Decorators and IDisposable
c#, decorator, ioc-container, simple-injector
Solution
what should [decorators] do with regard to implementing IDisposable
This comes back to the general principle of ownership. Ask yourself: "who owns that disposable type?". The answer to this question is: He who owns the type is responsible for disposing of it.
Since a disposable type is passed on to the decorator from the outside, the decorator didn't create that type and should normally not be responsible for cleaning it up. The decorator has no way of knowing whether the type should be disposed of (since it doesn't control its lifetime) and this is very clear in your case, since the decorator can be registered as transient, while the decoratee has a much longer lifetime. In your case your system will simply break if you dispose the decoratee from within the decorator.
So the decorator should never dispose the decoratee, simply because it doesn't own the decoratee. It's the responsibility of your Composition Root to dispose that decoratee. It doesn't matter that we're talking about decorators in this case; it still comes down to the general principle of ownership.
Each decorator should only be concerned with disposing of itself and should never pass on the call to the decorated instance.
Correct. The decorator should dispose everything it owns though, but since you're using dependency injection, it typically doesn't create much stuff itself and therefore doesn't own that stuff.
Your `UnitOfWork` on the other hand creates a new `MyContext` class and therefor has the ownership of that instance and it should dispose of it.
There are exceptions to this rule, but it still comes down to ownership. Sometimes you do pass on ownership of a type to others. When using a factory method for instance, by convention the factory method passes on the ownership of the created object to the caller. Sometimes ownership is passed on to a created object, such as .NET's `StreamReader` class does. The API documentation is clear about this, but since the design is such unintuitive, developers keep tripping over this behavior. Most of the types in the .NET framework don't work this way. For instance, the `SqlCommand` class doesn't dispose the `SqlConnection`, and it would be very annoying if it did dispose of the connection.
A different way of looking at this issue is from perspective of the SOLID principles. By letting the `IUnitOfWork` implement `IDisposable` you are violating the Dependency Inversion Principle, because "Abstractions should not depend on details; Details should depend on abstractions". By implementing `IDisposable` you are leaking implementation details into the `IUnitOfWork` interface. Implementing `IDisposable` means that the class has unmanaged resources that need disposal, such as file handles and connection strings. These are implementation details, because it can't hardly ever be the case that each implementation of such interface actually needs disposal at all. You just have to create one fake or mock implementation for your unit tests and you have proof of an implementation that doesn't need disposal.
So when you fix this DIP violation by removing the `IDisposable` interface from `IUnitOfWork` -and moving it to the implementation-, it becomes impossible for the decorator to dispose the decoratee, because it has no way of knowing whether or not the decoratee implements `IDisposable`. And this is good, because according to the DIP, the decorator shouldn't know -and- we already established that the decorator should not dispose the decoratee.
Problem
I have a subclass of `DbContext` ``` public class MyContext : DbContext { } ``` and I have an `IUnitOfWork` abstraction around `MyContext` that implements `IDisposable` to ensure that references such as `MyContext` are disposed of at the appropriate time ``` public interface IUnitOfWork : IDisposable { } public class UnitOfWork : IUnitOfWork { private readonly MyContext _context; public UnitOfWork() { _context = new MyContext(); } ~UnitOfWork() { Dispose(false); } public void Dispose() { Dispose(true); GC.SuppressFinalize(this); } private bool _disposed; protected virtual void Dispose(bool disposing) { if (_disposed) return; if (disposing) { if (_context != null) _context.Dispose(); } _disposed = true; } } ``` My `UnitOfWork` is registered with a lifetime scope of per (web) request. I have decorators of `IUnitOfWork` that could be registered as transient or lifetime scoped and my question is what should they do with regard to implementing `IDisposable` - specifically should they or should they not pass on the call to `Dispose()`. ``` public class UnitOfWorkDecorator : IUnitOfWork { private readonly IUnitOfWork _decorated; public UnitOfWorkDecorator(IUnitOfWork decorated) { _decorated = decorated; } public void Dispose() { //do we pass on the call? _decorated.Dispose(); } } ``` I see 2 options (I'm guessing option 2 is the correct answer): - It is expected that each Decorator will know whether it is transient or lifetime scoped. If a decorator is transient then it should not call `Dispose()` on the decorated instance. If it is lifetime scoped it should. - Each decorator should only be concerned with disposing of itself and should never pass on the call to the decorated instance. The container will manage the call to `Dispose()` for each object in the call chain at the appropriate time. An object should only `Dispose()` of instances that it encapsulates and decorating is not encapsulation.