Turning #ifdef's into a template metaprogram in C++

Question

I have the following code inside a C++ class:

class Features
{
    #define Feature_Size_A 12345
    #define Feature_Size_B 45678
    #define Feature_Size_C 78901
    //#define Feature_Size_D 14725

    const int Feature_Sum = 0
    #ifdef Feature_Size_A
        + Feature_Size_A
    #endif
    #ifdef Feature_Size_B
        + Feature_Size_B
    #endif
    #ifdef Feature_Size_C
        + Feature_Size_C
    #endif
    #ifdef Feature_Size_D
        + Feature_Size_D
    #endif
        ;

    #ifdef Feature_Size_A
        static float Feature_A[Feature_Size_A];
    #endif
    #ifdef Feature_Size_B
        static float Feature_B[Feature_Size_B];
    #endif
    #ifdef Feature_Size_C
        static float Feature_C[Feature_Size_C];
    #endif
    #ifdef Feature_Size_D
        static float Feature_D[Feature_Size_D];
    #endif
};

I used to comment out features, like line 4, to compile and run different tests. But now I'd like to have the class as a template, so I can instantiate several versions with different features turned on or off in the same program.

I'm thinking of something like this:

template <bool Feature_A, bool Feature_B, bool Feature_C, bool Feature_D>
class Features
{
    ...
};

Features<true, true, true, false> f;

I tried with boost::mpl:vector's but I'm struggling harshly.

BTW: This is not the complete code. The original code has 25 features.

I'm thankful for every idea not involving macros :-)

Answer 1

Type lists can be used to solve this problem.

template<unsigned num, unsigned size, typename T>
class Feature : public T
{
public:
    static float feature[size];
    static const unsigned int feature_sum = size + T::feature_sum;
};
template<unsigned num, unsigned size, typename T>
float Feature<num, size, T>::feature[size];
class Tail { 
public:
    static const unsigned feature_sum = 0; 
};

template<unsigned num, unsigned size, typename T>
float* get_feature_arr(Feature<num, size, T>& ref) 
{
    return ref.feature;
}

int main() {
    Feature<1, 12345, Feature<2, 45678, Feature<4, 78901, Tail>>> TripleFeatures;
    auto first = get_feature_arr<1>(TripleFeatures);
    auto third = get_feature_arr<4>(TripleFeatures);
    auto size = TripleFeatures.feature_sum;
}

This can also be used to access any feature, regardless of what other features are or aren't in the list.

Edit: I nubbed up some of the details, like not defining the array and trying to have "3features" as an identifier. Le fixed. Code compiles GCC 4.7.1.

Answer 2

Why not use statically allocated arrays?

#include <stdio.h>

template <bool Feature_A, bool Feature_B, bool Feature_C, bool Feature_D>
class Features
{
    static const int Feature_Size_A = 12345;
    static const int Feature_Size_B = 45678;
    static const int Feature_Size_C = 78901;
    static const int Feature_Size_D = 14725;
    static const int Feature_Sum = 0
        + Feature_A ? Feature_Size_A : 0
        + Feature_B ? Feature_Size_B : 0
        + Feature_C ? Feature_Size_C : 0
        + Feature_D ? Feature_Size_D : 0
    ;

public:
    static float Feature_Vector_A[Feature_A ? Feature_Size_A : 0];
    static float Feature_Vector_B[Feature_B ? Feature_Size_B : 0];
    static float Feature_Vector_C[Feature_C ? Feature_Size_C : 0];
    static float Feature_Vector_D[Feature_D ? Feature_Size_D : 0];
};

Features<true, true, true, true> f1;
Features<true, true, true, false> f2;

int main()
{
    printf("%d %d\n", sizeof(f1.Feature_Vector_D), sizeof(f2.Feature_Vector_D));
}

Output:

58900 0

Answer 3

It is not clear exactly what the features will be exactly, but here is a solution that allows you to conditionnaly include member functions as well as member data:

namespace mpl = boost::mpl;

// Define your features
struct FeatureA
{
    static const int size = 12345;
    static float Feature_A[size];

    static void methodA() {}
};
float FeatureA::Feature_A[12345];

struct FeatureB
{
    static const int size = 45678;
    static char Feature_B[size]; // possibly different types of data (?)

    static void methodB() {}
};
float FeatureB::Feature_B[45678];

struct FeatureC
{
    static const int size = 78901;
    static int Feature_C[size];

    static void methodC() {}
};
float FeatureC::Feature_C[78901];


// Helper metafunction
template <typename T>
struct get_size 
  : mpl::int_<T::size>
{};


template <typename FeaturesSeq>
struct Features_impl 
  : mpl::inherit_linearly<
        FeaturesSeq,
        mpl::inherit<mpl::_, mpl::_>
    >::type
{
    static const int Feature_Sum = 
        mpl::accumulate<
            FeaturesSeq,
            mpl::int_<0>,
            mpl::plus<
                mpl::_1,
                get_size<mpl::_2>
            >
        >::type::value;
};

template <typename... F>
using Features = Features_impl<mpl::vector<F...>>;


#include <iostream>

int main()
{
    typedef Features<FeatureA, FeatureC> F;

    std::cout << F::Feature_Sum << '\n';

    F::Feature_A[0] = 12.0f;
    F::methodA();

    F::methodC();
}

If all your features are really just float arrays, as in your example, you can use a generic Feature class

template <int Size>
struct Feature
{
    static float data[Size];
};
template <int Size>
float Feature::data[Size];

And store specializations of this class into mpl vectors:

typedef mpl::vector<Feature<1234>, Feature<5678>> Features;

mpl::at_c<Features, 0>::type::data[0] = 12.0f;
// could be encapsulated into a helper function

Without more information about the purpose of these so-called features, it is hard to provide a more complete answer.