Boost C++ Libraries

...one of the most highly regarded and expertly designed C++ library projects in the world. Herb Sutter and Andrei Alexandrescu, C++ Coding Standards

This is the documentation for an old version of Boost. Click here to view this page for the latest version.

boost/regex/v4/basic_regex_parser.hpp

/*
 *
 * 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_parser.cpp
  *   VERSION      see <boost/version.hpp>
  *   DESCRIPTION: Declares template class basic_regex_parser.
  */

#ifndef BOOST_REGEX_V4_BASIC_REGEX_PARSER_HPP
#define BOOST_REGEX_V4_BASIC_REGEX_PARSER_HPP

#ifdef BOOST_HAS_ABI_HEADERS
#  include BOOST_ABI_PREFIX
#endif

namespace boost{
namespace re_detail{

#ifdef BOOST_MSVC
#pragma warning(push)
#pragma warning(disable:4244)
#endif

template <class charT, class traits>
class basic_regex_parser : public basic_regex_creator<charT, traits>
{
public:
   basic_regex_parser(regex_data<charT, traits>* data);
   void parse(const charT* p1, const charT* p2, unsigned flags);
   void fail(regex_constants::error_type error_code, std::ptrdiff_t position);

   bool parse_all();
   bool parse_basic();
   bool parse_extended();
   bool parse_literal();
   bool parse_open_paren();
   bool parse_basic_escape();
   bool parse_extended_escape();
   bool parse_match_any();
   bool parse_repeat(std::size_t low = 0, std::size_t high = (std::numeric_limits<std::size_t>::max)());
   bool parse_repeat_range(bool isbasic);
   bool parse_alt();
   bool parse_set();
   bool parse_backref();
   void parse_set_literal(basic_char_set<charT, traits>& char_set);
   bool parse_inner_set(basic_char_set<charT, traits>& char_set);
   bool parse_QE();
   bool parse_perl_extension();
   bool add_emacs_code(bool negate);
   bool unwind_alts(std::ptrdiff_t last_paren_start);
   digraph<charT> get_next_set_literal(basic_char_set<charT, traits>& char_set);
   charT unescape_character();
   regex_constants::syntax_option_type parse_options();

private:
   typedef bool (basic_regex_parser::*parser_proc_type)();
   typedef typename traits::string_type string_type;
   typedef typename traits::char_class_type char_class_type;
   parser_proc_type           m_parser_proc;    // the main parser to use
   const charT*               m_base;           // the start of the string being parsed
   const charT*               m_end;            // the end of the string being parsed
   const charT*               m_position;       // our current parser position
   unsigned                   m_mark_count;     // how many sub-expressions we have
   std::ptrdiff_t             m_paren_start;    // where the last seen ')' began (where repeats are inserted).
   std::ptrdiff_t             m_alt_insert_point; // where to insert the next alternative
   bool                       m_has_case_change; // true if somewhere in the current block the case has changed
#if defined(BOOST_MSVC) && defined(_M_IX86)
   // This is an ugly warning suppression workaround (for warnings *inside* std::vector
   // that can not otherwise be suppressed)...
   BOOST_STATIC_ASSERT(sizeof(long) >= sizeof(void*));
   std::vector<long>           m_alt_jumps;      // list of alternative in the current scope.
#else
   std::vector<std::ptrdiff_t> m_alt_jumps;      // list of alternative in the current scope.
#endif

   basic_regex_parser& operator=(const basic_regex_parser&);
   basic_regex_parser(const basic_regex_parser&);
};

template <class charT, class traits>
basic_regex_parser<charT, traits>::basic_regex_parser(regex_data<charT, traits>* data)
   : basic_regex_creator<charT, traits>(data), m_mark_count(0), m_paren_start(0), m_alt_insert_point(0), m_has_case_change(false)
{
}

template <class charT, class traits>
void basic_regex_parser<charT, traits>::parse(const charT* p1, const charT* p2, unsigned l_flags)
{
   // pass l_flags on to base class:
   this->init(l_flags);
   // set up pointers:
   m_position = m_base = p1;
   m_end = p2;
   // empty strings are errors:
   if(p1 == p2)
   {
      fail(regex_constants::error_empty, 0);
      return;
   }
   // select which parser to use:
   switch(l_flags & regbase::main_option_type)
   {
   case regbase::perl_syntax_group:
      m_parser_proc = &basic_regex_parser<charT, traits>::parse_extended;
      break;
   case regbase::basic_syntax_group:
      m_parser_proc = &basic_regex_parser<charT, traits>::parse_basic;
      break;
   case regbase::literal:
      m_parser_proc = &basic_regex_parser<charT, traits>::parse_literal;
      break;
   }

   // parse all our characters:
   bool result = parse_all();
   //
   // Unwind our alternatives:
   //
   unwind_alts(-1);
   // reset l_flags as a global scope (?imsx) may have altered them:
   this->flags(l_flags);
   // if we haven't gobbled up all the characters then we must
   // have had an unexpected ')' :
   if(!result)
   {
      fail(regex_constants::error_paren, ::boost::re_detail::distance(m_base, m_position));
      return;
   }
   // if an error has been set then give up now:
   if(this->m_pdata->m_status)
      return;
   // fill in our sub-expression count:
   this->m_pdata->m_mark_count = 1 + m_mark_count;
   this->finalize(p1, p2);
}

template <class charT, class traits>
void basic_regex_parser<charT, traits>::fail(regex_constants::error_type error_code, std::ptrdiff_t position)
{
   if(0 == this->m_pdata->m_status) // update the error code if not already set
      this->m_pdata->m_status = error_code;
   m_position = m_end; // don't bother parsing anything else
   // get the error message:
   std::string message = this->m_pdata->m_ptraits->error_string(error_code);
   // and raise the exception, this will do nothing if exceptions are disabled:
#ifndef BOOST_NO_EXCEPTIONS
   if(0 == (this->flags() & regex_constants::no_except))
   {
      boost::regex_error e(message, error_code, position);
      e.raise();
   }
#else
   (void)position; // suppress warnings.
#endif
}

template <class charT, class traits>
bool basic_regex_parser<charT, traits>::parse_all()
{
   bool result = true;
   while(result && (m_position != m_end))
   {
      result = (this->*m_parser_proc)();
   }
   return result;
}

#ifdef BOOST_MSVC
#pragma warning(push)
#pragma warning(disable:4702)
#endif
template <class charT, class traits>
bool basic_regex_parser<charT, traits>::parse_basic()
{
   switch(this->m_traits.syntax_type(*m_position))
   {
   case regex_constants::syntax_escape:
      return parse_basic_escape();
   case regex_constants::syntax_dot:
      return parse_match_any();
   case regex_constants::syntax_caret:
      ++m_position;
      this->append_state(syntax_element_start_line);
      break;
   case regex_constants::syntax_dollar:
      ++m_position;
      this->append_state(syntax_element_end_line);
      break;
   case regex_constants::syntax_star:
      if(!(this->m_last_state) || (this->m_last_state->type == syntax_element_start_line))
         return parse_literal();
      else
      {
         ++m_position;
         return parse_repeat();
      }
   case regex_constants::syntax_plus:
      if(!(this->m_last_state) || (this->m_last_state->type == syntax_element_start_line) || !(this->flags() & regbase::emacs_ex))
         return parse_literal();
      else
      {
         ++m_position;
         return parse_repeat(1);
      }
   case regex_constants::syntax_question:
      if(!(this->m_last_state) || (this->m_last_state->type == syntax_element_start_line) || !(this->flags() & regbase::emacs_ex))
         return parse_literal();
      else
      {
         ++m_position;
         return parse_repeat(0, 1);
      }
   case regex_constants::syntax_open_set:
      return parse_set();
   case regex_constants::syntax_newline:
      if(this->flags() & regbase::newline_alt)
         return parse_alt();
      else
         return parse_literal();
   default:
      return parse_literal();
   }
   return true;
}

