Boost Spirit X3 AST not working with semantic actions when using separate rule definition and instantiation

Question

I am trying to use Boost Spirit X3 with semantic actions while parsing the structure to an AST. If I use a rule without separate definition and instantiation it works just fine, for example:

#include <vector>
#include <string>
#include <iostream>
#include <boost/fusion/include/adapt_struct.hpp>
#include <boost/spirit/home/x3.hpp>

namespace ast 
{

struct ast_struct
{
  int number;
  std::vector<int> numbers;
};

}

BOOST_FUSION_ADAPT_STRUCT(
    ast::ast_struct,
    (int, number)
    (std::vector<int>, numbers)
)

namespace x3 = boost::spirit::x3;
using namespace std;

void parse( const std::string &data )
{
  string::const_iterator begin = data.begin();
  string::const_iterator end = data.end();

  unsigned n(0);

  auto f = [&n]( auto &ctx )
    {
      n = x3::_attr(ctx);
    };

  ast::ast_struct ast;
  bool r = x3::parse( begin, end, 
                      x3::int_[f] >> +( x3::omit[+x3::blank] >> x3::int_ ), ast );

  if ( r && begin == end )
  {
    cout << "n: " << n << ", ";
    std::copy(ast.numbers.begin(), ast.numbers.end(), 
              std::ostream_iterator<int>(std::cout << ast.numbers.size() << " elements: ", " "));
    cout << endl;
  }
  else
    cout << "Parse failed" << endl;
}

int main()
{
  parse( "3 1 2 3" );
  parse( "4 1 2 3 4" );
  return 0;
}

Running the code above (compiled with flags -std=c++14) outputs the expected result:

n: 3, 3 elements: 1 2 3 
n: 4, 4 elements: 1 2 3 4

Now I am trying to have my Spirit X3 parser organized more or less the same way as the calc 9 example from Boost Spirit X3, but it does not work:

ast.hxx: defines the abstract syntax tree.
grammar.hxx: user interface exposing the parser methods.
grammar.cxx: instantiates the rules.
grammar_def.hxx: parser grammar definition.
config.hxx: parser configuration.
main.cxx: parser usage example.

ast.hxx:

#ifndef AST_HXX
#define AST_HXX

#include <vector>
#include <boost/fusion/include/adapt_struct.hpp>

namespace ast 
{

struct ast_struct
{
  int number;
  std::vector<int> numbers;
};

}

BOOST_FUSION_ADAPT_STRUCT(
    ast::ast_struct,
    (int, number)
    (std::vector<int>, numbers)
)

#endif

grammar.hxx:

#ifndef GRAMMAR_HXX
#define GRAMMAR_HXX

#include "ast.hxx"
#include <boost/spirit/home/x3.hpp>

namespace parser 
{

namespace x3 = boost::spirit::x3;

using my_rule_type = x3::rule<class my_rule_class, ast::ast_struct>;

BOOST_SPIRIT_DECLARE( my_rule_type );

const my_rule_type &get_my_rule();

}

#endif

grammar.cxx:

#include "grammar_def.hxx"
#include "config.hxx"

namespace parser 
{

BOOST_SPIRIT_INSTANTIATE( my_rule_type, iterator_type, context_type )

}

grammar_def.hxx:

#ifndef GRAMMAR_DEF_HXX
#define GRAMMAR_DEF_HXX

#include <iostream>
#include <boost/spirit/home/x3.hpp>
#include "grammar.hxx"
#include "ast.hxx"

namespace parser 
{
namespace x3 = boost::spirit::x3;

const my_rule_type  my_rule( "my_rule" );

unsigned n;

auto f = []( auto &ctx )
{
  n = x3::_attr(ctx);
};

auto my_rule_def =  x3::int_[f] >> +( x3::omit[+x3::blank] >> x3::int_ );

BOOST_SPIRIT_DEFINE( my_rule )

const my_rule_type &get_my_rule()
{
  return my_rule;
}

}

#endif

config.hxx:

#ifndef CONFIG_HXX
#define CONFIG_HXX

#include <string>
#include <boost/spirit/home/x3.hpp>

namespace parser 
{

namespace x3 = boost::spirit::x3;

using iterator_type = std::string::const_iterator;
using context_type = x3::unused_type;

}

#endif

main.cxx:

#include "ast.hxx"
#include "grammar.hxx"
#include "config.hxx"
#include <iostream>
#include <boost/spirit/home/x3.hpp>
#include <string>

namespace x3 = boost::spirit::x3;
using namespace std;

void parse( const std::string &data )
{
  parser::iterator_type begin = data.begin();
  parser::iterator_type end = data.end();

  ast::ast_struct ast;
  cout << "Parsing [" << string(begin,end) << "]" << endl;

  bool r = x3::parse( begin, end, parser::get_my_rule(), ast );

  if ( r && begin == end )
  {
    std::copy(ast.numbers.begin(), ast.numbers.end(), 
              std::ostream_iterator<int>(std::cout << ast.numbers.size() << " elements: ", " "));
    cout << endl;
  }
  else
    cout << "Parse failed" << endl;
}

int main()
{
  parse( "3 1 2 3" );
  parse( "4 1 2 3 4" );
  return 0;
}

Compiling main.cxx and grammar.cxx (flags: -std=c++14) and running the code above prints:

Parsing [3 1 2 3]
0 elements: 
Parsing [4 1 2 3 4]
0 elements:

I apologize for the long source code, I tried to make it as small as possible.

