Two Aspect Class,two around,get emptystack error #128
Comments
|
Please reformat your question, using code blocks correctly. It is hard to read your code without copying it into an editor. Please also do not post pseudo code. What you posted would not compile (e.g. lower-case annotation names). Post complete classes with package names and imports, then others can at least try to reproduce your problem. If you want to learn how to ask good questions (not just here, but anywhere), please read the MCVE article. Thank you. |
|
Edited by @kriegaex Thank you very much I use an android project which use aspectj1.9.6 When i call method testMethod(), i get the error： This is my demo @Target(ElementType.METHOD)
@Retention(RetentionPolicy.RUNTIME)
public @interface AA {
}@Aspect
@DeclarePrecedence("com.example.myapplication.AAAspect,com.example.myapplication.BBAspect")
public class AAAspect {
@Pointcut("execution(@com.example.myapplication.AA * *..*(..))")
public void allMethod(){
}
@Around("allMethod() && @annotation(param)")
public Object arounddd(ProceedingJoinPoint point, AA param) throws Throwable{
Log.d("blue", "AA before time = " + System.currentTimeMillis());
Object[] args = point.getArgs();
Object obj = point.proceed(args);
Log.d("blue", "AA after time = " + System.currentTimeMillis());
return obj;
}
}@Target(ElementType.METHOD)
@Retention(RetentionPolicy.RUNTIME)
public @interface BB {
}@Aspect
public class BBAspect {
@Pointcut("execution(@com.example.myapplication.BB * *..*(..))")
public void allMethod(){
}
@Around("allMethod() && @annotation(param)")
public Object arounddd(ProceedingJoinPoint point, AA param) throws Throwable{
Log.d("blue", "BB before time = " + System.currentTimeMillis());
new Thread(new Runnable() {
@Override
public void run() {
try {
Log.d("blue", "BB 2");
point.proceed();
} catch (Throwable throwable) {
throwable.printStackTrace();
}
}
}).start();
return null;
}
}public class Test {
@AA
@BB
public void testMethod(){
Log.d("blue", "testMethod");
}
} |
PrefaceSeriously? I asked you to edit your question and add more information, not to post more pseudo code in bogus, unreadable formatting. Sorry for lecturing you, but if you cannot even do that, I have low motivation to read your code and error log, because it simply hurts my eyes. Am I asking too much for you to click on the formatting icons on the top of the input field or learn markdown? Besides, your code with all the falsely lower-case AspectJ annotations would still not compile and is just pseudo code to me. Both pointcuts also have syntax errors and would never compile with the AspectJ compiler. Do you ever test your own code before posting it? Let me show you what I mean: correct annotation names starting with upper-case characters like Fixed, complete example codepackage com.example.myapplication;
import static java.lang.annotation.ElementType.METHOD;
import static java.lang.annotation.RetentionPolicy.RUNTIME;
import java.lang.annotation.Retention;
import java.lang.annotation.Target;
@Retention(RUNTIME)
@Target(METHOD)
public @interface AA {}package com.example.myapplication;
import static java.lang.annotation.ElementType.METHOD;
import static java.lang.annotation.RetentionPolicy.RUNTIME;
import java.lang.annotation.Retention;
import java.lang.annotation.Target;
@Retention(RUNTIME)
@Target(METHOD)
public @interface BB {}package com.example.myapplication;
import org.aspectj.lang.ProceedingJoinPoint;
import org.aspectj.lang.annotation.Around;
import org.aspectj.lang.annotation.Aspect;
import org.aspectj.lang.annotation.DeclarePrecedence;
import org.aspectj.lang.annotation.Pointcut;
@Aspect
@DeclarePrecedence("com.example.myapplication.AAAspect, com.example.myapplication.BBAspect")
public class AAAspect {
@Pointcut("execution(@com.example.myapplication.AA * *(..))")
