Creating your Features and Actions

Learn how to define your app's functionality in terms of features and how to perform the actions.


The definition of your app’s Features and the Actions that they provide is the fundamental starting point of Flint’s implementation of Feature Driven Development.

When starting out, if you are on iOS it can be very useful to use the FlintUI framework to add a debug Feature Browser so you can visualise what your app’s Feature graph looks like.

Defining your first Feature Group

Your application normally has a primary feature group that has all the root-level features your app supports. You pass this to Flint when you set it up during application startup.

It is a class that you can give any name, but typically you would use something like AppFeatures. It just needs to conform to FeatureGroup and define a list of subfeatures. We’ll start with that list empty:

final class AppFeatures: FeatureGroup {
    static var description = "My main app features"
    static var subfeatures: [FeatureDefinition.Type] = []

You use this when calling setup or quickSetup at startup:

func application(_ application: UIApplication, didFinishLaunchingWithOptions launchOptions: [UIApplicationLaunchOptionsKey: Any]?) -> Bool {

This will get Flint examining all your features and their actions, and setting up all the other conventions including the URL routes.

You can register more than one root-level feature group, for example if you have some frameworks that also expose features. Register these manually with Flint.register(). You should only call setup() or quickSetup() once.

Defining your first Feature

Now you can define a real feature and add it to the root feature’s subfeatures.

A single Feature in Flint lets you describe some functionality that your app offers, the actions that can be performed on that feature, and other conventions that alter app behaviour. Information about this feature is output in logs and you can use Focus to limit all logging to only a subset of features.

A regular Feature is always available, it cannot be turned off. This means your code can always perform the actions that belong to it, without you having to think twice. Later we will cover conditional features that can be feature flagged or enabled only through purchases or user preferences.

class DocumentManagementFeature: Feature {
    static let description = "Create, Open and Save documents"

    static func prepare(actions: FeatureActionsBuilder) {
    	// We'll add actions here soon

Note that there are more things required by the Feature protocol, but the class will receive default inmplementations of most requirements for the Feature and FeatureDefinition protocols by the magic of Swift protocol extensions.

Once you have this in place you can define an action and hook it up to the Feature using an action binding.

Note that it seems to work out nicely if you have an Xcode project Group for each Feature, with the Feature’s .swift file in there, along side other types needed for the feature, such as the Action(s) and Presenter(s).

Now we have to go back and edit the feature group to include this subfeature:

final class AppFeatures: FeatureGroup {
    static var description = "My main app features"
    static var subfeatures: [FeatureDefinition.Type] = [

That’s it for the feature part.

Defining your first Action

We’ve now got a feature so let’s add an Action. An Action in Flint is a small piece of logic that represents something the application can do – usually in response to some user-driven event or external stimulus (such as a push notification or location change). Sometimes you will need actions that are not directly invoked by the user, but are required because you need to track when something happens in the app, or you need to respond to a URL.

An Action receives a single input and a presenter, which it will use to perform its work and then update the presentation. The input is any type you specify when defining the action, as is the presenter. Using the power of Swift associated types, we will then restrict the inputs and presenters passed to the action to only those types.

What we need to do here is receive a reference to a document as input, and then pass it to the presenter for display. We’re not actually loading the document at this point, for simplicity.

Let’s implement it.

final class DocumentOpenAction: Action {
	// Define the type of input this action expects
    typealias InputType = DocumentRef

	// Define the type of presenter this action expects
    typealias PresenterType = DocumentPresenter

    static var description = "Open a document"
    static func perform(with context: ActionContext<DocumentRef>, using presenter: DocumentPresenter, completion: @escaping ((ActionPerformOutcome) -> ())) {
        completion(.success(closeActionStack: false))

The action’s perform function receives a context, which includes the input and access to context specific loggers, a presenter and a completion callback. Note that your perform function must use your InputType and PresenterType. So the context argument must be an ActionContext<...> with the ellipsis replaced with your specific input type or you can go with the InputType alias if you so prefer. The same applies to the presenter argument. You will get confusing Swift compiler errors if these types do not match up, so always check that, or always explicitly use InputType and PresenterType if you want to be safe, if a little less clear.

