模板函数重载无法在递归模板调用中调用。

Question

对于某些模板类型，我遇到了获取适当的函数模板重载要执行的问题。下面显示了我所遇到的最小示例：

#include <cstdio>
#include <vector>

template<typename id_type>
struct B {
    id_type ID;
    std::vector<int> values;
};

template<typename id_type>
struct A {
    id_type ID;
    std::vector<struct B<id_type>> b_elems;
};

// forward declarations
namespace aSDG {
    namespace meshing {
        template<typename id_type> size_t byte_content(const struct B<id_type>& instance);
        template<typename id_type> size_t serialize(const struct B<id_type>& instance, unsigned char* buffer, size_t start_idx = 0);
        template<typename id_type> size_t deserialize(struct B<id_type>& instance, const unsigned char* buffer, size_t start_idx = 0);
        template<typename id_type> size_t byte_content(const struct A<id_type>& instance);
        template<typename id_type> size_t serialize(const struct A<id_type>& instance, unsigned char* buffer, size_t start_idx = 0);
        template<typename id_type> size_t deserialize(struct A<id_type>& instance, const unsigned char* buffer, size_t start_idx = 0);
    }
}

namespace aSDG {
    namespace meshing {

        // serialization for primitive types
        template<typename T> size_t byte_content(const T& data){
            return sizeof(T);
        }

        template<typename T> size_t serialize(const T& data, unsigned char* buffer, size_t start_idx = 0)
        {
            std::memcpy((void*)(buffer + start_idx), (void*)&data, sizeof(data));
            return start_idx + sizeof(data);
        }
        template<typename T> size_t deserialize(T& data, const unsigned char* buffer, size_t start_idx = 0)
        {
            std::memcpy((void*)&data, (void*)(buffer + start_idx), sizeof(data));
            return start_idx + sizeof(data);
        }

        // serialization for vector containers
        template<typename T> size_t byte_content(const std::vector<T>& data){

            // get number of bytes for the size variable
            size_t num_req_bytes = sizeof(size_t);

            // get the number of bytes for each element of the vector
            for(size_t i = 0; i < data.size(); ++i){
                num_req_bytes += byte_content(data[i]);
            }// end for i

            // return the total number of required bytes
            return num_req_bytes;
        }

        template<typename T> size_t serialize(const std::vector<T>& data, unsigned char* buffer, size_t start_idx = 0)
        {
            // add the number of elements in the data
            const size_t size_ = data.size();
            start_idx = serialize(size_, buffer, start_idx);

            // add the actual data elements
            for(size_t i = 0; i < size_; ++i){
                start_idx = serialize(data[i], buffer, start_idx);
            }// end for i

            // return the final index after adding all the data
            return start_idx;
        }

        template<typename T> size_t deserialize(std::vector<T>& data, const unsigned char* buffer, size_t start_idx = 0)
        {
            // get the number of elements in the array
            size_t size_ = 0;
            start_idx = deserialize(size_, buffer, start_idx);

            // resize the input array
            data.resize(size_);

            // fill the array with the data in the buffer
            for(size_t i = 0; i < size_; ++i){
                start_idx = deserialize(data[i], buffer, start_idx);
            }// end for i

            // return the number of bytes we are at in the array
            return start_idx;
        }

    } // end namespace meshing
} // end namespace aSDG

namespace aSDG {
    namespace meshing {

        // serialization for B
        template<typename id_type>
        size_t byte_content(const struct B<id_type>& instance) {
            return byte_content(instance.ID) + byte_content(instance.values);
        }

        template<typename id_type>
        size_t serialize(const struct B<id_type>& instance, unsigned char* buffer, size_t start_idx){
            start_idx = serialize(instance.ID, buffer, start_idx);
            return serialize(instance.values, buffer, start_idx);
        }

        template<typename id_type>
        size_t deserialize(struct B<id_type>& instance, const unsigned char* buffer, size_t start_idx){
            start_idx = deserialize(instance.ID, buffer, start_idx);
            return deserialize(instance.values, buffer, start_idx);
        }

        // serialization functions for A
        template<typename id_type>
        size_t byte_content(const struct A<id_type>& instance) {
            return byte_content(instance.ID) + byte_content(instance.b_elems);
        }

        template<typename id_type>
        size_t serialize(const struct A<id_type>& instance, unsigned char* buffer, size_t start_idx){
            start_idx = serialize(instance.ID, buffer, start_idx);
            return serialize(instance.b_elems, buffer, start_idx);
        }

        template<typename id_type>
        size_t deserialize(struct A<id_type>& instance, const unsigned char* buffer, size_t start_idx){
            start_idx = deserialize(instance.ID, buffer, start_idx);
            return deserialize(instance.b_elems, buffer, start_idx);
        }


    } // end namespace meshing
} // end namespace aSDG



int main(int argc, const char * argv[]) {

    struct A<size_t> a1, a2;
    a1.b_elems.emplace_back();
    a1.b_elems.emplace_back();
    a1.b_elems.emplace_back();
    a1.b_elems[0].ID = 5;
    a1.b_elems[0].values.push_back(1);

    // get the number of bytes to be serialized
    size_t num_req_bytes = aSDG::meshing::byte_content(a1);

    // allocate the buffer
    std::vector<unsigned char> buf( num_req_bytes );

    // serialize the data in a1
    size_t serial_bytes = aSDG::meshing::serialize(a1, &buf[0]);

    // deserialize data into a2
    size_t deserial_bytes= aSDG::meshing::deserialize(a2, &buf[0]);

    // check that the bytes match
    printf("in_bytes = %zu vs. out_bytes = %zu\n", serial_bytes, deserial_bytes );

    return 0;
}

在本例中，我将序列化A类型的实例，而这种序列化又需要序列化A中包含的B元素的向量。A的所有序列化函数都会运行，这意味着它的风格为byte_content、serialize和deserialize，并使用适当的定义进行调用。但是，当程序递归到这些方法的通用std::vector定义以序列化A的std::vector<struct B>数据成员时，它无法调用为B定义的方法，而是调用基本原语的序列化函数(在代码示例顶部定义的前三个方法)。我不明白为什么在这种情况下不调用B的序列化方法(byte_content、serialize、deserialize)，因为它们是定义的。

我怀疑我缺少一些如何选择函数模板重载的基本规则，但我真的不确定。任何洞察力都将不胜感激。

编辑1

更准确地说，关键问题是当A的序列化发生时，它实际上将调用下面预期的方法

template<typename id_type>
size_t aSDG::meshing::serialize(const struct A<id_type>& instance, unsigned char* buffer, size_t start_idx = 0){
    start_idx = serialize(instance.ID, buffer, start_idx);
    return serialize(instance.b_elems, buffer, start_idx);
}

问题是，当它序列化b_elems时，它首先使用T = struct B调用泛型std::vector序列化方法。

template<typename T> size_t serialize(const std::vector<T>& data, unsigned char* buffer, size_t start_idx = 0)
{
    // add the number of elements in the data
    const size_t size_ = data.size();
    start_idx = serialize(size_, buffer, start_idx);

    // add the actual data elements
    for(size_t i = 0; i < size_; ++i){
        start_idx = serialize(data[i], buffer, start_idx);
    }// end for i

    // return the final index after adding all the data
    return start_idx;
}

但是，当它执行serialize(data[i], buffer, start_idx)时，函数不会调用

template<typename id_type>
size_t serialize(const struct B<id_type>& instance, unsigned char* buffer, size_t start_idx = 0){
    start_idx = serialize(instance.ID, buffer, start_idx);
    return serialize(instance.values, buffer, start_idx);
}

但相反，打电话

template<typename T> size_t serialize(const T& data, unsigned char* buffer, size_t start_idx = 0)
{
    std::memcpy((void*)(buffer + start_idx), (void*)&data, sizeof(data));
    return start_idx + sizeof(data);
}

我真的很困惑为什么会发生这种情况。

编辑2

在添加了@Evg推荐的前向声明之后，代码几乎和我预期的一样工作。现在唯一的问题是没有调用byte_content专门化的B。可以通过以下方法验证这一点:将B的上述专门化定义替换为

template<typename id_type>
size_t byte_content(const struct B<id_type>& instance) {
    printf("B byte_content\n");
    return byte_content(instance.ID) + byte_content(instance.values);
}

template<typename id_type>
size_t serialize(const struct B<id_type>& instance, unsigned char* buffer, size_t start_idx){
    printf("B serialize\n");
    start_idx = serialize(instance.ID, buffer, start_idx);
    return serialize(instance.values, buffer, start_idx);
}

template<typename id_type>
size_t deserialize(struct B<id_type>& instance, const unsigned char* buffer, size_t start_idx){
    printf("B deserialize\n");
    start_idx = deserialize(instance.ID, buffer, start_idx);
    return deserialize(instance.values, buffer, start_idx);
}

并证明"B byte_content“消息从未显示过。现在，也许我只是累了，没有看到一些错误，但我不明白为什么即使在前向声明之后，B的正确的B专门化也没有被调用。

Answer 1

注:这个答案指的是编辑前的问题(没有前向声明)。

在serialize(const std::vector<T>& data...)内部，您使用不限定的名称serialize。编译器应该决定要调用哪个serialize。它将考虑在定义点上可见的函数1，以及ADL在实例化点上可以找到的函数。两个查找都将无法找到serialize(const B<id_type>&...)。

一个可能的解决办法是提出声明。

template<typename id_type>
size_t byte_content(const B<id_type>&);

template<typename id_type>
size_t serialize(const B<id_type>&, unsigned char*, size_t = 0);

template<typename id_type>
size_t deserialize(B<id_type>&, const unsigned char*, size_t = 0);

刚开始的时候。