template <class charT, class traits>
bool basic_regex_parser<charT, traits>::parse_extended()
{
   bool result = true;
   switch(this->m_traits.syntax_type(*m_position))
   {
   case regex_constants::syntax_open_mark:
      return parse_open_paren();
   case regex_constants::syntax_close_mark:
      return false;
   case regex_constants::syntax_escape:
      return parse_extended_escape();
   case regex_constants::syntax_dot:
      return parse_match_any();
   case regex_constants::syntax_caret:
      ++m_position;
      this->append_state(
         (this->flags() & regex_constants::no_mod_m ? syntax_element_buffer_start : syntax_element_start_line));
      break;
   case regex_constants::syntax_dollar:
      ++m_position;
      this->append_state(
         (this->flags() & regex_constants::no_mod_m ? syntax_element_buffer_end : syntax_element_end_line));
      break;
   case regex_constants::syntax_star:
      if(m_position == this->m_base)
      {
         fail(regex_constants::error_badrepeat, 0);
         return false;
      }
      ++m_position;
      return parse_repeat();
   case regex_constants::syntax_question:
      if(m_position == this->m_base)
      {
         fail(regex_constants::error_badrepeat, 0);
         return false;
      }
      ++m_position;
      return parse_repeat(0,1);
   case regex_constants::syntax_plus:
      if(m_position == this->m_base)
      {
         fail(regex_constants::error_badrepeat, 0);
         return false;
      }
      ++m_position;
      return parse_repeat(1);
   case regex_constants::syntax_open_brace:
      ++m_position;
      return parse_repeat_range(false);
   case regex_constants::syntax_close_brace:
      fail(regex_constants::error_brace, this->m_position - this->m_end);
      return false;
   case regex_constants::syntax_or:
      return parse_alt();
   case regex_constants::syntax_open_set:
      return parse_set();
   case regex_constants::syntax_newline:
      if(this->flags() & regbase::newline_alt)
         return parse_alt();
      else
         return parse_literal();
   case regex_constants::syntax_hash:
      //
      // If we have a mod_x flag set, then skip until
      // we get to a newline character:
      //
      if((this->flags() 
         & (regbase::no_perl_ex|regbase::mod_x))
         == regbase::mod_x)
      {
         while((m_position != m_end) && !is_separator(*m_position++)){}
         return true;
      }
      // Otherwise fall through:
   default:
      result = parse_literal();
      break;
   }
   return result;
}
#ifdef BOOST_MSVC
#pragma warning(pop)
#endif

template <class charT, class traits>
bool basic_regex_parser<charT, traits>::parse_literal()
{
   // append this as a literal provided it's not a space character
   // or the perl option regbase::mod_x is not set:
   if(
      ((this->flags() 
         & (regbase::main_option_type|regbase::mod_x|regbase::no_perl_ex)) 
            != regbase::mod_x)
      || !this->m_traits.isctype(*m_position, this->m_mask_space))
         this->append_literal(*m_position);
   ++m_position;
   return true;
}

template <class charT, class traits>
bool basic_regex_parser<charT, traits>::parse_open_paren()
{
   //
   // skip the '(' and error check:
   //
   if(++m_position == m_end)
   {
      fail(regex_constants::error_paren, m_position - m_base);
      return false;
   }
   //
   // begin by checking for a perl-style (?...) extension:
   //
   if(
         ((this->flags() & (regbase::main_option_type | regbase::no_perl_ex)) == 0)
         || ((this->flags() & (regbase::main_option_type | regbase::emacs_ex)) == (regbase::basic_syntax_group|regbase::emacs_ex))
     )
   {
      if(this->m_traits.syntax_type(*m_position) == regex_constants::syntax_question)
         return parse_perl_extension();
   }
   //
   // update our mark count, and append the required state:
   //
   unsigned markid = 0;
   if(0 == (this->flags() & regbase::nosubs))
      markid = ++m_mark_count;
   re_brace* pb = static_cast<re_brace*>(this->append_state(syntax_element_startmark, sizeof(re_brace)));
   pb->index = markid;
   std::ptrdiff_t last_paren_start = this->getoffset(pb);
   // back up insertion point for alternations, and set new point:
   std::ptrdiff_t last_alt_point = m_alt_insert_point;
   this->m_pdata->m_data.align();
   m_alt_insert_point = this->m_pdata->m_data.size();
   //
   // back up the current flags in case we have a nested (?imsx) group:
   //
   regex_constants::syntax_option_type opts = this->flags();
   bool old_case_change = m_has_case_change;
   m_has_case_change = false; // no changes to this scope as yet...
   //
   // now recursively add more states, this will terminate when we get to a
   // matching ')' :
   //
   parse_all();
   //
   // Unwind pushed alternatives:
   //
   if(0 == unwind_alts(last_paren_start))
      return false;
   //
   // restore flags:
   //
   if(m_has_case_change)
   {
      // the case has changed in one or more of the alternatives
      // within the scoped (...) block: we have to add a state
      // to reset the case sensitivity:
      static_cast<re_case*>(
         this->append_state(syntax_element_toggle_case, sizeof(re_case))
         )->icase = opts & regbase::icase;
   }
   this->flags(opts);
   m_has_case_change = old_case_change;
   //
   // we either have a ')' or we have run out of characters prematurely:
   //
   if(m_position == m_end)
   {
      this->fail(regex_constants::error_paren, ::boost::re_detail::distance(m_base, m_end));
      return false;
   }
   BOOST_ASSERT(this->m_traits.syntax_type(*m_position) == regex_constants::syntax_close_mark);
   ++m_position;
   //
   // append closing parenthesis state:
   //
   pb = static_cast<re_brace*>(this->append_state(syntax_element_endmark, sizeof(re_brace)));
   pb->index = markid;
   this->m_paren_start = last_paren_start;
   //
   // restore the alternate insertion point:
   //
   this->m_alt_insert_point = last_alt_point;
   //
   // allow backrefs to this mark:
   //
   if((markid > 0) && (markid < sizeof(unsigned) * CHAR_BIT))
      this->m_backrefs |= 1u << (markid - 1);

