Containers.md

Resolver: Containers

What are containers?

In a Dependency Injection system, a container contains all of the service registrations. When a service is being resolved, the container is searched to find the correct registration and corresponding factory.

In Resolver, a resolver instance contains its registration code, its resolution code, and a corresponding container. Put another way, each and every instance of a Resolver is a container.

Resolver's Main Container

Inspect Resolver's code and you'll see the following.

public final class Resolver {
    public static let main: Resolver = Resolver()
    public static var root: Resolver = main
}

Resolver creates a main container that it uses as its default container for all static registrations. It also defines a root contrainer that defaults to pointing to the main container.

Static Registration Functions

This basically means that when you do....

extension Resolver: ResolverRegistering {
    static func registerAllServices() {
        register { XYZNetworkService(session: resolve()) }
        register { XYZSessionService() }
    }
}

You're effectively doing...

extension Resolver: ResolverRegistering {
    static func registerAllServices() {
        main.register { XYZNetworkService(session: root.resolve()) }
        main.register { XYZSessionService() }
    }
}

The static (class) register and resolve functions simply pass the buck to main and root, respectively.

public static func register<Service>(_ type: Service.Type = Service.self, name: Resolver.Name? = nil,
                                     factory: @escaping ResolverFactoryArgumentsN<Service>) -> ResolverOptions<Service> {
    return main.register(type, name: name, factory: factory)
}

public static func resolve<Service>(_ type: Service.Type = Service.self, name: String? = nil, args: Any? = nil) -> Service {
    return root.resolve(type, name: name, args: args)
}

Creating your own containers

Creating your own container is simple, and similar to creating your own scope caches.

extension Resolver {
    static let mock = Resolver()
}

It could then be used as follows.

extension Resolver: ResolverRegistering {
    static func registerAllServices() {
        mock.register { XYZNetworkService(session: mock.resolve()) }
        mock.register { XYZSessionService() }
    }
}

class MyClass {
    var session: XYZNetworkService = Resolver.mock.resolve()
}

By itself, however, Resolver's main container can handle thousands of registrations, so the above, while possible, doesn't tend to be particularly useful. On its own, that is.

Nested Containers

So once again, and just to be perfectly clear, when you do Resolver.resolve(SomeClass.self), you're effectively doing Resolver.root.resolve(SomeClass.self).

This implies that if root were to point to a different container, like our mock container above, that container would then become the default container used for all resolutions. While true, however, that also means that any services registered in main are now lost.

Now consider the following:

extension Resolver {
    static let mock = Resolver(child: main)

    static func registerAllServices() {
        register { XYZNetworkService(session: resolve()) }
        register { XYZSessionService() }
        
        #if DEBUG
        mock.register { XYZMockSessionService() as XYZSessionService }
        root = mock
        #end
    }
}

Now, when in DEBUG mode, root is switched out and points to the new parent container mock. When a service is resolved, mock will now be searched first.

This means that the XYZSessionService service we just defined in mock will be used over any matching service defined in main.

If a service is not found in mock, the main child container will be searched automatically, thanks to our adding the child parameter to mock = Resolver(child: main).

This lookup mechanism is recursive. Should main not have what we're looking for, any child containers that might have been added to it would be searched, effectively resulting in a depth-first tree search.

Party Trick?

One might ask why we simply don't do the following:

extension Resolver {
    static func registerAllServices() {
        register { XYZNetworkService(session: resolve()) }
        register { XYZSessionService() }

        #if DEBUG
        mock.register { XYZMockSessionService() as XYZSessionService }
        #end
    }
}

Here, if we're in DEBUG mode our later registration of XYZSessionService overwrites the earler one and the mocked version is constructed and used instead.

Isn't swapping out the containers just a party trick?

But what if, for example, we want to keep both registrations and use the proper one at the proper time?

Consider the following:

extension Resolver {
    #if DEBUG
    static let mock = Resolver(child: main)
    #end
    
    static func registerAllServices() {
        register { XYZNetworkService(session: resolve()) }
        register { XYZSessionService() }

        #if DEBUG
        mock.register { XYZMockSessionService() as XYZSessionService }
        #end
    }
}

And then somewhere in our code we do this before we enter a given section:

#if DEBUG
Resolver.root = Resolver.mock
#end

And then when exiting that section:

#if DEBUG
Resolver.root = Resolver.main
#end

Now the app behaves normally up until we enter that section of our app. It then switches to using the mocked services for all injections past that point.

Returning, we switch back and the app again behaves normally.

Nice party trick, don't you think?

Multiple Child Containers

Resolver 1.4.3 adds support for multiple child containers.

As stated above, you can put thousands of registrations into a single container but, should you desire to do so, you can now segment your registrations into smaller groups of containiners and then add each subcontainer to the main container.

Consider...

extension Resolver {
    static let containerA = Resolver()
    static let containerB = Resolver()

    static func registerAllServices() {
        main.add(child: containerA)
        main.add(child: containerB)
        ...
    }
}

Now when main is asked to resolve a given service, it will first search its own registrations and then, if not found, will search each of the included child containers to see if one of them contains the needed registration. First match will return, and containers will be searched in the order in which they're added.

This is basically a small change that reworks the search mechanism to support multiple children. Nested (or "parent/child") containers still work as before.