Aftermath is a stateless message-driven micro-framework in Swift, which is based on the concept of the unidirectional data flow architecture.
At first sight Aftermath may seem to be just a type-safe implementation of the publish-subscribe messaging pattern, but actually it could be considered as a distinct mental model in application design, different from familiar MVC, MVVM or MVP approaches. Utilizing the ideas behind Event Sourcing and Flux patterns it helps to separate concerns, reduce code dependencies and make data flow more predictable.
The following diagram demonstrates the data flow in Aftermath architecture in details:
- Core components
- The flow
- Extra
- Engine
- Life hacks
- Tools
- Summary
- Installation
- Examples
- Extensions
- Alternatives
- Author
- Influences
- Contributing
- License
Command is a message with a set of instructions describing an intention to execute the corresponding behavior. Command could lead to data fetching, data mutation and any sort of sync or async operation that produces desirable output needed to update application/view state.
Every command can produce only one Output type.
Command Handler layer is responsible for business logic in the application. The submission of a command is received by a command handler, which usually performs short- or long-term operation, such as network request, database query, cache read/white process, etc. Command handler can be sync and publish the result immediately. On the other hand it's the best place in the application to write asynchronous code.
The restriction is to create only one command handler per command.
Command Handler is responsible for publishing events that will be consumed by reactions. There are 3 types of events:
progress
event indicates that the operation triggered by command has been started and is in the pending state at the moment.data
event holds the output produced by the command execution.error
notifies that an error has been occurred during the command execution.
Reaction responds to event published by command handler. It is supposed to handle 3 possible event types by describing the desired behavior in the each scenario:
wait
function reacts onprogress
type of the eventconsume
function reacts ondata
type of the event.rescue
function is a fallback for the case whenerror
event has been received.
Normally reaction performs UI updates, but could also be used for other kinds of output processing.
Taking 4 core components described before, we can build a simplified version of the data flow:
The first step is to declare a command. Your command type has to conform to the
Aftermath.Command
protocol and the Output
type must be implicitly specified.
Let's say we want to fetch a list of books from some untrusted resource and correct typos in titles and author names 🤓.
// This is our model we are going to work with.
struct Book {
let id: Int
let title: String
let author: String
}
struct BookListCommand: Command {
// Result of this command will be a list of books.
typealias Output = [Book]
}
struct BookUpdateCommand: Command {
// Result of this command will be an updated book.
typealias Output = Book
// Let's pass the entire model to the command to simplify this example.
// Ideally we wouldn't do that because a command is supposed to be as simple
// as possible, only with attributes that are needed for handler.
let book: Book
}
Note that any type can play the role of Output
, so if we want to add a
date to our BookUpdateCommand
it could look like the following:
typealias Output = (Book, Date)
In order to execute a command you have to conform to CommandProducer
protocol:
class ViewController: UITableViewController, CommandProducer {
// Fetch a list of books.
func load() {
execute(command: BookListCommand())
}
// Update a single book with corrected title and/or author name.
func update(book: Book) {
execute(command: BookUpdateCommand(book: book))
}
}
Command is an intention that needs to be translated into an action by a handler.
The command handler is responsible for publishing events to notify about results of
the operation it performs. The command handlers type has to conform to
Aftermath.CommandHandler
protocol, that needs to know about the command type
it will work with:
struct BookListCommandHandler: CommandHandler {
func handle(command: BookListCommand) throws -> Event<BooksCommand> {
// Start network request to fetch data.
fetchBooks { books, error in
if let error = error {
// Publish error.
self.publish(error: error)
return
}
// Publish fetched data.
self.publish(data: books)
}
// Load data from local database/cache.
let localBooks = loadLocalBooks()
// If the list is empty let the listeners know that operation is in the process.
return Book.list.isEmpty ? Event.progress : Event.data(localBooks)
}
}
Note that every command handler needs to be registered on Aftermath Engine.
Engine.shared.use(handler: BookListCommandHandler())
The last step, but not the least, is to react to events published by the command
handlers. Just conform to ReactionProducer
protocol, implement reaction
behavior and you're ready to go:
class ViewController: UITableViewController, CommandProducer, ReactionProducer {
var books = [Book]()
deinit {
// Don't forget to dispose all reaction tokens.
disposeAll()
}
override func viewDidLoad() {
super.viewDidLoad()
// React to events.
react(to: BookListCommand.self, with: Reaction(
wait: { [weak self] in
// Wait for results to come.
self?.refreshControl?.beginRefreshing()
},
consume: { [weak self] books in
// We're lucky, there are some books to display.
self?.books = books
self?.refreshControl?.endRefreshing()
self?.tableView.reloadData()
},
rescue: { [weak self] error in
// Well, seems like something went wrong.
self?.refreshControl?.endRefreshing()
print(error)
}))
}
// ...
