tzlaine
commented
9 months ago I'm fine describing backtracking more fully (and I'm going to), but I don't understand the statement that X3 does not have backtracking on by default. Consider the case you reported earlier, slightly modified:
#include <boost/parser/parser.hpp>
#include <vector>
struct X {
char a;
int b;
};
struct Y {
std::vector<X> x;
int c;
};
int main()
{
namespace bp = boost::parser;
auto parse_x = bp::int_ >> bp::int_;
auto parse_y = +parse_x >> bp::int_;
Y y;
auto b = bp::parse("1 3 4", parse_y, bp::ws, y);
}I changed it so that it parses ints all the time, instead of a char at the beginning of parse_x.
This still works because the second iteration of parse_x, having eaten "1 3" fails to eat two more ints, since it sees only "4" next in the input. parse_y, in which +parse_x is a subparser, must then backtrack a bit, and parse the "4" using the bp::int_ at the end of parse_y. This is how Spriit 1 and 2 worked. Is this different from how X3 works? If so, I can't find that in the X3 docs.
akrzemi1
I only learned from the JSON example that there is a system for backtracking incomparible with Boost.Spirit.
I am trying to port my parser code from Boost.Spirit X3 to Boost.Parser and stumbled upon the problem, how to rewrite the usage of
boost::spirit::qi::hold. Boost.Spirit X3 has backtracking disabled by default -- for parsing performance, I guess -- and when defining alternatives one has to exlicitly enable it.Because the port from Boost.Spirit X3 will be a common use case I would recommend:
>and>>.