At the core of the Avalon framework is the component. We define it as "a passive entity that performs a specific role". This is important to grasp because it requires a specific way of thinking.
A passive entity must employ a passive API. A passive API is one that is acted upon, versus one that acts itself. See the Inversion of Control pattern for an explanation.
The concept of roles comes from the theater. A play, musical, or movie will have a certain number of roles that actors play. Although there never seems to be a shortage of actors, there are a finite number of roles. I am not going to make reference to different types of roles at this point, but simply bring the concept to light. The function or action of a role is defined by its script.
We are introducing this concept now because you need to have it in mind when you are designing your system architecture. Think of the different roles in your system, and you will have your "cast" of components so to speak.
For each role, you need to specify its script, or interface to the rest of the system. To be honest the interface is not enough. There are specific contracts that you must define and keep in mind when you specify your interfaces. In other words, what users of the component must provide, and what the component produces. When the interfaces and contracts are defined, you can work on your implementation.
John Donne wrote, "No man is an island." to communicate that we are all interdependent. The same is true for the component. That is why there are different concerns regarding the component. In the section on roles we specified one of the concerns: the role. The concerns directly supported by the Avalon Framework are: configuration, external component use, management, and execution.
We used to have a marker interface component. This has been deprecated because requiring all components extend this interface makes integrating Avalon with other component systems like CORBA very cumbersome.
As you might have guessed, each one of these concerns has a separate interface that describes that concern. We will delve deeper into the interfaces and the reasoning behind them in other sections. It is important to know the order of precedence for the concerns so that you know the overall contracts of how they are put together.
The contract surrounding this order means that the methods defined by each of those interfaces are called in a specific order by the object that created the component. Each interface represents a narrow view of the component or object being controlled.
Notice that each interface is separate from Component, so you can use them for simple objects.
In Avalon, Serviceable is defined as an active entity that controls or uses components. Its best analogy is that of a musical composer. The musical composer chooses instruments (components) by their role in the symphony (system) and tells them which notes to play.
The Avalon Serviceable follows the principles of Inversion of Control, and is assigned a Service Manager. Within this section we will discuss how to look up specific components, and then how to prepare the ServiceManager for the Serviceable.
compose method must ignore all subsequent
requests to set the ServiceManager after it is successfully set.
For the majority of all cases, you will need to use the ServiceManager to get the instance of the component you need. If you recall the discussion on component roles in the component documentation, you already have a head start. In Avalon, roles are defined by the work interface a component has. A work interface is different from any other interface because it is the interface that defines the component's role. Serviceable and Component are concern interfaces because they address specific concerns about the component.
The ServiceManager has one method to retrieve all of your components.
The lookup method will look up the component based on the
fully qualified name (FQN) of the work interface (Role). See the following
example:
final MyComponent component = (MyComponent)manager.
lookup( "com.mycompany.myproject.MyComponent" );
It is important to note that role is not the same thing as functional equivalence. In other words, if you have a MailSpooler that is functionally equivalent to a FileStore (they do the same thing), it does not mean that they perform the same role. The FileStore is used to store objects to files, and the MailSpooler is used to temporarily store messages until they are sent. Thus they are separate roles. Some containers require that the interface name match the key used to lookup component. In this situation you may need to create a new interface that does nothing more than extend another interface and add a new role.
Sometimes you will have several components that function in the same role. For those cases, you will use the ServiceSelector to choose the exact one you need. The best way to describe its proper use is the scenario described here. You have several formatters that have the same role: to take an input document and format it according to the rules in the individual component implementations. One formatter may take a text file and remove all tabs and replace them with four spaces. Another formatter may reverse the formerly mentioned one. Yet another takes the text file and formats it for a canvas object. For the Serviceable, it makes no difference what the implementation does--just that it formats the text.
Using the processing chain example in the previous paragraph, we realize the unsuitability of the ServiceManager for getting the right component. The component addresses the concern of one component per role. Fortunately, the ServiceSelector is a component. That means we use the ServiceManager to lookup the ServiceSelector. The ServiceSelector is designed to choose the specific component out of many that perform the same role. The following code will help:
final ServiceSelector selector = (ServiceSelector)manager.
lookup( "org.mycompany.myproject.FormatterSelector" );
final Formatter formatter = (Formatter)selector.select( myURL );
The selector does not discriminate against lookup keys. In that respect it acts much like a hashtable lookup. Keep in mind that the implementation of the selector does not limit you to a hashtable lookup--you can dynamically instantiate objects as well. It takes an object (a hint), and returns the specific component based on that hint.
Discussions are currently taking place about the ServiceSelector interface and concept. It might be deprecated at some point in the future. There are indications that usage of the ServiceSelector is only required when an application is badly designed.
Both the ServiceManager and the ServiceSelector require you to release your component when you are done with it. The method used to do this is "release". One way of handling this is to use the try/catch/finally construct. For your convenience, the following code can help:
MyComponent component = null;
try
{
component = (MyComponent) manager.lookup("org.mycom.MyComponent");
component.myMethod();
}
catch (Exception e)
{
getLogger().debug("Error using MyComponent", e);
}
finally
{
if (component != null) manager.release(component);
}
The reason for this is so that smart component managers that select components from a pool can properly manage the resources.
It is the responsibility of the entity that creates the Serviceable to give it a ServiceManager with all of the Roles populated. If you create your own implementations of the ServiceManager and ServiceSelector then you have the liberty of deciding how to populate them. Keep in mind that there are default implementations included, and you should model their behavior as much as possible.
The DefaultComponentManager is nothing more than a Hashtable lookup of roles
and Components. It even gives you the method put to populate
the ServiceManager. One feature of the DefaultComponentManager is that
it can cascade. In other words, if the role is not found in this ServiceManager,
the default implementation will look in the parent ServiceManager.
For the paranoid developer, the cascading feature of the ServiceManager can be seen as a security hole as opposed to a usability enhancement. You are free to create your own implementation that does not use the cascading feature--but you have to manually populate it with anything that would have been in the parent ServiceManager that your child Serviceable needs. Truth be told, there is very little risk due to the set-once contract for ComponentManagers. The method is never exposed to hostile agents before the ServiceManager is set.
The DefaultComponentSelector again is simply a Hashtable selection of components
based on hints. It gives the method put to populate the ServiceSelector.
The ServiceSelector does not have the cascading feature of the ServiceManager,
nor should it. A ServiceSelector simply holds a number of components that
implement the same role--there really is no need to cascade.
After the ServiceSelector is populated, you must put it in the ServiceManager. Please use the role of the component you are selecting, not the role of the selector itself. An acceptable convention is to add the "Selector" name to the end of the role you are looking up. Just be consistent.
The container is the entity that manages your components. It handles things like loading of configuration files, resolution of dependencies, component management, component isolation, and lifecycle support.
The container is not formalized in the form of an interface or contract within Avalon Framework, though it might be at some point in the future. The informal contract for the container is that it has the ability to host any fully Avalon-Framework compliant component. Most current containers place additional requirements on the component.