A Promise library for Swift, based partially on Javascript's A+ spec.
A Promise is a way to represent a value that will exist (or will fail with an error) at some point in the future. This is similar to how an Optional
represents a value that may or may not be there.
Using a special type to represent values that will exist in the future means that those values can be combined, transformed, and built in systematic ways. If the system knows what success and what failure look like, composing those asynchronous operations becomes much easier. For example, it becomes trivial to write reusable code that can:
- perform a chain of dependent asynchronous operations with one completion block at the end
- perform many independent asynchronous operations simultaneously with one completion block
- race many asynchronous operations and return the value of the first to complete
- retry asynchronous operations
- add a timeout to asynchronous operations
Promises are suited for any asynchronous action that can succeed or fail exactly once, such as HTTP requests. If there is an asynchronous action that can "succeed" more than once, or delivers a series of values over time instead of just one, take a look at Signals or Observables.
To access the value once it arrives, you call the then
method with a block.
let usersPromise = fetchUsers() // Promise<[User]>
usersPromise.then({ users in
self.users = users
})
All usage of the data in the users
Promise is gated through the then
method.
In addition to performing side effects (like setting the users
instance variable on self
), then
enables you do two other things. First, you can transform the contents of the Promise, and second, you can kick off another Promise, to do more asynchronous work. To do either of these things, return something from the block you pass to then
. Each time you call then
, the existing Promise will return a new Promise.
let usersPromise = fetchUsers() // Promise<[User]>
let firstUserPromise = usersPromise.then({ users in // Promise<User>
return users[0]
})
let followersPromise = firstUserPromise.then({ firstUser in //Promise<[Follower]>
return fetchFollowers(of: firstUser)
})
followersPromise.then({ followers in
self.followers = followers
})
Based on whether you return a regular value or a promise, the Promise
will determine whether it should transform the internal contents, or fire off the next promise and await its results.
As long as the block you pass to then
is one line long, its type signature will be inferred, which will make Promises much easier to read and work with.
Since each call to then
returns a new Promise
, you can write them in a big chain. The code above, as a chain, would be written:
fetchUsers()
.then({ users in
return users[0]
})
.then({ firstUser in
return fetchFollowers(of: firstUser)
})
.then({ followers in
self.followers = followers
})
To catch any errors that are created along the way, you can add a catch
block as well:
fetchUsers()
.then({ users in
return users[0]
})
.then({ firstUser in
return fetchFollowers(of: firstUser)
})
.then({ followers in
self.followers = followers
})
.catch({ error in
displayError(error)
})
If any step in the chain fails, no more then
blocks will be executed. Only failure blocks are executed. This is enforced in the type system as well. If the fetchUsers()
promise fails (for example, because of a lack of internet), there's no way for the promise to construct a valid value for the users
variable, and there's no way that block could be called.
To create a promise, there is a convenience initializer that takes a block and provides functions to fulfill
or reject
the promise:
let promise = Promise<Data>(work: { fulfill, reject in
try fulfill(Data(contentsOf: someURL)
})
It will automatically run on a global background thread.
You can use this initializer to wrap a completion block-based API, like URLSession
.
let promise = Promise<(Data, HTTPURLResponse)>(work: { fulfill, reject in
self.dataTask(with: request, completionHandler: { data, response, error in
if let error = error {
reject(error)
} else if let data = data, let response = response as? HTTPURLResponse {
fulfill((data, response))
} else {
fatalError("Something has gone horribly wrong.")
}
}).resume()
})
If the API that you're wrapping is sensitive to which thread it's being run on, like any UIKit code, be sure to pass add a queue: .main
parameter to the work:
initializer, and it will be executed on the main queue.
For delegate-based APIs, you can can create a promise in the .pending
state with the default initializer.
let promise = Promise()
and use the fulfill
and reject
instance methods to change its state.
Because promises formalize how success and failure blocks look, it's possible to build behaviors on top of them.
For example, if you want to execute code when the promise fulfills — regardless of whether it succeeds or fails – you can use always
.
activityIndicator.startAnimating()
fetchUsers()
.then({ users in
self.users = users
})
.always({
self.activityIndicator.stopAnimating()
})
Even if the network request fails, the activity indicator will stop. Note that the block that you pass to always
has no parameters. Because the Promise
doesn't know if it will succeed or fail, it will give you neither a value nor an error.
ensure
is a method that takes a predicate, and rejects the promise chain if that predicate fails.
URLSession.shared.dataTask(with: request)
.ensure({ data, httpResponse in
return httpResponse.statusCode == 200
})
.then({ data, httpResponse in
// the statusCode is valid
})
.catch({ error in
// the network request failed, or the status code was invalid
})
Static methods, like zip
, race
, retry
, all
, kickoff
live in a namespace called Promises
. Previously, they were static functions in the Promise
class, but this meant that you had to specialize them with a generic before you could use them, like Promise<()>.all
. This is ugly and hard to type, so as of v2.0, you now write Promises.all
.
