How do you leverage the power of Swift to write great asynchronous code? BrightFutures is our answer.
BrightFutures implements proven functional concepts (i.e. futures and promises) to provide a powerful alternative to completion blocks and NSErrorPointers.
The goal of BrightFutures is to be the idiomatic Swift implementation of futures and promises. Our Big Hairy Audacious Goal (BHAG) is to be copy-pasted into the Swift standard library.
BrightFutures 2.0 is now available! It removes the direct dependency on NSError and takes a Swiftier approach. The tests (97% coverage) and documentation (100% coverage) have been improved as well. Please check the Migration guide for help on how to migrate your project to BrightFutures 2.0.
A Swift 2.0 compatible version is in the works and will be released as BrightFutures 3.0. Pre-release builds are available.
Latest releases:
- 3.0.0-beta.2
- Built for Swift 2.0
- Runs on iOS 8 / OS X 10.10 and above
- 2.0.1
- Built for Swift 1.2
- Runs on iOS 8 / OS X 10.10 and above
- 1.0.1
- Superseded by 2.0.0, 1.x is in maintenance mode
- Built for Swift 1.2
- Runs on iOS 8 / OS X 10.10 and above
Version 0.36 or higher is required. Add the following to your Podfile:
pod 'BrightFutures'Make sure that you are integrating your dependencies using frameworks: add use_frameworks! to your Podfile. Then run pod install.
Add the following to your Cartfile:
github "Thomvis/BrightFutures"
Run carthage update and follow the steps as described in Carthage's README.
- API documentation is available at the wonderful cocoadocs.org (100% documentation coverage)
- This README covers almost all features of BrightFutures
- The tests contain (trivial) usage examples for every feature (97% test coverage)
- The primary author, Thomas Visser, gave a talk at the April 2015 CocoaHeadsNL meetup
- The Highstreet Watch App is an Open Source WatchKit app that makes extensive use of BrightFutures
We write a lot of asynchronous code. Whether we're waiting for something to come in from the network or want to perform an expensive calculation off the main thread and then update the UI, we often do the 'fire and callback' dance. Here's a typical snippet of asynchronous code:
User.logIn(username, password) { user, error in
if !error {
Posts.fetchPosts(user, success: { posts in
// do something with the user's posts
}, failure: handleError)
} else {
handleError(error) // handeError is a custom function to handle errors
}
}Now let's see what BrightFutures can do for you:
User.logIn(username,password).flatMap { user in
Posts.fetchPosts(user)
}.onSuccess { posts in
// do something with the user's posts
}.onFailure { error in
// either logging in or fetching posts failed
}Both User.logIn and Posts.fetchPosts now immediately return a Future. A future can either fail with an error or succeed with a value, which can be anything from an Int to your custom struct, class or tuple. You can keep a future around and register for callbacks for when the future succeeds or fails at your convenience.
When the future returned from User.logIn fails, e.g. the username and password did not match, flatMap and onSuccess are skipped and onFailure is called with the error that occurred while logging in. If the login attempt succeeded, the resulting user object is passed to flatMap, which 'turns' the user into an array of his or her posts. If the posts could not be fetched, onSuccess is skipped and onFailure is called with the error that occurred when fetching the posts. If the posts could be fetched successfully, onSuccess is called with the user's posts.
This is just the tip of the proverbial iceberg. A lot more examples and techniques can be found in this readme, by browsing through the tests or by checking out the official companion framework FutureProofing.
If you already have a function (or really any expression) defined that you just want to execute asynchronously, you can easily wrap it in a future block:
future {
fibonacci(50)
}.onSuccess { num in
// value is 12586269025
}While this is really short and simple, it is equally limited. In many cases, you will need a way to indicate that the task failed. To do this, instead of returning the value, you can return a Result. Results can indicate either a success or a failure:
let f = future { () -> Result<NSDate, ReadmeError> in
let now: NSDate? = serverTime()
if let now = now {
return Result(value: now)
}
return Result(error: ReadmeError.TimeServiceError)
}
f.onSuccess { value in
// value will the NSDate from the server
}The future block needs an explict type because the Swift compiler is not able to deduce the type of multi-statement blocks. ReadmeError is an enum consisting of all errors that can happen in this readme.
Now let's assume the role of an API author who wants to use BrightFutures. The 'producer' of a future is called a Promise. A promise contains a future that you can immediately hand to the client. The promise is kept around while performing the asynchronous operation, until calling Promise.success(result) or Promise.failure(error) when the operation ended. Futures can only be completed through a Promise.
func asyncCalculation() -> Future<String, NoError> {
let promise = Promise<String, NoError>()
Queue.global.async {
// do a complicated task and then hand the result to the promise:
promise.success("forty-two")
}
return promise.future
}Queue is a simple wrapper around a dispatch queue. NoError indicates that the Future cannot fail. This is guaranteed by the type system, since NoError has no initializers.
You can be informed of the result of a Future by registering callbacks: onComplete, onSuccess and onFailure. The order in which the callbacks are executed upon completion of the future is not guaranteed, but it is guaranteed that the callbacks are executed serially. It is not safe to add a new callback from within a callback of the same future.
