Setting Swift compiler flags in CocoaPods

Lately I’ve been working on a Swift framework that I’m integrating into an existing app with CocoaPods. The framework relies on an #if DEBUG macro to run one of two code paths, depending on whether we’re building with the Debug configuration or something else.

public var baseURL: NSURL {
    #if DEBUG
        return NSURL(string:"https://coolapp-staging.herokuapp.com")!
    #else
        return NSURL(string:"https://coolapp.com")!
    #endif
}

In order for our code to trigger an #if <something> macro, we need to set a -D<something> flag in the other Swift flags section of our target’s build settings. There’s nothing special about “debug” as it’s used here, we’re just creating a global variable and naming it DEBUG.

Target build settings

So far so good – we can see that the #if DEBUG branch runs when we build with our Debug configuration and the #else branch runs otherwise. But this changes when we consume our framework in another project. Firstly, CocoaPods doesn’t look at the build settings in our library’s Xcode project at all. The xcconfig files that CocoaPods generates for our framework are entirely independent of the project file in our framework’s own Xcode project. (Thanks to Caleb Davenport at North for pointing this out.) This means that even if we specify a -DDEBUG flag for the Debug build configuration in our framework’s build settings, they won’t be there when CocoaPods installs the framework into our app’s workspace.

So let’s set the flag higher up, say in our app target’s build settings. Well it turns out that those flags don’t trickle down to our framework targets at compile time. Any flags you set on the app target only apply to the app target.

OK, different idea – why don’t we make the changes in our podspec instead, using pod_target_xcconfig? Unfortunately, it doesn’t seem possible to set flags for only our Debug configuration, which is the whole point. And besides, we don’t want to be beholden to the consumer of our API — what if they’re using a different naming convention for their build configurations?

Fortunately, we can use CocoaPods’s post_install_hooks to get what we want. As you can see in the docs, each framework target holds an array of build_configurations representing the xcconfig files generated for each of our project’s build configurations. Each of these build_configuration objects then holds a hash of build_settings representing the structured data inside the xcconfig file. Using post_install_hooks we can just write out the relevant flags for the configurations we care about.

post_install do |installer|
    installer.pods_project.targets.each do |target|
        if target.name == 'CoolFramework'
            target.build_configurations.each do |config|
                if config.name == 'Debug'
                    config.build_settings['OTHER_SWIFT_FLAGS'] = '-DDEBUG'
                    else
                    config.build_settings['OTHER_SWIFT_FLAGS'] = ''
                end
            end
        end
    end
end

Bong bong.

Getting Started with Moya

Moya is functional networking library, built on top of Alamofire, that applies the best of Swift’s language and compiler features to your network requests. Its key insight is tying URL request-building logic to an enumerated type, both guaranteeing that you don’t miss anything (switch statements need to be exhaustive), and allowing you to cleanly factor code out of your main suite of public “call this endpoint” functions. On top of that, there are subspecs supporting both ReactiveCocoa and RxSwift, making it relatively easy to support functional reactive programming in your project.

What’s in the Box?

Target

At its heart, Moya revolves around a single protocol called MoyaTarget. This is the enumerated type alluded to earlier, and encompasses your base URL and paths, supplied parameters, and HTTP methods. Let’s use the Twitter API for this example, and call our implementation TwitterTarget.

public enum Target: MoyaTarget {
    case .HomeTimeline
    case .NewPost(text: String)

    var baseURL: NSURL { return NSURL(string: "https://api.twitter.com/1.1/")! }

    var path: String { 
        switch self {
            case .HomeTimeline:
                return "statuses/user_timeline.json"
            case .NewPost:
                return "statuses/update.json"
        } 
    }

    var method: Moya.Method {
        switch self {
            case .HomeTimeline:
                return .GET
            case .NewPost:
                return .POST
        }
    }

    var parameters: [String: AnyObject] {
        switch self {
            case .HomeTimeline:
                return [:]
            case .NewPost(let text):
                return ["status": text]
        }
    }

    var sampleData: NSData { return NSData() }  // We just need to return something here to fully implement the protocol
}

So for each of the protocol methods we can switch on self (enums ftw), using let bindings to pull out associated values where we need them. The only non-obvious method is sampleData — it’s related to testing, we’ll come back to it later.

