How to using boost::multi_index with struct in struct?

Question

I have a vector containing information called ST_ThepInfo.

My problem is when using struct ST_ThepInfo in struct Infovalue_t.

struct ST_ThepInfo
{
    int length;
    string ex;
    int weight;
};

struct Infovalue_t {
    ST_ThepInfo s;
    int i;
};
struct ST_ThepInfo_tag {};

   typedef boost::multi_index_container<
    Infovalue_t,
    boost::multi_index::indexed_by<
    boost::multi_index::random_access<>, // this index represents insertion order
    boost::multi_index::hashed_unique<
        boost::multi_index::tag<ST_ThepInfo_tag>,
        boost::multi_index::member<Infovalue_t, ST_ThepInfo,&Infovalue_t::s>>
    >
> myvalues_t;

then i call these code:

myvalues_t s;
ST_ThepInfo k;
....
auto t = count.emplaceback(k, 0); 

However I get an error like this:

How do I fix it?

Answer 1

Hashed indexes require the key type to be hashable and equality-comparable.

You need to provide these for the info struct:

Live On Coliru

#include <boost/multi_index/hashed_index.hpp>
#include <boost/multi_index/member.hpp>
#include <boost/multi_index/random_access_index.hpp>
#include <boost/multi_index_container.hpp>
namespace bmi = boost::multi_index;

struct STInfo {
    int         length;
    std::string ex;
    int         weight;

    auto key_fields() const { return std::tie(length, ex, weight); }

    friend size_t hash_value(STInfo const& info) {
        using boost::hash_value;
        return hash_value(info.key_fields());
    }

    bool operator==(STInfo const& other) const {
        return key_fields() == other.key_fields();
    }
};

struct Infovalue_t {
    STInfo s;
    int    i;
};

using Table = bmi::multi_index_container< //
    Infovalue_t,                          //
    bmi::indexed_by<                      //
        bmi::random_access<>,             // represents insertion order
        bmi::hashed_unique<               //
            bmi::tag<struct byInfo>,      //
            bmi::member<Infovalue_t, STInfo, &Infovalue_t::s>> //
        >>;

#include <fmt/ranges.h>
#include <fmt/ostream.h>
struct Format : fmt::formatter<int> {
    auto format(STInfo const& info, auto& ctx) const {
        return fmt::format_to(ctx.out(), "{}", info.key_fields());
    }
    auto format(Infovalue_t const& iv, auto& ctx) const {
        return fmt::format_to(ctx.out(), "({}, {})", iv.s, iv.i);
    }
};
template<> struct fmt::formatter<Infovalue_t> : Format{};
template<> struct fmt::formatter<STInfo>      : Format{};

int main() {
    Table  count;
    STInfo k;
    count.push_back({STInfo{42, "LtUaE", 99}, 30});
    count.push_back({STInfo{43, "SomethingElse", 98}, 40});
    count.push_back({STInfo{44, "SomethingElse", 97}, 30});

    fmt::print("{}\n", count);
}

Prints:

[((42, "LtUaE", 99), 30), ((43, "SomethingElse", 98), 40), ((44, "SomethingElse", 97), 30)]

Note that it is very important that equality matches hashing. If they don't agree, there's going to be Undefined Behaviour. This is the main reason why I don't recommend defaulting the equality operator as you can in c++20:

#ifdef __cpp_impl_three_way_comparison
    auto operator<=>(STInfo const&) const = default;
#endif

This makes it less explicit that hash_value needs to agree with the members and risks they go out of sync when you e.g. add a member.