DDD: Where to keep domain Interfaces, the Infrastructure?

Question

Does it make sense to group all Interfaces of your Domain Layer (Modules, Models, Entities, Domain Services, etc) all within the Infrastructure layer? If not, does it make sense to create a "shared" project/component that groups all of these into a shared library? After all, the definition of "Infrastructure Layer" includes "shared libraries for Domain, Application, and UI layers".

I am thinking of designing my codebase around the DDD layers: UI, Application, Domain, Infrastructure. This would create 4 projects respectfully. My point is, you reference the Infrastructure Layer from the Domain Layer. But if you define the interfaces in the Domain Layer project, say for IPost, then you'll have a circulur reference when you have to reference the Domain Layer project from the Infrastructure project when you are defining the IPostRepository.Save(IPost post) method. Hence, the idea of "define all Interfaces in the Shared library".

Perhaps the repositories should not expect an object to save (IPostRepository.Save(IPost post); but instead, expect the params of the object (that could be a long set of params in the Save() though). Given, this could be an ideal situation that shows when an object is getting overly complex, and additional Value Objects should be looked into for it.

Thoughts?

Answer 1

Hi Eric,

I’m quiet new in DDD so don’t hesitate to comment if you disagree, as you I’m here to learn.

Personally I don’t understand why you should reference the infrastructure layer from your domain. In my opinion the domain shouldn’t be dependent on the infrastructure. The Domain objects should be completely ignorant on which database they are running on or which type of mail server is used to send mails. By abstracting the domain from the infrastructure it is easier to reuse; because the domain don’t know on which infrastructure its running.

What I do in my code is reference the domain layer from my infrastructure layer (but not the opposite). Repositories know the domain objects because their role is to preserve state for the domain. My repositories contains my basic CRUD operations for my root aggregates (get(id), getall(), save(object), delete(object) and are called from within my controllers.

What I did on my last project (my approach isn’t purely DDD but it worked pretty well) is that I abstracted my Repositories with interfaces. The root aggregates had to be instantiated by passing a concrete type of a Repository:

The root aggregate had to be instantiated through a repository by using the Get(ID) or a Create() method of the repository. The concrete Repository constructing the object passed itself so that the aggregate could preserve his state and the state of his child objects but without knowing anything of the concrete implementation of the repository. e.g.:

public class PostRepository:IPostRepository
{
     ...
    public Post Create()
    {
     Post post=new Post(this);
     dbContext.PostTable.Insert(post);
     return post;
     }
     public  Save(post)
     {
         Post exitingPost=GetPost(post.ID);
         existingPost = post;
         dbContext.SubmitChanges();
     }
}

public class Post
{

     private IPostRepository _repository
     internal  Post(IPostRepository repository)
     {
        _repository = repository;
     }
     ...
    Public Save()
    {
     _repository.Save(this);
     }

}

Answer 2

I would advise you to consider Onion architecture. It fits with DDD very nicely. The idea is that your repository interfaces sit in a layer just outside Domain and reference Entities directly:

IPostRepository.Save(Post post)

Domain does not need to know about repositories at all.

Infrastructure layer is not referenced by Domain, or anybody else and contains concrete implementations of repositories among other I/O-related stuff. The common library with various helpers is called Application Core in this case, and it can be referenced by anyone.

Answer 3

Eric, I was away for a couple, so forgive me for answering so late. Concerning where to put the repositories, personally I always put the repositories in a dedicated infrastructure layer (e.g . MyApp.Data.Oracle) but declare the interfaces to which the repositories have to conform to in the domain layer.
In my projects the Application Layer has to access the Domain and Infrastructure layer because it’s responsible to configure the domain and infrastructure layer.
The application layer is responsible to inject the proper infrastructure into the domain. The domain doesn’t know to which infrastructure it’s talking to, it only knows how to call it. Of course I use IOC containers like Structuremap to inject the dependencies into the domain. Again I do not state that this is the way DDD recommends to structure your projects, it’s just the way, I architecture my apps. Cheers.