Re: Are Limitations in AspectJ 1.1 satisfied with1.5?

On 6 Sep 2006, at 02:19, Ron Bodkin wrote:

> Ajc 1.1 used binary weaving: the change occurred after ajc 1.0
>
> It is true that the AspectJ team deliberately decided to not allow  
> access to
> local variables in pointcuts on the grounds that it wouldn't be  
> stable.
>
> I will testify that the Glassbox open source project (see  
> Glassbox.com) is
> using AspectJ successfully to measure performance on and report  
> problems
> with black box applications. Most performance management tools and  
> profilers
> don't go deeper than method and constructor execution. I believe  
> that is
> sufficient depth for measuring performance.

I guess the actual reason is a technical limitation in Java: In Java,  
the symbolic information and (IIRC) the static type information for  
local variables is stripped away in class files / bytecode, and it is  
hard to reconstruct useful information about local variables. (Unless  
you compile with debug information switched on, but that's not a  
black-box approach anymore.)

> In my view, the major use case where AspectJ isn't fine-grained  
> enough is
> measuring code coverage for tests. I believe the right direction  
> there is to
> be able to write pointcuts that match branching without picking out  
> line
> numbers or variable names. That might also be of some value for really
> fine-grained spotlighting when profiling too. I believe that the EOS
> research project offers this kind of more fine grained yet anonymous
> pointcut although I haven't looked it in a while...

JMangler has already been used successfully in industrial projects  
for doing code coverage. I think it's an advantage in this case that  
you don't have to worry about a dedicated pointcut language, but can  
use the full Java programming language for expressing what you are  
interested in at a very low level, at least for the time being.  
Approaches based on logic metaprogramming are probably also appropriate.


Pascal


> -----Original Message-----
> From: [email protected] [mailto:[email protected]] On  
> Behalf
> Of Eric Bodden
> Sent: Tuesday, September 05, 2006 9:20 AM
> To: Fabiano Ricci
> Cc: [email protected]
> Subject: Re: [aosd-discuss] Are Limitations in AspectJ 1.1 satisfied
> with1.5?
>
>>
>> 1) It doesn't provide a full support to access variables in the  
>> target
>> program.
>>     A pointcut can create a reference to all variables  in a  
>> joinpoint.
>> Possible variables are: this, target. variables passed as  
>> parameters to
>> the actual function in object and the returning value of the  
>> method. If
>> we want catch other variables we must change the Source Code ( not
>> available in black-box application ) and include this variables in  
>> that
>> jointpoint
>
> This has not changed in AspectJ, but is available via the "let"  
> pointcut,
> which lets you bind expressions in the form of let(x,someExpr())  
> and is
> available in the "EAJ" extension to AspectJ (www.aspectbench.org).
>
>> 2) If we want to access an object at runtime, we need to know the  
>> type
>> of that object. So if we don't know the source code we can't  
>> access to
>> that object. There is a manner to avoid this?
>
> I don't understand that statement. Of course you can access objects  
> which
> were created in code which you don't have the source code for.  
> Maybe they
> are referring to the fact that ajc 1.1 used source code  
> translation. So yes,
> that should be solved.
>
>> This two issues make very difficult create an aspect that  
>> automatically
>> profiling black-box applications
>
> Many more subtle problems remain, however that's another topic.
>
> Eric
>
> --
> Eric Bodden
> Sable Research Group, McGill University
> Montréal, Québec, Canada
>
>
>
>
> _______________________________________________
> discuss mailing list    -    [email protected]
>
> To unsubscribe and change options, go to:
> http://aosd.net/mailman/listinfo/discuss_aosd.net
>
> Check out the AOSD.net Wiki: http://aosd.net/wiki
>
>
> _______________________________________________
> discuss mailing list    -    [email protected]
>
> To unsubscribe and change options, go to:
> http://aosd.net/mailman/listinfo/discuss_aosd.net
>
> Check out the AOSD.net Wiki: http://aosd.net/wiki

-- 
Pascal Costanza, mailto:[email protected], http://p-cos.net
Vrije Universiteit Brussel, Programming Technology Lab
Pleinlaan 2, B-1050 Brussel, Belgium





_______________________________________________
discuss mailing list    -    [email protected]

To unsubscribe and change options, go to:
http://aosd.net/mailman/listinfo/discuss_aosd.net

Check out the AOSD.net Wiki: http://aosd.net/wiki