Changing Dependency context for UI Testing

[YamiDaisuke]

We have a project using the old Reducer and wanted to migrate to the ReducerProtocol. But we have a problem running our UI Tests. Previously we use this:

    static var environment: AppEnvironment {
        #if RELEASE
            return AppEnvironment.live
        #else
            if ProcessInfo.isUITesting || ProcessInfo.isUnitTesting {
                return AppEnvironment.test
            } else {
                return AppEnvironment.live
            }
        #endif
    }

Now we are trying to rely on the Dependency context which works perfect for Previews and Unit tests but for UI Testing we had to do this to make it work:

private lazy var store: StoreOf<AppReducer> = {
        if ProcessInfo.isUITesting {
            return StoreOf<AppReducer>(
                initialState: .loading,
                reducer: AppReducer()
                    .dependency(\.context, .test)
                    .dependency(\.date, DateGenerator { Date() })
                    .dependency(\.mainQueue, .immediate)
            )
        } else {
            return StoreOf<AppReducer>(
                initialState: .loading,
                reducer: AppReducer()
            )
        }
    }()

We are not sure if there is a better approach. We don't like that we have to provide alternatives to the dependencies provided by TCA (E.G.: mainQueue, date) to avoid falling into the unimplemented versions, so maybe there is a better way to do it. We use the test context for UI Test because we don't want to reach our backend we want to have fixed and predictable values

[mbrandonw]

Hi @YamiDaisuke, thanks for bringing this up. We have only used XCUITest a small amount, so we have some questions.

What are the problems you are having while running the UI test? Are you getting test failures from the dependencies? Or is it because the test value of @Dependency(\.mainQueue) doesn't actually execute its work?

[mbrandonw]

Interesting, since UI tests work by running two processes, one for the host app and one for the test, I'm surprised that XCTFails in the host process can cause failures in the test process.

Can you try pointing TCA to main and let us know how it differs, if at all? We made a few small changes to how test dependencies are accessed, and are curious if it changes anything for you.

Independent of that, I'm realizing that we should update our unimplemented scheduler to behave more like an immediately scheduler. Right now it just performs an XCTFail and then does nothing.

[mbrandonw]

We just cut a release of combine-schedulers that make unimplemented schedulers behave like immediate schedulers, so that may help a bit too.

[YamiDaisuke]

Interesting, since UI tests work by running two processes, one for the host app and one for the test, I'm surprised that XCTFails in the host process can cause failures in the test process.

Can you try pointing TCA to main and let us know how it differs, if at all? We made a few small changes to how test dependencies are accessed, and are curious if it changes anything for you.

Independent of that, I'm realizing that we should update our unimplemented scheduler to behave more like an immediately scheduler. Right now it just performs an XCTFail and then does nothing.

It's weird it nos actually breaking by the XCTFail but I caught that in the logs, the app is just not doing the work so the test fails. I'll try with main and let you know if that helps, but the change in combine-schedulers sounds like it will work.

[mbrandonw]

It's weird it nos actually breaking by the XCTFail but I caught that in the logs, the app is just not doing the work so the test fails.

Ah ok, that's how I expected things to work. The XCTFails are not causing the failures (you can ignore the logs), and instead what is causing problems is that the "unimplemented" main queue doesn't ever execute its work. I think if you update your application to use 0.9.0 of combine-schedulers things should work.

[YamiDaisuke]

ok! got it, it makes sense. I'll try the new version! thanks for the help!