Let’s look at an example trait first, then re-implement the same behavior using an AspectJ aspect, and finally compare the two approaches.

Observing with Traits

In a previous post on Scala, I gave an example of the Observer Pattern implemented using a trait. Chris Shorrock and James Iry provided improved versions in the comments. I’ll use James’ example here.

To keep things as simple as possible, let’s observe a simple Counter, which increments an internal count variable by the number input to an add method.

    
package example

class Counter {
    var count = 0
    def add(i: Int) = count += i
}
    

The count field is actually public, but I will only write to it through add.

Here is James’ Subject trait that implements the Observer Pattern.

    
package example

trait Subject {
  type Observer = { def receiveUpdate(subject:Any) }

  private var observers = List[Observer]()
  def addObserver(observer:Observer) = observers ::= observer
  def notifyObservers = observers foreach (_.receiveUpdate(this))
}
    

Effectively, this says that we can use any object as an Observer as long as it matches the structural type { def receiveUpdate(subject:Any) }. Think of structural types as anonymous interfaces. Here, a valid observer is one that has a receiveUpdate method taking an argument of Any type.

The rest of the trait manages a list of observers and defines a notifyObservers method. The expression observers ::= observer uses the List :: (“cons”) operator to prepend an item to the list. (Note, I am using the default immutable List, so a new copy is created everytime.)

The notifyObservers method iterates through the observers, calling receiveUpdate on each one. The _ that gets replaced with each observer during the iteration.

Finally, here is a specs file that exercises the code.

    
package example

import org.specs._

object CounterObserverSpec extends Specification {
    "A Counter Observer" should {
        "observe counter increments" in {
            class CounterObserver {
                var updates = 0
                def receiveUpdate(subject:Any) = updates += 1
            }
            class WatchedCounter extends Counter with Subject {
                override def add(i: Int) = { 
                    super.add(i)
                    notifyObservers
                }
            }
            var watchedCounter = new WatchedCounter
            var counterObserver = new CounterObserver
            watchedCounter.addObserver(counterObserver)
            for (i <- 1 to 3) watchedCounter.add(i)
            counterObserver.updates must_== 3
            watchedCounter.count must_== 6
    }
  }
}
    

The specs library is a BDD tool inspired by rspec in Rubyland.

I won’t discuss it all the specs-specific details here, but hopefully you’ll get the general idea of what it’s doing.

Inside the "observe counter increments" in {...}, I start by declaring two classes, CounterObserver and WatchedCounter. CounterObserver satisfies our required structural type, i.e., it provides a receiveUpdate method.

WatchedCounter subclasses Counter and mixes in the Subject trait. It overrides the add method, where it calls Counter’s add first, then notifies the observers. No parentheses are used in the invocation of notifyObservers because the method was not defined to take any!

Next, I create an instance of each class, add the observer to the WatchedCounter, and make 3 calls to watchedCounter.add.

Finally, I use the “actual must_== expected” idiom to test the results. The observer should have seen 3 updates, while the counter should have a total of 6.

The following simple bash shell script will build and run the code.

    
SCALA_HOME=...
SCALA_SPECS_HOME=...
CP=$SCALA_HOME/lib/scala-library.jar:$SCALA_SPECS_HOME/specs-1.3.1.jar:bin
rm -rf bin
mkdir -p bin
scalac -d bin -cp $CP src/example/*.scala
scala -cp $CP example/CounterObserverSpec
    

Note that I put all the sources in a src/example directory. Also, I’m using v1.3.1 of specs, as well as v2.7.1 of Scala. You should get the following output.

    
Specification "CounterObserverSpec" 
  A Counter Observer should
  + observe counter increments

Total for specification "CounterObserverSpec":
Finished in 0 second, 60 ms
1 example, 2 assertions, 0 failure, 0 error
    

Observing with Aspects

Because Scala compiles to Java byte code, I can use AspectJ to advice Scala code! For this to work, you have to be aware of how Scala represents its concepts in byte code. For example, object declarations, e.g., object Foo {...} become static final classes. Also, method names like + become $plus in byte code.

However, most Scala type, method, and variable names can be used as is in AspectJ. This is true for my example.

Here is an aspect that observes calls to Counter.add.

    
package example

public aspect CounterObserver {
    after(Object counter, int value): 
        call(void *.add(int)) && target(counter) && args(value) {

        RecordedObservations.record("adding "+value);
    }
}
    

You can read this aspect as follows, after calling Counter.add (and keeping track of the Counter object that was called, and the value passed to the method), call the static method record on the RecordedObservations.