Provider

In order to interact with our TwitterTarget type we’ll need an instance of MoyaProvider; its request method is the gateway between TwitterTarget and Twitter’s servers. While we’re at it, let’s write a public function to wrap all of this stuff up.

    let provider = MoyaProvider<TwitterTarget>()

    public func getTimeline() -> (tweets: [Tweet]?, error: ErrorType?) {
        provider.request(.HomeTimeline) { [unowned self] (data: NSData?, statusCode: Int?, response: NSURLResponse?, error: ErrorType?) in
            guard self.requestSucceeded(statusCode) else {
                let error = // some error
                return (nil, error)
            }

            guard let data = data,
            json = try? NSJSONSerialization(data: data, options: NSJSONReadingOptions(),
            jsonArray = json as? [[String: AnyObject]] else {
                let error = // some error
                return (nil, error)
            }

            let tweets = jsonArray.flatMap { Tweet(json: $0) }
            return (tweets, nil)  
        }
    }

Using Moya we’ve factored everything but the bare essentials out of our public getTimeline function. The resulting code is simple, safe, and easy to reason about.

Testing

An awesome feature of Moya is the ability to test our logic with canned API responses with just a couple of small changes to our code. First, let’s implement that sampleData method for real using the sample responses in Twitter’s API docs:

public enum Target: MoyaTarget {

    (...)

    var sampleData: NSData {
        switch self {
            case .HomeTimeline:
                let jsonStr = "[{\"coordinates\": null,\"truncated\": false,\"created_at\": \"Tue Aug 28 21:16:23 +0000 2012\",\"favorited\": false,\"id_str\": \"240558470661799936\",\"in_reply_to_user_id_str\": null,\"entities\": {\"urls\": [],\"hashtags\": [],\"user_mentions\": []},\"text\": ...]"    // truncated because massive
                if let data = jsonStr.dataUsingEncoding(NSUTF8StringEncoding) {
                    return data
                } else {
                    print("Couldn't serialize string")
                    return NSData()
                }

            case .NewPost:
                (...)
        }
    }
}

Now in order to use this in our tests, we’re going to need an instance of MoyaProvider that we’ve configured for testing:

    let testProvider = MoyaProvider<TwitterTarget>(endpointClosure: MoyaProvider.DefaultEndpointMapping, endpointResolver: MoyaProvider.DefaultEndpointResolution, stubBehavior: MoyaProvider.ImmediateStubbingBehaviour)

This is almost the default configuration we get with a new MoyaProvider, except for the stubBehavior parameter. We’re using it to tell the provider that we want to use the stubbed responses we specified above, and not actually call out to the network. Now we can write tests like this (using Quick and Nimble, of course):

    describe("Getting the user's home timeline") {
        it("Should have some tweets") {
            getTimeline() { (tweets, error) in
                expect(tweets).toNot(beNil())
            }
        }
    }

Where to go from Here?

Headers

In order to add custom headers (say, for authentication), you’re going to need to provide a value for the endpointClosure parameter of MoyaProvider.


let ourEndpointClosure = { (target: TwitterTarget) -> Endpoint in
        let url = target.baseURL.URLByAppendingPathComponent(target.path).absoluteString
        let endpoint = Endpoint(URL: url, sampleResponse: .Success(200, {target.sampleData}), method: target.method, parameters: target.parameters)
        let headers = headersForTarget(target)
        return endpoint.endpointByAddingHTTPHeaderFields(headers)
}

let provider = MoyaProvider<TwitterTarget>(endpointClosure: ourEndpointClosure)

Functional Reactive Programming

If you’d rather use signals than completion closures you can still use everything we’ve done so far, only swapping the base MoyaProvider out with RxMoyaProvider (for RxSwift) or ReactiveCocoaMoyaProvider (for Reactive Cocoa). Now the provider’s request method returns an Observable (or Signal) that you can map, filter, and bind to stuff.

Happy coding!

Welcome

What is Marginal Futility?

In the world of economics, marginal utility is the amount of benefit provided by some small change — another slice of pizza, an extra minute of leisure time, or that one last beer. With Marginal Futility, my goal is to log the little things I figure out as I try to get better at programming, hopefully giving others a leg up along the way.

The blog is inspired by awesome people like Julia Evans, Natasha the Robot, and Ash Furrow, who advocate writing at all costs. Also, a massive thank you to the Recurse Center, the greatest place on earth.