Question

I’ve noticed that a lot of developers define an interface for EVERY class that is going to be injected using DI framework. What are the advantages of defining Interfaces for every class?

Answer 1

Letting your application components (the classes that contain the application logic) implement an interface is important, since this promotes the concept of:

Program to an interface, not an implementation.

This is effectively the Dependency Inversion Principle (DIP). Doing so allows you to replace, intercept or decorate dependencies without the need to change consumers of such dependency.

But still, the DIP is not the only principle to consider. In many cases developers will be violating other parts of the SOLID principles (DIP is one out of five SOLID principles) when having primarily one-to-one mappings between classes and an interfaces in their applications. One of the principles that is almost certainly violated in that case is the Open/closed principle, because when every class has its own interface, it is not possible to extend (decorate) a set of classes with cross-cutting concerns (without dynamic proxy generation trickery that is).

In the systems I write, I define two generic interfaces that cover the bulk of the code of the business layer. They are called ICommandHandler<TCommand> and an IQueryHandler<TQuery, TResult>:

public interface ICommandHandler<TCommand>
{
    void Handle(TCommand command);
}

public interface IQueryHandler<TQuery, TResult> where TQuery : IQuery<TResult>
{
    TResult Handle(TQuery query);
}

Besides the nice side effect of not having to define many interfaces, this allows great flexibility and ease of testing. You can read more about it here and here.

Depending on the system I write, I might also use interfaces such as:

• IValidator<T> for validating messages
• ISecurityValidator<T> for applying security restrictions on messages
• IRepository<T>, the repository pattern
• IAuthorizationFilter<T> for applying authorization/security filtering on IQueryable<T> queries.

Depending on the system I write, somewhere between 80% and 98% of all components implement one of these generic interfaces I define. This makes applying most cross-cutting concerns to those so called joinpoints trivial.