Using the andThen function on a Future, the order of callbacks can be explicitly defined. The closure passed to andThen is meant to perform side-effects and does not influence the result. andThen returns a new Future with the same result as this future.
var answer = 10
let f = Future<Int, NoError>.succeeded(4).andThen { result in
switch result {
case .Success(let val):
answer *= val.value
case .Failure(_):
break
}
}.andThen { result in
if let val = result.value {
answer += 2
}
}
// answer will be 42 (not 48)map returns a new Future that contains the error from this Future if this Future failed, or the return value from the given closure that was applied to the value of this Future. There's also a flatMap function that can be used to map the result of a future to the value of a new Future.
future {
fibonacci(10)
}.map { number -> String in
if number > 5 {
return "large"
}
return "small"
}.map { sizeString in
sizeString == "large"
}.onSuccess { numberIsLarge in
// numberIsLarge is true
}let f = future(1)
let f1 = future(2)
f.zip(f1).onSuccess { (let a, let b) in
// a is 1, b is 2
}future(3).filter { $0 > 5 }.onComplete { result in
// failed with error NoSuchElementError
}
future("Swift").filter { $0.hasPrefix("Sw") }.onComplete { result in
// succeeded with value "Swift"
}If a Future fails, use recover to offer a default or alternative value and continue the callback chain.
let f = future {
// imagine a request failed
return Result<Int, ReadmeError>(error: ReadmeError.RequestFailed)
}.recover { _ in // provide an offline default
return 5
}.onSuccess { value in
// value is 5 if the request failed or 10 if the request succeeded
}In addition to recover, recoverWith can be used to provide a Future that will provide the value to recover with.
BrightFutures also comes with a number of utility functions that simplify working with multiple futures. These are implemented as free (i.e. global) functions to work around current limitations of Swift.
The built-in fold function allows you to turn a list of values into a single value by performing an operation on every element in the list that consumes it as it is added to the resulting value. A trivial usecase for fold would be to calculate the sum of a list of integers.
Folding a list of Futures is not very convenient with the built-in fold function, which is why BrightFutures provides one that works especially well for our use case. BrightFutures' fold turns a list of Futures into a single Future that contains the resulting value. This allows us to, for example, calculate the sum of the first 10 Future-wrapped elements of the fibonacci sequence:
let fibonacciSequence = [future(fibonacci(1)), future(fibonacci(2)), ..., future(fibonacci(10))]
// 1+1+2+3+5+8+13+21+34+55
fold(fibonacciSequence, 0, { $0 + $1 }).onSuccess { sum in
// sum is 143
}With sequence, you can turn a list of Futures into a single Future that contains a list of the results from those futures.
let fibonacciSequence = [future(fibonacci(1)), future(fibonacci(2)), ..., future(fibonacci(10))]
sequence(fibonacciSequence).onSuccess { fibNumbers in
// fibNumbers is an array of Ints: [1, 1, 2, 3, etc.]
}traverse combines map and fold in one convenient function. traverse takes a list of values and a closure that takes a single value from that list and turns it into a Future. The result of traverse is a single Future containing an array of the values from the Futures returned by the given closure.
traverse(1...10) {
i in future(fibonacci(i))
}.onSuccess { fibNumbers in
// fibNumbers is an array of Ints: [1, 1, 2, 3, etc.]
}BrightFutures tries its best to provide a simple and sensible default threading model. In theory, all threads are created equally and BrightFutures shouldn't care about which thread it is on. In practice however, the main thread is more equal than others, because it has a special place in our hearts and because you'll often want to be on it to do UI updates.
A lot of the methods on Future accept an optional execution context and a block, e.g. onSuccess, map, recover and many more. The block is executed (when the future is completed) in the given execution context, which in practice is a GCD queue. When the context is not explicitly provided, the following rules will be followed to determine the execution context that is used:
- if the method is called from the main thread, the block is executed on the main queue (
Queue.main) - if the method is not called from the main thread, the block is executed on a global queue (
Queue.global)
The future keyword uses a much simpler threading model. The block (or expression) given to future is always executed on the global queue. You can however provide an explicit execution context to override the default behavior.
If you want to have custom threading behavior, skip do do not the section. next 😉
The default threading behavior can be overridden by providing explicit execution contexts. By default, BrightFutures comes with three contexts: Queue.main, Queue.global, and ImmediateExecutionContext. You can also create your own by implementing the ExecutionContext protocol.
let f = future(context: ImmediateExecutionContext) {
fibonacci(10)
}
f.onComplete(context: Queue.main.context) { value in
// update the UI, we're on the main thread
}The calculation of the 10nth Fibonacci number is now performed on the same thread as where the future is created.
An invalidation token can be used to invalidate a callback, preventing it from being executed upon completion of the future. This is particularly useful in cases where the context in which a callback is executed changes often and quickly, e.g. in reusable views such as table views and collection view cells. An example of the latter:
class MyCell : UICollectionViewCell {
var token = InvalidationToken()
public override func prepareForReuse() {
super.prepareForReuse()
token.invalidate()
token = InvalidationToken()
}
public func setModel(model: Model) {
ImageLoader.loadImage(model.image).onSuccess(token: token) { [weak self] UIImage in
self.imageView.image = UIImage
}
}
}By invalidating the token on every reuse, we prevent that the image of the previous model is set after the next model has been set.
Invalidation tokens do not cancel the task that the future represents. That is a different problem. With invalidation tokens, the result is merely ignored. The callbacks are invoked as soon as the token is invalidated, which is typically before the original future is completed, or if the original future is completed. Invalidating a token after the original future completed does nothing.
If you are looking for a way to cancel a running task, you should look into using NSProgress (or https://github.com/Thomvis/GoodProgress if you're looking for a nice Swift wrapper).
BrightFutures' primary author is Thomas Visser. He is lead iOS Engineer at Highstreet. We welcome any feedback and pull requests. Get your name on this list!
BrightFutures was inspired by Facebook's BFTasks, the Promises & Futures implementation in Scala and Max Howell's PromiseKit.
BrightFutures is available under the MIT license. See the LICENSE file for more info.