axiom-swift-testing（axiom Swift 测试）

Swift Testing

Overview

Swift Testing is Apple's modern testing framework introduced at WWDC 2024. It uses Swift macros (@Test, #expect) instead of naming conventions, runs tests in parallel by default, and integrates seamlessly with Swift concurrency.

Core principle: Tests should be fast, reliable, and expressive. The fastest tests run without launching your app or simulator.

The Speed Hierarchy

Tests run at dramatically different speeds depending on how they're configured:

Key insight: Move testable logic into Swift Packages or frameworks, then test with swift test or "None" host application.

Building Blocks

@Test Functions

import Testing

@Test func videoHasCorrectMetadata() {
    let video = Video(named: "example.mp4")
    #expect(video.duration == 120)
}

Key differences from XCTest:

• No test prefix required — @Test attribute is explicit
• Can be global functions, not just methods in a class
• Supports async, throws, and actor isolation
• Each test runs on a fresh instance of its containing suite

#expect and #require

// Basic expectation — test continues on failure
#expect(result == expected)
#expect(array.isEmpty)
#expect(numbers.contains(42))

// Required expectation — test stops on failure
let user = try #require(await fetchUser(id: 123))
#expect(user.name == "Alice")

// Unwrap optionals safely
let first = try #require(items.first)
#expect(first.isValid)

Why #expect is better than XCTAssert:

• Captures source code and sub-values automatically
• Single macro handles all operators (==, >, contains, etc.)
• No need for specialized assertions (XCTAssertEqual, XCTAssertNil, etc.)

Error Testing

// Expect any error
#expect(throws: (any Error).self) {
    try dangerousOperation()
}

// Expect specific error type
#expect(throws: NetworkError.self) {
    try fetchData()
}

// Expect specific error value
#expect(throws: ValidationError.invalidEmail) {
    try validate(email: "not-an-email")
}

// Custom validation
#expect {
    try process(data)
} throws: { error in
    guard let networkError = error as? NetworkError else { return false }
    return networkError.statusCode == 404
}

@Suite Types

@Suite("Video Processing Tests")
struct VideoTests {
    let video = Video(named: "sample.mp4")  // Fresh instance per test

    @Test func hasCorrectDuration() {
        #expect(video.duration == 120)
    }

    @Test func hasCorrectResolution() {
        #expect(video.resolution == CGSize(width: 1920, height: 1080))
    }
}

Key behaviors:

• Structs preferred (value semantics, no accidental state sharing)
• Each @Test gets its own suite instance
• Use init for setup, deinit for teardown (actors/classes only)
• Nested suites supported for organization

Traits

Traits customize test behavior:

// Display name
@Test("User can log in with valid credentials")
func loginWithValidCredentials() { }

// Disable with reason
@Test(.disabled("Waiting for backend fix"))
func brokenFeature() { }

// Conditional execution
@Test(.enabled(if: FeatureFlags.newUIEnabled))
func newUITest() { }

// Time limit
@Test(.timeLimit(.minutes(1)))
func longRunningTest() async { }

// Bug reference
@Test(.bug("https://github.com/org/repo/issues/123", "Flaky on CI"))
func sometimesFailingTest() { }

// OS version requirement
@available(iOS 18, *)
@Test func iOS18OnlyFeature() { }

Tags for Organization

// Define tags
extension Tag {
    @Tag static var networking: Self
    @Tag static var performance: Self
    @Tag static var slow: Self
}

// Apply to tests
@Test(.tags(.networking, .slow))
func networkIntegrationTest() async { }

// Apply to entire suite
@Suite(.tags(.performance))
struct PerformanceTests {
    @Test func benchmarkSort() { }  // Inherits .performance tag
}

Use tags to:

• Run subsets of tests (filter by tag in Test Navigator)
• Exclude slow tests from quick feedback loops
• Group related tests across different files/suites

Parameterized Testing

Transform repetitive tests into a single parameterized test:

// ❌ Before: Repetitive
@Test func vanillaHasNoNuts() {
    #expect(!IceCream.vanilla.containsNuts)
}
@Test func chocolateHasNoNuts() {
    #expect(!IceCream.chocolate.containsNuts)
}
@Test func almondHasNuts() {
    #expect(IceCream.almond.containsNuts)
}

// ✅ After: Parameterized
@Test(arguments: [IceCream.vanilla, .chocolate, .strawberry])
func flavorWithoutNuts(_ flavor: IceCream) {
    #expect(!flavor.containsNuts)
}

@Test(arguments: [IceCream.almond, .pistachio])
func flavorWithNuts(_ flavor: IceCream) {
    #expect(flavor.containsNuts)
}

