Boost C++ Libraries: Ticket #6043: Spirit insists that user types are adapted to Fusion

attachment set

Vladimir Prus — Thu, 20 Oct 2011 16:41:28 GMT

attachment → main.cpp

testcase

Vladimir Prus — Thu, 20 Oct 2011 16:44:39 GMT

On IRC, Thomas produced somewhat easier example. Code the does not work:

#include <boost/spirit/home/qi.hpp>
#include <boost/spirit/home/phoenix.hpp>
namespace fusion = boost::fusion;
namespace qi = boost::spirit::qi;
namespace phoenix = boost::phoenix;
struct foo
{
    foo(fusion::vector<int, double>) {}
};
int main()
{
    typedef std::string::iterator iterator;
    qi::rule<iterator, foo()> foo_rule;
    foo_rule = (qi::int_ >> qi::double_);
}

Code that does work:

#include <boost/spirit/home/qi.hpp>
#include <boost/spirit/home/phoenix.hpp>
namespace fusion = boost::fusion;
namespace qi = boost::spirit::qi;
namespace phoenix = boost::phoenix;
struct foo
{
    foo() {}
    foo(int, double) {}
    foo(fusion::vector<int, double>) {}
};
int main()
{
    typedef std::string::iterator iterator;
    qi::rule<iterator, fusion::vector<int, double>()> tuple_rule;
    qi::rule<iterator, foo()> foo_rule;
    tuple_rule = (qi::int_ >> qi::double_);
    foo_rule %= tuple_rule.alias();
}

I frankly don't understand what's going on here, but trying the same workaround for me does not work (with Boost trunk, and gcc 4.5.3), while Thomas reported this new example (main-with-workaround.cpp) compiles for him just fine with 1.47.

attachment set

Vladimir Prus — Thu, 20 Oct 2011 16:45:17 GMT

attachment → main-with-workaround.cpp

testcase with attempted workaround, still does not work for me.

component changed; owner set

anonymous — Tue, 08 Nov 2011 16:02:34 GMT

owner set to Joel de Guzman
component None → spirit

anonymous — Wed, 09 Nov 2011 00:16:14 GMT

Here's what's happening:

When you have A >> B, the rules are: http://tinyurl.com/bqch2m6

So, as far as compatibility goes, the attribute either has to be a fusion sequence *or* a std container. That is why your code does not compile.

The thing is, spirit is optimal and walks the sequence and builds its elements in-situ, instead of synthesising a temporary sequence in memory and assign later. That is why it has to iterate over existing instances. In this case, it cannot walk a foo. Heller's workaround works because the inner rule is able to walk the fusion sequence and then attribute transformation kicks in whereby it assigns the fusion sequence to foo. That works but is sub optimal.

status changed; resolution set

Joel de Guzman — Wed, 09 Nov 2011 23:43:29 GMT

status new → closed
resolution → wontfix

Volodya, I'm going to close this as no-fix. If you want to discuss, we can do it in IRC or the mailing lists.