Using mpl::vector to define boost::variant types

Question

I'm using the library boost::variant to store a large number of types. As the number of type is growing, I will soon reach the limit of 20 types. In the documentation it seems possible to define the variant using a mpl::vector, which allows more than 20 types (up to 50 if I'm correct). I tried to replace my variant definition like this:

#include <boost/variant.hpp>
#include <boost/mpl/vector.hpp>

typedef boost::mpl::vector<
    float,
    math::float2,
    math::float3,
    relative_point<1>,
    relative_point<2>,
    relative_point<3>,
    std::string,
    color,
    group,
    dictionnary,
    reference,
    line,
    strip,
    text,
    font
> variant_mpl_vec;

typedef boost::make_variant_over<variant_mpl_vec>::type data_type;

// This is the old definition
/*typedef boost::variant<
    float,
    math::float2,
    math::float3,
    relative_point<1>,
    relative_point<2>,
    relative_point<3>,
    std::string,
    color,
    group,
    dictionnary,
    reference,
    line,
    strip,
    text,
    font
> data_type;*/

I'm putting directly my code. Most of types are structures containing very few data.

When compiling, I got a strange:

error: no matching function for call to ‘boost::detail::variant::make_initializer_node::apply<boost::mpl::pair< ... and lots more ...

Previous variant definition was working fine, so I'm surprised my replacement doesn't work. I'm new to mpl so maybe I'm missing something - but can't find what ! Am'I doing well ?

Thanks in advance.

Answer 1

Variant type definition is correct, the problem was due to a generic function in the program taking an arbitrary variant as parameter. Indeed, make_variant_over<mpl::vector<T0, T1, ...>> behaves like variant<T0, T1, ...> but is not the same type: it is a variant<over_sequence<vector<T0, T1, ...>>> (so T0 corresponds to over_sequence<vector<T0, T1, ...>>.