}
It's important to dispose all reaction tokens when your ReactionProducer
instance is about to be deallocated or reaction needs to be unsubscribed from
events.
// Disposes all reaction tokens for the current `ReactionProducer`.
disposeAll()
// Disposes a specified reaction token.
let token = react(to: BookListCommand.self, with: reaction)
dispose(token: token)
Action is a variation of command that handles itself. It's a possibility to simplify the code when command itself or business logic are super tiny. There is no need to register an action, it will be automatically added to the list of active command handlers on the fly, when it's executed as a command.
import Sugar
struct WelcomeAction: Action {
typealias Output = String
let userId: String
func handle(command: WelcomeAction) throws -> Event<WelcomeAction> {
fetchUser(id: userId) { user in
self.publish(data: "Hello \(user.name)")
}
return Event.progress
}
}
// Execute action
struct WelcomeManager: CommandProducer {
func salute() {
execute(action: WelcomeAction(userId: 11))
}
}
Fact works like notification, with no async operations involved. It can
be used when there is no need for a handler to generate an output. Fact is an
output itself, so the only thing you want to do is notify all
subscribers that something happened in the system, and they will react
accordingly. In this sense it's closer to a type-safe alternative to
Notification
.
struct LoginFact: Fact {
let username: String
}
class ProfileController: UIViewController, ReactionProducer {
override func viewDidLoad() {
super.viewDidLoad()
// React
next { (fact: LoginFact) in
title = fact.username
}
}
}
struct AuthService: FactProducer {
func login() {
let fact = LoginFact(username: "John Doe")
// Publish
post(fact: fact)
}
}
Middleware is a layer where commands and events can be intercepted before they reach their listeners.
It means you can modify/cancel/extend the executed command in Command Middleware before it's processed by the command handler:
Or you can do appropriate operation in Event Middleware before the published event is received by its reactions.
It's handy for logging, crash reporting, aborting particular commands or events, etc.
// Command middleware
struct ErrorCommandMiddleware: CommandMiddleware {
func intercept(command: AnyCommand, execute: Execute, next: Execute) throws {
do {
// Don't forget to call `next` to invoke the next function in the chain.
try next(command)
} catch {
print("Command failed with error -> \(command)")
throw error
}
}
}
Engine.shared.pipeCommands(through: [ErrorCommandMiddleware()])
// Event middleware
struct LogEventMiddleware: EventMiddleware {
// Don't forget to call `next` to invoke the next function in the chain.
func intercept(event: AnyEvent, publish: Publish, next: Publish) throws {
print("Event published -> \(event)")
try next(event)
}
}
Engine.shared.pipeEvents(through: [LogEventMiddleware()])
Note that it's necessary to call next
to invoke the next function in the
chain while building your custom middleware.
AnyCommand
and AnyEvent
are special protocols that every Command
or
Event
conform to. They are used mostly in middleware to
workaround restrictions of working with Swift generic protocols that have
associatedtype
.
Engine is the main entry point for Aftermath configuration:
- Register command handlers:
Engine.shared.use(handler: BookListCommandHandler())
- Add command and event middleware:
// Commands
Engine.shared.pipeCommands(through: [LogCommandMiddleware(), ErrorCommandMiddleware()])
// Events
Engine.shared.pipeEvents(through: [LogEventMiddleware(), ErrorEventMiddleware()])
- Set global error handler to catch all unexpected errors and framework warnings:
struct EngineErrorHandler: ErrorHandler {
func handleError(error: Error) {
if let error = error as? Failure {
print("Engine error -> \(error)")
} else if let warning = error as? Warning {
print("Engine warning -> \(warning)")
} else {
print("Unknown error -> \(error)")
}
}
}
Engine.shared.errorHandler = EngineErrorHandler()
- Dispose all registered command handlers and event listeners (reactions):
Engine.shared.invalidate()
Naming is hard. It doesn't feel right to have names like BookListCommand
,
BookListCommandHandler
and BookListWhatever
, does it? If you agree, then
you can work around this issue by introducing a new idea into the mix.
You can group all related types into stories, which make the flow more concrete.
struct BookListStory {
struct Command: Aftermath.Command {
// ...
}
struct Handler: Aftermath.CommandHandler {
// ...
}
}
In this sense, it's close to user stories used in agile software development methodologies.
