JUnit 5 integration (rift-java-junit5)

@RiftTest starts one Rift engine per test class, creates any @RiftImposter-declared imposters against it, and injects @InjectRift/@InjectImposter fields and parameters. It's a plain ExtendWith(RiftTestExtension.class) meta-annotation — no base class required.

import io.github.achirdlabs.rift.Imposter;
import io.github.achirdlabs.rift.dsl.ImposterSpec;
import io.github.achirdlabs.rift.junit5.*;
import static io.github.achirdlabs.rift.dsl.RiftDsl.*;

@RiftTest
class UserClientTest {

    @RiftImposter
    static ImposterSpec usersSpec = imposter("users").record()
            .stub(onGet("/api/users/1").willReturn(okJson("{\"id\":1}")));

    @InjectImposter("users")
    Imposter users;

    @Test
    void fetchesUser(@InjectRift Rift rift) {
        // point your SUT at users.uri(), then:
        users.verify(onGet("/api/users/1"), times(1));
    }
}

@RiftTest

Attribute	Default	Meaning
transport()	Transport.AUTO	How the engine is obtained — see below.
adminUri()	""	Admin API URI for Transport.CONNECT (and AUTO, once set). Supports a ${property} placeholder resolved against a JVM system property of the same name.
reset()	Reset.PER_TEST	When configured imposters are reset during the class run — see the table below.
dumpRecordedOnFailure()	false	On a failing test, publish each imposter's recorded requests to the JUnit report — see Failure diagnostics.

Transport values:

• AUTO — prefers an embedded (in-process FFM) engine when rift-java-embedded/the native library is available, otherwise spawns and manages a rift binary for the class lifetime.
• CONNECT — connects to an already-running admin API at adminUri(). Requires a non-blank resolved adminUri; throws IllegalStateException otherwise.
• SPAWN — always spawns and manages a rift binary for the class lifetime.
• EMBEDDED — always runs the engine in-process (requires JDK 22+ and rift-java-embedded).

The ${property} placeholder is useful when the admin URI is only known at runtime — e.g. a @BeforeAll/static initializer that starts a fake or real admin server and publishes its port:

@RiftTest(transport = Transport.CONNECT, adminUri = "${my.admin.uri}")
class SomeTest {
    static { System.setProperty("my.admin.uri", startFakeAdmin().toString()); }
}

(JUnit only guarantees a test class's static initializer has run once the class is loaded, which RiftTestExtension forces before resolving adminUri().)

@RiftImposter static specs

@RiftImposter marks a static ImposterSpec field; RiftTestExtension builds and creates one imposter per such field before any test method runs, keyed by the spec's own declared name (the name passed to RiftDsl.imposter(String), read client-side — no engine round-trip required to echo it back). Declaring two fields with the same name is a class-level configuration error (IllegalStateException at beforeAll), as is a null or unnamed spec.

@RiftImposter
static ImposterSpec usersSpec = imposter("users").record()
        .stub(onGet("/api/users/1").willReturn(okJson("{\"id\":1}")));

@RiftImposter
static ImposterSpec paymentsSpec = imposter("payments").record();

@InjectImposter / @InjectRift

Both work on instance fields and on test-method parameters:

@InjectImposter("users") Imposter users;   // field

@Test
void aTest(@InjectRift Rift rift, @InjectImposter("payments") Imposter payments) {
    ...
}

@InjectImposter("name") looks up the imposter created from the @RiftImposter field with that declared name; an unknown name fails parameter/field resolution. @InjectRift hands back the class's single shared Rift client.

Reset semantics

Reset value	When it runs	What it clears
PER_TEST (default)	Before every test method	Recorded requests, scenario state, and proxy responses on every @RiftImposter-configured imposter.
PER_CLASS	Once, before the first test method	Same as above, but only once for the whole class.
NONE	Never (automatically)	Nothing — the test manages imposter state itself, e.g. across intentionally stateful test methods.

Reset only touches imposters declared via @RiftImposter; imposters created ad hoc inside a test method (rift.create(...)) are the test's own responsibility to clean up.

