Swift Mocking

SwiftMocking

Mocking for Swift protocols: @Mockable generates mock classes; when(...) stubs, verify(...) asserts calls. When the macro can't generate a mock — most notably protocol inheritance (protocol B: A — the macro drops inherited requirements and the mock fails to conform) — hand-write the mock following the exact generated-code shape.

The macro is not the only entry point. Two escape hatches cover everything it can't reach, and both are fully supported (see spies-and-composition.md):

• Spy used directly — a standalone recorder needing no Mock subclass and no macro. This is how you mock classes, which @Mockable cannot touch at all: subclass the class and back each override with a Spy.
• Composition over inheritance — hold a let mock = Mock() property instead of inheriting from Mock. Mock's @dynamicMemberLookup subscript is public, so mock.name resolves spies from outside the class. Required when the type already has a superclass (UIViewController, a legacy base class) or isn't a class at all (struct, actor). Substitute super.name → self.mock.name and adapt(...) → Mock.adapt(...); everything else in manual-mocking.md is unchanged.

The macro generates the composed form itself with @Mockable([.composition]) — no hand-writing — when the blocker is a protocol whose conformers must inherit a class:

@Mockable([.composition])
protocol ViewControllerService: SampleBase { ... }

Hand-write composition only for what the macro still can't express: structs, actors, and mocking a concrete class.

When to use what

API interface

Machine-generated from the compiled modules — the authoritative signature reference. Check these before guessing at an overload or a generic constraint:

• Core library: references/interface/SwiftMocking.swiftinterface
• XCTest/swift-testing helpers: references/interface/SwiftMockingTestSupport.swiftinterface

One caveat: Sendable is a marker protocol and the compiler elides it from where clauses in emitted interfaces — thenThrow<E: Error & Sendable> prints as where E : Error. The constraint is still enforced. For Sendable questions trust sendable.md, not these files.

Deterministic generation: the mockable CLI

When a mock must exist as written source (macro plugin unavailable, codegen pipeline, review), the mockable CLI is the fastest exact path. Hand-writing per manual-mocking.md is equally correct and needs nothing; use the CLI when it's available, hand-write when it isn't or when you're already customizing.

echo 'protocol P { func price(_ item: String) throws -> Int }' | mockable

Invoke as mockable when it is on PATH; otherwise .build/release/mockable inside a swift-mocking checkout (build once with swift build -c release --product mockable, then optionally copy the binary onto PATH).

• Input needs no annotation — paste the protocol as it is written in the source you're mocking.
• Pass options with --options, comma-separated (--options composition,suffixMock). Use it for protocols you can't or shouldn't annotate — most importantly --options composition for a protocol constrained to a class, the fastest way to see the composed shape:

  echo 'protocol P: UIViewController { func load() }' | mockable --options composition
  

• Options may still ride the input as @Mockable([.suffixMock]) protocol P {...}, which wins over --options. Prefer the flag: it keeps the stdin text identical to the real declaration.
• Default output keeps the macro's #if DEBUG wrapper; pass --no-debug-wrap when pasting into a test target (DEBUG is per build configuration — a wrapped mock vanishes under swift test -c release).
• A stderr warning about inherited requirements means the output will not conform — hand-write per manual-mocking.md instead.
• Output keeps the macro's zero-arg/property-getter shape: when(...) silently stubs a disconnected spy, and verify(...) reports zero calls for those members — apply the pinned-spy fix from manual-mocking.md before relying on them.

The one rule for manual mocks

Every protocol requirement gets two members in the mock class:

1. Runtime member — fulfills the protocol, forwards to the spy: adapt(super.method, args)
2. Interaction member — same name, ArgMatcher<T> parameters, returns Interaction<Inputs..., Effect, Output> — what when(...)/verify(...) consume

class FooMock: Mock, @unchecked Sendable, Foo {
    func price(_ item: String) throws -> Int {
        return try adaptThrowing(super.price, item)
    }
    func price(_ item: ArgMatcher<String>) -> Interaction<String, Throws, Int> {
        Interaction(item, spy: super.price)
    }
}

Class shell: inherit Mock first, restate @unchecked Sendable, conform to the most-derived protocol only, match access levels. If inheriting Mock isn't possible, keep both members and compose instead — let mock = Mock(), super.price → mock.price, adaptThrowing(...) → Mock.adaptThrowing(...) (spies-and-composition.md).