Actions that do not require an input must use NoInput as their InputType. If no presenter is required they must use NoPresenter as the PresenterType. These are special Flint types that let you indicate that you don’t need either the input or presenter. Here’s an example of an action with neither an input nor a presenter:

final class BeepAction: Action {
    typealias InputType = NoInput
    typealias PresenterType = NoPresenter

    static func perform(with context: ActionContext<InputType>, using presenter: PresenterType, completion: @escaping ((ActionPerformOutcome) -> ())) {
        completion(.success(closeActionStack: false))

Note that actions must call completion at some point, usually synchronously, indicating the outcome of the action.

You’ll notice that there is a closeActionStack argument set on the outcome passed to the completion handler. This flag indicates whether or not the action “closes off” a run of actions on the feature, which is used by Action Stacks. A perfect example is closing a document, where you can have Flint discard the Action Stack. An editing feature or a drawing feature however will not typically close the Action Stack (unless implied by closing the parent Action Stack). Don’t worry about the details of this too much for now. Usually you pass false.

Now we have to add the action to the feature, so we edit the feature declaration from earlier, adding a static property for the binding of action and feature, and prepare the feature by declaring that action.

class DocumentManagementFeature: Feature {
    static let description = "Create, Open and Save documents"

    static let openDocument = action(DocumentOpenAction.self)

    static func prepare(actions: FeatureActionsBuilder) {

Things to note about this:

First, the static action property createNew shown above is how you access the action to perform it later. Second, the action values must be assigned the result of the action() function, which is a binding of the action and the feature. We use this to know which feature the action belongs to when you invoke it.

Finally, the prepare function must call actions.declare for every action you want to use. You can call publish on these instead, to mark them as “visible to the user” if you have UI that shows actions, such as a painting app with a tool palette.

Calling your Action

Now comes the really easy part! In your app, you need to perform this action. Actions are performed within an ActionSession, which gives you a conceptual grouping for actions in your logging and debugging (e.g. one session per window, or currently open document), tracks the Action Stacks and ensures threading hygiene. By default actions are performed in the main session, ActionSession.main, which expects to be called on the main queue. It’ll crash with fatalError if you don’t call it on that queue, to keep you safe.

Somewhere in a view controller, you can add:

DocumentManagementFeature.openDocument.perform(using: self, with: document)

What we are doing here is calling a convenience function called perform on the action binding stored in the static property openDocument that we defined on the feature.

The using argument is the presenter, so in this case the scope we’re in — likely a View Controller — must conform to our own DocumentPresenter protocol. The with argument is the input, which must be a DocumentRef we have in scope.

This is where the magic of Swift really helps us. Those strongly typed arguments mean we can’t pass the wrong thing to the action and get weird errors. It also means Xcode can autocomplete the arguments for us using only compatible types.

Furthermore, if the Action uses NoInput or NoPresenter or both, you can omit those arguments when performing the action:


As mentioned, the above invocation will use the main ActionSession. If we have background processing in the app, we could create another ActionSession and dispatch to that instead, using a slightly different syntax:

// Do this at startup or "session setup" time, stash it globally somewhere
let bgProcessingSession = ActionSession(name: "bg",
                                      	userInitiatedActions: false,
                                      	callerQueue: myBackgroundQueue)

// Later on your background queue"
bgProcessionSession.perform(MachineLearningFeature.processImages, using: nil, with: imageStore)

All of the previous code samples just declare actions and allow you to perform them. However, all action dispatch goes through Flint’s ActionDispatcher which is observable. This means that even at this level of simplicity, if an ActionDispatchObserver is registered, it will be able to do something whenever these actions are performed – such as emitting an analytics tracking event for each action. Flint provides such an observer called AnalyticsReporting which you can use to route analytics to whatever backend you use.

Next steps