I’m using a separate Scala object RecordedObservations

    
package example

object RecordedObservations {
    private var messages = List[String]()
    def record(message: String):Unit = messages ::= message
    def count() = messages.length
    def reset():Unit = messages = Nil
}
    

Recall that this is effectively a static final Java class. I need this separate object, rather than keeping information in the aspect itself, because of the simple-minded way I’m building the code. ;) However, it’s generally a good idea with aspects to delegate most of the work to Java or Scala code anyway.

Now, the “spec” file is:

    
package example

import org.specs._

object CounterObserverSpec extends Specification {
    "A Counter Observer" should {
        "observe counter increments" in {
            RecordedObservations.reset()
            var counter = new Counter
            for (i <- 1 to 3) counter.add(i)
            RecordedObservations.count() must_== 3
            counter.count must_== 6
    }
  }
}
    

This time, I don’t need two more classes for the adding a mixin trait or defining an observer. Also, I call RecordedObservations.count to ensure it was called 3 times.

The build script is also slightly different to add the AspectJ compilation.

    
SCALA_HOME=...
SCALA_SPECS_HOME=...
ASPECTJ_HOME=...
CP=$SCALA_HOME/lib/scala-library.jar:$SCALA_SPECS_HOME/specs-1.3.1.jar:$ASPECTJ_HOME/lib/aspectjrt.jar:bin
rm -rf bin app.jar
mkdir -p bin
scalac -d bin -cp $CP src/example/*.scala 
ajc -1.5 -outjar app.jar -cp $CP -inpath bin src/example/CounterObserver.aj
aj -cp $ASPECTJ_HOME/lib/aspectjweaver.jar:app.jar:$CP example.CounterObserverSpec
    

The ajc command not only compiles the aspect, but it “weaves” into the compiled Scala classes in the bin directory. Actually, it only affects the Counter class. Then it writes all the woven and unmodified class files to app.jar, which is used to execute the test. Note that for production use, you might prefer load-time weaving.

The output is the same as before (except for the milliseconds), so I won’t show it here.

Comparing Traits with Aspects

So far, both approaches are equally viable. The traits approach obviously doesn’t require a separate language and corresponding tool set.

However, traits have one important limitation with respect to aspects. Aspects let you define pointcuts that are queries over all possible points where new behavior or modifications might be desired. These points are called join points in aspect terminology. The aspect I showed above has a simple pointcut that selects one join point, calls to the Counter.add method.

However, what if I wanted to observe all state changes in all classes in a package? Defining traits for each case would be tedious and error prone, since it would be easy to overlook some cases. With an aspect framework like AspectJ, I can implement observation at all the points I care about in a modular way.

Aspect frameworks support this by providing wildcard mechanisms. I won’t go into the details here, but the * in the previous aspect is an example, matching any type. Also, one of the most powerful techniques for writing robust aspects is to use pointcuts that reference only annotations, a form of abstraction. As a final example, if I add an annotation Adder to Counter.add,

    
package example

class Counter {
    var count = 0
    @Adder def add(i: Int) = count += i
}
    

Then I can rewrite the aspect as follows.

    
package example

public aspect CounterObserver {
    after(Object counter, int value): 
        call(@Adder void *.*(int)) && target(counter) && args(value) {

        RecordedObservations.record("adding "+value);
    }
}
    

Now, there are no type and method names in the pointcut. Any instance method on any visible type that takes one int (or Scala Int) argument and is annotated with Adder will get matched.

Note: Scala requires that you create any custom annotations as normal Java annotations. Also, if you intend to use them with Aspects, use runtime retention policy, which will be necessary if you use load-time weaving.

Conclusion

If you need to mix in behavior in a specific, relatively-localized set of classes, Scala traits are probably all you need and you don’t need another language. If you need more “pervasive” modifications (e.g., tracing, policy enforcement, security), consider using aspects.

Acknowledgements

Thanks to Ramnivas Laddad, whose forthcoming 2^nd Edition of AspectJ in Action got me thinking about this topic.

Aquarium uses Ruby’s metaprogramming API extensively and the JRuby team has done some pretty sophisticated work to integrate Java types with Ruby. Hence, it’s not too surprising there are some gotchas. Hopefully, workarounds will be possible for all of them.

The details are discussed on the JRuby page, the README on the Aquarium site, and of course the “specs” in the distribution’s jruby/spec directory. I’ll summarize them here, after discussing the pros and cons of Aquarium vs. the venerable AspectJ and showing you an example of using Aquarium for Java.

Briefly, Aquarium’s advantages over AspectJ are these:

1. You can add and remove advice dynamically at runtime. You can’t remove AspectJ advice.
2. You can advise JDK types easily with Aquarium. AspectJ won’t do this by default, but this is really more of a legacy licensing issue than a real technical limitation.
3. You can advise individual objects, not just types.

