AspectJ, Intertype Definition - aop

Question

I am receiving this error when I compile
The type XXX must implement the inherited abstract method
I have three files
A default implementation [com.SafeReaderIMPL.java]
public class SafeReaderIMPL implements ISafeReader {
private boolean successfulRead;
public SafeReaderIMPL() {
successfulRead = true;
}
protected void fail() {
successfulRead = false;
}
#Override
public boolean isSuccessfulRead() {
return successfulRead;
}
}
An interface file [com.ISafeReader.java]
public interface ISafeReader {
public boolean isSuccessfulRead();
}
An apsect (using annotations) [com.SafeReaderAspect.java]
#Aspect
public class SafeReaderAspect {
#DeclareParents(value = "com.BadReader", defaultImpl = SafeReaderIMPL.class)
public ISafeReader implementedInterface;
#AfterThrowing(pointcut = "execution(* *.*(..)) && this(m)", throwing = "e")
public void handleBadRead(JoinPoint joinPoint, ISafeReader m, Throwable e) {
((SafeReaderIMPL)m).fail();
}
}
And a Test Class [com.BadReader]
public class BadReader {
public void fail() throws Throwable {
throw new Throwable();
}
}
I compile the first three files in a separate jar using
ajc -source 1.8 -sourceroots . -outjar aspectLib.jar
I then compile the second file using the aspectLib like so
ajc -source 1.8 -sourceroots . -aspectpath ./aspectLib.jar -outjar common.jar
When I go to compile the second jar I get the error. I am using the latest stable version of AspectJ 1.8.3
BadReader.java:10 [error] The type BadReader must implement the
inherited abstract method ISafeReader.isSuccessfulRead() public class
BadReader {
^^^^^^^^

Answer 1

The problem is not two-step compilation as such, but the fact that #DeclareParents in #AspectJ syntax in not 100% compatible with declare parents in native syntax. Actually, #DeclareParents for introducing default interface implementations is superseded by #DeclareMixin (see this bug ticket), but the downside of the mixin approach is that you do not have a real A implements B scenario there, i.e. you cannot cast as you wish in your after-throwing advice, so this is also not a good option in your case.
So what do you do if you want to keep the two-step compilation approach? Just use native syntax:
Interface:
package com;
public interface ISafeReader {
boolean isSuccessfulRead();
}
Default implementation:
package com;
public class SafeReaderIMPL implements ISafeReader {
private boolean successfulRead;
public SafeReaderIMPL() { successfulRead = true; }
public void fail() { successfulRead = false; }
#Override public boolean isSuccessfulRead() { return successfulRead; }
}
ITD aspect:
package com;
public aspect SafeReaderAspect {
declare parents : com.BadReader implements SafeReaderIMPL;
after(ISafeReader safeReader) throwing : execution(* *(..)) && this(safeReader) {
System.out.println(thisJoinPoint + " - calling 'fail()' before rethrowing error");
((SafeReaderIMPL) safeReader).fail();
}
}
ITD target class with sample main method:
package com;
public class BadReader {
public void doSomething() {
throw new RuntimeException("my error");
}
public static void main(String[] args) {
BadReader badReader = new BadReader();
System.out.println("badReader.isSuccessfulRead() = " + badReader.isSuccessfulRead());
try { badReader.doSomething(); }
catch(Throwable t) { System.out.println(t); }
System.out.println("badReader.isSuccessfulRead() = " + badReader.isSuccessfulRead());
}
}
Now you can use the two-stage compilation approach.
Console output:
badReader.isSuccessfulRead() = true
execution(void com.BadReader.doSomething()) - calling 'fail()' before rethrowing error
java.lang.RuntimeException: my error
badReader.isSuccessfulRead() = false

Answer 2

The problem is due to the two-step compilation. During the second step, ajc needs the source code of SafeReaderIMPL to be able to weave BadReader, but it cannot find it into aspectLib.jar
In fact, if you try compiling in a single step (I did), it compiles and runs.
Unfortunately I don't know a way to fix this without providing the source code during the second compile step, which I suppose would render the whole two-step approach a bit pointless.