Testing and Quality

Testing and Quality

Herald uses Swift Testing (not XCTest). Tests are organized by feature, with injected fakes for all dependencies.

Test structure

• HeraldTests/ — App-level tests (view models, compose, notifications, analytics)
• HeraldKit/Tests/HeraldKitTests/ — Kit-level tests (sync, auth, API, drafts)

Test suites use the @Suite / @Test macro syntax. Each test file is a struct with test methods.

Testing conventions

Swift Testing conventions:

• Swift Testing only (@Suite / @Test / #expect); no XCTest
• Protocol-oriented fakes (injected via nonisolated protocols)
• Protocols meant to be adopted by fakes are declared nonisolated protocol … : Sendable
• No mocking frameworks; hand-written fakes
• Assertions use #expect(condition), not XCTAssertTrue

Common fakes

• FakeMailAPIClient (HeraldKit/Tests/HeraldKitTests/Sync/FakeMailAPIClient.swift:9) — Records calls and returns fixtures
• FakeServer (HeraldKit/Tests/HeraldKitTests/Support/FakeServer.swift:55) — URLProtocol-based fake HTTP server
• FakeTokenProvider (HeraldKit/Tests/HeraldKitTests/Support/FakeTokenProvider.swift:8) — In-memory account and token store
• FakeOutbox (HeraldTests/ComposeViewModelTests.swift:10) — Records sent messages

Running tests

From Xcode, select the test target and press Cmd+U.

From the command line:

xcodebuild test -project Herald.xcodeproj -scheme Herald -destination 'platform=macOS' -derivedDataPath DerivedData

The Herald scheme runs both HeraldTests and HeraldKitTests (roughly 370 tests today). Analytics tests run inside the sandboxed host, so they use throwaway app ids and temp queue directories — never the production UserDefaults suite.

UI Tests — The herald-ui-tests scheme exercises the full app with keyboard and mouse automation. Do NOT run UI tests automatically or in CI; they take over the screen and keyboard. Run scripts/ui-tests.sh only with explicit approval from the project owner, as it locks the machine for the duration of the test.

Discriminating tests

Tests are "discriminating" — they exercise a real feature end-to-end using fakes at the boundary, not mocks scattered throughout. For example:

• An auth test uses a real AuthCoordinator with a FakeAuthorizationPresenter
• A sync test uses a real SyncEngine with a FakeMailAPIClient
• A view-model test uses a real MailViewModel with a FakeMailAPIClient and FakeOutbox

This style catches integration bugs that mocks would hide.

---

Last updated: 2026-09-29 — testing and quality runbook; added UI test warning and runbook reference