   return true;
}

template <class charT, class traits>
bool basic_regex_parser<charT, traits>::parse_basic_escape()
{
   ++m_position;
   bool result = true;
   switch(this->m_traits.escape_syntax_type(*m_position))
   {
   case regex_constants::syntax_open_mark:
      return parse_open_paren();
   case regex_constants::syntax_close_mark:
      return false;
   case regex_constants::syntax_plus:
      if(this->flags() & regex_constants::bk_plus_qm)
      {
         ++m_position;
         return parse_repeat(1);
      }
      else
         return parse_literal();
   case regex_constants::syntax_question:
      if(this->flags() & regex_constants::bk_plus_qm)
      {
         ++m_position;
         return parse_repeat(0, 1);
      }
      else
         return parse_literal();
   case regex_constants::syntax_open_brace:
      if(this->flags() & regbase::no_intervals)
         return parse_literal();
      ++m_position;
      return parse_repeat_range(true);
   case regex_constants::syntax_close_brace:
      if(this->flags() & regbase::no_intervals)
         return parse_literal();
      fail(regex_constants::error_brace, this->m_position - this->m_base);
      return false;
   case regex_constants::syntax_or:
      if(this->flags() & regbase::bk_vbar)
         return parse_alt();
      else
         result = parse_literal();
      break;
   case regex_constants::syntax_digit:
      return parse_backref();
   case regex_constants::escape_type_start_buffer:
      if(this->flags() & regbase::emacs_ex)
      {
         ++m_position;
         this->append_state(syntax_element_buffer_start);
      }
      else
         result = parse_literal();
      break;
   case regex_constants::escape_type_end_buffer:
      if(this->flags() & regbase::emacs_ex)
      {
         ++m_position;
         this->append_state(syntax_element_buffer_end);
      }
      else
         result = parse_literal();
      break;
   case regex_constants::escape_type_word_assert:
      if(this->flags() & regbase::emacs_ex)
      {
         ++m_position;
         this->append_state(syntax_element_word_boundary);
      }
      else
         result = parse_literal();
      break;
   case regex_constants::escape_type_not_word_assert:
      if(this->flags() & regbase::emacs_ex)
      {
         ++m_position;
         this->append_state(syntax_element_within_word);
      }
      else
         result = parse_literal();
      break;
   case regex_constants::escape_type_left_word:
      if(this->flags() & regbase::emacs_ex)
      {
         ++m_position;
         this->append_state(syntax_element_word_start);
      }
      else
         result = parse_literal();
      break;
   case regex_constants::escape_type_right_word:
      if(this->flags() & regbase::emacs_ex)
      {
         ++m_position;
         this->append_state(syntax_element_word_end);
      }
      else
         result = parse_literal();
      break;
   default:
      if(this->flags() & regbase::emacs_ex)
      {
         bool negate = true;
         switch(*m_position)
         {
         case 'w':
            negate = false;
            // fall through:
         case 'W':
            {
            basic_char_set<charT, traits> char_set;
            if(negate)
               char_set.negate();
            char_set.add_class(this->m_word_mask);
            if(0 == this->append_set(char_set))
            {
               fail(regex_constants::error_ctype, m_position - m_base);
               return false;
            }
            ++m_position;
            return true;
            }
         case 's':
            negate = false;
            // fall through:
         case 'S':
            return add_emacs_code(negate);
         case 'c':
         case 'C':
            // not supported yet:
            fail(regex_constants::error_escape, m_position - m_base);
            return false;
         default:
            break;
         }
      }
      result = parse_literal();
      break;
   }
   return result;
}

template <class charT, class traits>
bool basic_regex_parser<charT, traits>::parse_extended_escape()
{
   ++m_position;
   bool negate = false; // in case this is a character class escape: \w \d etc
   switch(this->m_traits.escape_syntax_type(*m_position))
   {
   case regex_constants::escape_type_not_class:
      negate = true;
      // fall through:
   case regex_constants::escape_type_class:
      {
         typedef typename traits::char_class_type mask_type;
         mask_type m = this->m_traits.lookup_classname(m_position, m_position+1);
         if(m != 0)
         {
            basic_char_set<charT, traits> char_set;
            if(negate)
               char_set.negate();
            char_set.add_class(m);
            if(0 == this->append_set(char_set))
            {
               fail(regex_constants::error_ctype, m_position - m_base);
               return false;
            }
            ++m_position;
            return true;
         }
         //
         // not a class, just a regular unknown escape:
         //
         this->append_literal(unescape_character());
         break;
      }
   case regex_constants::syntax_digit:
      return parse_backref();
   case regex_constants::escape_type_left_word:
      ++m_position;
      this->append_state(syntax_element_word_start);
      break;
   case regex_constants::escape_type_right_word:
      ++m_position;
      this->append_state(syntax_element_word_end);
      break;
   case regex_constants::escape_type_start_buffer:
      ++m_position;
      this->append_state(syntax_element_buffer_start);
      break;
   case regex_constants::escape_type_end_buffer:
      ++m_position;
      this->append_state(syntax_element_buffer_end);
      break;
   case regex_constants::escape_type_word_assert:
      ++m_position;
      this->append_state(syntax_element_word_boundary);
      break;
   case regex_constants::escape_type_not_word_assert:
      ++m_position;
      this->append_state(syntax_element_within_word);
      break;
   case regex_constants::escape_type_Z:
      ++m_position;
      this->append_state(syntax_element_soft_buffer_end);
      break;
   case regex_constants::escape_type_Q:
      return parse_QE();
   case regex_constants::escape_type_C:
      return parse_match_any();
   case regex_constants::escape_type_X:
      ++m_position;
      this->append_state(syntax_element_combining);
      break;
   case regex_constants::escape_type_G:
      ++m_position;
      this->append_state(syntax_element_restart_continue);
      break;
   case regex_constants::escape_type_not_property:
      negate = true;
      // fall through:
   case regex_constants::escape_type_property:
      {
         ++m_position;
         char_class_type m;
         if(m_position == m_end)
         {
            fail(regex_constants::error_escape, m_position - m_base);
            return false;
         }
         // maybe have \p{ddd}
         if(this->m_traits.syntax_type(*m_position) == regex_constants::syntax_open_brace)
         {
            const charT* base = m_position;
            // skip forward until we find enclosing brace:
            while((m_position != m_end) && (this->m_traits.syntax_type(*m_position) != regex_constants::syntax_close_brace))
               ++m_position;
            if(m_position == m_end)
            {
               fail(regex_constants::error_escape, m_position - m_base);
               return false;
            }
            m = this->m_traits.lookup_classname(++base, m_position++);
         }
         else
         {
            m = this->m_traits.lookup_classname(m_position, m_position+1);
            ++m_position;
         }
         if(m != 0)
         {
            basic_char_set<charT, traits> char_set;
            if(negate)
               char_set.negate();
            char_set.add_class(m);
            if(0 == this->append_set(char_set))
            {
               fail(regex_constants::error_ctype, m_position - m_base);
               return false;
            }
            return true;
         }
         fail(regex_constants::error_ctype, m_position - m_base);
      }
   default:
      this->append_literal(unescape_character());
      break;
   }
   return true;
}

template <class charT, class traits>
bool basic_regex_parser<charT, traits>::parse_match_any()
{
   //
   // we have a '.' that can match any character:
   //
   ++m_position;
   static_cast<re_dot*>(
      this->append_state(syntax_element_wild, sizeof(re_dot))
      )->mask = static_cast<unsigned char>(this->flags() & regbase::no_mod_s 
      ? re_detail::force_not_newline 
         : this->flags() & regbase::mod_s ?
            re_detail::force_newline : re_detail::dont_care);
   return true;
}

template <class charT, class traits>
bool basic_regex_parser<charT, traits>::parse_repeat(std::size_t low, std::size_t high)
{
   bool greedy = true;
   std::size_t insert_point;
   // 
   // when we get to here we may have a non-greedy ? mark still to come:
   //
   if((m_position != m_end) 
      && (
            (0 == (this->flags() & (regbase::main_option_type | regbase::no_perl_ex)))
            || ((regbase::basic_syntax_group|regbase::emacs_ex) == (this->flags() & (regbase::main_option_type | regbase::emacs_ex)))
         )
      )
   {
      // OK we have a perl regex, check for a '?':
      if(this->m_traits.syntax_type(*m_position) == regex_constants::syntax_question)
      {
         greedy = false;
         ++m_position;
      }
   }
   if(0 == this->m_last_state)
   {
      fail(regex_constants::error_badrepeat, ::boost::re_detail::distance(m_base, m_position));
      return false;
   }
   if(this->m_last_state->type == syntax_element_endmark)
   {
      // insert a repeat before the '(' matching the last ')':
      insert_point = this->m_paren_start;
   }
   else if((this->m_last_state->type == syntax_element_literal) && (static_cast<re_literal*>(this->m_last_state)->length > 1))
   {
      // the last state was a literal with more than one character, split it in two:
      re_literal* lit = static_cast<re_literal*>(this->m_last_state);
      charT c = (static_cast<charT*>(static_cast<void*>(lit+1)))[lit->length - 1];
      --(lit->length);
      // now append new state:
      lit = static_cast<re_literal*>(this->append_state(syntax_element_literal, sizeof(re_literal) + sizeof(charT)));
      lit->length = 1;
      (static_cast<charT*>(static_cast<void*>(lit+1)))[0] = c;
      insert_point = this->getoffset(this->m_last_state);
   }
   else
   {
      // repeat the last state whatever it was, need to add some error checking here:
      switch(this->m_last_state->type)
      {
      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_alt:
      case syntax_element_soft_buffer_end:
      case syntax_element_restart_continue:
      case syntax_element_jump:
      case syntax_element_startmark:
         // can't legally repeat any of the above:
         fail(regex_constants::error_badrepeat, m_position - m_base);
         return false;
      default:
         // do nothing...
         break;
      }
      insert_point = this->getoffset(this->m_last_state);
   }
   //
   // OK we now know what to repeat, so insert the repeat around it:
   //
   re_repeat* rep = static_cast<re_repeat*>(this->insert_state(insert_point, syntax_element_rep, re_repeater_size));
   rep->min = low;
   rep->max = high;
   rep->greedy = greedy;
   rep->leading = false;
   // store our repeater position for later:
   std::ptrdiff_t rep_off = this->getoffset(rep);
   // and append a back jump to the repeat:
   re_jump* jmp = static_cast<re_jump*>(this->append_state(syntax_element_jump, sizeof(re_jump)));
   jmp->alt.i = rep_off - this->getoffset(jmp);
   this->m_pdata->m_data.align();
   // now fill in the alt jump for the repeat:
   rep = static_cast<re_repeat*>(this->getaddress(rep_off));
   rep->alt.i = this->m_pdata->m_data.size() - rep_off;
   return true;
}