Aquarium’s disadvantages compared to AspectJ include:

1. Aquarium will be slower than using AspectJ (although this has not been studied in depth yet).
2. Aquarium’s pointcut language is not as full-featured as AspectJ’s.
3. There are the bugs and limitations I mentioned above in this initial V0.4.0 release, which I’ll elaborate shortly.

Here is an example of adding tracing calls to a method doIt in all classes that implement the Java interface com.foo.Work.

Aspect.new :before, :calls_to => [:doIt, :do_it], :in_types_and_descendents => Java::com.foo.Work do |jp, obj, *args|
  log "Entering: #{jp.target_type.name}##{jp.method_name}: object = #{object}, args = #{args.inspect}" 
end

There are two important points to notice in this example:

1. You can choose to refer to the method as do_it (Ruby style) or doIt, but these variants are effectively treated as separate methods; advice on one will not affect invocations of the other. So, if you want to be sure to catch all invocations, use both forms. There is a bug (18326) that happens in certain conditions if you use just the Java naming convention.
2. If the type is an interface, you must use :types_and_descendents (or one of the supported variants on the word types...). Since interfaces don’t have method implementations, you will match no join points unless you use the _and_descendents clause. (By default, Aquarium warns you when no join points are matched by an aspect.) However, there is a bug (18325) with this approach if Java types are subtyped in Ruby.

Limitations and Bugs

Okay, here’s the “fine print”...

In this (V0.4.0) release, there are some important limitations.

1. Aquarium advice on a method in a Java type will only be invoked when the method is called directly from Ruby.
2. To have the advice invoked when the method is called from either Java or Ruby, it is necessary to create a Ruby subclass of the Java type and override the method(s) you want to advise. These overrides can just call super. Note that it will also be necessary for instances of this Ruby type to be used throughout the application, in both the Java and Ruby code. So, you’ll have to instantiate the object in your Ruby code.

Yea, this isn’t so great, but if you’re motivated… ;)

There are also a few outstanding Aquarium bugs (which could actually be JRuby bugs or quirks of the Aquarium-JRuby “interaction”; I’m not yet sure which).

1. Bug #18325: If you have Ruby subclasses of Java types and you advise a Java method in the hierarchy using :types_and_descendents => MyJavaBaseClassOrInterface and you call unadvise on the aspect, the advice “infrastructure” is not correctly removed from the Ruby types. Workaround: Either don’t “unadvise” such Ruby types or only advise methods in such Ruby types where the method is explicitly overridden in the Ruby class. (The spec and the Rubyforge bug report provide examples.)
2. Bug #18326: Normally, you can use either Java- or Ruby-style method names (e.g., doSomething vs. do_something), for Java types. However, if you write an aspect using the Java-style for a method name and a Ruby subclass of the Java type where the method is actually defined (i.e., the Ruby class doesn’t override the method), Aquarium acts like the JoinPoint is advised, but the advice is never actually called. Workaround: Use the Ruby-style name in this scenario.

So, there is still some work to do, but it’s promising that you can use an aspect framework in one language with another. A primary goal of Aquarium is to make it easy to write simple aspects. My hope is that people who might find AspectJ daunting will still give Aquarium a try.

Just as for regular classes, TDD can drive aspects to a better design.

Assume that I’m testing a logging aspect that logs when certain methods are called. Here’s the JUnit 4 test:

package logging;
import static org.junit.Assert.*;
import org.junit.Test;
import app.TestApp;

public class LoggerTest {
    @Test
    public void FakeLoggerShouldBeCalledForAllMethodsOnTestClasses() {
        String message = "hello!";
        new TestApp().doFirst(message);
        assertTrue(FakeLogger.messageReceived().contains(message));
        String message2 = "World!";
        new TestApp().doSecond(message, message2);
        assertTrue(FakeLogger.messageReceived().contains(message));
    }
}

Already, you might guess that FakeLogger will be a test-only version of something, in this case, my logging aspect. Similarly, TestApp is a simple class that I’m using only for testing. You might choose to use one or more production classes, though.

package app;
@Watchable
public class TestApp {
    public void doFirst(String message) {}
    public void doSecond(String message1, String message2) {}
}

and @Watchable is a marker annotation that allows me to define pointcuts in my logger aspect without fragile coupling to concrete names, etc. You could also use an interface.

package app;
public @interface Watchable {}

I made up @Watchable as a way of marking classes where the public methods might be of “interest” to particular observers of some kind. It’s analogous to the EJB 3 annotations that mark classes as “persistable” without implying too many details of what that might mean.

