CLI to generate Java POJOs and interfaces from Conjure API definitions.

Note that all Java APIs in this repository (including conjure-java-core) are considered "internal" in the sense that they may change at any time and without prior warning or deprecation notice. These packages are published for convenience, but users must assume responsibility for adapting their code when dependencies change.

The recommended way to use conjure-java is via a build tool like gradle-conjure. However, if you don't want to use gradle-conjure, there is also an executable which conforms to RFC 002, published on bintray.

Usage: conjure-java generate [...options] <input> <output>

Generate Java bindings for a Conjure API
    <input>      Path to the input IR file
    <output>     Output directory for generated source

Options:
    --objects    Generate POJOs for Conjure type definitions
    --jersey     Generate jax-rs annotated interfaces for client or server-usage
    --undertow   Generate undertow handlers and interfaces for server-usage
    --retrofit   Generate retrofit interfaces for streaming/async clients
    --requireNotNullAuthAndBodyParams
                 Generate @NotNull annotations for AuthHeaders and request body params
    --retrofitCompletableFutures
                 Generate retrofit services which return Java8 CompletableFuture instead of OkHttp Call (deprecated)
    --retrofitListenableFutures
                 Generate retrofit services which return Guava ListenableFuture instead of OkHttp Call
    --undertowServicePrefixes
                 Generate service interfaces for Undertow with class names prefixed 'Undertow'
    --undertowListenableFutures
                 Generate Undertow services which return Guava ListenableFuture for asynchronous processing
    --useImmutableBytes
                 Generate binary fields using the immutable 'Bytes' type instead of 'ByteBuffer'

Conjure-java supports feature flags to enable additional opt-in features. To enable features provided by a feature flag, simply supply the feature flag as an additional parameter when executing conjure-java. Available feature flags can be found in the FeatureFlags class.

Conjure-java objects are always immutable and thread-safe. Fields are never null; instead, Java 8 Optionals are used. JSON serialization is handled using Jackson annotations.

• Conjure object: ManyFieldExample

  Objects can only be instantiated using the builder pattern:

  ManyFieldExample example = ManyFieldExample.builder()
          .string("foo")
          .integer(123)
          .optionalItem("bar")
          .addAllItems(iterable)
          .build();

  Or using Jackson via conjure-jackson-serialization:

  ObjectMapper mapper = ObjectMappers.newServerObjectMapper();
  ManyFieldExample fromJson = mapper.readValue("{\"string\":\"foo\", ...}", ManyFieldExample.class);

• Conjure union: UnionTypeExample

  Union types can be one of a few variants. To interact with a union value, users should use the .accept method and define a Visitor that handles each of the possible variants, including the possibility of an unknown variant.

  Foo output = unionTypeExample.accept(new Visitor<Foo>() {
  
      public Foo visitStringExample(StringExample value) {
          // your logic here!
      }
  
      public Foo visitSet(Set<String> value) {}
  
      // ...
  
      public Foo visitUnknown(String unknownType) {}
  
  });

  Visitors may seem clunky in Java, but they have the upside of compile-time assurance that you've handled all the possible variants. If you upgrade an API dependency and the API author added a new variant, the Java compiler will force you to explicitly deal with this new variant. We intentionally avoid switch statements and instanceof checks for this exact reason.

• Conjure enum: EnumExample

  Enums are subtly different from regular Java enums because they tolerate deserializing unknown values. This is important because it ensures server authors can add new variants to an enum without causing runtime errors for consumers that use older API jars.

  EnumExample one = EnumExample.ONE;
  System.out.println(one); // prints: 'ONE'
  
  EnumExample fromJson = mapper.readValue("\"XYZ\"", EnumExample.class);
  System.out.println(fromJson); // prints: 'XYZ'

• Conjure alias: StringAliasExample

  Aliases have exactly the same JSON representation as their inner type, so they are useful for making error-prone function signatures more bulletproof:

  -doSomething(String, String, String);
  +doSomething(ProductId, UserId, EmailAddress);

Conjure-java generates Java interfaces with JAX-RS annotations, so they can be used on the client side or on the server-side.

Example jersey interface: EteService

@Consumes(MediaType.APPLICATION_JSON)
@Produces(MediaType.APPLICATION_JSON)
@Path("/")
@Generated("com.palantir.conjure.java.services.JerseyServiceGenerator")
public interface EteService {
    @GET
    @Path("base/string")
    String string(@HeaderParam("Authorization") AuthHeader authHeader);
}

Use conjure-java-runtime's JaxRsClient which configures Feign with sensible defaults:

RecipeBookService recipes = JaxRsClient.create(
        RecipeBookService.class,
        userAgent,
        hostMetrics,
        clientConfiguration);

List<Recipe> results = recipes.getRecipes();