Two-Collection Parameterization

// Test all combinations (4 × 3 = 12 test cases)
@Test(arguments: [1, 2, 3, 4], ["a", "b", "c"])
func allCombinations(number: Int, letter: String) {
    // Tests: (1,"a"), (1,"b"), (1,"c"), (2,"a"), ...
}

// Test paired values only (3 test cases)
@Test(arguments: zip([1, 2, 3], ["one", "two", "three"]))
func pairedValues(number: Int, name: String) {
    // Tests: (1,"one"), (2,"two"), (3,"three")
}

Benefits Over For-Loops

Fast Tests: Architecture for Testability

Strategy 1: Swift Package for Logic (Fastest)

Extract app logic into a Swift Package. Tests run with swift test (~0.4s) instead of xcodebuild test (~25s) — no simulator, no app launch. This is the key enabler for TDD in Claude Code hooks.

Step 1: Create Package.swift

Create the package directory alongside your .xcodeproj:

// MyAppCore/Package.swift
// swift-tools-version: 6.0
import PackageDescription

let package = Package(
    name: "MyAppCore",
    platforms: [.iOS(.v18), .macOS(.v15)],
    products: [
        .library(name: "MyAppCore", targets: ["MyAppCore"]),
    ],
    targets: [
        .target(name: "MyAppCore"),
        .testTarget(name: "MyAppCoreTests", dependencies: ["MyAppCore"]),
    ]
)

Step 2: Link Package to App

Create an .xcworkspace containing both the app project and the package:

1. File → New → Workspace
2. Drag your .xcodeproj into the workspace
3. File → Add Package Dependencies → Add Local → select MyAppCore/
4. Add MyAppCore framework to your app target's "Frameworks, Libraries, and Embedded Content"

Step 3: Move Logic, Expose Root View

Move models, services, and view models into MyAppCore/Sources/MyAppCore/. Types used by the app must be public. Create a public root view that accepts dependencies via injection:

// In MyAppCore
public struct MyAppRootView: View {
    @State private var appState: AppStateController

    public init(modelContainer: ModelContainer) {
        _appState = State(initialValue: AppStateController(container: modelContainer))
    }

    public var body: some View { /* ... */ }
}

Step 4: Thin-Shell App.swift

The app target becomes a thin shell that imports the package and delegates (see axiom-app-composition for the full thin-shell principle):

import SwiftUI
import MyAppCore

@main
struct MyApp: App {
    let container = try! ModelContainer(for: /* schemas */)

    var body: some Scene {
        WindowGroup {
            MyAppRootView(modelContainer: container)
        }
    }
}

What Stays vs What Moves

Tests use @testable import MyAppCore for internal access.

Running Tests

cd MyAppCore
swift test                              # All tests (~0.4s)
swift test --filter MyAppCoreTests.UserTests  # Single suite

For project-level scripts separating unit from UI tests:

# script/test
#!/bin/bash
case "${1:-unit}" in
    unit) cd MyAppCore && swift test ;;
    ui)   xcodebuild test -workspace MyApp.xcworkspace \
            -scheme MyApp -destination 'platform=iOS Simulator,name=iPhone 16' ;;
esac

Progressive Extraction for Existing Projects

For apps that can't extract everything at once, move modules incrementally:

Phase 1: Leaf Modules First

Start with code that has no dependencies on the app target:

• Data models and DTOs
• Networking layer (API clients, request builders)
• Business logic and validation rules
• Utility extensions

Phase 2: Break Circular Dependencies

If package code needs to call back into app-owned types:

1. Define a protocol in the package (the package owns the abstraction)
2. Inject a conforming implementation from the app target at startup
3. Move the implementation into the package once all its dependencies are in the package

Phase 3: Maintain Both Test Targets

During transition, keep two test targets:

• MyAppCoreTests — runs with swift test (extracted logic)
• MyAppTests — runs with xcodebuild test (remaining app-level tests)

Gradually migrate tests from MyAppTests to MyAppCoreTests as you extract their source files.

Goal: Each extraction should leave the app building and all tests passing. Never extract more than one module boundary at a time.

Strategy 2: Framework with No Host Application

For code that must stay in the app project:

1. Create a framework target (File → New → Target → Framework)
2. Move model code into the framework
3. Make types public that need external access
4. Add imports in files using the framework
5. Set Host Application to "None" in test target settings

Project Settings → Test Target → Testing
  Host Application: None  ← Key setting
  ☐ Allow testing Host Application APIs

Build+test time: ~3 seconds vs 20-60 seconds with app launch.

