Using boost::spirit, how do I require part of a record to be on its own line?
Asked Answered
G

1

6

I have a record parser that throws one of several exceptions to indicate which rule failed.

Front matter:

#include <iostream>
#include <sstream>
#include <stdexcept>
#include <string>

#include <boost/spirit/include/qi.hpp>
#include <boost/spirit/include/phoenix.hpp>
#include <boost/spirit/include/classic_position_iterator.hpp>

using namespace boost::spirit;
using namespace boost::spirit::ascii;
using namespace boost::spirit::qi;
using namespace boost::spirit::qi::labels;

using boost::phoenix::function;
using boost::phoenix::ref;
using boost::spirit::qi::eol;
using boost::spirit::qi::fail;
using boost::spirit::qi::lit;
using boost::spirit::qi::on_error;

using BOOST_SPIRIT_CLASSIC_NS::file_position;
using BOOST_SPIRIT_CLASSIC_NS::position_iterator;

We use the position_iterator from Spirit.Classic, so the following stream-insertion operator is handy.

std::ostream&
operator<<(std::ostream& o, const file_position &fp)
{
  o << fp.file << ": " << fp.line << ',' << fp.column;
  return o;
}

The template err_t factors out the boilerplate for throwing the exceptions associated with different forms of parse failure.

template <typename Exception>
struct err_t {
  template <typename, typename, typename>
  struct result { typedef void type; };

  template <typename Iterator>
  void operator() (info const &what, Iterator errPos, Iterator last) const
  {
    std::stringstream ss;
    ss << errPos.get_position()
       << ": expecting " << what
       << " near '" << std::string(errPos, last) << "'\n";
    throw Exception(ss.str());
  }
};

The exceptions used along with their err_t wrappers:

class MissingA : public std::runtime_error {
  public: MissingA(const std::string &s) : std::runtime_error(s) {}
};

class MissingB : public std::runtime_error {
  public: MissingB(const std::string &s) : std::runtime_error(s) {}
};

class MissingC : public std::runtime_error {
  public: MissingC(const std::string &s) : std::runtime_error(s) {}
};

function<err_t<MissingA> > const missingA = err_t<MissingA>();
function<err_t<MissingB> > const missingB = err_t<MissingB>();
function<err_t<MissingC> > const missingC = err_t<MissingC>();
function<err_t<std::runtime_error> > const other_error =
  err_t<std::runtime_error>();

The grammar looks for simple sequences. Without eps, the start rule fails rather than a on an empty input.

template <typename Iterator, typename Skipper>
struct my_grammar
  : grammar<Iterator, Skipper>
{
  my_grammar(int &result)
    : my_grammar::base_type(start)
    , result(result)
  {
    a = eps > lit("Header A") > eol;
    b = eps > lit("Header B") > eol;
    c = eps > lit("C:") > int_[ref(result) = _1] > eol;
    start = a > b > c;

    a.name("A");
    b.name("B");
    c.name("C");

    on_error<fail>(start, other_error(_4, _3, _2));
    on_error<fail>(a, missingA(_4, _3, _2));
    on_error<fail>(b, missingB(_4, _3, _2));
    on_error<fail>(c, missingC(_4, _3, _2));
  }

  rule<Iterator, Skipper> start;
  rule<Iterator, Skipper> a;
  rule<Iterator, Skipper> b;
  rule<Iterator, Skipper> c;
  int &result;
};

In my_parse, we dump the contents of the stream into a std::string and use position_iterator to track the parse's location.

int
my_parse(const std::string &path, std::istream &is)
{
  std::string buf;
  is.unsetf(std::ios::skipws);
  std::copy(std::istream_iterator<char>(is),
            std::istream_iterator<char>(),
            std::back_inserter(buf));

  typedef position_iterator<std::string::const_iterator> itertype;
  typedef my_grammar<itertype, boost::spirit::ascii::space_type> grammar;
  itertype it(buf.begin(), buf.end(), path);
  itertype end;

  int result;
  grammar g(result);

  bool r = phrase_parse(it, end, g, boost::spirit::ascii::space);
  if (r && it == end) {
    std::cerr << "success!\n";
    return result;
  }
  else {
    file_position fpos = it.get_position();
    std::cerr << "parse failed at " << fpos << '\n';
    return -9999;
  }
}

Finally, the main program

int main()
{
  std::stringstream ss;
  ss << "Header A\n"
     << "Header B\n"
     << "C: 3\n";

  int val = my_parse("path", ss);
  std::cout << "val = " << val << '\n';

  return 0;
}

The code above throws MissingA:

terminate called after throwing an instance of 'MissingA'
  what():  path: 2,1: expecting  near 'Header B
C: 3
'

I thought the skipper might have consumed the newline, but attempting lexeme[eol] instead produced the same result.

I must be missing something obvious because this seems one of the most trivial sort of parsers to write. What am I doing wrong?

Gangboard answered 11/3, 2010 at 20:18 Comment(0)
F
7

Yes, the skipper eats the newline characters. lexeme[eol] doesn't help either because the lexeme directive invokes the skipper before switching to no-skipper mode (see here for more details).

In order to avoid skipping newlines, either use a different skipper type, or wrap the eol components into no_skip[eol], which is semantically equivalent to lexeme[], except it does not invoke the skipper. Note though, that no_skip[] has been added recently only, so it will be available with the next release only (Boost V1.43). But it is in the Boost SVN already (see here for the preliminary docs).

Fubsy answered 12/3, 2010 at 3:47 Comment(0)

© 2022 - 2024 — McMap. All rights reserved.