-
Notifications
You must be signed in to change notification settings - Fork 0
EventBusExplained
EventBus
allows publish-subscribe-style communication between components
without requiring the components to explicitly register with one another (and
thus be aware of each other). It is designed exclusively to replace traditional
Java in-process event distribution using explicit registration. It is not a
general-purpose publish-subscribe system, nor is it intended for interprocess
communication.
// Class is typically registered by the container.
class EventBusChangeRecorder {
@Subscribe public void recordCustomerChange(ChangeEvent e) {
recordChange(e.getChange());
}
}
// somewhere during initialization
eventBus.register(new EventBusChangeRecorder());
// much later
public void changeCustomer()
ChangeEvent event = getChangeEvent();
eventBus.post(event);
}
Converting an existing EventListener
-based system to use the EventBus
is
easy.
To listen for a specific flavor of event (say, a CustomerChangeEvent
)...
-
...in traditional Java events: implement an interface defined with the
event -- such as
CustomerChangeEventListener
. -
...with
EventBus
: create a method that acceptsCustomerChangeEvent
as its sole argument, and mark it with the@Subscribe
annotation.
To register your listener methods with the event producers...
-
...in traditional Java events: pass your object to each producer's
registerCustomerChangeEventListener
method. These methods are rarely defined in common interfaces, so in addition to knowing every possible producer, you must also know its type. -
...with
EventBus
: pass your object to theEventBus.register(Object)
method on anEventBus
. You'll need to make sure that your object shares anEventBus
instance with the event producers.
To listen for a common event supertype (such as EventObject
or Object
)...
- ...in traditional Java events: not easy.
-
...with
EventBus
: events are automatically dispatched to listeners of any supertype, allowing listeners for interface types or "wildcard listeners" forObject
.
To listen for and detect events that were dispatched without listeners...
- ...in traditional Java events: add code to each event-dispatching method (perhaps using AOP).
-
...with
EventBus
: subscribe toDeadEvent
. TheEventBus
will notify you of any events that were posted but not delivered. (Handy for debugging.)
To keep track of listeners to your events...
-
...in traditional Java events: write code to manage a list of listeners
to your object, including synchronization, or use a utility class like
EventListenerList
. -
...with
EventBus
:EventBus
does this for you.
To dispatch an event to listeners...
- ...in traditional Java events: write a method to dispatch events to each event listener, including error isolation and (if desired) asynchronicity.
-
...with
EventBus
: pass the event object to anEventBus
'sEventBus.post(Object)
method.
The EventBus
system and code use the following terms to discuss event
distribution:
Event | Any object that may be posted to a bus. |
---|---|
Subscribing | The act of registering a listener with an EventBus , so that its handler methods will receive events. |
Listener | An object that wishes to receive events, by exposing handler methods. |
Handler method | A public method that the EventBus should use to deliver posted events. Handler methods are marked by the @Subscribe annotation. |
Posting an event | Making the event available to any listeners through the EventBus . |
EventBus
doesn't specify how you use it; there's nothing stopping your
application from having separate EventBus
instances for each component, or
using separate instances to separate events by context or topic. This also makes
it trivial to set up and tear down EventBus
objects in your tests.
Of course, if you'd like to have a process-wide EventBus
singleton, there's
nothing stopping you from doing it that way. Simply have your container (such as
Guice) create the EventBus
as a singleton at global scope (or stash it in a
static field, if you're into that sort of thing).
In short, EventBus
is not a singleton because we'd rather not make that
decision for you. Use it how you like.
Yes, using EventBus.unregister
, but we find this is needed only rarely:
- Most listeners are registered on startup or lazy initialization, and persist for the life of the application.
- Scope-specific
EventBus
instances can handle temporary event distribution (e.g. distributing events among request-scoped objects) - For testing,
EventBus
instances can be easily created and thrown away, removing the need for explicit unregistration.
Why use an annotation to mark handler methods, rather than requiring the listener to implement an interface?
We feel that the Event Bus's @Subscribe
annotation conveys your intentions
just as explicitly as implementing an interface (or perhaps more so), while
leaving you free to place event handler methods wherever you wish and give them
intention-revealing names.
Traditional Java Events use a listener interface which typically sports only a handful of methods -- typically one. This has a number of disadvantages:
- Any one class can only implement a single response to a given event.
- Listener interface methods may conflict.
