Fusion framework aims at providing a very light framework.

Core Values

For cloud applications, being the most reactive possible is a key criteria so Fusion chose to:

  • Be build time oriented: only delegate to runtime the bean resolution (to enable dynamic module aggregation) and not the bean and model discovery nor proxy generation,

  • Stay flexible: even if the model is generated at build time you can generally still customize it by removing a bean and adding your own one,

  • Be native friendly: some applications need to be native to start very fast and bypass the classloading and a bunch of JVM init, for that purpose we ensure the framework is GraalVM friendly - using Apache Geronimo Arthur or not.

Features

  • Field/constructor injections

  • Contexts/scopes

    • Default scope (@DefaultScoped) is to create an instance per lookup/injection

    • Application scope (@ApplicationScoped) creates an instance per container and instantiates it lazily (at first call)

  • Event bus

    • Start/Stop events are fired with container related lifecycle hooks,

  • Lifecycle: @Init/@Destroy to react to the bean lifecycle,

  • Optional<MyClass> injections,

  • Basic cloud friendly configuration,

  • Adding fusion-json module, you can get JSON records mapping support without reflection,

  • Adding fusion-http-server module, you get an Apache Tomcat abstraction enabling to write any HTTP endpoint in an efficient and GraalVM friendly manner,

  • Adding fusion-testing module, you get some JUnit 5 integration enabling to test easily your code.

  • Adding fusion-cli module, you can generate simple CLI applications just writing commands.

Tip
you can find examples on the examples page.

No interceptor support

Since some years we got used to see declarative interceptors (annotations) like in this snippet:

public class MyBean {
    @Traced
    public void doSomething() {
        // ...
    }
}

These are great and the container is actually linking the annotation to an implementation (a bean in general) which intercepts the call. This is not bad but has some design pitfalls:

  • Most interceptors will use parameters and for such a generic approach to work, it needs an Object[] (or List) of parameters. This is really not fast (it requires to allocate an array for that purpose).

  • It requires to know and understand the rules between class interceptors, method interceptors, appending/overriding when relevant plus the same with parent classes. All that can quickly become complex.

  • It is often static: once put on a method disabling an interceptor requires the underlying library to be able to do that or to use some advanced customization at startup to do it.

For these reasons, we think that we don’t need an interceptor solution in Fusion framework but we don’t say the underlying feature is pointless, not at all. However, thanks to a more modern programming style, we can use a more functional approach to solve the same problem. Therefore, previous example would rather become:

public class MyBean {
    public void doSomething() {
        tracing(() -> {
            // ...
        });
    }
}

The big advantage is you can use some static utility if you want but also rely on beans and even combine more efficiently interceptions in a custom and configurable fashion:

public class MyBean {
    public void doSomething() {
        tracing(() -> timed(() -> logged(() -> {
            // ...
        })));
    }
}

can become:

public class MyBean {
    @Injection
    MyObservabilityService obs;

    public void doSomething() {
        obs.instrumented(() -> {
            // ...
        });
    }
}

If you compare the case with parameters it is way more efficient in general since you just do a standard parameter passing call:

public class MyBean {
    @Injection
    EntityManager em; // assume the application used JPA - not required

    @Injection
    JpaService jpa; // custom bean to handle transactions for ex

    public void store(final Transaction tx) {
        tracing(
          // no Object[] created for an interceptor
          // and no reflection to extract the id
          tx.id(),
          () -> jpa.tx(() -> em.persist(tx)));
    }
}
Tip

going with this solution can, however, get the chaining lambda pitfall (a.k.a. callback hell in JavaScript), to solve this one we encourage you to ensure your "interceptor" can be chained properly using the same kind of callback.

Here is an example (the important part is more the signature than the fact it is a static method or a bean method):

public static <T> Supplier<T> interceptor1(String marker, Map<String, String> data, Supplier<T> nested) {
    return () -> {
        logger.info(message(marker, data)); // interceptor role
        return task.get();  // intercepted business, "ic.proceed()" in jakarta interceptor API
    };
}


public static <T> Supplier<T> interceptor12(Params params, Supplier<T> nested) {
    // same kind of logic for the impl
}

Thanks this definition which commonly agreed to use Supplier<T> as the intercepted call and the fact interceptor methods return a call and not execute it directly, you can chain them more easily:

public void storeCustomer(final Customer customer) {
    interceptor2(
            Params.of(customer),
            interceptor1(
                "incoming-customer", Map.of("id", customer.id()),
                () -> {
                    // business code
                }))
    .get(); // trigger the actual execution, it is the terminal operation for the chain
}

If you want to go further you can use a Stream to represent that. Now an interceptor is a Function<Supplier<T>, Supplier<T>> so if you define the list of interceptors in a Stream, then you can just reduce them using the business function/logic as identity to have the actual invocation and execute it. Only detail to take care: ensure to reverse the stream to call the interceptor in order:

public void storeCustomer(final Customer customer) {
    Stream.<Function<Supplier<Void>, Supplier<Void>>>of(
                // reversed chain of interceptor (i1 will be executed before i2)
                delegate -> interceptor2(Params.of(customer), delegate),
                delegate -> interceptor1("incoming-customer", Map.of("id", customer.id()), delegate)
        )
        // merge the stream of interceptors as one execution wrapper
        .reduce(identity(), Function::andThen)
        .apply(() -> { // apply to the actual business logic
            System.out.println(">Business");
            return null;
        })
        .get(); // execute it
}

Indeed in practise you can extract that kind of code in an utility and use something like:

// utility
public static <T> T intercepted(final Supplier<T> execution, final Function<Supplier<T>, Supplier<T>>... interceptors) {
    return Stream.of(interceptors)
            .reduce(identity(), Function::andThen)
            .apply(execution)
            .get();
}

// usage
intercepted(
    () -> { // business logic
        System.out.println(">Business");
        return null;
    },
    // interceptors
    delegate -> interceptor2(Params.of(customer), delegate),
    delegate -> interceptor1("incoming-customer", Map.of("id", customer.id()), delegate)
);

This is what the class io.yupiik.fusion.framework.api.composable.Wraps does.

Last tip: you interceptor can work with CompletionStage to add some behavior before/after the call even if the result is not computed synchronously ;).

Limitations

Note
these are limitations as of today, none are technically strong limitations we can’t fix at a later point if desired.
  • A no-arg constructor must be available for any class bean,

  • If a method producer bean is AutoCloseable then it will be automatically closed,

  • Event methods can not be package scope if the enclosing bean uses a subclass proxy (like @ApplicationScoped context),

  • Constructor injections are supported but for proxied scopes (@ApplicationScoped for ex) it requires a default no-arg constructor (in scope protected or public) in the class (if not existing the instantiation constructor will be called with null parameters),

  • Event bus listeners can only have the event as method parameter,

  • Only classes are supported exception for method producers which can return a ParameterizedType (ex: List<String>) but injections must exactly match this type and List/Set injections are handled by looking up all beans matching the parameter.

Setup

See setup page to see how to get your project started.