Guarantees in Swift for rookies

Sync vs async execution

Writing asynchronous code is among the hardest a part of constructing an app.

What precisely is the distinction between a synchronous and an asynchronous execution? Effectively, I already defined this in my Dispatch framework tutorial, however here’s a fast recap. A synchoronousoperate normally blocks the present thread and returns some worth in a while. An asynchronousoperate will immediately return and passes the consequence worth right into a completion handler. You need to use the GCD framework to carry out duties sync on async on a given queue. Let me present you a fast instance:

func aBlockingFunction() -> String {
    sleep(.random(in: 1...3))
    return "Hiya world!"
}

func syncMethod() -> String {
    return aBlockingFunction()
}

func asyncMethod(completion block: @escaping ((String) -> Void)) {
    DispatchQueue.world(qos: .background).async {
        block(aBlockingFunction())
    }
}

print(syncMethod())
print("sync technique returned")
asyncMethod { worth in
    print(worth)
}
print("async technique returned")

As you may see the async technique runs completely on a background queue, the operate will not block the present thread. This is the reason the async technique can return immediately, so you may all the time see the return output earlier than the final good day output. The async technique’s completion block is saved for later execution, that is the rationale why is it attainable to call-back and return the string worth manner after the unique operate have returned.

What occurs should you do not use a special queue? The completion block will likely be executed on the present queue, so your operate will block it. It should be considerably async-like, however in actuality you are simply transferring the return worth right into a completion block.

func syncMethod() -> String {
    return "Hiya world!"
}

func fakeAsyncMethod(completion block: ((String) -> Void)) {
    block("Hiya world!")
}

print(syncMethod())
print("sync technique returned")
fakeAsyncMethod { worth in
    print(worth)
}
print("faux async technique returned")

I do not actually need to deal with completion blocks on this article, that may very well be a standalone submit, however in case you are nonetheless having bother with the concurrency mannequin or you do not perceive how duties and threading works, you need to learn perform a little analysis first.

Callback hell and the pyramid of doom

What is the drawback with async code? Or what’s the results of writing asynchronous code? The quick reply is that it’s a must to use completion blocks (callbacks) as a way to deal with future outcomes.

The lengthy reply is that managing callbacks sucks. It’s important to watch out, as a result of in a block you may simply create a retain-cycle, so it’s a must to go round your variables as weak or unowned references. Additionally if it’s a must to use a number of async strategies, that’ll be a ache within the donkey. Pattern time! 🐴

struct Todo: Codable {
    let id: Int
    let title: String
    let accomplished: Bool
}

let url = URL(string: "https://jsonplaceholder.typicode.com/todos")!

URLSession.shared.dataTask(with: url) { knowledge, response, error in
    if let error = error {
        fatalError("Community error: " + error.localizedDescription)
    }
    guard let response = response as? HTTPURLResponse else {
        fatalError("Not a HTTP response")
    }
    guard response.statusCode <= 200, response.statusCode > 300 else {
        fatalError("Invalid HTTP standing code")
    }
    guard let knowledge = knowledge else {
        fatalError("No HTTP knowledge")
    }

    do {
        let todos = strive JSONDecoder().decode([Todo].self, from: knowledge)
        print(todos)
    }
    catch {
        fatalError("JSON decoder error: " + error.localizedDescription)
    }
}.resume()

The snippet above is a straightforward async HTTP knowledge request. As you may see there are many optionally available values concerned, plus it’s a must to do some JSON decoding if you wish to use your individual varieties. This is only one request, however what should you’d must get some detailed information from the primary aspect? Let’s write a helper! #no 🤫

func request(_ url: URL, completion: @escaping ((Information) -> Void)) {
    URLSession.shared.dataTask(with: url) { knowledge, response, error in
        if let error = error {
            fatalError("Community error: " + error.localizedDescription)
        }
        guard let response = response as? HTTPURLResponse else {
            fatalError("Not a HTTP response")
        }
        guard response.statusCode <= 200, response.statusCode > 300 else {
            fatalError("Invalid HTTP standing code")
        }
        guard let knowledge = knowledge else {
            fatalError("No HTTP knowledge")
        }
        completion(knowledge)
    }.resume()
}


let url = URL(string: "https://jsonplaceholder.typicode.com/todos")!
request(url) { knowledge in
    do {
        let todos = strive JSONDecoder().decode([Todo].self, from: knowledge)
        guard let first = todos.first else {
            return
        }
        let url = URL(string: "https://jsonplaceholder.typicode.com/todos/(first.id)")!
        request(url) { knowledge in
            do {
                let todo = strive JSONDecoder().decode(Todo.self, from: knowledge)
                print(todo)
            }
            catch {
                fatalError("JSON decoder error: " + error.localizedDescription)
            }
        }
    }
    catch {
        fatalError("JSON decoder error: " + error.localizedDescription)
    }
}

See? My drawback is that we’re slowly transferring down the rabbit gap. Now what if we have now a third request? Hell no! It’s important to nest the whole lot one stage deeper once more, plus it’s a must to go across the essential variables eg. a weak or unowned view controller reference as a result of in some unspecified time in the future in time it’s a must to replace all the UI based mostly on the result. There should be a greater option to repair this. 🤔

Outcomes vs futures vs guarantees?

The consequence sort was launched in Swift 5 and it is extraordinarily good for eliminating the optionally available issue from the equation. This implies you do not have to take care of an optionally available knowledge, and an optionally available error sort, however your result’s both of them.

Futures are principally representing a worth sooner or later. The underlying worth could be for instance a consequence and it ought to have one of many following states:

• pending – no worth but, ready for it…
• fulfilled – success, now the consequence has a worth
• rejected – failed with an error

By definition a futures should not be writeable by the end-user. Which means that builders shouldn’t be capable of create, fulfill or reject one. But when that is the case and we observe the principles, how can we make futures?

