } return state->type; } template void basic_regex_creator::probe_leading_repeat(re_syntax_base* state) { // enumerate our states, and see if we have a leading repeat // for which failed search restarts can be optimised; do { switch(state->type) { case syntax_element_startmark: if(static_cast(state)->index >= 0) { state = state->next.p; continue; } if((static_cast(state)->index == -1) || (static_cast(state)->index == -2)) { // skip past the zero width assertion: state = static_cast(state->next.p)->alt.p->next.p; continue; } if(static_cast(state)->index == -3) { // Have to skip the leading jump state: state = state->next.p->next.p; continue; } return; case syntax_element_endmark: case syntax_element_start_line: case syntax_element_end_line: case syntax_element_word_boundary: case syntax_element_within_word: case syntax_element_word_start: case syntax_element_word_end: case syntax_element_buffer_start: case syntax_element_buffer_end: case syntax_element_restart_continue: state = state->next.p; break; case syntax_element_dot_rep: case syntax_element_char_rep: case syntax_element_short_set_rep: case syntax_element_long_set_rep: if(this->m_has_backrefs == 0) static_cast(state)->leading = true; BOOST_FALLTHROUGH; default: return; } }while(state); } } // namespace re_detail } // namespace boost #ifdef BOOST_MSVC # pragma warning(pop) #endif #ifdef BOOST_MSVC #pragma warning(push) #pragma warning(disable: 4103) #endif #ifdef BOOST_HAS_ABI_HEADERS # include BOOST_ABI_SUFFIX #endif #ifdef BOOST_MSVC #pragma warning(pop) #endif #endif