Why swapping subparser around >> will stop the program from compiling?

[liubing]

This program fails to compile on both MSVC & GCC under C++20 mode:

#include <boost/parser/parser.hpp>
#include <string>

namespace bp = boost::parser;

struct url_foo_attr
{
    char zero_or_one_;
    std::string digits_;
};

bp::rule<struct url_foo_tag, url_foo_attr> url_foo = "url_foo";
auto url_foo_def = bp::lit("/foo/") >> bp::char_('0', '1') >> '/' >> *bp::digit;
BOOST_PARSER_DEFINE_RULES(url_foo);

int main()
{
    bp::parse("/foo/0/1", url_foo);
    return 0;
}

But if you swap the members inside the struct and adjust url_foo_def accordingly, it suddenly works correctly:

#include <boost/parser/parser.hpp>
#include <string>

namespace bp = boost::parser;

struct url_foo_attr
{
    std::string digits_;
    char zero_or_one_;
};

bp::rule<struct url_foo_tag, url_foo_attr> url_foo = "url_foo";
auto url_foo_def = bp::lit("/foo/") >> *bp::digit >> '/' >> bp::char_('0', '1');
BOOST_PARSER_DEFINE_RULES(url_foo);

int main()
{
    bp::parse("/foo/0/1", url_foo);
    return 0;
}

Why?

[d5ch4k]

My Observations: Both rules return a std::optional<std::string>. You can verify it by using the following code with the help of a debugger:

#include <boost/parser/parser.hpp>
#include <string>

namespace bp = boost::parser;

auto url_foo_def1 = bp::lit("/foo/") >> bp::char_('0', '1') >> '/' >> *bp::digit;
auto url_foo_def2= bp::lit("/foo/") >> *bp::digit >> bp::lit("/") >> bp::char_('0', '1');

int main()
{
    auto result1 = bp::parse("/foo/0/1", url_foo_def1);
    if (result1) {
        std::cout << "success 1" << std::endl;;
        std::cout << *result1 << std::endl;;
    } else {
        std::cout << "error 1";
    }
    auto result2 = bp::parse("/foo/0/1", url_foo_def2);
    if (result2) {
        std::cout << "success 2" << std::endl;;
        std::cout << *result2 << std::endl;;
    } else {
        std::cout << "error 2";
    }
    return 0;
}

the output is:

success 1
01
success 2
01

If you examine the output of your second version by adding an output:

#include <boost/parser/parser.hpp>
#include <string>

namespace bp = boost::parser;

struct url_foo_attr
{
    std::string digits_;
    char zero_or_one_;
};

bp::rule<struct url_foo_tag, url_foo_attr> url_foo = "url_foo";
auto url_foo_def = bp::lit("/foo/") >> *bp::digit >> '/' >> bp::char_('0', '1');
BOOST_PARSER_DEFINE_RULES(url_foo);

int main()
{
    auto result2 = bp::parse("/foo/0/1", url_foo);
    if (result2) {
        std::cout << "success!" << std::endl;
        std::cout << "digits: '" << result2->digits_ << "' expected '0'" << std::endl;
        std::cout << "zero_or_one: '" << result2->zero_or_one_ << "' expected '1'"  << std::endl;
    } else {
        std::cout << "failed!";
    }

    return 0;
}

the corresponding output is:

digits: '01' expected '0'
zero_or_one: ' ' expected '1'

this means it compiled, because your second structure begins with a std::string as c++20 seems to allow a partial initialization list but the result is not, what you expected (in c++17 both version do not compile which indicates that the parser result and the structure do not match). For forcing a separation you can place bp::separate[...] around your rule

Then your first example will compile and work as expected:

#include <boost/parser/parser.hpp>
#include <string>

namespace bp = boost::parser;

struct url_foo_attr
{
    char zero_or_one_;
    std::string digits_;
};

bp::rule<struct url_foo_tag, url_foo_attr> url_foo = "url_foo";
auto url_foo_def = bp::lit("/foo/") >> bp::separate[bp::char_('0', '1') >> '/' >> *bp::digit];
BOOST_PARSER_DEFINE_RULES(url_foo);

int main()
{
    auto result1 = bp::parse("/foo/0/1", url_foo);
    if (result1) {
        std::cout << "success!" << std::endl;
        std::cout << "zero_or_one: '" << result1->zero_or_one_ << "' expected '0'"  << std::endl;
        std::cout << "digits: '" << result1->digits_ << "' expected '1'" << std::endl;
    } else {
        std::cout << "failed!";
    }
    return 0;
}

gives

success!
zero_or_one: '0' expected '0'
digits: '1' expected '1'

Process finished with exit code 0

regards Martin

[liubing]

Got it, thank you very much!