public void allMethod() {}
@Around("allMethod() && @annotation(param)")
public Object arounddd(ProceedingJoinPoint point, AA param) throws Throwable {
System.out.println("blue - AA before time = " + System.currentTimeMillis());
Object[] args = point.getArgs();
Object obj = point.proceed(args);
System.out.println("blue - AA after time = " + System.currentTimeMillis());
return obj;
}
}package com.example.myapplication;
import org.aspectj.lang.ProceedingJoinPoint;
import org.aspectj.lang.annotation.Around;
import org.aspectj.lang.annotation.Aspect;
import org.aspectj.lang.annotation.Pointcut;
@Aspect
public class BBAspect {
@Pointcut("execution(@com.example.myapplication.BB * *(..))")
public void allMethod() {}
@Around("allMethod() && @annotation(param)")
public Object arounddd(ProceedingJoinPoint point, AA param) throws Throwable {
System.out.println("blue - BB before time = " + System.currentTimeMillis());
new Thread(new Runnable() {
@Override
public void run() {
try {
System.out.println("blue - BB 2");
point.proceed();
}
catch (Throwable throwable) {
throwable.printStackTrace();
}
}
}).start();
return null;
}
}package com.example.myapplication;
public class Test {
@AA
@BB
public void testMethod() {
System.out.println("blue - testMethod");
}
public static void main(String[] args) {
new Test().testMethod();
}
}Now I can read the code, compile and execute it as well as reproduce your problem. I also do not need an unknown log framework to the example project. Fixing a small bug (not the error root cause, but anyway)First, please note that there is a bug in public class BBAspect {
@Pointcut("execution(@com.example.myapplication.BB * *(..))")
public void allMethod() {}
@Around("allMethod() && @annotation(param)")
public Object arounddd(ProceedingJoinPoint point, AA param) throws Throwable {It should be Let us also add another log line in the same aspect before the }).run();
System.out.println("blue - BB after time = " + System.currentTimeMillis());
return null;
}
}The console log becomes: So we see that both around-aspects do their jobs and return correctly. What is obviously still missing is the Experimentally executing without spawning off a new threadI can already tell you that this is because of your asynchronous There we have out The root causeNow why is that? Because in And later we have: What does that mean? Because around-advice methods need to be executed onion-skin style, i.e. entered outside-in and exited inside-out, we use a ... both around-aspects have already finished doing their jobs when suddenly in the new thread you are trying to proceed - but the stack of around-closures is empty already. This is simply a user error. Around advices are not supposed to proceed asynchronously, if there is a nesting situation, i.e. more than one around-advice advising the exact same joinpoint. If you temporarily comment out the advice annotation in // @Around("allMethod() && @annotation(param)")
public Object arounddd(ProceedingJoinPoint point, AA param) throws Throwable {and also add a little waiting time in the main method in order to wait for the asynchronous thread to finish so we can see its log output, ... public static void main(String[] args) throws InterruptedException {
new Test().testMethod();
Thread.sleep(2000);
}the program with Lesson learned: Please do not proceed asynchronously in a nested around-advice situation! I am sorry to have no better news for you, but you simply cannot make nested around-advices asynchronous. From AspectJ's perspective, both advice methods have terminated already and the closures discarded when suddenly you want to proceed. Besides, even in the situation where you only have one around-advice per method, you can execute the target method asynchronously, but if the advice already returned a result to the calling method (if the target method has non-void return type), the result of the Last, but not least: Questions like this belong on Stack Overflow or the AspectJ users mailing list. GitHub issues are for bugs. I am answering it, because I do believe you that you thought there is a bug in AspectJ. |