template <class charT, class traits>
bool basic_regex_parser<charT, traits>::parse_repeat_range(bool isbasic)
{
   //
   // parse a repeat-range:
   //
   std::size_t min, max;
   int v;
   // skip whitespace:
   while((m_position != m_end) && this->m_traits.isctype(*m_position, this->m_mask_space))
      ++m_position;
   // fail if at end:
   if(this->m_position == this->m_end)
   {
      fail(regex_constants::error_brace, this->m_position - this->m_base);
      return false;
   }
   // get min:
   v = this->m_traits.toi(m_position, m_end, 10);
   // skip whitespace:
   while((m_position != m_end) && this->m_traits.isctype(*m_position, this->m_mask_space))
      ++m_position;
   if(v < 0)
   {
      fail(regex_constants::error_badbrace, this->m_position - this->m_base);
      return false;
   }
   else if(this->m_position == this->m_end)
   {
      fail(regex_constants::error_brace, this->m_position - this->m_base);
      return false;
   }
   min = v;
   // see if we have a comma:
   if(this->m_traits.syntax_type(*m_position) == regex_constants::syntax_comma)
   {
      // move on and error check:
      ++m_position;
      // skip whitespace:
      while((m_position != m_end) && this->m_traits.isctype(*m_position, this->m_mask_space))
         ++m_position;
      if(this->m_position == this->m_end)
      {
         fail(regex_constants::error_brace, this->m_position - this->m_base);
         return false;
      }
      // get the value if any:
      v = this->m_traits.toi(m_position, m_end, 10);
      max = (v >= 0) ? v : (std::numeric_limits<std::size_t>::max)();
   }
   else
   {
      // no comma, max = min:
      max = min;
   }
   // skip whitespace:
   while((m_position != m_end) && this->m_traits.isctype(*m_position, this->m_mask_space))
      ++m_position;
   // OK now check trailing }:
   if(this->m_position == this->m_end)
   {
      fail(regex_constants::error_brace, this->m_position - this->m_base);
      return false;
   }
   if(isbasic)
   {
      if(this->m_traits.syntax_type(*m_position) == regex_constants::syntax_escape)
      {
         ++m_position;
         if(this->m_position == this->m_end)
         {
            fail(regex_constants::error_brace, this->m_position - this->m_base);
            return false;
         }
      }
      else
      {
         fail(regex_constants::error_badbrace, this->m_position - this->m_base);
         return false;
      }
   }
   if(this->m_traits.syntax_type(*m_position) == regex_constants::syntax_close_brace)
      ++m_position;
   else
   {
      fail(regex_constants::error_badbrace, this->m_position - this->m_base);
      return false;
   }
   //
   // finally go and add the repeat, unless error:
   //
   if(min > max)
   {
      fail(regex_constants::error_badbrace, this->m_position - this->m_base);
      return false;
   }
   return parse_repeat(min, max);
}

template <class charT, class traits>
bool basic_regex_parser<charT, traits>::parse_alt()
{
   //
   // error check: if there have been no previous states,
   // or if the last state was a '(' then error:
   //
   if((this->m_last_state == 0) || (this->m_last_state->type == syntax_element_startmark))
   {
      fail(regex_constants::error_empty, this->m_position - this->m_base);
      return false;
   }
   ++m_position;
   //
   // we need to append a trailing jump: 
   //
   re_syntax_base* pj = this->append_state(re_detail::syntax_element_jump, sizeof(re_jump));
   std::ptrdiff_t jump_offset = this->getoffset(pj);
   //
   // now insert the alternative:
   //
   re_alt* palt = static_cast<re_alt*>(this->insert_state(this->m_alt_insert_point, syntax_element_alt, re_alt_size));
   jump_offset += re_alt_size;
   this->m_pdata->m_data.align();
   palt->alt.i = this->m_pdata->m_data.size() - this->getoffset(palt);
   //
   // update m_alt_insert_point so that the next alternate gets
   // inserted at the start of the second of the two we've just created:
   //
   this->m_alt_insert_point = this->m_pdata->m_data.size();
   //
   // the start of this alternative must have a case changes state
   // if the current block has messed around with case changes:
   //
   if(m_has_case_change)
   {
      static_cast<re_case*>(
         this->append_state(syntax_element_toggle_case, sizeof(re_case))
         )->icase = this->m_icase;
   }
   //
   // push the alternative onto our stack, a recursive
   // implementation here is easier to understand (and faster
   // as it happens), but causes all kinds of stack overflow problems
   // on programs with small stacks (COM+).
   //
   m_alt_jumps.push_back(jump_offset);
   return true;
}

template <class charT, class traits>
bool basic_regex_parser<charT, traits>::parse_set()
{
   ++m_position;
   if(m_position == m_end)
   {
      fail(regex_constants::error_brack, m_position - m_base);
      return false;
   }
   basic_char_set<charT, traits> char_set;

   const charT* base = m_position;  // where the '[' was
   const charT* item_base = m_position;  // where the '[' or '^' was

   while(m_position != m_end)
   {
      switch(this->m_traits.syntax_type(*m_position))
      {
      case regex_constants::syntax_caret:
         if(m_position == base)
         {
            char_set.negate();
            ++m_position;
            item_base = m_position;
         }
         else
            parse_set_literal(char_set);
         break;
      case regex_constants::syntax_close_set:
         if(m_position == item_base)
         {
            parse_set_literal(char_set);
            break;
         }
         else
         {
            ++m_position;
            if(0 == this->append_set(char_set))
            {
               fail(regex_constants::error_range, m_position - m_base);
               return false;
            }
         }
         return true;
      case regex_constants::syntax_open_set:
         if(parse_inner_set(char_set))
            break;
         return true;
      case regex_constants::syntax_escape:
         {
            // 
            // look ahead and see if this is a character class shortcut
            // \d \w \s etc...
            //
            ++m_position;
            if(this->m_traits.escape_syntax_type(*m_position)
               == regex_constants::escape_type_class)
            {
               char_class_type m = this->m_traits.lookup_classname(m_position, m_position+1);
               if(m != 0)
               {
                  char_set.add_class(m);
                  ++m_position;
                  break;
               }
            }
            else if(this->m_traits.escape_syntax_type(*m_position)
               == regex_constants::escape_type_not_class)
            {
               // negated character class:
               char_class_type m = this->m_traits.lookup_classname(m_position, m_position+1);
               if(m != 0)
               {
                  char_set.add_negated_class(m);
                  ++m_position;
                  break;
               }
            }
            // not a character class, just a regular escape:
            --m_position;
            parse_set_literal(char_set);
            break;
         }
      default:
         parse_set_literal(char_set);
         break;
      }
   }
   return m_position != m_end;
}

