The full list of resolved issues in 1.6.12 is available here.
Prior to this change it was only possible to specify an annotation match like this:
get(@Anno(someValue=1) * *) || get(@Anno(someValue=3) * *)
Now it is possible to use != and write this:
get(@Anno(someValue!=2) * *)
This can enable a group of annotated elements to be more easily identified.
Prior to this version the wiring was quite limited. In order to wire a pointcut to a piece of code the user needed to write an abstract aspect that included an abstract pointcut and some advice attached to that abstract pointcut. Then compile this aspect and finally write the XML to concretize the abstract pointcut. With 1.6.12 more flexibility has been added and for some cases there can be no need for that abstract aspect.
This is a work in progress but now you can write this in the aop.xml:
<concrete-aspect name="MyAspect">
<before pointcut="execution(* Hello.say2(..)) AND args(message)"
invokeClass="SomeRegularJavaClass"
invokeMethod="someMethod(JoinPoint tjp, java.lang.String message)"/>
<after pointcut="execution(* Hello.say2(..)) AND args(message)"
invokeClass="SomeRegularJavaClass"
invokeMethod="someOtherMethod(JoinPoint tjp, java.lang.String message)"/>
</concrete-aspect>
public class SomeRegularJavaClass {
public static void someMethod(org.aspectj.lang.JoinPoint tjp, String s) {
System.out.println("in advice4: s="+s+" at "+tjp);
}
public static void someOtherMethod(org.aspectj.lang.JoinPoint tjp, String s) {
System.out.println("in advice5: s="+s+" at "+tjp);
}
}
In this example there is a simple regular java class containing some static methods. In the XML these can be joined to pointcuts, kind as if they were advice. Notice in the XML it specifies:
Due to the method specification being in XML the parameter types must be fully specified. The only exception to this rule is that the AspectJ core types JoinPoint (and JoinPoint.StaticPart) do not have to be fully qualified (see the example above). Important: notice that in the case above which does argument binding, the names are bound according to the XML specification, not according to the parameter names in the Java code.
Around advice is also supported (the return type of the method must match the joinpoint return type). The example shows after advice, currently there is no way to specify either after returning or after finally, there is only after.
Expanding this further would enable support for all the code style features in the XML. Some of the language features like declare annotation cannot be done in annotation style aspects but the XML doesn't have the same kind of restrictions. If anyone wants to help out by fleshing this area of the weaver out, let me know and I'll help you get started!
There is now a new well known name that you can use in the if clauses in your aspects. thisAspectInstance provides access to the aspect instance.
Here is an example:
aspect X {
boolean doit() {
System.out.println("In instance check method doit()");
return true;
}
before():execution(* m(..)) && if(thisAspectInstance.doit()){
System.out.println(thisJoinPoint);
}
}
Now why not just use X.aspectOf() instead of thisAspectInstance? Well thisAspectInstance is quite useful
when working with abstract/concrete aspects:
abstract aspect X {
abstract pointcut p();
boolean doit() {
return true;
}
before():p() && if(thisAspectInstance.doit()){
System.out.println(thisJoinPoint);
}
}
aspect Y extends X {
pointcut p(): execution(* m(..));
}
Now thisAspectInstance will be an instance of the Y, not X. It enables the aspect instance to be used in some kind of check/guard that will avoid the costly creation of a thisJoinPoint object if the advice isn't going to run. Note: right now this only works for singleton aspects. If you have need of it with other instantiation models, please comment on https://bugs.eclipse.org/bugs/show_bug.cgi?id=239649
Although we have been successfully weaving groovy for a long time, it is becoming more popular and a few issues have been uncovered when using non-singleton aspects with groovy code. These have been fixed.
The release notes for the last few versions of AspectJ have mentioned two options (minimalModel and typeDemotion) which can be switched on to reduce memory consumption. They have had enough field testing now and from 1.6.12.M2 onwards they are on by default. Users should see a reduction in memory consumed by AspectJ projects in AJDT. It won't affect load time weaving. It may also help command line (or Ant) compile time weaving. If these options cause a problem then please raise a bugzilla but in the interim you could work around the problem by actively turning them off by specifying -Xset:minimalModel=false,typeDemotion=false in the project properties for your AspectJ project.
Some preliminary work has been done to support Java7. Java7 class files must contain the necessary extra verifier support attributes in order to load successfully on a Java7 VM - the attributes were only optional in Java6. It is possible to force loading of classes missing the attributes but that requires use of a -XX option. AspectJ 1.6.12.M2 should create these for you if you weave Java7 level class files. Nothing has been done yet to rebase AspectJ on a version of the Eclipse compiler that supports Java7 language constructs - that will happen after Eclipse 3.7.1 is out.
It is now possible to specify synthetic in pointcuts:
pointcut p(): execution(!synthetic * *(..));
This enables us to weave signed jars correctly. AspectJ sometimes generates closure classes during weaving and these must be defined with the same protection domain as the jar that gave rise to them. In 1.6.12.M1 this should now work correctly.
Starting with Eclipse 3.6, the Eclipse compiler no longer suppresses raw type warnings with @SuppressWarnings("unchecked"). You need to use @SuppressWarnings("rawtypes") for that. AspectJ has now been updated with this rule too.
The optimized annotation value binding now supports ints - this is for use when you want to match upon the existence of an annotation but you don't need the annotation, you just need a value from it. This code snippet shows an example:
@interface SomeAnnotation {
int i();
}
before(int i): execution(* *(..)) && @annotation(SomeAnnotation(i)) {
Binding values in this way will result in code that runs *much* faster than using pointcuts that bind the annotation itself then pull out the value.
Under that same bug some changes were made to match values by name when binding too. Suppose the annotation had multiple int values, how would we select which int to bind? AspectJ will now use the name (if it can) to select the right value:
@interface SomeAnnotation {
int mods();
int flags();
}
before(int flags): execution(* *(..)) && @annotation(SomeAnnotation(flags)) {
Here the use of 'flags' as the name of the value being bound will ensure the 'flags' value from any SomeAnnotation is bound and not the 'mods' value.