Please notice I have some usage for the unsigned n global variable, it will be used with a custom repeat directive (see question here and one of the solutions here). In order to keep the question focused I removed the repeat part from this question, so even though I could remove the semantic action in this example, it is not a possible solution.

I would appreciate some help to get this issue uncovered, it is not clear to me why the code above does not work. Thank you in advance.

sehe sehe · Accepted Answer · 2015-11-26T11:11:43

I must admit actually reconstructing your sample was a bit too much work for me (call me lazy...).

However, I know the answer and a trick to make your life simpler.

The Answer

Semantic actions on a rule definition inhibit automatic attribute propagation. From the Qi docs (the same goes for X3, but I always lose the link to the docs):

r = p; Rule definition
This is equivalent to r %= p (see below) if there are no semantic actions attached anywhere in p.

r %= p; Auto-rule definition
The attribute of p should be compatible with the synthesized attribute of r. When p is successful, its attribute is automatically propagated to r's synthesized attribute.

The Trick

You can inject state (your n reference, in this case) using the x3::with<> directive. That way you don't have the namespace global (n) and can make the parser reentrant, threadsafe etc.

Here's my "simplist" take on things, in a single file:

namespace parsing {
    x3::rule<struct parser, ast::ast_struct> parser {"parser"};

    struct state_tag { };

    auto record_number = [](auto &ctx) { 
        unsigned& n = x3::get<state_tag>(ctx);
        n = x3::_attr(ctx); 
    };

    auto parser_def = x3::rule<struct parser_def, ast::ast_struct> {} 
                   %= x3::int_[record_number] >> +(x3::omit[+x3::blank] >> x3::int_);

    BOOST_SPIRIT_DEFINE(parser)
}

Tip: run the demo with = instead of the %= to see the difference in behaviour!

Note that get<state_tag>(ctx) returns a reference_wrapper<unsigned> just because we use the parser as follows:

void parse(const std::string &data) {
    using namespace std;

    ast::ast_struct ast;
    unsigned n;
    auto parser = x3::with<parsing::state_tag>(ref(n)) [parsing::parser] >> x3::eoi;

    if (x3::parse(data.begin(), data.end(), parser, ast)) {
        cout << "n: " << n << ", ";
        copy(ast.numbers.begin(), ast.numbers.end(), ostream_iterator<int>(cout << ast.numbers.size() << " elements: ", " "));
        cout << "\n";
    } else
        cout << "Parse failed\n";
}

Live Demo

Live On Coliru

#include <boost/fusion/include/adapt_struct.hpp>
#include <boost/spirit/home/x3.hpp>
#include <iostream>

namespace ast {
    struct ast_struct {
        int number;
        std::vector<int> numbers;
    };
}

BOOST_FUSION_ADAPT_STRUCT(ast::ast_struct, number, numbers)

namespace x3 = boost::spirit::x3;

namespace parsing {
    x3::rule<struct parser, ast::ast_struct> parser {"parser"};

    struct state_tag { };

    auto record_number = [](auto &ctx) { 
        unsigned& n = x3::get<state_tag>(ctx); // note: returns reference_wrapper<T>
        n = x3::_attr(ctx); 
    };

    auto parser_def = x3::rule<struct parser_def, ast::ast_struct> {} 
                   %= x3::int_[record_number] >> +(x3::omit[+x3::blank] >> x3::int_);

    BOOST_SPIRIT_DEFINE(parser)
}

void parse(const std::string &data) {
    using namespace std;

    ast::ast_struct ast;
    unsigned n = 0;
    auto parser = x3::with<parsing::state_tag>(ref(n)) [parsing::parser] >> x3::eoi;

    if (x3::parse(data.begin(), data.end(), parser, ast)) {
        cout << "n: " << n << ", ";
        copy(ast.numbers.begin(), ast.numbers.end(), ostream_iterator<int>(cout << ast.numbers.size() << " elements: ", " "));
        cout << "\n";
    } else
        cout << "Parse failed\n";
}

int main() {
    parse("3 1 2 3");
    parse("4 1 2 3 4");
}

Prints

n: 3, 3 elements: 1 2 3 
n: 4, 4 elements: 1 2 3 4

Boost Spirit X3 AST not working with semantic actions when using separate rule definition and instantiation

1 Answers

The Answer

The Trick

Live Demo