swift-architecture（Swift 架构）

Swift Architecture

Select and implement the right architecture pattern for Apple platform apps built with Swift 6.3 and SwiftUI or UIKit.

Contents

• Architecture Selection
• MV Pattern (Model-View with @Observable)
• MVVM
• MVI (Model-View-Intent)
• TCA (The Composable Architecture)
• Clean Architecture
• Coordinator Pattern
• Migration Between Patterns
• Common Mistakes
• Review Checklist

Architecture Selection

Choose based on feature complexity, team size, and testing requirements.

Default recommendation for new SwiftUI apps: Start with MV (Model-View with @Observable). Escalate to MVVM or TCA only when the feature's complexity demands it.

Decision Framework

1. Is the feature a simple CRUD screen? → MV pattern
2. Does the screen have complex business logic separate from the view? → MVVM
3. Do you need deterministic state transitions and side-effect management? → MVI or TCA
4. Is the app large with many independent feature modules? → TCA or Clean Architecture
5. Is navigation complex with deep linking and conditional flows? → Add Coordinator pattern

MV Pattern

The simplest SwiftUI architecture. The view observes @Observable models directly. No intermediate view model layer.

Docs: @Observable)

import Observation
import SwiftUI

@Observable
class TripStore {
    var trips: [Trip] = []
    var isLoading = false
    var error: Error?

    private let service: TripService

    init(service: TripService) {
        self.service = service
    }

    func loadTrips() async {
        isLoading = true
        defer { isLoading = false }
        do {
            trips = try await service.fetchTrips()
        } catch {
            self.error = error
        }
    }

    func deleteTrip(_ trip: Trip) async throws {
        try await service.delete(trip)
        trips.removeAll { $0.id == trip.id }
    }
}

struct TripsView: View {
    @State private var store = TripStore(service: .live)

    var body: some View {
        List(store.trips) { trip in
            TripRow(trip: trip)
        }
        .task { await store.loadTrips() }
    }
}

When MV is enough: Single-screen features, prototype/MVP, small teams, straightforward data flow.

When to upgrade: Business logic grows complex, unit testing the view's behavior becomes difficult, multiple views need to share and transform the same state differently.

MVVM

Separates view logic into a ViewModel that the view observes. The view model transforms model data for display and handles user actions.

@Observable
class TripListViewModel {
    private(set) var trips: [TripRowItem] = []
    private(set) var isLoading = false
    var searchText = ""

    var filteredTrips: [TripRowItem] {
        guard !searchText.isEmpty else { return trips }
        return trips.filter { $0.name.localizedStandardContains(searchText) }
    }

    private let repository: TripRepository

    init(repository: TripRepository) {
        self.repository = repository
    }

    func loadTrips() async {
        isLoading = true
        defer { isLoading = false }
        let models = (try? await repository.fetchAll()) ?? []
        trips = models.map { TripRowItem(from: $0) }
    }

    func delete(at offsets: IndexSet) async {
        let toDelete = offsets.map { filteredTrips[$0] }
        for item in toDelete {
            try? await repository.delete(id: item.id)
        }
        await loadTrips()
    }
}

struct TripRowItem: Identifiable {
    let id: UUID
    let name: String
    let dateRange: String

    init(from trip: Trip) {
        self.id = trip.id
        self.name = trip.name
        self.dateRange = trip.startDate.formatted(.dateTime.month().day())
            + " – " + trip.endDate.formatted(.dateTime.month().day())
    }
}

struct TripListView: View {
    @State private var viewModel: TripListViewModel

    init(repository: TripRepository) {
        _viewModel = State(initialValue: TripListViewModel(repository: repository))
    }

    var body: some View {
        List {
            ForEach(viewModel.filteredTrips) { item in
                Text(item.name)
            }
            .onDelete { offsets in
                Task { await viewModel.delete(at: offsets) }
            }
        }
        .searchable(text: $viewModel.searchText)
        .task { await viewModel.loadTrips() }
    }
}

Testing a ViewModel:

@Test func filteredTripsMatchesSearch() async {
    let repo = MockTripRepository(trips: [
        Trip(name: "Paris"), Trip(name: "Tokyo"), Trip(name: "Paris TX")
    ])
    let vm = TripListViewModel(repository: repo)
    await vm.loadTrips()
    vm.searchText = "Paris"
    #expect(vm.filteredTrips.count == 2)
}

MVI

Unidirectional data flow: views dispatch intents, a reducer produces new state, and side effects are handled explicitly.

@Observable
class TripListStore {
    private(set) var state = State()

    struct State {
        var trips: [Trip] = []
        var isLoading = false
        var error: String?
    }

    enum Intent {
        case loadTrips
        case deleteTrip(Trip)
        case clearError
    }

    private let service: TripService

    init(service: TripService) {
        self.service = service
    }

    func send(_ intent: Intent) {
        Task { await handle(intent) }
    }

    @MainActor
    private func handle(_ intent: Intent) async {
        switch intent {
        case .loadTrips:
            state.isLoading = true
            do {
                state.trips = try await service.fetchTrips()
            } catch {
                state.error = error.localizedDescription
            }
            state.isLoading = false

        case .deleteTrip(let trip):
            try? await service.delete(trip)
            state.trips.removeAll { $0.id == trip.id }

        case .clearError:
            state.error = nil
        }
    }
}

Advantages: Predictable state transitions, easy to log/replay intents, clear separation of "what happened" from "what changed."