Now, the actual logging is divided into an abstract base aspect and a test-only concrete sub-aspect>

package logging;

import org.aspectj.lang.JoinPoint;
import app.Watchable;

abstract public aspect AbstractLogger {
    // Limit the scope to the packages and types you care about.
    public abstract pointcut scope();

    // Define how messages are actually logged.
    public abstract void logMessage(String message);

    // Notice the coupling is to the @Watchable abstraction.
    pointcut watch(Object object):
        scope() && call(* (@Watchable *).*(..)) && target(object);

    before(Object watchable): watch(watchable) {
        logMessage(makeLogMessage(thisJoinPoint));
    }

    public static String makeLogMessage(JoinPoint joinPoint) {
        StringBuffer buff = new StringBuffer();
        buff.append(joinPoint.toString()).append(", args = ");
        for (Object arg: joinPoint.getArgs())
            buff.append(arg.toString()).append(", ");
        return buff.toString();
    }
}

and

package logging;

public aspect FakeLogger extends AbstractLogger {
    // Only match on calls from the unit tests.
    public pointcut scope(): within(logging.*Test);

    public void logMessage(String message) {
        lastMessage = message; 
    }

    static String lastMessage = null;
    public static String messageReceived() {
        return lastMessage;
    }
}

Pointcuts in aspects are like most other dependencies, best avoided ;) ... or at least minimized and based on abstractions, just like associations and inheritance relationships.

So, my test “pressure” drove the design in terms of where I needed abstraction in the Logger aspect: (i) how a message is actually logged and (ii) what classes get “advised” with logging behavior.

Just as for TDD of regular classes, the design ends up with minimized dependencies and flexibility (abstraction) where it’s most useful.

I can now implement the real, concrete logger, which will also be a sub-aspect of AbstractLogger. It will define the scope() pointcut to be a larger section of the system and it will send the message to the real logging subsystem.

The Industry Track had a number of interesting papers, including, for example, a paper that describes how aspects are used in the innovative Terracotta JVM clustering tool. Also, the last keynote by Adrian Colyer of Interface 21 recounted his personal involvement in the AspectJ and Spring communities, as well as the impact that aspects are having at major Interface 21 clients. It’s worth noting that many of the important Java middleware systems, e.g., Spring, JBoss, and Weblogic have embraced aspects to one degree or another.

AOSD tools like AspectJ and Spring AOP (AO Programming) solve obvious “cross-cutting concerns” (CCCs) like object-relational mapping, transactions, security, etc. in Java. However, I feel that AOSD needs some breakthrough innovations to move beyond its current role as a useful niche technology to a more central role in the software development process. I get the impression that industry interest in AOP has reached a plateau.

The academic community is doing some interesting work on fundamentals of AOSD theory (type theory, modeling, categorizing types of aspects, etc.), on “early aspects” (e.g., cross-cutting requirements), and on tooling. There is also a lot of minor iterating around the edges, but that’s the nature of academic research (speaking as one who has been there…). Most of the research work is a long ways from practical applicability.

However, I’m seeing too much emphasis on extending the work of AspectJ-like approaches in statically-typed languages, rather than innovating in new areas like new applications of aspects and the nature of AOSD in dynamic languages.

My recent work with Ruby has made me think about these two topics lately. I’ll blog about these topics later, but for now, I’ll just say that I anticipate a fruitful growth area for AOSD will be to facilitate the implementation of powerful DSLs (Domain Specific Languages), a popular topic in the Ruby community.

Here are some other observations from the conference.

All other engineering disciplines recognize cross-cutting concerns

Gregor Kiczales (the father of AspectJ and one of the fathers of AOSD) made this remark in a panel discussion on “early aspects”. He cited the examples of electrical engineers considering systemic issues in circuit design, like capacitance, current leakage, etc. and mechanical engineers who evaluate the stresses and strains of the entire structure they are designing (buildings, brake assemblies in cars, etc.).

Gregor also remarked that CCCs that are evident in the requirements may disappear in the implementation and vice-versa.

Gerald Sussman keynote

In the first keynote, Gerald Sussman argued that robust systems are adaptable for uses that were not anticipated by their designers. These systems often have multiple organizational ideas and a “component” structure that promotes “combinatorial behavior”.

He doesn’t like formalisms such as Dykstra’s A Discipline of Programming. Provable correctness and rigor aren’t very compatible with rich programs. Sussman prefers a more “exploratory” model, very much analogous to Test Driven Development (TDD), and what he calls Paranoid Programming, which he defined as “I won’t be at fault if it fails.”

More on AOSD

I will blog further about aspect-oriented design and aspects in dynamic languages.