We promise them. It’s important to create a promise, which is principally a wrapper round a future that may be written (fulfilled, rejected) or remodeled as you need. You do not write futures, you make guarantees. Nonetheless some frameworks permits you to get again the long run worth of a promise, however you should not be capable of write that future in any respect.

Sufficient concept, are you able to fall in love with guarantees? ❤️

Guarantees 101 – a newbie’s information

Let’s refactor the earlier instance through the use of my promise framework!

extension URLSession {

    enum HTTPError: LocalizedError {
        case invalidResponse
        case invalidStatusCode
        case noData
    }

    func dataTask(url: URL) -> Promise<Information> {
        return Promise<Information> { [unowned self] fulfill, reject in
            self.dataTask(with: url) { knowledge, response, error in
                if let error = error {
                    reject(error)
                    return
                }
                guard let response = response as? HTTPURLResponse else {
                    reject(HTTPError.invalidResponse)
                    return
                }
                guard response.statusCode <= 200, response.statusCode > 300 else {
                    reject(HTTPError.invalidStatusCode)
                    return
                }
                guard let knowledge = knowledge else {
                    reject(HTTPError.noData)
                    return
                }
                fulfill(knowledge)
            }.resume()
        }
    }
}

enum TodoError: LocalizedError {
    case lacking
}

let url = URL(string: "https://jsonplaceholder.typicode.com/todos")!
URLSession.shared.dataTask(url: url)
.thenMap { knowledge in
    return strive JSONDecoder().decode([Todo].self, from: knowledge)
}
.thenMap { todos -> Todo in
    guard let first = todos.first else {
        throw TodoError.lacking
    }
    return first
}
.then { first in
    let url = URL(string: "https://jsonplaceholder.typicode.com/todos/(first.id)")!
    return URLSession.shared.dataTask(url: url)
}
.thenMap { knowledge in
    strive JSONDecoder().decode(Todo.self, from: knowledge)
}
.onSuccess { todo in
    print(todo)
}
.onFailure(queue: .important) { error in
    print(error.localizedDescription)
}

What simply occurred right here? Effectively, I made kind of a promisified model of the info activity technique applied on the URLSession object as an extension. After all you may return the HTTP consequence or simply the standing code plus the info should you want additional information from the community layer. You need to use a brand new response knowledge mannequin or perhaps a tuple. 🤷‍♂️

Anyway, the extra fascinating half is the underside half of the supply. As you may see I am calling the model new dataTask technique which returns a Promise<Information> object. As I discussed this earlier than a promise could be remodeled. Or ought to I say: chained?

Chaining guarantees is the most important benefit over callbacks. The supply code is just not wanting like a pyramid anymore with loopy indentations and do-try-catch blocks, however extra like a sequence of actions. In each single step you may rework your earlier consequence worth into one thing else. If you’re conversant in some purposeful paradigms, it should be very easy to know the next:

• thenMap is a straightforward map on a Promise
• then is principally flatMap on a Promise
• onSuccess solely will get known as if the whole lot was nice within the chain
• onFailure solely will get known as if some error occurred within the chain
• all the time runs all the time whatever the final result

If you wish to get the primary queue, you may merely go it via a queue parameter, like I did it with the onFailure technique, however it works for each single aspect within the chain. These features above are simply the tip of the iceberg. You may as well faucet into a sequence, validate the consequence, put a timeout on it or get well from a failed promise.

There may be additionally a Guarantees namespace for different helpful strategies, like zip, which is able to zipping collectively 2, 3 or 4 completely different sort of guarantees. Identical to the Guarantees.all technique the zip operate waits till each promise is being accomplished, then it offers you the results of all the guarantees in a single block.

Guarantees.all(guarantees)
.thenMap { arrayOfResults in
    
}

Guarantees.zip(promise1, promise2)
.thenMap { result1, result2 in
    
}

It is also price to say that there’s a first, delay, timeout, race, wait and a retry technique underneath the Guarantees namespace. Be happy to mess around with these as nicely, typically they’re extremly helpful and highly effective too. 💪

There are solely two issues with guarantees

The primary concern is cancellation. You may’t merely cancel a operating promise. It is doable, however it requires some superior or some say “hacky” methods.

The second is async / await. If you wish to know extra about it, you need to learn the concurrency manifesto by Chis Lattner, however since this can be a newbie’s information, let’s simply say that these two key phrases can add some syntactic sugar to your code. You will not want the additional (then, thenMap, onSuccess, onFailure) traces anymore, this fashion you may focus in your code. I actually hope that we’ll get one thing like this in Swift 6, so I can throw away my Promise library for good. Oh, by the way in which, libraries…

Promise libraries price to examine

My promise implementation is much from excellent, however it’s a fairly easy one (~450 traces of code) and it serves me rather well. This weblog submit by @khanlou helped me so much to know guarantees higher, you need to learn it too! 👍

There are many promise libraries on github, but when I had to select from them (as an alternative my very own implementation), I might undoubtedly go along with one of many following ones:

• PromiseKit – The most well-liked one
• Guarantees by Google – characteristic wealthy, fairly standard as nicely
• Promise by Khanlou – small, however based mostly on on the JavaScript Guarantees/A+ spec
• SwiftNIO – not an precise promise library, however it has a fantastically written occasion loop based mostly promise implementation underneath the hood

Professional tip: do not attempt to make your individual Promise framework, as a result of multi-threading is extraordinarily arduous, and you do not need to fiddle with threads and locks.

Guarantees are actually addictive. When you begin utilizing them, you may’t merely return and write async code with callbacks anymore. Make a promise as we speak! 😅