@aclement, this made me think about what has changed between then and now. 1.9.1 was published in April 2018, 1.9.6 in July 2020. Other than you mavenising the project and moving the class to another folder, actually there was only one relevant change: 8819fad, see also Bugzilla #333274. This is where you introduced the stack in order to be able to handle problems related to proceeding multiple times, e.g. for aspects re-trying things. The change became effective in 1.9.3. Since then, what the OP was trying here no longer works. I did not try, but I believe him when he says that in 1.9.1 he could do this. Do you see a way to fix this without reverting the useful change in the commit I just mentioned? |
Running the program under AspectJ 1.9.2@aclement, quick update: Like @walt2021 said, this particular scenario works in 1.9.2. I added some stack trace info to the target method, so we see where it was called from: @AA
@BB
public void testMethod() {
System.out.println("blue - testMethod - called by " + new Exception().getStackTrace()[3]);
}Like I said, the following logs are all from AspectJ 1.9.2. Proceeding exactly once from both aspects like in the example above yields: Proceeding twice in the synchronously proceeding I.e. interestingly, only one target method execution comes from "inner" of the two nested around-aspects Proceeding twice in the asynchronously proceeding Here, everything is as expected, i.e. both target method executions are triggered by the "inner" asynchronous aspect. Proceeding twice in both aspects yields: Now we do not have anything like 2*2=4 executions, but 2 from the "inner" asynchronous aspect plus one more from the "outer" synchronous one, all in all 3 executions. Running the program under AspectJ 1.9.8 (1.9.3+)Because of the stack problem, we cannot spawn the extra thread, but if again I am replacing This is wonderful, perfect nesting. We see that the "inner" aspect is triggered twice completely and each time proceeds twice. The only downside is that this only works synchronously. I am going to play with the source code and see if I can synchronise on the stack object somehow. Sorry for the lengthy post, but I do want to document my findings for reference. |
If we do not synchronise, we will run into problems with nested around-aspects in combination with proceeding asynchronously (in another thread). Fixes eclipse-aspectj#128. Signed-off-by: Alexander Kriegisch <[email protected]>
Otherwise, we would jeopardise asynchronous proceeding, see eclipse-aspectj#129 (comment) Instead, we - keep the stack, - but treat it like a list (Stack implements the List interface), - keeping a current closure index, - using 'get(currentArcIndex)' instead of `peek()` and `pop()`. Fixes eclipse-aspectj#128. Signed-off-by: Alexander Kriegisch <[email protected]>
Fixes eclipse-aspectj#128. Signed-off-by: Alexander Kriegisch <[email protected]>
In case of asynchronous proceeding for nested around-advice, child threads need copies of the original around-closure stack. Otherwise, the target method thread will have popped the stack empty after unwinding. Fixes eclipse-aspectj#128. Signed-off-by: Alexander Kriegisch <[email protected]>
Relates to eclipse-aspectj#128. Signed-off-by: Alexander Kriegisch <[email protected]>
|
Thank you very much again. |
|
Let me know if you agree to test a snapshot version with the bugfix for your situation and provide feedback, testing on Android. I can publish it on the Sonatype OSSRH snapshot Maven repository later today when I have some spare time. |
In case of asynchronous proceeding for nested around-advice, child threads need copies of the original around-closure stack. Otherwise, the target method thread will have popped the stack empty after unwinding. Fixes #128. Signed-off-by: Alexander Kriegisch <[email protected]>
Relates to #128. Signed-off-by: Alexander Kriegisch <[email protected]>
…after merge Signed-off-by: Alexander Kriegisch <[email protected]>
Leak was introduced in commit 3c80a36, fixing #128, but introducing #288 instead, which was the lesser of two evils, but still bad for some users unwilling to use native AspectJ syntax for their aspects, avoiding the problem. Relates to #288. Signed-off-by: Alexander Kriegisch <[email protected]>
Leak was introduced in commit 3c80a36, fixing #128, but introducing #288 instead, which was the lesser of two evils, but still bad for some users unwilling to use native AspectJ syntax for their aspects, avoiding the problem. Relates to #288. Signed-off-by: Alexander Kriegisch <[email protected]>
Edited by @kriegaex
When i use the code , i get the error（aspectj is 1.9.6）
However，i use 1.9.1，it is ok
I need help,thank you.
The text was updated successfully, but these errors were encountered: