序列化map<fs::path，fs::path>？

Question

我正在使用谷物库将我的类序列化为文件，但是在使用std::map时遇到了麻烦--特别是使用std::filesystem::path的地图。

假设我有一个Object类，它只包含一个map<fs::path, fs::path>，以及谷类所需的序列化函数：

struct Object
{
    map<fs::path, fs::path> _map;

    template<class Archive>
    void serialize(Archive & archive)
    {
        archive(_map);
    }

    friend ostream& operator<<(ostream& outs, const Object c)
    {
        for (auto const &pair: c._map)
            std::cout << "{" << pair.first << ": " << pair.second << "}\n";
        return outs;
    }
};

在我的main函数中，我有：

int main()
{
    // create Object
    cout << "Creating object..." << endl;
    Object o;
    fs::path a = "bye.txt";
    fs::path b = "hello.txt";
    o._map[a] = b;
    o._map[b] = a;
    cout << "Object created: " << endl;
    cout << o;

    // serialize
    cout << "Serializing object...." << endl;
    stringstream ss;
    cereal::BinaryOutputArchive oarchive(ss);
    oarchive(o);
    cout << "Object serialized." << endl;

    // write to file
    cout << "Writing serialized object to file...." << endl;
    ofstream file("serialized_object");
    file << ss.str();
    file.close();
    cout << "Object written to file." << endl;

    // read from file
    cout << "Reading from file..." << endl;
    stringstream ss2;
    fs::path ins = "serialized_object";
    ifstream file_stream(ins, ios::binary);
    ss2 << file_stream.rdbuf();
    cereal::BinaryInputArchive iarchive(ss2);
    Object out;
    iarchive(out);
    cout << "Object read from file." << endl;
    cout << out;
}

在我的输出中，我看到了从序列化文件中读取时出现的错误：

Creating object...
Object created:
{"bye.txt": "hello.txt"}
{"hello.txt": "bye.txt"}
Serializing object....
Object serialized.
Writing serialized object to file....
Object written to file.
Reading from file...
terminate called after throwing an instance of 'cereal::Exception'
  what():  Failed to read 2573 bytes from input stream! Read 28

我包含的内容包括：

#include <iostream>
#include <filesystem>
#include <fstream>
#include <string.h>
#include <map>
#include "cereal/types/map.hpp"
#include "cereal/archives/binary.hpp"
#include "cereal/types/string.hpp"

为了能够序列化fs::path，我在开头包含了以下代码

namespace std
{
  namespace filesystem
  {
    template<class Archive>
    void CEREAL_LOAD_MINIMAL_FUNCTION_NAME(const Archive&, path& out, const string& in)
    {
        out = in;
    }

    template<class Archive>
    string CEREAL_SAVE_MINIMAL_FUNCTION_NAME(const Archive& ar, const path& p)
    {
      return p.string();
    }
  }
}

我确信我遗漏了一些明显的东西，因为这是我第一次使用谷类食品。有人知道我为什么会遇到这个问题吗？

Answer 1

根据cereal文档，在使用BinaryArchive时必须始终使用ios::binary标志。以下是其文档中this页面的特定部分：

当使用二进制存档和文件流(std::fstream)时，请记住在构造流时指定二进制标志(std::ios:: binary )。这可防止流将数据解释为ASCII字符并对其进行更改。

另一个问题是Cereal如何保证序列化已经完成。根据下面的link，在尝试读取存档之前，输出存档对象应该超出作用域(调用析构函数)。它们使用RAII方法，就像ifstream和here中提到的ofstream一样。

通过允许Cereal库在内部执行写入和读取操作，您可以极大地简化代码。以下是main函数的修改版本：

int main()
{
    // create Object
    cout << "Creating object..." << endl;
    Object o;
    fs::path a = "bye.txt";
    fs::path b = "hello.txt";
    o._map[a] = b;
    o._map[b] = a;
    cout << "Object created: " << endl;
    cout << o;

    
    cout << "Serializing object...." << endl;
    // scope boundary (a function call would be cleaner)
    {
        ofstream ofstm("serialized_object", ios::binary);
        cereal::BinaryOutputArchive oarchive(ofstm);
        // serialize
        oarchive(o);
    }
    cout << "Object serialized and written." << endl;

    // deserialize the object by reading from the archive
    Object out;

    // scope boundary (a function would be cleaner)
    {
        ifstream istm("serialized_object", ios::binary);
        cereal::BinaryInputArchive iarchive(istm);
        //deserialize
        iarchive(out);
    }
    cout << "Object read from file." << endl;
    cout << out;
}

我希望这能帮到你。请在使用前对其进行测试。