@RiftTest(reset = Reset.PER_CLASS)
class SharedStateTest {
    @RiftImposter
    static ImposterSpec usersSpec = imposter("users").record();

    @InjectImposter("users") Imposter users;

    @Test
    @Order(1)
    void firstCallIsRecorded() {
        // ... drive the SUT ...
        users.verify(onGet("/api/users/1"), times(1));
    }

    @Test
    @Order(2)
    void recordingPersistsAcrossMethods() {
        // still 1, not reset between methods under PER_CLASS
        users.verify(onGet("/api/users/1"), times(1));
    }
}

Parallel execution

RiftTestExtension starts exactly one engine per test class (in beforeAll) and tears it down in afterAll. That makes test classes safe to run in parallel with each other under JUnit 5's parallel execution — each gets its own engine, its own imposters, and its own admin connection, with no shared mutable state between classes. Test methods within the same class share that one engine and its @RiftImposter imposters, so:

• Don't enable method-level parallelism within a @RiftTest class unless your test methods only read/verify disjoint imposters or paths — concurrent methods hitting the same imposter's recorded requests will race against each other's traffic and against PER_TEST resets.
• Prefer class-level parallelism (JUnit 5's default granularity) and keep Reset.PER_TEST (the default) so each method starts from a clean slate.

Tier-2: programmatic @RegisterExtension builder

When transport or imposters are computed at runtime (or you want to avoid the class annotation), build the extension programmatically and register it as a static field with @RegisterExtension. It supports exactly the same field/parameter injection and reset semantics as the annotation:

class UserClientTest {
    static final FakeAdmin admin = FakeAdmin.start();

    @RegisterExtension
    static final RiftTestExtension rift = RiftTestExtension.newInstance()
            .transport(Transport.CONNECT)
            .adminUri(admin.uri().toString())      // computed at runtime
            .imposter(imposter("users").record())  // repeatable; one per call
            .reset(Reset.PER_TEST)
            .dumpRecordedOnFailure(true)
            .build();

    @Test
    void fetchesUser(@InjectRift Rift r, @InjectImposter("users") Imposter users) { ... }
}

newInstance() defaults match the annotation (Transport.AUTO, Reset.PER_TEST, dumpRecordedOnFailure=false). adminUri(...) also honours the ${property} placeholder.

Configuring the engine

There are two ways to configure the engine the extension constructs, matching the two tiers:

Annotation tier — launch properties. @RiftTest stays deliberately small; point the embedded engine at a specific library and relax the version preflight from the launch command (no code change), which is exactly what a dev/vendored engine needs:

-Drift.ffi.lib=native/librift_ffi.dylib -Drift.versionCheck=off

(rift.ffi.lib/RIFT_FFI_LIB set the library path; rift.versionCheck/RIFT_VERSION_CHECK accept off/warn/fail.)

Builder tier — options objects. For full programmatic control, hand the builder the options for the transport it will use. Each is validated against the selected transport (AUTO may resolve to embedded or spawn, never connect):

@RegisterExtension
static final RiftTestExtension rift = RiftTestExtension.newInstance()
        .transport(Transport.EMBEDDED)
        .embeddedOptions(EmbeddedOptions.builder()
                .libraryPath(Path.of("native/librift_ffi.dylib"))
                .versionCheck(VersionCheck.OFF)
                .build())
        .build();
// .connectOptions(ConnectOptions.builder(uri)...) for CONNECT (carries its own adminUri)
// .spawnOptions(SpawnOptions.builder()...)        for SPAWN / AUTO

A mismatched pairing (e.g. transport(SPAWN) with embeddedOptions(...)) fails fast at build().

