I think I have a solution, but I'm not sure whether it violates a best practice (or is just beneath a good aspect developer):
I created a ThreadLocal<Boolean> to determine which invocations should be advised and which should not: private static final ThreadLocal<Boolean> tlAdvise = new ThreadLocal<Boolean>() { @Override protected Boolean initialValue() { return true; } }; Then I set the ThreadLocal to false when I want to run the real methods and to true when I want them to be short-circuited with my own advise: ... tlAdvise.set(false); if( ctx.hasNextIncoming() ) { // call the real Sonic code System.out.println("In beforeCallToService: hasNextIncoming returned true"); XQEnvelope env = ctx.getNextIncoming(); // call the real Sonic code tlAdvise.set(true); ... My pointcuts use if: pointcut hasIncoming() : execution(boolean XQServiceContext.hasNextIncoming()) && if(tlAdvise.get()); and pointcut getIncoming() : execution(XQEnvelope XQServiceContext.getNextIncoming()) && if(tlAdvise.get()); Is this a brute-force solution to something that could be done far more elegantly and efficiently with just a little more pointcut knowledge? Is there anything that makes this a bad idea? Thanks for your insights, Lee ________________________________ From: aspectj-users-boun...@eclipse.org [mailto:aspectj-users-boun...@eclipse.org] On Behalf Of Grey, Lee Sent: Wednesday, February 02, 2011 11:45 PM To: aspectj-users@eclipse.org Subject: Re: [aspectj-users] Examining a destructive read in an aspect I've done what you suggested, and I think it would work fine, except that I can't get the pointcut right for the two calls I need to intercept. I seem to be getting all or nothing. What pointcut would intercept the calls to ctx.hasNextIncoming() and ctx.getNextIncoming() when they come from the actual XQServiceEx.service() method but would still make the real, unadulterated calls when made from inside my aspect? My failed attempts include pointcut hasIncoming() : call(boolean XQServiceContext.hasNextIncoming()) && !within(com.sonicsw..*); pointcut getIncoming(XQServiceContext ctx) : call(XQEnvelope XQServiceContext.getNextIncoming()) && target(ctx) && (within(XQService) || within(XQServiceEx)); and other unsuccessful flailing. I find creating successful pointcuts to be black magic. Thanks, Lee ________________________________ From: aspectj-users-boun...@eclipse.org [mailto:aspectj-users-boun...@eclipse.org] On Behalf Of Ramnivas Laddad Sent: Monday, January 31, 2011 8:50 PM To: aspectj-users@eclipse.org Subject: Re: [aspectj-users] Examining a destructive read in an aspect You could do something along the following lines: 1. In your before advice read destructively as needed and store away the result 'env' in a ThreadLocal. 2. Advise ctx.getNextIncoming() with an around advice to return the stored result (and don't call proceed() in it). -Ramnivas On Mon, Jan 31, 2011 at 2:59 PM, Grey, Lee <lee.g...@ncr.com<mailto:lee.g...@ncr.com>> wrote: I've been working on intercepting the service() method invocation in a Sonic ESB container. I had just started to write my before() advice when the bad news dawned on me. Just about every ESB service() method starts with the following... public void service(XQServiceContext ctx) throws XQServiceException { XQEnvelope env = null; while (ctx.hasNextIncoming()) { env = ctx.getNextIncoming(); if (env != null) { XQMessage msg = env.getMessage(); That call to ctx.getNextIncoming() is a destructive read that returns a different XQEnvelope every time it's called. The problem is that I need to evaluate the contents of the message that comes from ctx.getNextIncoming() in my before() advice. What that means is that the call to ctx.getNextIncoming() in the service() method is not going to get the message, because the before() advice already got it. Now I'm wondering if there's a way to put an aspect around ctx.getNextIncoming() to make it deliver the message again. Or if I can somehow clone ctx so that I can read the cloned message in before() and then read it from the real XQServiceContext object in service(). XQServiceContext doesn't offer any way to peek or browse, and it doesn't have a method to put a message, either. I would imagine I'm not the first person to run into this kind of issue with AOP before. I'm hoping that there's a pattern to address it. Thanks, Lee Grey _______________________________________________ aspectj-users mailing list aspectj-users@eclipse.org<mailto:aspectj-users@eclipse.org> https://dev.eclipse.org/mailman/listinfo/aspectj-users
_______________________________________________ aspectj-users mailing list aspectj-users@eclipse.org https://dev.eclipse.org/mailman/listinfo/aspectj-users