template <class charT, class traits>
bool basic_regex_parser<charT, traits>::parse_inner_set(basic_char_set<charT, traits>& char_set)
{
   //
   // we have either a character class [:name:]
   // a collating element [.name.]
   // or an equivalence class [=name=]
   //
   if(m_end == ++m_position)
   {
      fail(regex_constants::error_brack, m_position - m_base);
      return false;
   }
   switch(this->m_traits.syntax_type(*m_position))
   {
   case regex_constants::syntax_dot:
      //
      // a collating element is treated as a literal:
      //
      --m_position;
      parse_set_literal(char_set);
      return true;
   case regex_constants::syntax_colon:
      {
      // check that character classes are actually enabled:
      if((this->flags() & (regbase::main_option_type | regbase::no_char_classes)) 
         == (regbase::basic_syntax_group  | regbase::no_char_classes))
      {
         --m_position;
         parse_set_literal(char_set);
         return true;
      }
      // skip the ':'
      if(m_end == ++m_position)
      {
         fail(regex_constants::error_brack, m_position - m_base);
         return false;
      }
      const charT* name_first = m_position;
      // skip at least one character, then find the matching ':]'
      if(m_end == ++m_position)
      {
         fail(regex_constants::error_brack, m_position - m_base);
         return false;
      }
      while((m_position != m_end) 
         && (this->m_traits.syntax_type(*m_position) != regex_constants::syntax_colon)) 
         ++m_position;
      const charT* name_last = m_position;
      if(m_end == m_position)
      {
         fail(regex_constants::error_brack, m_position - m_base);
         return false;
      }
      if((m_end == ++m_position) 
         || (this->m_traits.syntax_type(*m_position) != regex_constants::syntax_close_set))
      {
         fail(regex_constants::error_brack, m_position - m_base);
         return false;
      }
      //
      // check for negated class:
      //
      bool negated = false;
      if(this->m_traits.syntax_type(*name_first) == regex_constants::syntax_caret)
      {
         ++name_first;
         negated = true;
      }
      typedef typename traits::char_class_type mask_type;
      mask_type m = this->m_traits.lookup_classname(name_first, name_last);
      if(m == 0)
      {
         if(char_set.empty() && (name_last - name_first == 1))
         {
            // maybe a special case:
            ++m_position;
            if( (m_position != m_end) 
               && (this->m_traits.syntax_type(*m_position) 
                  == regex_constants::syntax_close_set))
            {
               if(this->m_traits.escape_syntax_type(*name_first) 
                  == regex_constants::escape_type_left_word)
               {
                  ++m_position;
                  this->append_state(syntax_element_word_start);
                  return false;
               }
               if(this->m_traits.escape_syntax_type(*name_first) 
                  == regex_constants::escape_type_right_word)
               {
                  ++m_position;
                  this->append_state(syntax_element_word_end);
                  return false;
               }
            }
         }
         fail(regex_constants::error_ctype, name_first - m_base);
         return false;
      }
      if(negated == false)
         char_set.add_class(m);
      else
         char_set.add_negated_class(m);
      ++m_position;
      break;
   }
   case regex_constants::syntax_equal:
      {
      // skip the '='
      if(m_end == ++m_position)
      {
         fail(regex_constants::error_brack, m_position - m_base);
         return false;
      }
      const charT* name_first = m_position;
      // skip at least one character, then find the matching '=]'
      if(m_end == ++m_position)
      {
         fail(regex_constants::error_brack, m_position - m_base);
         return false;
      }
      while((m_position != m_end) 
         && (this->m_traits.syntax_type(*m_position) != regex_constants::syntax_equal)) 
         ++m_position;
      const charT* name_last = m_position;
      if(m_end == m_position)
      {
         fail(regex_constants::error_brack, m_position - m_base);
         return false;
      }
      if((m_end == ++m_position) 
         || (this->m_traits.syntax_type(*m_position) != regex_constants::syntax_close_set))
      {
         fail(regex_constants::error_brack, m_position - m_base);
         return false;
      }
      string_type m = this->m_traits.lookup_collatename(name_first, name_last);
      if((0 == m.size()) || (m.size() > 2))
      {
         fail(regex_constants::error_collate, name_first - m_base);
         return false;
      }
      digraph<charT> d;
      d.first = m[0];
      if(m.size() > 1)
         d.second = m[1];
      else
         d.second = 0;
      char_set.add_equivalent(d);
      ++m_position;
      break;
   }
   default:
      --m_position;
      parse_set_literal(char_set);
      break;
   }
   return true;
}

template <class charT, class traits>
void basic_regex_parser<charT, traits>::parse_set_literal(basic_char_set<charT, traits>& char_set)
{
   digraph<charT> start_range(get_next_set_literal(char_set));
   if(m_end == m_position)
   {
      fail(regex_constants::error_brack, m_position - m_base);
      return;
   }
   if(this->m_traits.syntax_type(*m_position) == regex_constants::syntax_dash)
   {
      // we have a range:
      if(m_end == ++m_position)
      {
         fail(regex_constants::error_brack, m_position - m_base);
         return;
      }
      if(this->m_traits.syntax_type(*m_position) != regex_constants::syntax_close_set)
      {
         digraph<charT> end_range = get_next_set_literal(char_set);
         char_set.add_range(start_range, end_range);
         if(this->m_traits.syntax_type(*m_position) == regex_constants::syntax_dash)
         {
            if(m_end == ++m_position)
            {
               fail(regex_constants::error_brack, m_position - m_base);
               return;
            }
            if(this->m_traits.syntax_type(*m_position) == regex_constants::syntax_close_set)
            {
               // trailing - :
               --m_position;
               return;
            }
            fail(regex_constants::error_range, m_position - m_base);
            return;
         }
         return;
      }
      --m_position;
   }
   char_set.add_single(start_range);
}

template <class charT, class traits>
digraph<charT> basic_regex_parser<charT, traits>::get_next_set_literal(basic_char_set<charT, traits>& char_set)
{
   digraph<charT> result;
   switch(this->m_traits.syntax_type(*m_position))
   {
   case regex_constants::syntax_dash:
      if(!char_set.empty())
      {
         // see if we are at the end of the set:
         if((++m_position == m_end) || (this->m_traits.syntax_type(*m_position) != regex_constants::syntax_close_set))
         {
            fail(regex_constants::error_range, m_position - m_base);
            return result;
         }
         --m_position;
      }
      result.first = *m_position++;
      return result;
   case regex_constants::syntax_escape:
      // check to see if escapes are supported first:
      if(this->flags() & regex_constants::no_escape_in_lists)
      {
         result = *m_position++;
         break;
      }
      ++m_position;
      result = unescape_character();
      break;
   case regex_constants::syntax_open_set:
   {
      if(m_end == ++m_position)
      {
         fail(regex_constants::error_collate, m_position - m_base);
         return result;
      }
      if(this->m_traits.syntax_type(*m_position) != regex_constants::syntax_dot)
      {
         --m_position;
         result.first = *m_position;
         ++m_position;
         return result;
      }
      if(m_end == ++m_position)
      {
         fail(regex_constants::error_collate, m_position - m_base);
         return result;
      }
      const charT* name_first = m_position;
      // skip at least one character, then find the matching ':]'
      if(m_end == ++m_position)
      {
         fail(regex_constants::error_collate, name_first - m_base);
         return result;
      }
      while((m_position != m_end) 
         && (this->m_traits.syntax_type(*m_position) != regex_constants::syntax_dot)) 
         ++m_position;
      const charT* name_last = m_position;
      if(m_end == m_position)
      {
         fail(regex_constants::error_collate, name_first - m_base);
         return result;
      }
      if((m_end == ++m_position) 
         || (this->m_traits.syntax_type(*m_position) != regex_constants::syntax_close_set))
      {
         fail(regex_constants::error_collate, name_first - m_base);
         return result;
      }
      ++m_position;
      string_type s = this->m_traits.lookup_collatename(name_first, name_last);
      if(s.empty() || (s.size() > 2))
      {
         fail(regex_constants::error_collate, name_first - m_base);
         return result;
      }
      result.first = s[0];
      if(s.size() > 1)
         result.second = s[1];
      else
         result.second = 0;
      return result;
   }
   default:
      result = *m_position++;
   }
   return result;
}