You can find more detailed example in AftermathNotes demo project.
Some of the stories may seem very similar. Then in makes sense to make them more generic and reusable according to the DRY principle. For example, let's say we have the flow to fetch a single resource by id.
import Aftermath
import Malibu
// Generic feature
protocol DetailFeature {
associatedtype Model: Entity
var resource: String { get }
}
// Command
struct DetailCommand<Feature: DetailFeature>: Aftermath.Command {
typealias Output = Feature.Model
let id: Int
}
// Command handler
struct DetailCommandHandler<Feature: DetailFeature>: Aftermath.CommandHandler {
typealias Command = DetailCommand<Feature>
let feature: Feature
func handle(command: Command) throws -> Event<Command> {
fetchDetail("\(feature.resource)/\(command.id)") { json, error in
if let error = error {
self.publish(error: error)
return
}
do {
self.publish(data: try Feature.Model(json))
} catch {
self.publish(error: error)
}
}
return Event.progress
}
}
// Concrete feature
struct BookFeature: ListFeature, DeleteFeature, CommandProducer {
typealias Model = Todo
var resource = "books"
}
// Execute command to load a single resource.
execute(command: DetailCommand<BookFeature>(id: id))
// Register reaction listener.
react(to: DetailCommand<BookFeature>.self, with: reaction)
You can find more detailed example in AftermathNotesPlus demo project.
We believe that in iOS applications, in most of the cases, there is no real need for single global state (single source of truth) or multiple sub-states distributed between stores. Data is stored on disc in local persistence layer, such as database and cache, or it's fetched from network. Then this content, assembled piece by piece from different sources, is translated into the "view state", which is readable by the view to render it on the screen. This "view state" is kept in memory and valid at a given instant in time until we switch the context and the current view is deallocated.
Keeping that in mind, it makes more sense to dispose the "view state" together with the view it belongs to, rather than retain no longer used replication in any sort of global state.
It should be enough to restore a state by re-playing previous events from the history.
Advantages of Aftermath
- Separation of concerns
- Code reusability
- Unidirectional data flow
- Type safety
**Disadvantages of Aftermath
- No state (?)
- Focusing on command output instead of actual data
- Async command handler could confuse the flow
P.S. Even though Aftermath is a stateless framework at the moment, we have plans to introduce some sort of optional store(s) for better state management. It might be a new feature in v2, keep watching.
- Aftermath comes with a set of development tools, such as additional helpers, useful command and event middleware for logging, error handling, etc.
// Commands
Engine.sharedInstance.pipeCommands(through: [LogCommandMiddleware(), ErrorCommandMiddleware()])
// Events
Engine.sharedInstance.pipeEvents(through: [LogEventMiddleware(), ErrorEventMiddleware()])
Aftermath is available through CocoaPods. To install it, simply add the following line to your Podfile:
pod 'Aftermath'
Aftermath is also available through Carthage. To install just write into your Cartfile:
github "hyperoslo/Aftermath"
Aftermath can also be installed manually. Just download and drop Sources
folders in your project.
-
iOS Playground uses live view of interactive playground to show how to fetch data from network and display it in the
UITableView
. -
AftermathNotes is a simple application that demonstrates how to setup networking stack and data cache layer using Aftermath. It uses the concept of stories to group related types and make the
command -> event
flow more readable. -
AftermathNotesPlus is a more advanced example that extends AftermathNotes demo. It plays with generics and introduces the concept of features in order to reuse view controllers and RESTful network requests.
This repository aims to be the core implementation of framework, but there are also a range of extensions that integrate Aftermath with other libraries and extend it with more features:
-
AftermathCompass is a message-driven routing system built on top of Aftermath and Compass.
-
AftermathSpots is made to improve development routines of building component-based UIs using Spots cross-platform view controller framework. It comes with custom reactions and injectable behaviors that move code reusability to the next level and make your application even more decoupled and flexible.
Still not sure about state management? It's not that easy to cover all scenarios and find a silver bullet for all occasions. But if you think it's time to break conventions and try new architecture in your next application, there are some links for further reading and research:
-
ReSwift - Unidirectional data flow in Swift, inspired by Redux.
-
fantastic-ios-architecture - A list of resources related to iOS architecture topic.
Hyper Interaktiv AS, ios@hyper.no
Aftermath is inspired by the idea of unidirectional data flow in Flux and utilizes some concepts like sequence of commands and events from Event Sourcing.
We would love you to contribute to Aftermath, check the CONTRIBUTING file for more info.
Aftermath is available under the MIT license. See the LICENSE file for more info.