Full recipe (inheritance flattening, properties, subscripts, variadics, generics, statics, initializers): manual-mocking.md. Zero-parameter methods and property getters need the pinned-spy pattern described there — the macro-generated form silently mis-stubs them.

Quick verification checklist

A correct mock (manual or generated) round-trips:

let mock = FooMock()
when(mock.price(.any)).thenReturn(42)
let svc: Foo = mock                               // protocol-typed: avoids overload traps
_ = try svc.price("apple")
verify(mock.price(.equal("apple"))).called(1)

Known sharp edges

• @Mockable on protocol B: A → compile error does not conform to protocol 'A' (inherited requirements never generated). Hand-write per manual-mocking.md.
• @Mockable on protocol P: SomeClass → requires that 'MockP' inherit from 'SomeClass'. This is a class constraint, not protocol inheritance, and no hand-edit fixes the default output — the mock needs its superclass slot for SomeClass, and the default strategy needs it for Mock. Add [.composition]. The option is opt-in: the macro can't tell a class from a protocol by name, so it never infers it.
• @Mockable only accepts protocols — never a class. To fake a class, subclass it and back overrides with raw Spy properties (spies-and-composition.md). final classes/members can't be faked either way; extract a protocol.
• In a composed mock, plain adapt(...) doesn't resolve — it's an instance method on Mock and you didn't inherit it. Use the static Mock.adapt(...) / Mock.adaptThrowing(...). clear() does resolve: MockProviding supplies both the instance and static forms.
• Hand-written composed mocks must spell self.mock.name, not a bare mock.name, wherever the spy is read inside a closure — settable members do exactly that, and Swift rejects the bare form with requires explicit use of 'self' to make capture semantics explicit. super never needed the qualifier. [.composition] output already does this.
• Zero-arg members (func start(), property getters): when(mock.getX()).thenReturn(v) does not reach the runtime member in macro-generated mocks. Manual mocks fix this with the pinned-spy pattern.
• Stub API is thenReturn / thenThrow / do — there is no .then.
• Never write thenReturn(()). Void is in the default-value registry, so an unstubbed -> Void member just runs and records the call — stub it only to throw or to attach a do side effect. The registry also covers Bool/String/Int/Double/Float/Array/Set/Dictionary/Optional, so those members are callable unstubbed too (usage.md — Default values).
• Spy names come from the requirement, ignoring argument labels: methods use their name, subscripts are namespaced as subscript+ParameterNames (subscript(row:column:) → subscriptRowColumn), settable members add set+Name. The prefix keeps a subscript from colliding with a method or variable of the same name. The compiler already rejects most same-key cases (two subscripts differing only by argument label, or a var x beside a func x(), are both invalid redeclarations); the one that compiles but mocks incorrectly is two methods differing only by argument label (fetch(id:)/fetch(name:)), which silently share a spy — rename one or vary the parameter types. Same-name/different-signature overloads are fine.
• Settable members ({ get set }) record reads and writes on separate spies: verify(mock.x) counts reads, verify(mock.x <- v) counts writes. A write never registers as a read.
• Bare mock.start() (zero-arg) is ambiguous on the mock type — call via a protocol-typed reference.
• Literal arguments on the mock type dispatch to the interaction member (mock.fetchUser(id: "1") returns an Interaction instead of calling through) — call the mock via a protocol-typed reference.

References

• manual-mocking.md — hand-writing mock classes; inheritance chains; pinned-spy pattern
• spies-and-composition.md — using Spy directly; mocking classes; composing let mock = Mock() for structs/actors/existing superclasses
• usage.md — when/verify/matchers/stubbing reference
• sendable.md — Swift 6 concurrency contract and non-Sendable workarounds