- The method must be named after the event (e.g.
handleChangeEvent
), rather than its purpose (e.g.recordChangeInJournal
). - Each event usually has its own interface, without a common parent interface for a family of events (e.g. all UI events).
The difficulties in implementing this cleanly has given rise to a pattern, particularly common in Swing apps, of using tiny anonymous classes to implement event listener interfaces.
Compare these two cases:
class ChangeRecorder {
void setCustomer(Customer cust) {
cust.addChangeListener(new ChangeListener() {
public void customerChanged(ChangeEvent e) {
recordChange(e.getChange());
}
};
}
}
versus
// Class is typically registered by the container.
class EventBusChangeRecorder {
@Subscribe public void recordCustomerChange(ChangeEvent e) {
recordChange(e.getChange());
}
}
The intent is actually clearer in the second case: there's less noise code, and the event handler has a clear and meaningful name.
Some have proposed a generic
Handler<T>
interface forEventBus
listeners. This runs into issues with Java's use of type erasure, not to mention problems in usability.
Let's say the interface looked something like the following:
interface Handler<T> {
void handleEvent(T event);
}
Due to erasure, no single class can implement a generic interface more than once with different type parameters. This is a giant step backwards from traditional Java Events, where even if
actionPerformed
andkeyPressed
aren't very meaningful names, at least you can implement both methods!
Some have freaked out about EventBus
's register(Object)
and post(Object)
methods' use of the Object
type.
Object
is used here for a good reason: the Event Bus library places no
restrictions on the types of either your event listeners (as in
register(Object)
) or the events themselves (in post(Object)
).
Event handler methods, on the other hand, must explicitly declare their argument type -- the type of event desired (or one of its supertypes). Thus, searching for references to an event class will instantly find all handler methods for that event, and renaming the type will affect all handler methods within view of your IDE (and any code that creates the event).
It's true that you can rename your @Subscribed
event handler methods at will;
Event Bus will not stop this or do anything to propagate the rename because, to
Event Bus, the names of your handler methods are irrelevant. Test code that
calls the methods directly, of course, will be affected by your renaming -- but
that's what your refactoring tools are for. We see this as a feature, not a bug:
being able to rename your handler methods at will lets you make their meaning
clearer.
Nothing at all.
The Event Bus was designed to integrate with containers and module systems, with
Guice as the prototypical example. In these cases, it's convenient to have the
container/factory/environment pass every created object to an EventBus
's
register(Object)
method.
This way, any object created by the container/factory/environment can hook into the system's event model simply by exposing handler methods.
Any problem that can be unambiguously detected by Java's type system. For example, defining a handler method for a nonexistent event type.
Immediately upon invoking register(Object)
, the listener being registered is
checked for the well-formedness of its handler methods. Specifically, any
methods marked with @Subscribe
must take only a single argument.
Any violations of this rule will cause an IllegalArgumentException
to be
thrown.
(This check could be moved to compile-time using APT, a solution we're researching.)
If a component posts events with no registered listeners, it may indicate an
error (typically an indication that you missed a @Subscribe
annotation, or
that the listening component is not loaded).
(Note that this is not necessarily indicative of a problem. There are many cases where an application will deliberately ignore a posted event, particularly if the event is coming from code you don't control.)
To handle such events, register a handler method for the DeadEvent
class.
Whenever EventBus
receives an event with no registered handlers, it will turn
it into a DeadEvent
and pass it your way -- allowing you to log it or
otherwise recover.
Because handler methods on your listener classes are normal methods, you can
simply call them from your test code to simulate the EventBus
.
EventBus
is designed to deal with a large class of use cases really, really
well. We prefer hitting the nail on the head for most use cases to doing
decently on all use cases.
Additionally, making EventBus
extensible -- and making it useful and
productive to extend, while still allowing ourselves to make additions to the
core EventBus
API that don't conflict with any of your extensions -- is an
extremely difficult problem.
If you really, really need magic thing X, that EventBus
can't currently
provide, you should file an issue, and then design your own alternative.
- Introduction
- Basic Utilities
- Collections
- Graphs
- Caches
- Functional Idioms
- Concurrency
- Strings
- Networking
- Primitives
- Ranges
- I/O
- Hashing
- EventBus
- Math
- Reflection
- Releases
- Tips
- Glossary
- Mailing List
- Stack Overflow
- Footprint of JDK/Guava data structures