You need a Jersey-compatible server. We recommend Dropwizard (which is based on Jetty), but Grizzly, Tomcat, etc should also work fine. Use conjure-java-runtime's ConjureJerseyFeature to configure Jackson and Exception mappers appropriately. Then, you just need to implement the interface:

public final class RecipeBookResource implements RecipeBookService {

    @Override
    public List<Recipe> getRecipes() {
        // ...
    }
}

Then in your server main method, register your resource. For example, using Dropwizard:

 public void run(YourConfiguration configuration, Environment environment) {
     // ...
+    environment.jersey().register(new RecipeBookResource());
 }

As an alternative to the JAX-RS interfaces above, conjure-java can generate equivalent interfaces with Retrofit2 annotations. These clients are useful if you want to stream binary data or make non-blocking async calls:

Example Retrofit2 interface: EteServiceRetrofit

public interface EteServiceRetrofit {

    @GET("./base/binary")
    @Streaming
    Call<ResponseBody> binary(@Header("Authorization") AuthHeader authHeader);
}

Use conjure-java-runtime's Retrofit2Client which configures Retrofit2 with sensible defaults:

RecipeBookServiceRetrofit recipes = Retrofit2Client.create(
            RecipeBookServiceRetrofit.class,
            userAgent,
            hostMetrics,
            clientConfiguration);

Call<List<Recipe>> asyncResults = recipes.getRecipes();

In the undertow setting, for a ServiceName conjure defined service, conjure will generate an interface: ServiceName to be extended by your resource and an UndertowService named ServiceNameEndpoints

To avoid conflicts with the equivalent jersey interface (when you are generating both), use in your build.gradle:

conjure {
    java {
        undertowServicePrefixes = true
    }
}

With this option, Undertow generated interfaces will be prefixed with Undertow. Here the interface would be called: UndertowServiceName

To use the generated handlers:

public static void main(String[] args) {
    Undertow server = Undertow.builder()
            .addHttpListener(8080, "0.0.0.0")
            .setHandler(ConjureHandler.builder()
                    .addAllEndpoints(RecipeBookServiceEndpoints.of(new RecipeBookResource())
                            .endpoints(ConjureUndertowRuntime.builder().build()))
                    .build())
            .build();
    server.start();
}

The Conjure Undertow generator supports asynchronous request processing allowing all service methods to return a Guava ListenableFuture. These methods may be implemented synchronously by replacing return object with return Futures.immediateFuture(object).

Asynchronous request processing decouples the HTTP request lifecycle from server task threads, allowing you to replace waiting on shared resources with callbacks, reducing the number of required threads.

This feature is enabled using:

conjure {
    java {
        undertowListenableFutures = true
    }
}

Asynchronous request processing is helpful for endpoints which do not need a thread for the entirety of the request.

👍 Delegation to an asynchronous client, for instance either retrofit or dialogue 👍

@Override
public ListenableFuture<String> getValue() {
    // Assuming this retrofit client was compiled with --retrofitListenableFutures
    return retrofitClient.getValue();
}

👍 Long polling 👍

Long polling provides lower latency than simple repeated polling, but requests take a long time relative to computation. A single thread can often handle updating all polling requests without blocking N request threads waiting for results.

@Override
public ListenableFuture<String> getValue() {
    SettableFuture<String> result = SettableFuture.create();
    registerFuture(result);
    return result;
}

👎 Not for delegation to synchronous operations, Feign clients for example 👎

This example is less efficient than return Futures.immediateFuture(feignClient.getValue()) because you pay an additional cost to switch threads, and maintain an additional executor beyond the configured server thread pool.

ListeningExecutor executor = MoreExecutors.listeningDecorator(Executors.newCachedThreadPool());

@Override
public ListenableFuture<String> getValue() {
    // BAD: do not do this!
    return executor.submit(() -> feignClient.getValue());
}

👎 Not waiting on another thread 👎

This is even less efficient than the previous example because it requires two entire threads for the duration of the request. It's reasonable to defer computation to an executor bounded to the work, but you should wrap the executor with MoreExecutors.listeningDecorator and return the future to avoid blocking a server worker thread for both the queued and execution durations of the task.

ExecutorService executor = Executors.newFixedThreadPool(CORES);

@Override
public ListenableFuture<BigInteger> getValue() {
    // BAD: do not do this!
    Future<BigInteger> future = executor.submit(() -> complexComputation());
    BigInteger result = Uninterruptibles.getUninterruptibly(future);
    return Futures.immediateFuture(result);
}

By default, conjure-java will use java.nio.ByteByffer to represent fields of Conjure type binary. However, the ByteBuffer class has many subtleties, including interior mutability.

To avoid many of these foot-guns, we recommend using the useImmutableBytes feature flag to opt-in to using the com.palantir.conjure.java.lib.Bytes class. This will become the default behaviour in a future major release of conjure-java.

For instructions on how to set up your local development environment, check out the Contributing document.

This project is made available under the Apache 2.0 License.