Strategy 3: Bypass SwiftUI App Launch

If you can't use a framework, bypass the app launch:

// Simple solution (no custom startup code)
@main
struct ProductionApp: App {
    var body: some Scene {
        WindowGroup {
            if !isRunningTests {
                ContentView()
            }
        }
    }

    private var isRunningTests: Bool {
        NSClassFromString("XCTestCase") != nil
    }
}

// Thorough solution (custom startup code)
@main
struct MainEntryPoint {
    static func main() {
        if NSClassFromString("XCTestCase") != nil {
            TestApp.main()  // Empty app for tests
        } else {
            ProductionApp.main()
        }
    }
}

struct TestApp: App {
    var body: some Scene {
        WindowGroup { }  // Empty
    }
}

Async Testing

Basic Async Tests

@Test func fetchUserReturnsData() async throws {
    let user = try await userService.fetch(id: 123)
    #expect(user.name == "Alice")
}

Testing Callbacks with Continuations

// Convert completion handler to async
@Test func legacyAPIWorks() async throws {
    let result = try await withCheckedThrowingContinuation { continuation in
        legacyService.fetchData { result in
            continuation.resume(with: result)
        }
    }
    #expect(result.count > 0)
}

Confirmations for Multiple Events

@Test func cookiesAreEaten() async {
    await confirmation("cookie eaten", expectedCount: 10) { confirm in
        let jar = CookieJar(count: 10)
        jar.onCookieEaten = { confirm() }
        await jar.eatAll()
    }
}

// Confirm something never happens
await confirmation(expectedCount: 0) { confirm in
    let cache = Cache()
    cache.onEviction = { confirm() }
    cache.store("small-item")  // Should not trigger eviction
}

Reliable Async Testing with Concurrency Extras

Problem: Async tests can be flaky due to scheduling unpredictability.

// ❌ Flaky: Task scheduling is unpredictable
@Test func loadingStateChanges() async {
    let model = ViewModel()
    let task = Task { await model.loadData() }
    #expect(model.isLoading == true)  // Often fails!
    await task.value
}

Solution: Use Point-Free's swift-concurrency-extras:

import ConcurrencyExtras

@Test func loadingStateChanges() async {
    await withMainSerialExecutor {
        let model = ViewModel()
        let task = Task { await model.loadData() }
        await Task.yield()
        #expect(model.isLoading == true)  // Deterministic!
        await task.value
        #expect(model.isLoading == false)
    }
}

Why it works: Serializes async work to main thread, making suspension points deterministic.

Deterministic Time with TestClock

Use Point-Free's swift-clocks to control time in tests:

import Clocks

@MainActor
class FeatureModel: ObservableObject {
    @Published var count = 0
    let clock: any Clock<Duration>
    var timerTask: Task<Void, Error>?

    init(clock: any Clock<Duration>) {
        self.clock = clock
    }

    func startTimer() {
        timerTask = Task {
            while true {
                try await clock.sleep(for: .seconds(1))
                count += 1
            }
        }
    }
}

// Test with controlled time
@Test func timerIncrements() async {
    let clock = TestClock()
    let model = FeatureModel(clock: clock)

    model.startTimer()

    await clock.advance(by: .seconds(1))
    #expect(model.count == 1)

    await clock.advance(by: .seconds(4))
    #expect(model.count == 5)

    model.timerTask?.cancel()
}

Clock types:

• TestClock — Advance time manually, deterministic
• ImmediateClock — All sleeps return instantly (great for previews)
• UnimplementedClock — Fails if used (catch unexpected time dependencies)

Parallel Testing

Swift Testing runs tests in parallel by default.

When to Serialize

// Serialize tests in a suite that share external state
@Suite(.serialized)
struct DatabaseTests {
    @Test func createUser() { }
    @Test func deleteUser() { }  // Runs after createUser
}

// Serialize parameterized test cases
@Test(.serialized, arguments: [1, 2, 3])
func sequentialProcessing(value: Int) { }

Hidden Dependencies

// ❌ Bug: Tests depend on execution order
@Suite struct CookieTests {
    static var cookie: Cookie?

    @Test func bakeCookie() {
        Self.cookie = Cookie()  // Sets shared state
    }

    @Test func eatCookie() {
        #expect(Self.cookie != nil)  // Fails if runs first!
    }
}

// ✅ Fixed: Each test is independent
@Suite struct CookieTests {
    @Test func bakeCookie() {
        let cookie = Cookie()
        #expect(cookie.isBaked)
    }

    @Test func eatCookie() {
        let cookie = Cookie()
        cookie.eat()
        #expect(cookie.isEaten)
    }
}

Random order helps expose these bugs — fix them rather than serialize.

Known Issues

Handle expected failures without noise:

@Test func featureUnderDevelopment() {
    withKnownIssue("Backend not ready yet") {
        try callUnfinishedAPI()
    }
}

// Conditional known issue
@Test func platformSpecificBug() {
    withKnownIssue("Fails on iOS 17.0") {
        try reproduceEdgeCaseBug()
    } when: {
        ProcessInfo().operatingSystemVersion.majorVersion == 17
    }
}

Better than.disabled because:

• Test still compiles (catches syntax errors)
• You're notified when the issue is fixed
• Results show "expected failure" not "skipped"

Migration from XCTest

Comparison Table

Keep Using XCTest For

• UI tests (XCUIApplication)
• Performance tests (XCTMetric)
• Objective-C tests

Migration Tips

1. Both frameworks can coexist in the same target
2. Migrate incrementally, one test file at a time
3. Consolidate similar XCTests into parameterized Swift tests
4. Single-test XCTestCase → global @Test function

Common Mistakes

❌ Mixing Assertions

// Don't mix XCTest and Swift Testing
@Test func badExample() {
    XCTAssertEqual(1, 1)  // ❌ Wrong framework
    #expect(1 == 1)       // ✅ Use this
}

❌ Using Classes for Suites

// ❌ Avoid: Reference semantics can cause shared state bugs
@Suite class VideoTests { }

// ✅ Prefer: Value semantics isolate each test
@Suite struct VideoTests { }

❌ Forgetting @MainActor

// ❌ May fail with Swift 6 strict concurrency
@Test func updateUI() async {
    viewModel.updateTitle("New")  // Data race warning
}

// ✅ Isolate to main actor
@Test @MainActor func updateUI() async {
    viewModel.updateTitle("New")
}

❌ Over-Serializing

// ❌ Don't serialize just because tests use async
@Suite(.serialized) struct APITests { }  // Defeats parallelism

// ✅ Only serialize when tests truly share mutable state

❌ XCTestCase with Swift 6.2 MainActor Default

Swift 6.2's default-actor-isolation = MainActor breaks XCTestCase:

// ❌ Error: Main actor-isolated initializer 'init()' has different
// actor isolation from nonisolated overridden declaration
final class PlaygroundTests: XCTestCase {
    override func setUp() async throws {
        try await super.setUp()
    }
}

Solution: Mark XCTestCase subclass as nonisolated:

// ✅ Works with MainActor default isolation
nonisolated final class PlaygroundTests: XCTestCase {
    @MainActor
    override func setUp() async throws {
        try await super.setUp()
    }

    @Test @MainActor
    func testSomething() async {
        // Individual tests can be @MainActor
    }
}

Why: XCTestCase is Objective-C, not annotated for Swift concurrency. Its initializers are nonisolated, causing conflicts with MainActor-isolated subclasses.

Better solution: Migrate to Swift Testing (@Suite struct) which handles isolation properly.

Xcode Optimization for Fast Feedback

Turn Off Parallel XCTest Execution

Swift Testing runs in parallel by default; XCTest parallelization adds overhead:

Test Plan → Options → Parallelization → "Swift Testing Only"

Turn Off Test Debugger

Attaching the debugger costs ~1 second per run:

Scheme → Edit Scheme → Test → Info → ☐ Debugger

Delete UI Test Templates

Xcode's default UI tests slow everything down. Remove them:

1. Delete UI test target (Project Settings → select target → -)
2. Delete UI test source folder

Disable dSYM for Debug Builds

Build Settings → Debug Information Format
  Debug: DWARF
  Release: DWARF with dSYM File

Check Build Scripts

Run Script phases without defined inputs/outputs cause full rebuilds. Always specify:

• Input Files / Input File Lists
• Output Files / Output File Lists

Checklist

Before Writing Tests

• Identify what can move to a Swift Package (pure logic)
• Set up framework target if package isn't viable
• Configure Host Application: None for unit tests

Writing Tests

• Use @Test with clear display names
• Use #expect for all assertions
• Use #require to fail fast on preconditions
• Use parameterization for similar test cases
• Add .tags() for organization

Async Tests

• Mark test functions async and use await
• Use confirmation() for callback-based code
• Consider withMainSerialExecutor for flaky tests

Parallel Safety

• Avoid shared mutable state between tests
• Use fresh instances in each test
• Only use .serialized when absolutely necessary

Resources

WWDC: 2024-10179, 2024-10195

Docs: /testing, /testing/migratingfromxctest, /testing/testing-asynchronous-code, /testing/parallelization

GitHub: pointfreeco/swift-concurrency-extras, pointfreeco/swift-clocks

History: See git log for changes