Hi! Concerning JDK instrumentation, some work has been done in our research group [1] on that issue. We are mainly focusing on dynamic analysis aspects (for profiling, debugging) that needs to be woven in every class. From our experience the 2 main issues arising from the instrumentation of the JDK are: - during the bootstrapping phase the JVM expects to load the first classes in a given order. If they are woven, additional dependecies (e.g. to AspectJ runtime classes) are added and may break that order and causes a (pretty bad) JVM crash. - advice usually uses JDK classes as well, and this may cause infinite recursions when they call back into woven JDK classes.
We developed a system [2] that keeps 2 versions of each methods (original and woven with aspects) with a switch that executes woven code and fall backs to original code during the bootstrapping phase and when in the control-flow of an advice. Technical details are available in our publication "Aspect Weaving in Standard Java Class Libraries" [3]. For further information please contact us. [1] http://dag.inf.usi.ch/ [2] http://www.inf.usi.ch/projects/ferrari/MAJOR.html [3] http://www.inf.usi.ch/faculty/binder/documents/pppj08.pdf Cheers, Philippe On Thursday, March 17, 2011 03:14:10 pm Tiago Espinha wrote: > Just for future reference, if other people come across the same, I have > indeed managed to successfully instrument rt.jar. The exceptions I was > experiencing had actually to do with some double-weaving that I was doing. > I was weaving the rt.jar but I had forgotten that I had also weaved a > jaxws library which I was also using. > > That was causing some sort of problem, but it doesn't really matter in the > end as I didn't mean to have the jaxws library woven at this point. > > Thanks again for the clarification! > > Cheers, > Tiago > > 2011/3/16 Tiago Espinha <ti...@apache.org> > > > A-ha! When I faced those issues, I googled and I found a lot of people > > facing issues with weaving into the JDK so I thought it was an actual > > limitation of AspectJ. But from what you say, it seems like the only > > problem has to do with loadtime weaving but instrumenting a JAR should > > be fine. > > > > Well, that's certainly good news and tomorrow I'll give it another try. > > Maybe this exception I'm facing has to do with something else... > > > > Anyway, thanks a lot! > > > > Cheers, > > Tiago > > > > > > 2011/3/16 Andy Clement <andrew.clem...@gmail.com> > > > >> > What makes it so different from instrumenting any other JAR? > >> > >> These classes are loaded by a classloader that does not get a chance > >> to attach a weaver, when using javaagent - so loadtime weaving isn't > >> possible. With care, you can do what you have done, weave rt.jar and > >> use that new version. The trouble is that it can be easy to cause > >> yourself problems if you, for example, weave String but also use > >> String (perhaps inadvertently) in your aspect - easy to create > >> stackoverflow issues. I don't know why you have a NoSuchMethodError > >> though. > >> > >> But I know others have done the weaving successfully - hopefully one > >> of them can follow up my post :) > >> > >> cheers > >> Andy > >> > >> On 16 March 2011 06:19, Tiago Espinha <ti...@apache.org> wrote: > >> > Hi all, > >> > I've been looking around the Internet, and it seems like it's > >> > impossible > >> > >> (or > >> > >> > at least failure-prone) to weave aspects into JDKs. I did it myself - > >> > I instrumented the rt.jar file from the latest Oracle JDK for Linux - > >> > and, whilst the weaving goes seemingly fine, I hit failures on > >> > runtime with > >> > >> such > >> > >> > exceptions as NoSuchMethodError, relating to the aspect I wove. > >> > What I cannot understand is the reason why JDK instrumentation is > >> > impossible. What makes it so different from instrumenting any other > >> > JAR? > >> > >> And > >> > >> > does anyone know if there are workarounds to it? > >> > Thanks in advance! > >> > Cheers, > >> > Tiago > >> > _______________________________________________ > >> > aspectj-users mailing list > >> > 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 _______________________________________________ aspectj-users mailing list aspectj-users@eclipse.org https://dev.eclipse.org/mailman/listinfo/aspectj-users