//
// does a value fit in the specified charT type?
//
template <class charT>
bool valid_value(charT, int v, const mpl::true_&)
{
   return (v >> (sizeof(charT) * CHAR_BIT)) == 0;
}
template <class charT>
bool valid_value(charT, int, const mpl::false_&)
{
   return true; // v will alsways fit in a charT
}
template <class charT>
bool valid_value(charT c, int v)
{
   return valid_value(c, v, mpl::bool_<(sizeof(charT) < sizeof(int))>());
}

template <class charT, class traits>
charT basic_regex_parser<charT, traits>::unescape_character()
{
#ifdef BOOST_MSVC
#pragma warning(push)
#pragma warning(disable:4127)
#endif
   charT result(0);
   if(m_position == m_end)
   {
      fail(regex_constants::error_escape, m_position - m_base);
      return false;
   }
   switch(this->m_traits.escape_syntax_type(*m_position))
   {
   case regex_constants::escape_type_control_a:
      result = charT('\a');
      break;
   case regex_constants::escape_type_e:
      result = charT(27);
      break;
   case regex_constants::escape_type_control_f:
      result = charT('\f');
      break;
   case regex_constants::escape_type_control_n:
      result = charT('\n');
      break;
   case regex_constants::escape_type_control_r:
      result = charT('\r');
      break;
   case regex_constants::escape_type_control_t:
      result = charT('\t');
      break;
   case regex_constants::escape_type_control_v:
      result = charT('\v');
      break;
   case regex_constants::escape_type_word_assert:
      result = charT('\b');
      break;
   case regex_constants::escape_type_ascii_control:
      ++m_position;
      if(m_position == m_end)
      {
         fail(regex_constants::error_escape, m_position - m_base);
         return result;
      }
      /*
      if((*m_position < charT('@'))
            || (*m_position > charT(125)) )
      {
         fail(regex_constants::error_escape, m_position - m_base);
         return result;
      }
      */
      result = static_cast<charT>(*m_position % 32);
      break;
   case regex_constants::escape_type_hex:
      ++m_position;
      if(m_position == m_end)
      {
         fail(regex_constants::error_escape, m_position - m_base);
         return result;
      }
      // maybe have \x{ddd}
      if(this->m_traits.syntax_type(*m_position) == regex_constants::syntax_open_brace)
      {
         ++m_position;
         if(m_position == m_end)
         {
            fail(regex_constants::error_escape, m_position - m_base);
            return result;
         }
         int i = this->m_traits.toi(m_position, m_end, 16);
         if((m_position == m_end)
            || (i < 0)
            || ((std::numeric_limits<charT>::is_specialized) && (charT(i) > (std::numeric_limits<charT>::max)()))
            || (this->m_traits.syntax_type(*m_position) != regex_constants::syntax_close_brace))
         {
            fail(regex_constants::error_badbrace, m_position - m_base);
            return result;
         }
         ++m_position;
         result = charT(i);
      }
      else
      {
         std::ptrdiff_t len = (std::min)(static_cast<std::ptrdiff_t>(2), m_end - m_position);
         int i = this->m_traits.toi(m_position, m_position + len, 16);
         if((i < 0)
            || !valid_value(charT(0), i))
         {
            fail(regex_constants::error_escape, m_position - m_base);
            return result;
         }
         result = charT(i);
      }
      return result;
   case regex_constants::syntax_digit:
      {
      // an octal escape sequence, the first character must be a zero
      // followed by up to 3 octal digits:
      std::ptrdiff_t len = (std::min)(::boost::re_detail::distance(m_position, m_end), static_cast<std::ptrdiff_t>(4));
      const charT* bp = m_position;
      int val = this->m_traits.toi(bp, bp + 1, 8);
      if(val != 0)
      {
         // Oops not an octal escape after all:
         fail(regex_constants::error_escape, m_position - m_base);
         return result;
      }
      val = this->m_traits.toi(m_position, m_position + len, 8);
      if(val < 0) 
      {
         fail(regex_constants::error_escape, m_position - m_base);
         return result;
      }
      return static_cast<charT>(val);
      }
   case regex_constants::escape_type_named_char:
      {
         ++m_position;
         if(m_position == m_end)
         {
            fail(regex_constants::error_escape, m_position - m_base);
            return false;
         }
         // maybe have \N{name}
         if(this->m_traits.syntax_type(*m_position) == regex_constants::syntax_open_brace)
         {
            const charT* base = m_position;
            // skip forward until we find enclosing brace:
            while((m_position != m_end) && (this->m_traits.syntax_type(*m_position) != regex_constants::syntax_close_brace))
               ++m_position;
            if(m_position == m_end)
            {
               fail(regex_constants::error_escape, m_position - m_base);
               return false;
            }
            string_type s = this->m_traits.lookup_collatename(++base, m_position++);
            if(s.empty())
            {
               fail(regex_constants::error_collate, m_position - m_base);
               return false;
            }
            if(s.size() == 1)
            {
               return s[0];
            }
         }
         // fall through is a failure:
         fail(regex_constants::error_escape, m_position - m_base);
         return false;
      }
   default:
      result = *m_position;
      break;
   }
   ++m_position;
   return result;
#ifdef BOOST_MSVC
#pragma warning(pop)
#endif
}

template <class charT, class traits>
bool basic_regex_parser<charT, traits>::parse_backref()
{
   BOOST_ASSERT(m_position != m_end);
   const charT* pc = m_position;
   int i = this->m_traits.toi(pc, pc + 1, 10);
   if((i == 0) || (((this->flags() & regbase::main_option_type) == regbase::perl_syntax_group) && (this->flags() & regbase::no_bk_refs)))
   {
      // not a backref at all but an octal escape sequence:
      charT c = unescape_character();
      this->append_literal(c);
   }
   else if((i > 0) && (this->m_backrefs & (1u << (i-1))))
   {
      m_position = pc;
      re_brace* pb = static_cast<re_brace*>(this->append_state(syntax_element_backref, sizeof(re_brace)));
      pb->index = i;
   }
   else
   {
      fail(regex_constants::error_backref, m_position - m_end);
      return false;
   }
   return true;
}

template <class charT, class traits>
bool basic_regex_parser<charT, traits>::parse_QE()
{
#ifdef BOOST_MSVC
#pragma warning(push)
#pragma warning(disable:4127)
#endif
   //
   // parse a \Q...\E sequence:
   //
   ++m_position; // skip the Q
   const charT* start = m_position;
   const charT* end;
   do
   {
      while((m_position != m_end) 
         && (this->m_traits.syntax_type(*m_position) != regex_constants::syntax_escape))
         ++m_position;
      if(m_position == m_end)
      {
         //  a \Q...\E sequence may terminate with the end of the expression:
         end = m_position;
         break;  
      }
      if(++m_position == m_end) // skip the escape
      {
         fail(regex_constants::error_escape, m_position - m_base);
         return false;
      }
      // check to see if it's a \E:
      if(this->m_traits.escape_syntax_type(*m_position) == regex_constants::escape_type_E)
      {
         ++m_position;
         end = m_position - 2;
         break;
      }
      // otherwise go round again:
   }while(true);
   //
   // now add all the character between the two escapes as literals:
   //
   while(start != end)
   {
      this->append_literal(*start);
      ++start;
   }
   return true;
#ifdef BOOST_MSVC
#pragma warning(pop)
#endif
}

