Tapestry services – both those provided by Tapestry and those you write yourself – are configured using Java, not XML.
One of the key concepts in Tapestry IoC is distributed configuration. The distributed part refers to the fact that any module may configure a service. Distributed configuration is the key feature of Tapestry IoC that supports extensibility and modularity.
Modules configure a service by contributing to service configurations. This seems esoteric, but is quite handy, and is best explained by an example.
Let's say you've written a bunch of different services, each of which does something specific for a particular type of file (identified by the file's extension), and each implements the same interface, which we'll call FileServicer. And now let's say you need a central service that selects the one of your FileServicer implementations based on a given file extension. You start by providing a service builder method:
In order to provide a value for the contribution parameter, Tapestry will collect contributions from service contribution methods. It will ensure that the keys and values match the generic types shown (String for the key, FileServicer for the value). The map will be assembled and passed into the service builder method, and from there, into the FileServiceDispatcherImpl constructor.
So where do the values come from? Your service contributor methods, methods whose names start with "contribute":
Or, instead of instantiating those services ourselves, we could inject them:
The extensibility comes from the fact that multiple modules may all contribute to the same service configuration:
Now the FileServicerDispatcher builder method gets a Map with at least four entries in it.
Because Tapestry IoC is highly dynamic (it scans the visible JAR manifest files to identify module classes), the FileServicerDispatcher service may be in one module, and the other contributing modules (such as the one that contributes the Office file services) may be written at a much later date. With no change to the FileServicerDispatcher service or its module class, the new services "plug into" the overall solution, simply by having their JAR's on the runtime classpath.
|Added in 5.2|
If you prefer annotations over naming conventions you can use the @Contribute annotation. As of version 5.2 this annotation that may be placed on a contributor method of a module instead of starting the methods name with "contribute". The value of the annotation is the type of the service to contribute into.
The primary reasons to use @Contribute and marker annotations is twofold:
- There is no longer a linkage between the contribution method name and the service id, which is much more refactoring safe: if you change the service interface name, or the ID of the service, your method will still be invoked when using @Contribute.
- It makes it much easier for an override of the service to get the configuration intended for the original service.
The following example is an annotation-based alternative for the contribution method above.
If you have several implementations of a service interface, you have to disambiguate the services. For this purpose the marker annotations should be placed on the contributor method.
In this example, the method will only be invoked when constructing a service configuration where the service itself has both the Red and Blue marker annotations. Tapestry knows which annotations are marker annotations, and which marker annotations apply to the service, via the @Marker annotation on the service implementation.
If the special @Local annotation is present, then the contribution is made only to the configuration of a service being constructed in the same module.
Note that it is possible for the same contribution method to be invoked to contribute to the configuration of multiple different services.
There are three different styles of configurations (with matching contributions):
- Unordered Collection – Contributions are simply added in and order is not important.
- Ordered List – Contributions are provided as an ordered list. Contributions must establish the order by giving each contributed object a unique id, by establishing forward and backward dependencies between the values.
- Map – Contributions provide unique keys and corresponding values.
A service builder method can collect an unordered list of values by defining a parameter of type java.util.Collection. Further, you should parameterize the type of collection. Tapestry will identify the parameterized type and ensure that all contributions match.
One thing to remember is that the order in which contributions occur is unspecified. There will be a possibly large number of modules, each having zero or more methods that contribute into the service. The order in which these methods are invoked is unknown.
For example, here's a kind of Startup service that needs some Runnable objects. It doesn't care what order the Runnable objects are executed in.
Here we don't even need a separate class for the implementation, we use an inner class for the implementation. The point is, the configuration is provided to the builder method, which passes it along to the implementation of the service.
On the contribution side, a service contribution method sees a Configuration object:
The Configuration interface defines just a single method: add(). This is very intentional: the only thing you can do is add new items. If we passed in a Collection, you might be tempted to check it for values, or remove them ... but that flies in the face of the fact that the order of execution of these service contribution methods is entirely unknown.
For readability, we've parameterized the configuration parameter of the method, constraining it to instances of java.lang.Runnable, so as to match the corresponding parameter. This is optional, but often very helpful. In any case, attempting to contribute an object that doesn't extend or implement the type (Runnable) will result in a runtime warning (and the value will be ignored).
Tapestry supports only this simple form of parameterized types. Java generics supports a wider form, "wildcards", that Tapestry doesn't understand.
Ordered lists are much more common. With an ordered list, the contributions are sorted into a proper order before being provided to the service builder method.
Again, the order in which service contribution methods are invoked is unknown. Therefore, the order in which objects are added to the configuration is not known. Instead, we enforce an order on the items after all the contributions have been added. As with service decorators, we set the order by giving each contributed object a unique id, and identifying (by id) which items must preceded it in the list, and which must follow.
So, if we changed our Startup service to require a specific order for startup:
Notice that the service builder method is shielded from the details of how the items are ordered. It doesn't have to know about IDs and pre- and post-requisites. By using a parameter type of List, we've triggered Tapestry to collect all the ordering information.
For our service contribution methods, we must provide a parameter of type OrderedConfiguration:
Often, you don't care about ordering; the first form of the add method is used then. The ordering algorithm will find a spot for the object (here the JMSStartup instance) based on the constraints of other contributed objects.
For the "FileSystem" contribution, a constraint has been specified, indicating that FileSystem should be ordered after some other contribution named "CacheSetup". Any number of such ordering constraints may be specified (the add() method accepts a variable number of arguments).
The object passed in may be null; this is valid, and is considered a "join point": points of reference in the list that don't actually have any meaning of their own, but can be used when ordering other elements. TODO: Show example for chain of command, once that's put together.
Null values, once ordered, are edited out (the List passed to the service builder method does not include any nulls). Again, they are allowed as placeholders, for the actual contributed objects to organize themselves around.
|Added in 5.3|
As discussed in the earlier examples, mapped configurations are also supported. The keys passed in must be unique. When conflicts occur, Tapestry will log warnings (identifying the source of the conflict, in terms of invoked methods), and ignore the conflicting value.
Neither the key nor the value may be null.
For mapped configurations where the key type is String, a CaseInsensitiveMap will be automatically used (and passed to the service builder method), to help ensure that case insensitivity is automatic and pervasive.
All three configuration interfaces have a second method, addInstance(). This method takes a class, not an instance. The class is instantiated and contributed. If the constructor for the class takes dependencies, those are injected as well.
In addition to injecting services into a contributor method (via the @InjectService and @Inject annotations), Tapestry will key off of the parameter type to allow other things to be injected.
- ObjectLocator: access to other services visible to the contributing module
- org.slf4j.Logger: the Logger for the service being contributed to
No annotation is needed for these cases.
|Added in 5.1|
The OrderedConfiguration and MappedConfiguration interfaces now support overrides. An override is a replacement for a normally contributed object. An override must match a contributed object, and each contributed object may be overridden at most once.
The new object replaces the original object; alternately, you may override the original object with null.
This allows you to fine tune configuration values that are contributed from modules that you are using, rather than just those that you write yourself. It is powerful and a bit dangerous.
In Tapestry 5.0, services that wanted to support this kind of override behavior had to implement it on an ad-hoc basis, such as ApplicationDefaults overriding FactoryDefaults. In many cases, that is still useful.