TCA

The Composable Architecture (Point-Free) provides composable reducers, dependency injection, exhaustive testing, and structured side effects.

Docs: TCA

import ComposableArchitecture

@Reducer
struct TripList {
    @ObservableState
    struct State: Equatable {
        var trips: IdentifiedArrayOf<Trip> = []
        var isLoading = false
    }

    enum Action {
        case onAppear
        case tripsLoaded([Trip])
        case deleteTrip(Trip.ID)
    }

    @Dependency(\.tripClient) var tripClient

    var body: some ReducerOf<Self> {
        Reduce { state, action in
            switch action {
            case .onAppear:
                state.isLoading = true
                return .run { send in
                    let trips = try await tripClient.fetchAll()
                    await send(.tripsLoaded(trips))
                }
            case .tripsLoaded(let trips):
                state.trips = IdentifiedArray(uniqueElements: trips)
                state.isLoading = false
                return .none
            case .deleteTrip(let id):
                state.trips.remove(id: id)
                return .run { _ in try await tripClient.delete(id) }
            }
        }
    }
}

Use TCA when: Large team needs consistent patterns, exhaustive test coverage is a priority, features compose from smaller features, you need structured dependency injection across the app.

Clean Architecture

Layers: Domain (entities, use cases, repository protocols) → Data (repository implementations, network, persistence) → Presentation (views, view models). Dependencies point inward.

// Domain layer
protocol TripRepository: Sendable {
    func fetchAll() async throws -> [Trip]
    func save(_ trip: Trip) async throws
    func delete(id: UUID) async throws
}

struct FetchUpcomingTripsUseCase: Sendable {
    private let repository: TripRepository

    init(repository: TripRepository) {
        self.repository = repository
    }

    func execute() async throws -> [Trip] {
        try await repository.fetchAll()
            .filter { $0.startDate > .now }
            .sorted { $0.startDate < $1.startDate }
    }
}

// Data layer
struct RemoteTripRepository: TripRepository {
    private let client: APIClient

    func fetchAll() async throws -> [Trip] {
        try await client.request(.get, "/trips")
    }
    // ...
}

// Presentation layer
@Observable
class UpcomingTripsViewModel {
    private(set) var trips: [Trip] = []
    private let useCase: FetchUpcomingTripsUseCase

    init(useCase: FetchUpcomingTripsUseCase) {
        self.useCase = useCase
    }

    func load() async {
        trips = (try? await useCase.execute()) ?? []
    }
}

Use Clean Architecture when: Strict separation is required (enterprise, regulated domains), the domain layer must be testable without any framework dependencies, or multiple presentation targets share the same business logic.

Coordinator Pattern

Separates navigation logic from views. Especially useful in UIKit or hybrid apps with complex navigation flows.

@MainActor
protocol Coordinator: AnyObject {
    var navigationController: UINavigationController { get }
    func start()
}

@MainActor
final class TripCoordinator: Coordinator {
    let navigationController: UINavigationController
    private let repository: TripRepository

    init(navigationController: UINavigationController, repository: TripRepository) {
        self.navigationController = navigationController
        self.repository = repository
    }

    func start() {
        let vm = TripListViewModel(repository: repository)
        vm.onSelectTrip = { [weak self] trip in
            self?.showDetail(for: trip)
        }
        let vc = TripListViewController(viewModel: vm)
        navigationController.pushViewController(vc, animated: false)
    }

    private func showDetail(for trip: Trip) {
        let vm = TripDetailViewModel(trip: trip, repository: repository)
        vm.onEdit = { [weak self] trip in self?.showEditor(for: trip) }
        let vc = TripDetailViewController(viewModel: vm)
        navigationController.pushViewController(vc, animated: true)
    }

    private func showEditor(for trip: Trip) {
        // ...
    }
}

In pure SwiftUI apps, NavigationStack with path-based routing often replaces the Coordinator pattern. Use Coordinators when you need UIKit integration or shared navigation logic across platforms.

Migration Between Patterns

ObservableObject → @Observable

// Before (iOS 16)
class TripStore: ObservableObject {
    @Published var trips: [Trip] = []
}
// View uses @ObservedObject or @StateObject

// After (iOS 17+)
@Observable
class TripStore {
    var trips: [Trip] = []
}
// View uses @State for owned, plain property for injected

MVVM → MV (simplifying)

If a view model only passes through model data without transforming it, remove the view model and let the view observe the model directly.

MV → MVVM (scaling up)

Extract business logic and data transformation into a view model when:

• The view's body contains conditional logic for data formatting
• Multiple views need different projections of the same model
• You need to test logic without instantiating views

Any → TCA

TCA adoption is typically incremental: wrap one feature's state and actions in a Reducer, migrate its dependencies to @Dependency, and test.

Common Mistakes

Review Checklist

• Architecture choice is justified by feature complexity and team needs
• @Observable used instead of ObservableObject for iOS 17+ targets
• Dependencies are injected, not created internally (testability)
• Navigation logic is separated from business logic
• State mutations happen in a clear, auditable location
• View models (if present) are testable without views
• No god objects — responsibilities are distributed appropriately
• Pattern is consistent within each feature module

References

• Apple docs: Observation | Observable)