template <class charT, class traits>
bool basic_regex_parser<charT, traits>::parse_perl_extension()
{
   if(++m_position == m_end)
   {
      fail(regex_constants::error_badrepeat, m_position - m_base);
      return false;
   }
   //
   // treat comments as a special case, as these
   // are the only ones that don't start with a leading
   // startmark state:
   //
   if(this->m_traits.syntax_type(*m_position) == regex_constants::syntax_hash)
   {
      while((m_position != m_end) 
         && (this->m_traits.syntax_type(*m_position++) != regex_constants::syntax_close_mark))
      {}      
      return true;
   }
   //
   // backup some state, and prepare the way:
   //
   int markid = 0;
   std::ptrdiff_t jump_offset = 0;
   re_brace* pb = static_cast<re_brace*>(this->append_state(syntax_element_startmark, sizeof(re_brace)));
   std::ptrdiff_t last_paren_start = this->getoffset(pb);
   // back up insertion point for alternations, and set new point:
   std::ptrdiff_t last_alt_point = m_alt_insert_point;
   this->m_pdata->m_data.align();
   m_alt_insert_point = this->m_pdata->m_data.size();
   std::ptrdiff_t expected_alt_point = m_alt_insert_point;
   bool restore_flags = true;
   regex_constants::syntax_option_type old_flags = this->flags();
   bool old_case_change = m_has_case_change;
   m_has_case_change = false;
   //
   // select the actual extension used:
   //
   switch(this->m_traits.syntax_type(*m_position))
   {
   case regex_constants::syntax_colon:
      //
      // a non-capturing mark:
      //
      pb->index = markid = 0;
      ++m_position;
      break;
   case regex_constants::syntax_equal:
      pb->index = markid = -1;
      ++m_position;
      jump_offset = this->getoffset(this->append_state(syntax_element_jump, sizeof(re_jump)));
      this->m_pdata->m_data.align();
      m_alt_insert_point = this->m_pdata->m_data.size();
      break;
   case regex_constants::syntax_not:
      pb->index = markid = -2;
      ++m_position;
      jump_offset = this->getoffset(this->append_state(syntax_element_jump, sizeof(re_jump)));
      this->m_pdata->m_data.align();
      m_alt_insert_point = this->m_pdata->m_data.size();
      break;
   case regex_constants::escape_type_left_word:
      {
         // a lookbehind assertion:
         if(++m_position == m_end)
         {
            fail(regex_constants::error_badrepeat, m_position - m_base);
            return false;
         }
         regex_constants::syntax_type t = this->m_traits.syntax_type(*m_position);
         if(t == regex_constants::syntax_not)
            pb->index = markid = -2;
         else if(t == regex_constants::syntax_equal)
            pb->index = markid = -1;
         else
         {
            fail(regex_constants::error_badrepeat, m_position - m_base);
            return false;
         }
         ++m_position;
         jump_offset = this->getoffset(this->append_state(syntax_element_jump, sizeof(re_jump)));
         this->append_state(syntax_element_backstep, sizeof(re_brace));
         this->m_pdata->m_data.align();
         m_alt_insert_point = this->m_pdata->m_data.size();
         break;
      }
   case regex_constants::escape_type_right_word:
      //
      // an independent sub-expression:
      //
      pb->index = markid = -3;
      ++m_position;
      jump_offset = this->getoffset(this->append_state(syntax_element_jump, sizeof(re_jump)));
      this->m_pdata->m_data.align();
      m_alt_insert_point = this->m_pdata->m_data.size();
      break;
   case regex_constants::syntax_open_mark:
      {
      // a conditional expression:
      pb->index = markid = -4;
      if(++m_position == m_end)
      {
         fail(regex_constants::error_badrepeat, m_position - m_base);
         return false;
      }
      int v = this->m_traits.toi(m_position, m_end, 10);
      if(v > 0)
      {
         re_brace* br = static_cast<re_brace*>(this->append_state(syntax_element_assert_backref, sizeof(re_brace)));
         br->index = v;
         if(this->m_traits.syntax_type(*m_position) != regex_constants::syntax_close_mark)
         {
            fail(regex_constants::error_badrepeat, m_position - m_base);
            return false;
         }
         if(++m_position == m_end)
         {
            fail(regex_constants::error_badrepeat, m_position - m_base);
            return false;
         }
      }
      else
      {
         // verify that we have a lookahead or lookbehind assert:
         if(this->m_traits.syntax_type(*m_position) != regex_constants::syntax_question)
         {
            fail(regex_constants::error_badrepeat, m_position - m_base);
            return false;
         }
         if(++m_position == m_end)
         {
            fail(regex_constants::error_badrepeat, m_position - m_base);
            return false;
         }
         if(this->m_traits.syntax_type(*m_position) == regex_constants::escape_type_left_word)
         {
            if(++m_position == m_end)
            {
               fail(regex_constants::error_badrepeat, m_position - m_base);
               return false;
            }
            if((this->m_traits.syntax_type(*m_position) != regex_constants::syntax_equal)
               && (this->m_traits.syntax_type(*m_position) != regex_constants::syntax_not))
            {
               fail(regex_constants::error_badrepeat, m_position - m_base);
               return false;
            }
            m_position -= 3;
         }
         else
         {
            if((this->m_traits.syntax_type(*m_position) != regex_constants::syntax_equal)
               && (this->m_traits.syntax_type(*m_position) != regex_constants::syntax_not))
            {
               fail(regex_constants::error_badrepeat, m_position - m_base);
               return false;
            }
            m_position -= 2;
         }
      }
      break;
      }
   case regex_constants::syntax_close_mark:
      fail(regex_constants::error_badrepeat, m_position - m_base);
      return false;
   default:
      //
      // lets assume that we have a (?imsx) group and try and parse it:
      //
      regex_constants::syntax_option_type opts = parse_options();
      if(m_position == m_end)
         return false;
      // make a note of whether we have a case change:
      m_has_case_change = ((opts & regbase::icase) != (this->flags() & regbase::icase));
      pb->index = markid = 0;
      if(this->m_traits.syntax_type(*m_position) == regex_constants::syntax_close_mark)
      {
         // update flags and carry on as normal:
         this->flags(opts);
         restore_flags = false;
         old_case_change |= m_has_case_change; // defer end of scope by one ')'
      }
      else if(this->m_traits.syntax_type(*m_position) == regex_constants::syntax_colon)
      {
         // update flags and carry on until the matching ')' is found:
         this->flags(opts);
         ++m_position;
      }
      else
      {
         fail(regex_constants::error_badrepeat, m_position - m_base);
         return false;
      }

      // finally append a case change state if we need it:
      if(m_has_case_change)
      {
         static_cast<re_case*>(
            this->append_state(syntax_element_toggle_case, sizeof(re_case))
            )->icase = opts & regbase::icase;
      }

