c++ - boost::spirit::qi permutation parser and synthesized attributes

Question

I'm trying to put together a simple command line parser with SPIRIT without semantic actions. I'm using BOOST 1.52 but I would like to avoid using C++11 features. The grammar has the following syntax:

[-p num1] [-j] [--jobs num2] str1 str2

Optional parameters can be in any order. I successfully parsed only optional parameters. Once I add the additional mandatory two string parsers it breaks. It breaks even when I try to write down the "rstart" attributes explicitly and avoid type deduction using "auto". Any help or suggestion is very appreciated!

#include <iostream>
#include <string>
#include <vector>
#include <iterator>
#include <boost/spirit/include/qi.hpp>
#include <boost/optional.hpp>
#include <boost/fusion/include/boost_tuple.hpp>

bool parse_line( const std::string&str,bool useStart1 )
{
    bool rc=false;
    namespace qi = boost::spirit::qi;

    using boost::spirit::ascii::space_type;
    using boost::spirit::ascii::space;
    using boost::spirit::ascii::char_;

    std::string::const_iterator iter( str.begin() );

    std::size_t num1    =88;
    bool        bool1   =false;
    std::size_t num2    =88;
    std::string str1,str2;

    qi::rule< std::string::const_iterator,std::string() > rstring=+~space;
    qi::rule< std::string::const_iterator,std::size_t() ,space_type > 
        rOption1=qi::lit( "-p" ) >> qi::int_;
    qi::rule< std::string::const_iterator,bool()        ,space_type >
        rOption2=qi::lit( "-j" ) >> qi::attr(true);
    qi::rule< std::string::const_iterator,std::size_t() ,space_type >
        rOption3=qi::lit( "--jobs" ) >> qi::int_;

#if defined(AAA)
    qi::rule<   std::string::const_iterator,
                boost::spirit::ascii::space_type,
                boost::tuple<   boost::optional<std::size_t>,
                                boost::optional<bool>,
                                boost::optional<std::size_t >
                >
            >
#endif

    auto rstart1 = ( rOption1 ^ rOption2 ^ rOption3 ) ;
    auto rstart2 = ( rOption1 ^ rOption2 ^ rOption3 ) >> rstring >> rstring;

    if( useStart1 )
        qi::phrase_parse( iter,str.end(),
            ( qi::lit( "-p" ) >> qi::int_ ) ^
            ( qi::lit( "-j" ) >> qi::attr(true) ) ^
            ( qi::lit( "--jobs" ) >> qi::int_ ),space,num1,bool1,num2);
    else
    {
    //  qi::phrase_parse(
    //      iter,str.end(),rstart2,space,num1,bool1,num2,str1,str2);
    }

    if(iter==str.begin())
        iter=str.begin(); //NOP
    else
    if(iter!=str.end())
        std::cerr<<"syntax error: "<<std::string(iter,str.end())<<"!

";
    else
        rc=true;

    std::cout << "num1:" << num1 << std::endl;
    std::cout << "bool1:"<< bool1  << std::endl;
    std::cout << "num2:" << num2 << std::endl;
    std::cout << "str1:" << str1 << std::endl;
    std::cout << "str2:" << str2 << std::endl;

    return rc;
}

int main( int argc,char**argv )
{
    std::vector< std::string > testData1;
    testData1.push_back( "-p 100 -j" );
    testData1.push_back( "-j -p 100 --jobs 16" );
    testData1.push_back( "--jobs 16 -j -p 100" );

    for( std::vector< std::string >::const_iterator it=testData1.begin();
         it!=testData1.end(); ++it )
    {
        std::cout << "
parsing string:" << *it << std::endl;
        parse_line( *it,true );
    }

    std::vector< std::string > testData2;
    testData2.push_back( "-p 100 -j ifile ofile" );
    testData2.push_back( "-j -p 100 --jobs 16 ifile ofile" );
    testData2.push_back( "--jobs 16 -j -p 100 ifile ofile" );

    for( std::vector< std::string >::const_iterator it=testData2.begin();
         it!=testData2.end(); ++it )
    {
        std::cout << "
parsing string:" << *it << std::endl;
        parse_line( *it,false );
    }

    return 0;
}

Answer 1

The problem you are facing is that the attribute of your combined rule is basically:

tuple< tuple<size_t,bool,size_t>, std::string, std::string >

and by putting your variables one by one on the call to phrase_parse you have basically:

tuple< size_t, bool, size_t, std::string, std::string >

Because of the way attribute propagation works in spirit this is what is happening:

the whole tuple<size_t,bool,size_t> is assigned to your num1 (ignoring the bool and second size_t), after that spirit tries to assign the first string to your bool, resulting in the error you have.

