With dynamic grammar skipper doesn't work

[allopislozano]

I think this is a bug. I create dynamically a x3::any_parser that can parse strings like "t1" "t2" etc and when I compose it, it doesn't parse correctly unless I use a x3::lexeme[] around it. Looks like the skipper is not working properly in that case or something...

#include<vector>
#include<string>
#include <utility>
#include <iostream>
#define BOOST_SPIRIT_X3_DEBUG
#include <boost/spirit/home/x3.hpp>
using std::vector;
using std::string;
  namespace x3 = boost::spirit::x3;

x3::any_parser<string::iterator> makeComposedParser(vector<string> const & ss)
{

    x3::any_parser<string::iterator> p= x3::eps(false); //placeholder
    for (auto && s : ss)
        p = p | x3::lit(s) >> x3::int_;
     return p;
}

int main()
{
    vector<string> parserMustDetect = {"t", "b"};
    auto a = makeComposedParser(parserMustDetect);

    vector<string> toParse = {"t1","t2"};
    for (auto && s : toParse){
        string x;
        auto b = s.begin();
        auto e = s.end();
        std::cout << "parsed?" << x3::parse(b,e,a)<< "\n";
    }
    string x = "X t1";
    auto b = x.begin();
    auto e = x.end();

    //Should parse but it doesnt
    std::cout << "parsed ? " << x3::phrase_parse(b,e,x3::lit("X") >> a, x3::space) << "\n";
    //Lexeme helps
    std::cout << "parsed ? " << x3::phrase_parse(b,e,x3::lit("X") >> x3::lexeme[a], x3::space);

}

[Kojoley]

Unfortunately it is by design.

Sequence parser does not do skipping by itself, it relies on a fact that underlying parser will do skipping. The a parser does not do skipping because any_parser with default context template argument drops skipper, that's why in the first case no skipping happens, and in the second case lexeme does skipping.

[allopislozano]

So is this considered a bug ? Or is it a limitation that is not possible to solve? In that case I think it would be nice to document that limitation. In fact any_parser is undocumented AFAIK

[Kojoley]

It depends on a perspective. I see it as an unintuitive behavior that I would like to fix but: 1) it is forged into rule too (https://wandbox.org/permlink/QtXPYSDTJEqtHqre), fixing only any_parser will just bring inconsistency 2) it requires non-trivial changes to deal with non-preskipping parsers (like no_case, attr, etc.) 3) it will break backward compatibility.

@djowel do you have anything to say?

[Kojoley]

The only thing that could be done is to add a static assertion, but I think it will be too annoying to wrap a rule with no_skip[a].

Probably I was not clear enough in my first response, the example code could be fixed by replacing x3::any_parser<string::iterator> with x3::any_parser<string::iterator, x3::unused_type, x3::phrase_parse_context<x3::space_type>::type> or p = p | x3::lit(s) >> x3::int_ p = p | x3::skip(x3::space)[x3::lit(s) >> x3::int_].