Failure diagnostics (dumpRecordedOnFailure)

@RiftTest(dumpRecordedOnFailure = true) (or .dumpRecordedOnFailure(true) on the builder) makes a failing test publish each imposter's recorded requests to the JUnit report as an entry keyed rift.recorded.<name>, one METHOD path per line, capped at 20 requests per imposter (with a … N more note when truncated). Nothing is published for a passing test or when the flag is off. It's a fast way to see what traffic actually reached a mock when a test fails:

@RiftTest(dumpRecordedOnFailure = true)
class OrderFlowTest {
    @RiftImposter static ImposterSpec users = imposter("users").record();

    @Test
    void placesOrder(@InjectImposter("users") Imposter users) {
        // ... drive the SUT ...
        assertEquals(200, response.status());   // on failure, the report shows what hit `users`
    }
}

Most IDEs and the Surefire/Gradle reports surface report entries next to the failed test; only imposters that actually recorded a request produce an entry.

Golden files (@RiftGolden)

@RiftGolden records a real upstream once and replays it forever after — Hoverfly-style auto-capture with zero mode-switching. Add it to a @RiftTest class alongside the imposter to record:

@RiftTest
@RiftGolden(origin = "https://api.real-service.com",
            file = "src/test/resources/golden/users-api.json")
class UsersGoldenTest {
    @RiftImposter
    static ImposterSpec users = imposter("users").record();

    @InjectImposter("users") Imposter usersMock;

    @Test void fetchesUser() { /* drive your SUT through usersMock.uri() */ }
}

• file missing → CAPTURE: the imposter proxies to origin, records the traffic, and the recorded stubs are written to file when the class finishes.
• file present → REPLAY: the recorded stubs are loaded from file and served directly — no network, so CI (which has the committed golden file) never touches the origin.
• -Drift.golden=recapture forces CAPTURE even when the file exists (refresh the recording).

Commit the golden file. The persisted format is the engine's replayable imposter JSON — portable across the rift SDKs and loadable by rift --configfile. @RiftGolden targets the sole @RiftImposter on the class, or the one named by @RiftGolden(imposter = "...") when there is more than one. It builds on the core Recording API.

Intercept (@RiftIntercept)

@RiftIntercept starts a TLS-MITM intercept listener for the test class. The listener and its CA live for the class; only its rules reset per test (per the @RiftTest Reset policy). Declare rules with a @RiftInterceptRules static method and get the live handle with @InjectIntercept:

@RiftTest(transport = Transport.EMBEDDED)
@RiftIntercept(port = 18888, caCert = "certs/ca.pem", caKey = "certs/ca-key.pem",
               exportTruststore = "target/rift-truststore.jks", exportFormat = TruststoreFormat.JKS)
class FeatureFlagTest {

    @RiftImposter static ImposterSpec partner = imposter("partner").record()
            .stub(onGet("/api/quote").willReturn(okJson("{\"px\":1}")));

    @RiftInterceptRules
    static void rules(Intercept intercept, @InjectImposter("partner") Imposter partner) {
        intercept.serve("cdn.optimizely.com", okJson(DATAFILE));
        intercept.redirectTo("api.partner.com", partner);
    }

    @InjectIntercept Intercept intercept;

    @Test
    void resolvesFlag() {
        HttpClient sut = HttpClient.newBuilder()
                .sslContext(intercept.trust().sslContext())   // or sslContextWithSystemCAs()
                .proxy(intercept.proxySelector())
                .build();
        // … drive the SUT; its intercepted HTTPS calls are served by the rules above …
    }
}

• @RiftIntercept attributes: port/host (bind; host is an IP literal), caCert/caKey (committed CA, else ephemeral; ${property} placeholders resolved), and exportTruststore/exportFormat/ exportPassword (written during beforeAll, for a containerized SUT to mount — see docs/intercept.md).
• @RiftInterceptRules — a static void method invoked once the listener starts and re-invoked after each per-test rules reset. Its parameters resolve like test parameters: Intercept, @InjectRift Rift, @InjectImposter Imposter.
• @InjectIntercept — injects the live Intercept into a field or test parameter. Present without @RiftIntercept on the class → a clear resolution error.

The listener and CA are stable across a class's tests, so trust material handed to a SUT in one test stays valid in the next; only the rules reset.