How to safely access vector references in C++14

Question

I have a class managing vector data using the singleton pattern.

class DataManager
{
public:
    static DataManager& GetInstance() {
        static DataManager instance;
        return instance;
    }
    const vector<int>& GetData() const{return data;}
    void SetData(vector<int>& data_)
    {
        data.swap(data_);
    }
private:
    vector<int> data;
};

I need to access this data and modify it within a thread. However, if the data is released while accessing it, the program will throw an error.

I made the following attempts:

Pass the lock as a parameter:

class DataManager
{
prublic:
    const std::vector<int>& GetRects(std::unique_lock<std::mutex>& lock) const
    {
        lock = std::unique_lock<std::mutex>(m_mutex_data);
        return data;
    }
...
private:
    mutable std::mutex m_mutex_data;
}

This can work, but it's strange. Is there a better way?

Answer 1

Returning a lock and relying on the user of the class to unlock is very shaky.

You mentioned that you need to modify the data in a thread-safe way.

For this you can add a public method to DataManager that will accept a callable that will modify the data, and invoke it under the lock.

Something like (just a sketch):

class DataManager
{
public:
   template <typename T>
   void TransformDataUnderLock(T && transformer)
   {
       std::lock_guard<std::mutex> lock(m_mutex_data);
       transformer(data);
   }

   // ...

private:
   std::vector<int> data;
   mutable std::mutex m_mutex_data;
};

Notes:

1. You should also apply similar scope locking in your SetData method (or any other method that accesses data).

2. This patern might not protect against a mallicious user, but it will protect against accidental wrong usage of the accessed data and the lock (which can be much more common).

3. This pattern might cause a deadlock due to the fact that the lock is held during invocation of the transformer code which you have no control over. But as long as it is a simple transformation of the data you should be fine.