存储库模式通用应用

Question

When I started learning Repository Pattern with Unity few days ago I was under impression that the main benefit of this pattern is the separation of data layer from the business layer.

In other words, if there is a need to change the way, how application stores the data, it's very easy as only one main model takes care of the communication.

This means, that if application currently saves data into a serialized XML files, it would not be very difficult to change this logic to connect to database instead.

I have found few nice demos that are also using Unit Of Work layer, which seemed very handy. Let me show you a bit of the code I have.

public class UnitOfWork : IUnitOfWork
{
    private readonly RepositoryContext _context;
    public IEmployeeRepository Employees { get; set; }

    public UnitOfWork(RepositoryContext context)
    {
        _context = context;
        Employees = new EmployeeRepository(_context);
    }


    public int Complete()
    {
        return _context.SaveChanges();
    }

    public void Dispose()
    {
        _context.Dispose();
    }
}

Main Repository Context:

public class RepositoryContext : DbContext
{
    public RepositoryContext() : base("name=RepositoryContext")
    {
    }

    public virtual DbSet<Employee> Employees { get; set; }
    public virtual DbSet<Equipment> Furniture { get; set; }
}

And here is the demo EmployeeRepository:

public class EmployeeRepository:Repository<Employee>, IEmployeeRepository
{
    public EmployeeRepository(RepositoryContext context) : base(context) { }

    public Employee GetEmployeeByName(string sName)
    {
        return MyContext.Employees.FirstOrDefault(n => n.Name == sName);
    }

    public RepositoryContext MyContext
    {
        get { return Context as RepositoryContext; }
    }
}

Employee Repository derives from a generic Repository which looks like this:

public class Repository<T> : Interfaces.Repositories.IRepository<T> where T : class
{
    protected readonly DbContext Context;

    public Repository(DbContext context)
    {
        Context = context;
    }

    public void Add(T item)
    {
        Context.Set<T>().Add(item);
    }

    public IEnumerable<T> Find(Expression<Func<T, bool>> predicate)
    {
        return Context.Set<T>().Where(predicate);
    }

    public T Get(int ID)
    {
        return Context.Set<T>().Find(ID);
    }

    public IEnumerable<T> GetAll()
    {
        return Context.Set<T>().ToList();
    }

    public void Remove(T item)
    {
        Context.Set<T>().Remove(item);
    }
}

Here is the question:

As far as my understanding goes, we are directly declaring, that under our Repository expects in it's constructor DbContext , which is afterwards used under all Add / Remove / Find functions under that particular class.

Currently this model is communicating with the database, but if I wanted (for whatever reason) to change this model to save data in the XML file, I would have to completely rewrite all my Repository classes? Or am I missing something here?

If I am wrong and it is easily doable, could anyone show me how to change the code so that we are serializing values into the XML files, please? I am trying to better understand this Repository Pattern, yet for now it's one big chaos for me.

Any help / suggestions regarding this matter would be highly appreciated.

Answer 1

I'm reading the question as this:

How can I abstract DbContext so there are no dependencies to it?

I would abstract the context to an interface in an effort to embrace the Dependency inversion principle .

public interface IDbContext : IDisposable
{
    int SaveChanges();
    IDbSet<Employee> Employees { get; set; }
    IDbSet<Equipment> Furniture { get; set; }
}

public class RepositoryContext : DbContext, IDbContext
{
    public RepositoryContext() : base("name=RepositoryContext")
    {
    }

    public virtual DbSet<Employee> Employees { get; set; }
    public virtual DbSet<Equipment> Furniture { get; set; }
}

Then try to inject the IDbContext interface instead. As mentioned in your comments you will probably need to rewrite parts of your repo anyway, but if the new datalayer can expose an IDbSet you should be able to simply change the implementation of IDbContext .