Promises.all
is a static method that waits for all the promises you give it to fulfill, and once they have, it fulfills itself with the array of all fulfilled values. For example, you might want to write code to hit an API endpoint once for each item in an array. map
and Promises.all
make that super easy:
let userPromises = users.map({ user in
APIClient.followUser(user)
})
Promises.all(userPromises)
.then({
//all the users are now followed!
})
.catch ({ error in
//one of the API requests failed
})
Because the then
blocks of promises are a safe space for functions that throw
, its sometimes useful to enter those safe spaces even if you don't have asynchronous work to do. Promises.kickoff
is designed for that.
Promises
.kickoff({
return try initialValueFromThrowingFunction()
})
.then({ value in
//use the value from the throwing function
})
.catch({ error in
//if there was an error, you can handle it here.
})
This (coupled with Optional.unwrap()
) is particularly useful when you want to start a promise chain from an optional value.
These are some of the most useful behaviors, but there are others as well, like race
(which races multiple promises), retry
(which lets you retry a single promise multiple times), and recover
(which lets you return a new Promise
given an error, allowing you to recover from failure), and others.
You can find these behaviors in the Promises+Extras.swift file.
Each method on the Promise
that accepts a block accepts an execution context with the parameter name on:
. Usually, this execution context is a queue.
Promise<Void>(queue: .main, work: { fulfill, reject in
viewController.present(viewControllerToPresent, animated: flag, completion: {
fulfill()
})
}).then(on: DispatchQueue.global(), {
return try Data(contentsOf: someURL)
})
Because ExecutionContext
is a protocol, other things can be passed here. One particularly useful one is InvalidatableQueue
. When working with table cells, often the result of a promise needs to be ignored. To do this, each cell can hold on to an InvalidatableQueue
. An InvalidatableQueue
is an execution context that can be invalidated. If the context is invalidated, then the block that is passed to it will be discarded and not executed.
To use this with table cells, the queue should be invalidated and reset on prepareForReuse()
.
class SomeTableViewCell: UITableViewCell {
var invalidatableQueue = InvalidatableQueue()
func showImage(at url: URL) {
ImageFetcher(url)
.fetch()
.then(on: invalidatableQueue, { image in
self.imageView.image = image
})
}
override func prepareForReuse() {
super.prepareForReuse()
invalidatableQueue.invalidate()
invalidatableQueue = InvalidatableQueue()
}
}
Warning: don't chain blocks off anything that is executing on an invalidatable queue. then
blocks that return Void
won't stop the chain, but then
blocks that return values or promises will stop the chain. Because the block can't be executed, the result of the next value in the chain won't be calculable, and the next promise will remain in the pending
state forever, preventing resources from being released.
I made several design decisions when writing this Promise
library, erring towards making the library as easy to use as possible.
Other promise libraries use the functional names for then
, such as map
and flatMap
. The benefit to using these monadic functional terms is minor, but cost, in terms of understanding, is high. In this library, you call then
, and return anything you need to, and the library figures out how to handle it.
Other promise libraries allow you to define what type the error each promise will return. In theory, this a useful feature, allowing you to know what type the error will be in catch
blocks.
In practice, this is stifling. In practice, if you're using errors from two different domains, you have to either a) use a lowest common denominator error, like NSError
, or b) call a method like mapError
to convert the error from one domain to another.
Also note that Swift's built-in error handling system doesn't have typed errors, opting for pattern matching instead.
Lastly, you can use try
and throw
from within all the blocks, and the library will automatically translate it to a promise rejection. This makes working with APIs that throw much more easily. To extend our URLSession
example, we can use the throwing JSONSerialization API easily.
URLSession.shared.dataTask(with: request)
.ensure({ data, httpResponse in httpResponse.statusCode == 200 })
.then({ data, httpResponse in
return try JSONSerialization.jsonObject(with: data)
})
.then({ json in
// use the json
})
Working with optionals can be made simpler with a little extension.
struct NilError: Error { }
extension Optional {
func unwrap() throws -> Wrapped {
guard let result = self else {
throw NilError()
}
return result
}
}
Because you're in an environment where you can freely throw and it will be handled for you (in the form of a rejected Promise), you can now easily unwrap optionals. For example, if you need a specific key out of a json dictionary:
.then({ json in
return try (json["user"] as? [String: Any]).unwrap()
})
And you will transform your optional into a non-optional.
The threading model for this library is dead simple. init(work:)
happens on a background queue by default, and every other block-based method (then
, catch
, always
, etc) executes on the main thread. These can be overridden by passing in a DispatchQueue
object for the first parameter.
Promise<Void>(work: { fulfill, reject in
viewController.present(viewControllerToPresent, animated: flag, completion: {
fulfill()
})
}).then(on: DispatchQueue.global(), {
return try Data(contentsOf: someURL)
}).then(on: DispatchQueue.main, {
self.data = data
})
-
Add the following to your Podfile:
pod 'Promises'
-
Integrate your dependencies using frameworks: add
use_frameworks!
to your Podfile. -
Run
pod install
.
To get started playing with this library, you can use the included Promise.playground
. Simply open the .xcodeproj
, build the scheme, and then open the playground (from within the project) and start playing.