I believe the cleanest way to solve this is creating a custom struct to hold your result that reflects the structure of your rules.

#include <iostream>
#include <string>
#include <vector>
#include <iterator>
#include <boost/spirit/include/qi.hpp>
#include <boost/fusion/include/adapt_struct.hpp>

struct optional_command_line_options
{
   int num1;
    bool bool1;
    int num2;
};

struct command_line_options
{
    optional_command_line_options opt;
    std::string str1;
    std::string str2;
};

BOOST_FUSION_ADAPT_STRUCT(
    optional_command_line_options,
    (int, num1)
    (bool, bool1)
    (int, num2)
)

BOOST_FUSION_ADAPT_STRUCT(
    command_line_options,
    (optional_command_line_options, opt)
    (std::string, str1)
    (std::string, str2)
)

bool parse_line( const std::string&str )
{
    bool rc=false;
    namespace qi = boost::spirit::qi;

    using boost::spirit::ascii::space;
    using boost::spirit::ascii::char_;

    std::string::const_iterator iter( str.begin() );

    command_line_options options;
    options.opt.num1=88;
    options.opt.bool1=false;
    options.opt.num2=88;

    qi::rule< std::string::const_iterator, std::string() > rstring=+~space;


    qi::rule<std::string::const_iterator, boost::spirit::ascii::space_type,optional_command_line_options() >  trule; 
    trule=
        ( qi::lit( "-p" ) >> qi::int_ ) ^
        ( qi::lit( "-j" ) >> qi::attr(true) ) ^
        ( qi::lit( "--jobs" ) >> qi::int_ )
        ;

    qi::rule< std::string::const_iterator, boost::spirit::ascii::space_type, command_line_options() >arule;
    arule = -trule >> rstring >> rstring;

    bool result=qi::phrase_parse( iter,str.end(),
                                 arule,
                                 space,
                                 options
    );

    if(result && iter==str.end())
    {
        std::cout << "Parse successful." << std::endl;
        rc=true;
    }
    else
    {
        std::cerr<<"syntax error: "<<std::string(iter,str.end())<<"!

";
    }

     std::cout << std::boolalpha;
    std::cout << "num1:" << options.opt.num1 << std::endl;
    std::cout << "bool1:"<< options.opt.bool1  << std::endl;
    std::cout << "num2:" << options.opt.num2 << std::endl;
    std::cout << "str1:" << options.str1 << std::endl;
    std::cout << "str2:" << options.str2 << std::endl;

    return rc;
}

int main( int /*argc*/,char**/*argv*/ )
{
    std::vector< std::string > testData;
    testData.push_back( "-p 100 -j ifile ofile" );
    testData.push_back( "-j -p 100 --jobs 16 ifile ofile" );
    testData.push_back( "--jobs 16 -j -p 100 ifile ofile" );
     testData.push_back( "--jobs 16 -p 100 ifile ofile" );
     testData.push_back( "ifile ofile" );

    for( std::vector< std::string >::const_iterator it=testData.begin();
         it!=testData.end(); ++it )
    {
        std::cout << "
parsing string:" << *it << std::endl;
        parse_line( *it );
    }

    return 0;
}

Running on LWS.

PS: You can't assign directly to rules declared with auto that have literals embedded in them (strings or numbers for example) without using boost::proto::deep_copy;

auto trule = boost::proto::deep_copy(qi::lit( "-p" ) >> qi::int_);

There is a macro called BOOST_SPIRIT_AUTO that makes it easier to use:

#define BOOST_SPIRIT_AUTO(domain_, name, expr)                                  
    typedef boost::proto::result_of::                                           
        deep_copy<BOOST_TYPEOF(expr)>::type name##_expr_type;                   
    BOOST_SPIRIT_ASSERT_MATCH(                                                  
        boost::spirit::domain_::domain, name##_expr_type);                      
    BOOST_AUTO(name, boost::proto::deep_copy(expr));

BOOST_SPIRIT_AUTO(qi,trule,qi::lit( "-p" ) >> qi::int_);