   }
   //
   // now recursively add more states, this will terminate when we get to a
   // matching ')' :
   //
   parse_all();
   //
   // Unwind alternatives:
   //
   if(0 == unwind_alts(last_paren_start))
      return false;
   //
   // we either have a ')' or we have run out of characters prematurely:
   //
   if(m_position == m_end)
   {
      this->fail(regex_constants::error_paren, ::boost::re_detail::distance(m_base, m_end));
      return false;
   }
   BOOST_ASSERT(this->m_traits.syntax_type(*m_position) == regex_constants::syntax_close_mark);
   ++m_position;
   //
   // restore the flags:
   //
   if(restore_flags)
   {
      // append a case change state if we need it:
      if(m_has_case_change)
      {
         static_cast<re_case*>(
            this->append_state(syntax_element_toggle_case, sizeof(re_case))
            )->icase = old_flags & regbase::icase;
      }
      this->flags(old_flags);
   }
   //
   // set up the jump pointer if we have one:
   //
   if(jump_offset)
   {
      this->m_pdata->m_data.align();
      re_jump* jmp = static_cast<re_jump*>(this->getaddress(jump_offset));
      jmp->alt.i = this->m_pdata->m_data.size() - this->getoffset(jmp);
      if(this->m_last_state == jmp)
      {
         // Oops... we didn't have anything inside the assertion:
         fail(regex_constants::error_empty, m_position - m_base);
         return false;
      }
   }
   //
   // verify that if this is conditional expression, that we do have
   // an alternative, if not add one:
   //
   if(markid == -4)
   {
      re_syntax_base* b = this->getaddress(expected_alt_point);
      if(b->type != syntax_element_alt)
      {
         re_alt* alt = static_cast<re_alt*>(this->insert_state(expected_alt_point, syntax_element_alt, sizeof(re_alt)));
         alt->alt.i = this->m_pdata->m_data.size() - this->getoffset(alt);
      }
      else if(this->getaddress(static_cast<re_alt*>(b)->alt.i, b)->type == syntax_element_alt)
      {
         fail(regex_constants::error_bad_pattern, m_position - m_base);
         return false;
      }
   }
   //
   // append closing parenthesis state:
   //
   pb = static_cast<re_brace*>(this->append_state(syntax_element_endmark, sizeof(re_brace)));
   pb->index = markid;
   this->m_paren_start = last_paren_start;
   //
   // restore the alternate insertion point:
   //
   this->m_alt_insert_point = last_alt_point;
   //
   // and the case change data:
   //
   m_has_case_change = old_case_change;
   return true;
}

template <class charT, class traits>
bool basic_regex_parser<charT, traits>::add_emacs_code(bool negate)
{
   //
   // parses an emacs style \sx or \Sx construct.
   //
   if(++m_position == m_end)
   {
      fail(regex_constants::error_escape, m_position - m_base);
      return false;
   }
   basic_char_set<charT, traits> char_set;
   if(negate)
      char_set.negate();

   static const charT s_punct[5] = { 'p', 'u', 'n', 'c', 't', };

   switch(*m_position)
   {
   case 's':
   case ' ':
      char_set.add_class(this->m_mask_space);
      break;
   case 'w':
      char_set.add_class(this->m_word_mask);
      break;
   case '_':
      char_set.add_single(digraph<charT>(charT('$'))); 
      char_set.add_single(digraph<charT>(charT('&'))); 
      char_set.add_single(digraph<charT>(charT('*'))); 
      char_set.add_single(digraph<charT>(charT('+'))); 
      char_set.add_single(digraph<charT>(charT('-'))); 
      char_set.add_single(digraph<charT>(charT('_'))); 
      char_set.add_single(digraph<charT>(charT('<'))); 
      char_set.add_single(digraph<charT>(charT('>'))); 
      break;
   case '.':
      char_set.add_class(this->m_traits.lookup_classname(s_punct, s_punct+5));
      break;
   case '(':
      char_set.add_single(digraph<charT>(charT('('))); 
      char_set.add_single(digraph<charT>(charT('['))); 
      char_set.add_single(digraph<charT>(charT('{'))); 
      break;
   case ')':
      char_set.add_single(digraph<charT>(charT(')'))); 
      char_set.add_single(digraph<charT>(charT(']'))); 
      char_set.add_single(digraph<charT>(charT('}'))); 
      break;
   case '"':
      char_set.add_single(digraph<charT>(charT('"'))); 
      char_set.add_single(digraph<charT>(charT('\''))); 
      char_set.add_single(digraph<charT>(charT('`'))); 
      break;
   case '\'':
      char_set.add_single(digraph<charT>(charT('\''))); 
      char_set.add_single(digraph<charT>(charT(','))); 
      char_set.add_single(digraph<charT>(charT('#'))); 
      break;
   case '<':
      char_set.add_single(digraph<charT>(charT(';'))); 
      break;
   case '>':
      char_set.add_single(digraph<charT>(charT('\n'))); 
      char_set.add_single(digraph<charT>(charT('\f'))); 
      break;
   default:
      fail(regex_constants::error_ctype, m_position - m_base);
      return false;
   }
   if(0 == this->append_set(char_set))
   {
      fail(regex_constants::error_ctype, m_position - m_base);
      return false;
   }
   ++m_position;
   return true;
}

template <class charT, class traits>
regex_constants::syntax_option_type basic_regex_parser<charT, traits>::parse_options()
{
   // we have a (?imsx-imsx) group, convert it into a set of flags:
   regex_constants::syntax_option_type f = this->flags();
   bool breakout = false;
   do
   {
      switch(*m_position)
      {
      case 's':
         f |= regex_constants::mod_s;
         f &= ~regex_constants::no_mod_s;
         break;
      case 'm':
         f &= ~regex_constants::no_mod_m;
         break;
      case 'i':
         f |= regex_constants::icase;
         break;
      case 'x':
         f |= regex_constants::mod_x;
         break;
      default:
         breakout = true;
         continue;
      }
      if(++m_position == m_end)
      {
         fail(regex_constants::error_paren, m_position - m_base);
         return false;
      }
   }
   while(!breakout);

   if(*m_position == static_cast<charT>('-'))
   {
      if(++m_position == m_end)
      {
         fail(regex_constants::error_paren, m_position - m_base);
         return false;
      }
      do
      {
         switch(*m_position)
         {
         case 's':
            f &= ~regex_constants::mod_s;
            f |= regex_constants::no_mod_s;
            break;
         case 'm':
            f |= regex_constants::no_mod_m;
            break;
         case 'i':
            f &= ~regex_constants::icase;
            break;
         case 'x':
            f &= ~regex_constants::mod_x;
            break;
         default:
            breakout = true;
            continue;
         }
         if(++m_position == m_end)
         {
            fail(regex_constants::error_paren, m_position - m_base);
            return false;
         }
      }
      while(!breakout);
   }
   return f;
}

template <class charT, class traits>
bool basic_regex_parser<charT, traits>::unwind_alts(std::ptrdiff_t last_paren_start)
{
   //
   // If we didn't actually add any states after the last 
   // alternative then that's an error:
   //
   if((this->m_alt_insert_point == static_cast<std::ptrdiff_t>(this->m_pdata->m_data.size()))
      && m_alt_jumps.size() && (m_alt_jumps.back() > last_paren_start))
   {
      fail(regex_constants::error_empty, this->m_position - this->m_base);
      return false;
   }
   // 
   // Fix up our alternatives:
   //
   while(m_alt_jumps.size() && (m_alt_jumps.back() > last_paren_start))
   {
      //
      // fix up the jump to point to the end of the states
      // that we've just added:
      //
      std::ptrdiff_t jump_offset = m_alt_jumps.back();
      m_alt_jumps.pop_back();
      this->m_pdata->m_data.align();
      re_jump* jmp = static_cast<re_jump*>(this->getaddress(jump_offset));
      BOOST_ASSERT(jmp->type == syntax_element_jump);
      jmp->alt.i = this->m_pdata->m_data.size() - jump_offset;
   }
   return true;
}

#ifdef BOOST_MSVC
#pragma warning(pop)
#endif

} // namespace re_detail
} // namespace boost

#ifdef BOOST_HAS_ABI_HEADERS
#  include BOOST_ABI_SUFFIX
#endif

#endif