Testing methodology

[mgrider]

In my application, I have a single class DefaultNetworkService that handles all the network requests. (About the name: DefaultNetworkService conforms to a NetworkService protocol, so it can be mocked when testing other parts of the app.)

In general, I've tried to leave most of the app's logic out of DefaultNetworkService, but inevitably some has crept into it, and it would be nice to write tests for it.

Worth noting it has a client: Get.APIClient property. My initial thought was to write a protocol that Get.APIClient already conforms to, and inject a mock client. The mock client was feeling like it was going to be a lot of work (even though it really only needs the send function).

Instead I took inspiration from the tests in the Get -> Tests -> GetTests folder, specifically the APIClient extension in Helpers.swift.

I quickly made my own APIClient extension, but then realized that it's this line that was doing the magic there:

            $0.sessionConfiguration.protocolClasses = [MockingURLProtocol.self]

I was tempted to include the whole Mocker folder in my project, but I couldn't find a way to do that without copying it in. (Would be curious to hear if anyone knows if that is possible.) Instead I decided I could get away with just a customized version of MockingURLProtocol. Mine looks like this:

open class MockURLProtocol: URLProtocol {

    public var cacheStoragePolicy: URLCache.StoragePolicy = .notAllowed

    static var requestError: Error?
    static var responseValue: Data?
    static var responseStatus = 200
    static var responseHeaders = [String: String]()

    static var requestedUrlStrings = [String]()

    func mockResponse() -> HTTPURLResponse {
        let response = HTTPURLResponse(
            url: URL(string: "https://example.com")!,
            statusCode: MockURLProtocol.responseStatus,
            httpVersion: "HTTP/2.0",
            headerFields: MockURLProtocol.responseHeaders)!
        return response
    }

    private var responseWorkItem: DispatchWorkItem?

    /// Returns Mocked data based on the mocks register in the `Mocker`. Will end up in an error when no Mock data is found for the request.
    override public func startLoading() {
        if let data = MockURLProtocol.responseValue {
            self.client?.urlProtocol(self, didReceive: mockResponse(), cacheStoragePolicy: cacheStoragePolicy)
            self.client?.urlProtocol(self, didLoad: data)
            self.client?.urlProtocolDidFinishLoading(self)
        } else if let error = MockURLProtocol.requestError {
            self.client?.urlProtocol(self, didFailWithError: error)
        } else {
            self.client?.urlProtocol(self, didReceive: mockResponse(), cacheStoragePolicy: cacheStoragePolicy)
            self.client?.urlProtocol(self, didLoad: MockAPIResponseEmpty.data())
            self.client?.urlProtocolDidFinishLoading(self)
        }
    }

    /// Implementation does nothing, but is needed for a valid inheritance of URLProtocol.
    override public func stopLoading() {
        responseWorkItem?.cancel()
    }

    /// Simply sends back the passed request. Implementation is needed for a valid inheritance of URLProtocol.
    override public class func canonicalRequest(for request: URLRequest) -> URLRequest {
        return request
    }

    /// Overrides needed to define a valid inheritance of URLProtocol. We store the url string here.
    override public class func canInit(with request: URLRequest) -> Bool {
        if let urlString = request.url?.path {
            MockURLProtocol.requestedUrlStrings.append(urlString)
        }
        return true
}

This is working great, except that, because it's relying on static properties, my tests can't run in parallel. I tried to think of a way to key the canned responseValue and requestErrors, but just didn't find a good way to do it. Particularly if (as I do), I want to check for requestedUrlStrings after a request. (Essentially, some of my API requests kick off other requests, and this is some of the logic I'm testing for.)

For now, I've just decided that all my network tests will happen in one (rather large) test case. It's not great, but it's way better than no tests at all.

Curious if anyone has any feedback on this approach, or thoughts on how to avoid the monolithic test case situation. I'm happy to post more examples if anyone is interested in seeing more of the code.

[kean]

Hey,

I was tempted to include the whole Mocker folder in my project

You can install it as a Swift package from WeTransfer/Mocker. It's embedded in the project to make the dependency private.

I briefly looked into the Mocker code, and it appears to be thread-safe. Now, I'm not sure what's the best way to isolate the tests. The mocks are registered for specific URLs, so that's one potential avenue. I would suggest trying to use different domains for each unit test.

[mgrider]

YES! The URL idea is a good one! I will try it on Monday.

[mgrider]

And I do also appreciate the pointer to the official Mocker repo. I do generally try and keep dependencies to a minimum, but since this is just for the test target, maybe another is not so bad.