- Java Services (Not the RESTful Kind)
- Java ClassLoaders
- Managed Beans to CDI
- The Myriad Idioms For Finding Implementations In Java
When I was getting familiar with modern Java server development, one of the biggest conceptual stumbling blocks by far was CDI. Part of the trouble was that I kind of jumped in the deep end, by way of JNoSQL's examples. JNoSQL is a CDI citizen through and through, and so the docs would just toss out things like how you "create a repository" by just making an interface with no implementation.
Moreover, CDI has a bit of the "Maven" problem, where, once you do the work of getting familiar with it, the parts that are completely baffling to newcomers become more and more difficult to remember as being unusual.
Fortunately, like how coming to Maven by way of Tycho OSGi projects is "hard mode", coming to CDI by way of a toolkit that uses auto-created proxy objects is a more difficult path than necessary. Even better, XPages developers have a clean segue into it: managed beans.
JSF Managed Beans
XPages inherited the original JSF concept of managed beans, where you put definitions for your beans in
faces-config.xml like so:
Though the syntax isn't Faces-specific, the fact that it is defined in
faces-config.xml demonstrates what a JSF-ism it is. Newer versions of JSF (not XPages) let you declare your beans inline in the class, skipping the XML part:
These annotations were initially within the
javax.faces package, highlighting that, while they're a new developer convenience, it's still basically the same JSF-specific thing.
While all this was going on (and before it, really), the Enterprise JavaBeans (EJB) spec was chugging along, serving some similar concepts but it really is kind of its own, all-consuming beast. I won't talk about it much here, in large part because I've never used it, but it has an important part in this history, especially when we get to the "dependency injection" parts.
Move to CDI
Since it turns out that managed beans are a terrifically-useful concept beyond just JSF, Java EE siphoned concepts from JSF and EJB to make the obtusely named Contexts and Dependency Injection spec, or CDI. CDI is paired with some associated specs like Common Annotations and Inject to make a new bean system. With a switch to CDI, the bean above can be tweaked to something like:
Not wildly different - some same-named annotations in a different package, and some semantic switches, but the same basic idea. The difference here is that this is entirely divorced from JSF, and indeed from web apps in general. CDI specifically has a mode that works outside of a JEE/Servlet container and could work in e.g. a command-line program.
Newer versions of JSF (and other UI engines) deprecated their own version of this to allow for CDI to be the consistent pool of variable resolution and creation for the UI and for the business logic.
The Conceptual Leap
One of the things blocking me from properly grasping CDI at first was that
@Inject annotation on a property. If it's just some Java object, how would that property ever be set? Certainly, CDI couldn't be so magical that I could just do
new SomeBeanClass() and have
someProp populated, right? Well, yes, that's right. No matter how gussied up your class definition is with CDI annotations, constructing an instance with
new will pay no attention to any of it.
What got me over the hurdle is realizing that, in a modern web app in particular, almost everything you do runs through CDI. JSP request? That can resolve CDI. JAX-RS resource? That's managed by CDI. Filters? CDI. And, because those objects are all being instantiated by CDI, the CDI runtime can do whatever the heck it wants with them. That's why the managed property in the original example is so critical: it's the same idea, just managed by the JSF runtime instead of CDI.
That's how you can get to a class like the controller that manages the posts in this blog. It's annotated with all sorts of stuff: the JAX-RS
@Path, the MVC spec
@Controller, the CDI
@RequestScoped, and, importantly, the
@Inject'ed properties. Because the JAX-RS environment instantiates its resource classes through CDI in a JEE container, those will be populated from various sources.
HttpServletRequest comes from the servlet environment itself,
CommentRepository comes from JNoSQL as based on an interface in my non-JEE project (more on that in a bit), and
UserInfoBean is a by-the-numbers managed bean in the CDI style.
There's certainly more indirect "magic" going on here than in the
faces-config.xml starting point, but it's a clear line from there to here.
The Weird Stuff
CDI covers more ground, though, and this is the sort of thing that tripped me up when I saw the JNoSQL examples. Among CDI's toolset is the creation of "proxy" objects, which are dynamic objects that intercept normal method calls with new behavior. This is a language-level Java feature that I didn't even know this was a thing in this way, but it's been there since 1.3.
Dynamic scripting languages do this sort of thing as their bread and butter. In Ruby, you can define
method_missing to be called when code calls a method that wasn't already defined, and that can respond however you'd like. Years ago, I used this to let you do
doc.foo to get a document item value, for example. In Java, you get a mildly-less-loosey-goosey version of this kind of behavior with a proxy's
CDI does this extensively, even when you might think it's not. With CDI, all instances are dynamic proxy objects, which allows it to not only inject field values, but also add wrapper code around method calls. This allows tools like MicroProfile Metrics to do things like count invocations, measure timings, and so forth without requiring explicit code beyond the annotations.
And then there are the whole-cloth new objects, like the JNoSQL repositories. To take one of the examples from jnosql.org, here's a full definition of a JNoSQL repository as far as the app developer is concerned:
Without knowledge of CDI, this is absolute madness. How could it possibly work? There's no code! The trick to it is that CDI ends up creating a dynamic proxy implementation of the interface, which is in turn backed by an
InvocationHandler instance. That instance receives the incoming method call as a string and array of parameters, parses the method to look for a concept it handles, and either generates a result or throws an exception. Once you see the capabilities the stack has, the process to get from a JAX-RS class using
@Inject PersonRepository foo to having that actually work makes more sense:
- The JAX-RS servlet receives a request for the resource
- It asks the CDI environment to create a new instance of the resource class
- CDI runs through the fields and methods of the class to look for annotations it can handle, where it finds
- It looks through its contributed extensions and finds JNoSQL's
- One of the beans from that extension can handle creating
- That bean creates a proxy object, which handles method calls via
Still pretty weird, but at least there's a path to understanding.
The Overall Importance
The more I use modern JEE, the more I see CDI as the backbone of the whole development experience. It's even to the point where it feels unsafe to not have it present, managing objects, like everything is held together by shoestring. And its importance is further driven home by just how many specs depend on it. In addition to many existing technologies either switching to or otherwise supporting it, like JSF above, pretty much any new Jakarta EE or MicroProfile technology at least has it as the primary mechanism of interaction. Its importance can't be overstated, and it's worth taking some time either building an app with it or at least seeing some tutorials of it in action.