Testing
What each layer’s tests prove, how the doubles work, and the commands that run everything. All unit tests use Swift Testing; XCTest appears only in the XCUITest target, where it has no replacement.
The matrix
| Suite | Runs on | Command | Proves |
|---|---|---|---|
MVIKitTests |
macOS host | cd Package/MVIKit && swift test |
The loop itself: effects feed back, identified effects switch-latest, cancel stops work, bindings round-trip |
CinematicDomainTests |
macOS host | cd Package/CinematicDomain && swift test |
Business rules: genre ordering and degradation, query hygiene, snapshot-then-changes streaming |
CinematicDataTests |
macOS host | cd Package/CinematicData && swift test |
Wire truth: DTO decoding against captured payloads, error mapping per status and URLError, local search ranking, cache fallback, persistence round-trips |
CinematicDesignTests |
iOS simulator | xcodebuild test -scheme CinematicDesign -destination 'platform=iOS Simulator,name=iPhone 17' |
Token invariants |
CinematicPresentationTests |
iOS simulator | xcodebuild test -scheme CinematicPresentation -destination 'platform=iOS Simulator,name=iPhone 17' |
Every reducer’s state machine, pure and through the live loop |
CinematicTests + CinematicUITests |
iOS simulator | xcodebuild test -project Cinematic.xcodeproj -scheme Cinematic -destination 'platform=iOS Simulator,name=iPhone 17' |
Composition wiring, deep-link routing, and three black-box user flows |
The split is platform-driven: packages that never import UIKit or SwiftUI declare macOS support and test in seconds with swift test; UI packages test on a simulator.
Doubles, by layer
Domain and presentation test against hand-rolled stubs, not mocks. Catalog stubs are value types configured with Results at construction — no shared state, no expectations DSL. The favorites stub is a real actor with the same broadcast semantics the production repository honors, because the contract (subscribe-before-snapshot, emit-on-every-change) is exactly what the tests must exercise.
Data tests stub the network underneath a real URLSession via URLProtocol, so decoding, status handling, and error mapping run the production path. The stub is Mutex-guarded because Swift Testing runs in parallel; suites that share fixed URLs are marked .serialized.
Previews, presentation tests, and UI tests share one fixture set — PreviewSupport in CinematicPresentation. The app’s -uiTestMode launch argument swaps the composition root onto those same doubles, which is why the UI tests are deterministic and run offline.
Testing the loop
Request-shaped flows settle:
store.send(.task)
await store.settle()
#expect(store.state.catalog.value?.featured == PreviewCatalog.movies)
settle() waits for every in-flight effect, including effects spawned by the intents those effects send.
Stream-fed stores never settle — their observation effect is deliberately immortal — so those tests drive the stub and poll with a bounded helper:
store.send(.task)
await favorites.toggle(movie)
await waitUntil { store.state.isFavorite }
waitUntil yields between checks and records an issue on timeout. The pattern to remember: settle for pipelines, poll for subscriptions, and end a stub’s streams (finishAllStreams) if you do need a subscription store to settle.
Reading coverage
Every xcodebuild test invocation above accepts -resultBundlePath TestResults.xcresult; inspect with:
xcrun xccov view --report --only-targets TestResults.xcresult
The logic targets are the ones to watch — reducers, use cases, mappers, and repositories are the design’s claims, and the suites above hold them at high coverage (the presentation package sits near 90% including view code; the pure layers higher). Chasing percentage inside SwiftUI body code buys little; the UI tests cover those paths end to end instead.