public class Repository<T> : Interfaces.Repositories.IRepository<T> where T : class
{
    protected readonly IDbContext Context;

    public Repository(IDbContext context)
    {
        Context = context;
    }

    public void Add(T item)
    {
        Context.Set<T>().Add(item);
    }

    public IEnumerable<T> Find(Expression<Func<T, bool>> predicate)
    {
        return Context.Set<T>().Where(predicate);
    }

    public T Get(int ID)
    {
        return Context.Set<T>().Find(ID);
    }

    public IEnumerable<T> GetAll()
    {
        return Context.Set<T>().ToList();
    }

    public void Remove(T item)
    {
        Context.Set<T>().Remove(item);
    }
}

I would also look at the possibility to abstract the creating of the context in a separate class. As mentioned here: Entity Framework using Repository Pattern, Unit of Work and Unity

public interface IDbContextFactory
{
    IDbContext GetContext();
}

public class DbContextFactory : IDbContextFactory
{
    private readonly IDbContext _context;

    public DbContextFactory()
    {
        _context = new MyDbContext("ConnectionStringName");
    }

    public IDbContext GetContext()
    {
        return _context;
    }
}

This way you can inject IDbContextFactory into the Unit of work. Then you have abstracted the DbContext to the DbContextFactory , but you still have a dependency in DbContextFactory to DbContext . This will be enough for most people, but if you want to really go SOLID then you can abstract that as well with a generic IInstanceFactory .

    public interface IDbContextFactory
    {
        /// <summary>
        /// Creates a new context.
        /// </summary>
        /// <returns></returns>
        IDbContext GenerateContext();

        /// <summary>
        /// Returns the previously created context.
        /// </summary>
        /// <returns></returns>
        IDbContext GetCurrentContext();
    }

    public class DbContextFactory : IDbContextFactory
    {
        private readonly IInstanceFactory _instanceFactory;
        private IDbContext _context;

        public DbContextFactory(IInstanceFactory instanceFactory)
        {
            _instanceFactory = instanceFactory;
        }

        public IDbContext GenerateContext()
        {
            _context = _instanceFactory.CreateInstance<IDbContext>();
            return _context;
        }

        public IDbContext GetCurrentContext()
        {
            if (_context == null)
                _context = GenerateContext();
            return _context;
        }
    }

    /// <summary>
    /// Creates an instance of a specific model.
    /// </summary>
    public interface IInstanceFactory
    {
        /// <summary>
        /// Creates an instance of type T.
        /// </summary>
        T CreateInstance<T>();
    }

    /// <summary>
    /// Creates an instance based on the model defined by Unity.
    /// </summary>
    public class InstanceFactory : IInstanceFactory
    {
        private readonly IDictionary<Type, Func<object>> _funcs;

        public InstanceFactory(IEnumerable<Func<object>> createFunc)
        {
            // To remove the dependency to Unity we will receive a list of funcs that will create the instance.

            _funcs = new Dictionary<Type, Func<object>>();

            foreach (var func in createFunc)
            {
                var type = func.Method.ReturnType;
                _funcs.Add(type, func);
            }
        }

        /// <summary>
        /// Creates an instance of T.
        /// </summary>
        /// <typeparam name="T"></typeparam>
        /// <returns></returns>
        public T CreateInstance<T>()
        {
            var func = _funcs[typeof(T)];
            return (T) func();
        }
    }

And in my Unity registrations:

    container.RegisterType<IDbContext, YourDbContext>(new TransientLifetimeManager());
    container.RegisterType<IInstanceFactory, InstanceFactory>(
            new InjectionConstructor(new List<Func<object>>
            {
                new Func<IDbContext>(() => container.Resolve<IDbContext>())
            }
        ));

Now you have abstracted the DbContext all the way to the IoC, which in theory can be changed in web.config without even re-building. Considerations? Well, think about readability vs maintainability. I prefer a really abstracted layer, while others will argue that it's not necessary as EF already is a Unit of Work-pattern. Also, there will probably be an performance overhead instead of just creating the DbContext as you do now. From a more philosophical point of view one may argue that the abstraction of DbContext will be a Unit of Work itself since it's now at the abstracted layer with a SaveChanges() that can be "passed around" just as a Unit of Work. But i leave that discussion to you...

Most of this is wrote by hand, but I hope it will help you on the way, if you decide to abstract the DbContext as well.

EDIT: Added SaveChanges() to IDbContext .