Hi Andy.
Today I got a tricky one. I thought about opening a Bugzilla ticket, but this is actually more of a question, maybe a future feature request if what I want is technically possible at all. So it depends on your answer if I shall open a ticket or not.
The requirement is simple (to explain, not to implement): I want to around-advise constructors in order suppress any side effects from happening there. The purpose would be to (ab)use AspectJ in order to create some kind of mock object which does not do anything expensive while being constructed. BTW, actually I am thinking about doing this with a library like ASM, but first I wanted to play with AspectJ in order to see what is possible there and create a proof of concept. But let me not get ahead of myself and set the stage first:
package de.scrum_master.app;
public class Base {
protected final int id;
public Base(int id) {
System.out.println("Constructing Base -> " + this);
this.id = id;
}
}
package de.scrum_master.app;
public class Sub extends Base{
private final String name;
public Sub(int id, String name) {
super(id);
System.out.println("Constructing Sub -> " + this);
this.name = name;
}
@Override
public String toString() {
return "Sub@" + this.hashCode() + " [name=" + name + ", id=" + id + "]";
}
}
package de.scrum_master.app;
public class AnotherSub extends Base{
private final String name;
public AnotherSub(int id, String name) {
super(id);
System.out.println("Constructing AnotherSub -> " + this);
this.name = name;
}
@Override
public String toString() {
return "AnotherSub@" + this.hashCode() + " [name=" + name + ", id=" + id + "]";
}
}
package de.scrum_master.aspect;
import de.scrum_master.app.Base;
import de.scrum_master.app.Sub;
public aspect ConstructorAspect {
void around() : execution(Base+.new(..)) && target(Sub) {
return;
}
}
package de.scrum_master.app;
public class Application {
public static void main(String[] args) {
System.out.println(new Sub(11, "Xander"));
System.out.println("----------");
System.out.println(new AnotherSub(22, "Someone"));
}
}
As you can see, I did the following:
- Create a base class Base with two subclasses Sub and AnotherSub.
- The aspect tries to
- suppress constructor execution for one Sub (with the intent to "mock" it, just imagine its methods get advised by additional behaviour),
- but not for AnotherSub (not a mock target) of the subclasses.
- The aspect does this by
- making sure that super class constructors are also suppressed via execution(Base+.new(..)),
- excluding anything that is not the target type via target(Sub) and finally
- simply not proceeding to the constructor by just returning from the around advice.
Now this works beautifully whenever creating a Sub instance, but fails whenever creating an AnotherSub instance:
Sub@1211076369 [name=null, id=0]
----------
Constructing Base -> AnotherSub@1551870003 [name=null, id=0]
Exception in thread "main" java.lang.IllegalAccessError: Update to non-static final field de.scrum_master.app.Base.id attempted from a different method (init$_aroundBody0) than the initializer method <init>
at de.scrum_master.app.Base.init$_aroundBody0(Base.java:8)
at de.scrum_master.app.Base.<init>(Base.java:6)
at de.scrum_master.app.AnotherSub.<init>(AnotherSub.java:7)
at de.scrum_master.app.Application.main(Application.java:7)
The explanation is pretty straightforward if we look at the console log and the class definitions:
- The Base class has a final instance field.
- AspectJ factors the Base constructor code out into a private helper method Base.init$_aroundBody0 and dispatches to it from the instrumented constructor Base.<init> instead of initialising the field directly from there.
- While this does no harm for the "mock target" class Sub because there the helper method is never called (the no-op around advice kicks),
- it fails miserably when creating an AnotherSub instance because then the helper method does get called but a helper method must not violate the JVM rule that final fields can only be initialised directly from a constructor (or during declaration).
Of course, this is hard to detect by a compiler such as Ajc, but if there was an option to force inlining the around advice for the constructor, this scheme would work. So my questions are:
- Is there such an option? I did not find any obvious compiler switch.
- If not, do you see any chance to implement it, given technical feasibility?
Best regards
--
Alexander Kriegisch
https://scrum-master.de
_______________________________________________ aspectj-users mailing list aspectj-users@eclipse.org To unsubscribe from this list, visit https://www.eclipse.org/mailman/listinfo/aspectj-users
