On Mon, Nov 16, 2009 at 4:02 PM, Andy Clement
<andrew.clement@xxxxxxxxx> wrote:
Hi Jim,
> I made a minimal test app and ran it twice - one with it set to fail and one
> to pass (just commenting out a link that links to code that uses
> HttpSession).
so to make it pass you uncomment a reference to code that uses
HttpSession? Thus causing HttpSession to be actively referred to?
I might be interested in the 'debug' messages that indicate when types
are being passed to the weaver, rather than just the weaveinfo
messages.
It is possibly due to the use of call pointcuts and intertype
declarations with load time weaving. There are known issues here
because types don't always get loaded in the order we'd like and yet
we have no control because in LTW AspectJ is not in charge, it is at
the mercy of the classloader that is using it. Altering the loading
order (by doing what you have mentioned - making references to types
more explicit) can address this problem sometimes. However, your
scenario is a little bit complex, so I can't tell for certain that you
are suffering from this problem.
> For the success one, it looks like the code is only being weaved after the
> tests have started.
Don't you mean that in the success case HttpSession is woven before
the tests start? In the failure case it seems to be doing the weave
too late (after the tests have started).
I see a differing ordering in the weaveinfo messages relating to
method-execution and method-call. I think again it may help to have
the 'debug' messages as they will show the load ordering for the types
- seeing that difference may give us a bit more to go on.
I think it is worth raising a bug for this so we can sort it out:
https://bugs.eclipse.org/bugs/enter_bug.cgi?product=AspectJ
Andy.
