Discovering and Registering Programmatically-Generated Tests with Swift Testing

Hi all!

Long time lurker, first time poster. I'm the maintainer of Quick & Nimble.

I'm in the process of expanding Quick to also support/be built on top of Swift Testing. You can view my current progress in this draft PR to Quick.

One of the first problems I'm running in to is: how do I actually create and register tests with the swift testing runner? Earlier today, I filed an Issue based on a narrow interpretation of "well, I need some Test instances, then I somehow register those", but after some minor feedback, I don't know if that's actually the right approach.

There's a related issue of the entire bootstrapping process: How do I ensure that my test discovery code gets runs at the proper time so that my tests can get discovered and registered with Swift Testing? I've clearly not done too much research on this, but there doesn't appear to be an obvious (pure-swift) way to do this.



I can’t help you answer your question. But can say, thank you for your work on Quick and Nimble! And working on swift-testing integration :pray:t2:

1 Like

Fundamentally, Swift Testing relies on information about tests that's produced at compile time and discovered at runtime. It's intentional that no public initializer is offered for Test because just instantiating the type is not sufficient to make a test runnable.

A big part of Swift Testing's design philosophy is that we know what tests exist up front, before we start running any, so that we can plan out a test run (by skipping invalid tests, by serializing interdependent tests, etc. etc.) Having new instances of Test appear after a test plan is constructed is not something we want to support if we can avoid it.

I'm going to put aside the idea of adding a Test.register() function or similar, for the moment. I'm not ruling it out—just for the purposes of this forum post I'm ignoring it. :slight_smile:

In order for your test code to get discovered by Swift Testing with the implementation we have today, you need to create a type, give it a particular funky name, conform it to a private protocol, and add a static __tests property that enumerates your tests. This mechanism/interface for test discovery is not meant to be permanent.

Thoughts about a future test discovery ABI

Eventually, we intend to migrate to a dedicated runtime metadata section (swift5_tests). We're waiting for @kubamracek's @_section attribute to land in some form before we can do that.

We haven't defined the ABI for that section yet, but it's intended to be of a form that can be used by tools like Quick. It'll likely consist of a sequence of function pointers with signatures like:

@convention(thin) @Sendable () async -> [Test]

Since Quick still needs to be able to generate instances of Test for that to work, we could generalize it a bit and say "returns an instance of any TestGeneratorProtocolFactoryBeanWhatever" instead, and provide a public documented interface. This is appealing over simply providing Test.init because it's less likely to be accidentally used by a test author who thinks it'll cause a test to be dynamically run. You'd have to intentionally reach for it, meaning you probably have to read the documentation for it too!

Since the current solution (type with weird name and protocol conformance) is not permanent, we've prefixed all the public bits with double underscores to emphasize that developers shouldn't be using them. You could hypothetically write something within Quick that uses these symbols directly as a proof-of-concept.

Temporary hackish solution to the problem
struct GlueForQuick__🟠$test_container__: __TestContainer {
  static var __tests: [Test] {
    get async {
      var result = [Test]()

      for testData in ... {
            named: testData.programmaticFunctionName,
            in: testData.containingTypeIfAny,
            xcTestCompatibleSelector: nil,
            displayName: testData.humanReadableTestName,
            traits: [...],
            sourceLocation: testData.sourceLocationOfTest,
            testFunction: testData.testFunction

      return result

I want to again emphasize that this would not be a permanent solution, and at most would be an experimental implementation to allow you to move forward with your project. If you did write something like the above code, we'd ask that you be prepared to refactor it when we get our stable ABI at some point in the future, and to add lots of scary comments warning people not to copy the code because it will break someday.