Search code examples
c++boosttypesboost-multi-index

C++: Determine the class of a pointer type in the use of Boost MultiIndex


The answer to this question has general application, but I will motivate it with the following example:

I have the following template class:

template <typename V>
class Collection
{
public:
    struct id{};
    struct name{};
    // A collection parameterized by type V that is indexed by a string 'name'
    // and a integer 'id'
    // The type V must implement 'get_id()' and a 'get_name()' functions
    typedef multi_index_container <
        V,
        indexed_by<
            ordered_unique<
                tag<id>, const_mem_fun<V, unsigned int, &V::get_id> >,
            ordered_unique<
                tag<name>, const_mem_fun<V, std::string, &V::get_name> >
        >
    > ni_collection;
>

I want to modify this template so that I can create a collection with the objects, their pointers or their references: Collection<Obj>, Collection<std::unique_ptr<Obj>> or Collection<Obj *>.

How would I modify my template to achieve this?

--- UPDATE --- I had posted a related question here: Computing The Type Of A Function Pointer

Synthesizing the excellent answers in both places, I have finally achieved my original goal. Here are the details of my current implementation:

template <typename V>
class Collection
{
private:
// A type-level function that returns the undecorated type of the object
// So unrwap_object_type<Widget *> = Widget
//    unwrap_object_type<std::unique_ptr<Widget>> = Widget
//    unwrap_object_type<Widget> = Widget
template<typename T, typename = void>
struct unwrap_object_type { typedef T type; };

template <typename T>
struct unwrap_object_type<T *, void> { typedef T type; };

template<typename T>
struct unwrap_object_type<T,
    typename std::conditional<false,
        typename T::element_type, void>::type>
{
  typedef typename T::element_type type;
};
////

// So that QHNETO_COLLECTION<Widget>, QHNETO_COLLECTION<Widet *>,
// and QHNETO_COLLECTION<std::unique_ptr<Widget>> are valid
typedef typename unwrap_object_type<V>::type W;

// Tags for the two indices (id and name) of the collection
struct id;
struct name;

// A collection parameterized by type V that is indexed by a string 'name'
// and a integer 'id'
// The type V must implement 'get_id()' and a 'get_name()' functions
typedef multi_index_container <
    V,
    indexed_by<
        ordered_unique<
            tag<id>,
            const_mem_fun<W, unsigned int, &W::get_id> >,
        ordered_unique<
            tag<name>,
            const_mem_fun<W, std::string, &W::get_name> >
    >
> ni_collection;

ni_collection m_collection;
};

Solution

  • Elaborating on @sehe's answer: Boost.MultiIndex predefined key extractors handle dereferencing automatically (for instance, const_mem_fun<foo,bar,&foo::bar> can be used as is with a multi_index_container of foo*s). You can take advantage of this capability and write the following (without any user-provided key extractor):

    Live On Coliru

    #include <boost/multi_index_container.hpp>
    #include <boost/multi_index/mem_fun.hpp>
    #include <boost/multi_index/ordered_index.hpp>
    #include <memory>
    
    namespace bmi = boost::multi_index;
    
    template<typename T>
    struct remove_pointer{using type=T;};
    
    template<typename T>
    struct remove_pointer<T*>{using type=T;};
    
    template<typename T>
    struct remove_pointer<std::shared_ptr<T>>{using type=T;};
    
    template <typename V> class Collection {
      public:
        struct id;
        struct name;
    
        using W=typename remove_pointer<V>::type;
    
        typedef boost::multi_index_container<
            V,
            bmi::indexed_by<
                bmi::ordered_unique<
                    bmi::tag<id>,
                    bmi::const_mem_fun<W, unsigned int, &W::get_id>
                >,
                bmi::ordered_unique<
                    bmi::tag<name>,
                    bmi::const_mem_fun<W,const std::string&, &W::get_name> 
                > 
            >
        > ni_collection;
    };
    
    struct Demo {
        unsigned _id;
        std::string _name;
    
        Demo(unsigned _id,const std::string& _name):_id(_id),_name(_name){}
    
        unsigned           get_id() const { return _id; }
        std::string const& get_name() const { return _name; }
    };
    
    int main() {
        Collection<Demo>::ni_collection works{ { 42, "LTUAE" }, { 4, "PI" } };
        Collection<Demo *>::ni_collection also_works{ new Demo{ 42, "LTUAE" }, new Demo{ 4, "PI" } };
        Collection<std::shared_ptr<Demo>>::ni_collection this_too{ std::make_shared<Demo>( 42, "LTUAE" ), std::make_shared<Demo>( 4, "PI" ) };
    }
    

    The only tricky part is that const_mem_fun uses W=std::remove_pointer<V>::type (i.e. V if V is a plain type or the type it points to if it's a pointer).

    Edited: Instead of std::remove_pointer<V>, the updated code uses a handcrafted remove_pointer template class partially specialized to understand T* and std::shared_ptr<T>; you can extend this to cover, for instance, std::unique_ptr<T> or any other smart pointer class you need to cater to.