On Nov 14, 2007 12:49 AM, Xavier Hanin <[EMAIL PROTECTED]> wrote:
> On Nov 13, 2007 10:00 PM, Maarten Coene <[EMAIL PROTECTED]> wrote:
>
> > Xavier,
> >
> > this change broke some tests. Could you take a look at it?
>
> Sure, sorry about that.
>
> I've just checked in a fix, things should be ok now. For one of the two
> failing tests, I had to modify the test. Indeed the test makes use of
> several methods in Ivy and CacheManager, but now the IvyContext is clearer
> after the call to ivy methods, so during the later call to CacheManager
> method, the context is not set anymore, which is causing a problem in this
> particular case, because the retaining artifact mechanism relies on the
> context used during the resolve.
>
> Ideally due to the change in the Ivy "API" (requirement to push a context
> manually whenever you call several related methods consecutively) we should
> change almost all our unit tests, which most of the time call methods on ivy
> engines directly, and not on the Ivy instance itself. In practice this is
> only necessary when consecutive methods actually use the context, which was
> only the case in one unit test. But if we later make more extensive use of
> IvyContext, we may have to generalize context pushing/poping in all our unit
> tests. This could be done in the setup/teardown, but we need to push a
> context which ensure Ivy itself won't push another in its methods, which is
> not possible ATM (except if you know the Ivy instance when you push the
> context, but it's not possible in all unit tests). This means that we would
> have to create some kind of special IvyContext which Ivy would recognize in
> its pushContext() method to attach to it rather than creating a new sub
> context. Or maybe this could simply be done by checking in the pushContext()
> method not only if the current Ivy instance is the same as the current but
> also if it is null. Something like this:
>
> public void pushContext() {
> if (IvyContext.getContext().peekIvy() == null) {
> IvyContext.getContext().setIvy(this);
> IvyContext.pushContext(IvyContext.getContext());
> } else if (IvyContext.getContext().peekIvy() != this) {
> IvyContext.pushNewContext();
> IvyContext.getContext().setIvy(this);
> } else {
> IvyContext.pushContext(IvyContext.getContext());
> }
> }
>
The problem with that is that it defeats the whole purpose of context
pushing and popping: the context associated before calling an Ivy method
will then have this Ivy instance attached, so this is not a good answer to
the problem. A better solution would be to use something specific, for which
we could be sure the user has purposefully pushed a context and will thus
pop it.
Xavier
>
>
> WDYT?
>
> Xavier
>
> --
> Xavier Hanin - Independent Java Consultant
> http://xhab.blogspot.com/
> http://ant.apache.org/ivy/
> http://www.xoocode.org/
--
Xavier Hanin - Independent Java Consultant
http://xhab.blogspot.com/
http://ant.apache.org/ivy/
http://www.xoocode.org/
