Re: [aspectj-users] How to intercept proceed() in around() advice

[Andy Clement <andrew.clement@xxxxxxxxx>]

Isn't advising the proceed basically the same as advising the target
joinpoint again.

I can create two pieces of around advice that will apply at the same
joinpoint and they will be called in a reliable sequence (if from the
same aspect or ordering set with declare precedence).

===
aspect X {
  void around(): execution(* m(..)) {/*set message*/;proceed();}
  void around(): execution(* m(..)) {/*do indentation and log actual
message */proceed();/*do unindent*/}
}

public class A {
public static void main(String []argv) {
  new A().m();
}

public void m() {}
}
===
Andy


On 25 August 2012 02:57, Alexander Kriegisch <Alexander@xxxxxxxxxxxxxx> wrote:
> I have asked this question on StackOverflow [1] before, but received no
> answer so far. Maybe somebody here has one for me:
>
> Because it is impossible to bind a constant or dynamic parameter to an
> advice in a pointcut [2], I have a bunch of logging around() advice A
> which set a message to be logged (ThreadLocal private member of aspect)
> and then call proceed(). I would like to be able to intercept the
> proceed() in any of those advice from another advice B which does the
> actual logging and indentation work so as to avoid code duplication in
> the other advice, but there is no joinpoint available to do that. I
> verified this by logging *all* joinpoints within those advice.
>
> What I came up with is a rather complicated combination of
>   - intercepting the set() operations on the internal variable
>     containing the log message in an after() advice, doing indentation
>     and pre-proceed() logging there, and
>   - Java annotation tags on the many advice A plus another after()
>     advice capturing the end of each advice A, doing dedentation
>     and post-proceed() logging there.
>
> If anyone is interested I can post real or pseudo code, but I just
> wanted to explain *why* I need an interceptable joinpoint on proceed()
> if I do not want to use any tricks or trade one type of code repetition
> (the logging/indentation calls) for another (chaining all logging
> pointcuts used on advice of type A like "a() || b() || c() || ...",
> repeating them and having to update them if I add another logging
> pointcut + advice.
>
> Attention: I *cannot* just use another around() advice intercepting the
> other adviceexecution()s because I need to wait until the log message
> has been constructed inside of them. That would not be a problem either
> if I could intercept proceed() and would have the additional advantage
> that after proceed() I could even do other things in the advice, e.g.
> log more info not captured by the one-line message logged by my
> hypothetical advice B.
>
> I hope that all of this was understandable. Otherwise, please ask.
>
> Bottom line: My wish is a pointcut "proceedexecution()" or similar.
>
> [1]
> http://stackoverflow.com/questions/12018585/how-to-intercept-proceed-in-another-aspectj-aspect
> [2] https://bugs.eclipse.org/bugs/show_bug.cgi?id=92889
> --
> Alexander Kriegisch
>
> _______________________________________________
> aspectj-users mailing list
> aspectj-users@xxxxxxxxxxx
> https://dev.eclipse.org/mailman/listinfo/aspectj-users