I just got confirmation that the current trunk of the CXF-DOSGi passes the TCK again. It would be *really* good if we could release this version so that the Remote Services and RSA RI in the OSGi systems is a released version of CXF-DOSGi rather than a snapshot. Sergey, could I maybe ask you to do a release again?
Here's a summary of the fixes: * Fixed exports from Single Bundle Distro * Fixed deadlock * Fixed cleanup * Fixed ExportReferenceImpl.equals() and hashCode() * Fixed RemoteServiceAdminCore.exportService() As these issues are all exposed by the OSGi TCK I didn't write any additional tests for them in the CXF-DOSGi codebase. As an aside. Note that it is possible for Apache committers to run the OSGi TCK. If anyone is interested let me know and I'll dig up the details. Cheers, David On 13 March 2012 07:30, David Bosschaert <david.bosscha...@gmail.com> wrote: > Just a little update on this. > I made some more changes and at this point in time we're getting > close. For me the OSGi Remote Services and Remote Service Admin TCKs > are 100% passing but other people did report a failure that we have to > investigate a little further. > > I'll chime in again when everything is good, as we could do a release > at that point. > > Cheers, > > David > > On 20 February 2012 22:38, Sergey Beryozkin <sberyoz...@gmail.com> wrote: >> Hi David >> >> On 19/02/12 00:42, David Bosschaert wrote: >>> >>> Hi all, >>> >>> I was recently running the CXF-DOSGi 1.3 release through the OSGi TCK >>> to make sure it's still compliant with the spec. >>> It turned out that the changes made between 1.2 and 1.3 cause a number >>> of TCK failures, so I've been looking at fixing them. >>> Here's a quick summary. >>> * the single-bundle distro (which is used with the TCK) now includes >>> the org.osgi.enterprise-4.2.0.jar. This is fine, but it didn't >>> export/import the types defined in there which meant that these types >>> existed twice in the VM, once inside the single bundle distro and once >>> outside. This caused issues with ConfigAdmin and some event types >>> since communication with the outside world wasn't possible with these >>> types any more. >>> I fixed this for the single-bundle distro (it doesn't apply to the >>> multi-bundle distro). >>> * ExportReferenceImpl, which is really a wrapper, was used in a Map >>> but missing hashCode and equals(). I added these. >>> * There were some issues around close() calls not completely properly >>> behaving, I fixed those >>> * RemoteServiceAdminCore was putting objects of the wrong type in the >>> collection returned by exportService() >>> >>> Some more changes may be needed in order to fully pass the TCK, but >>> I've committed the above in r1290914. >>> >> Cool, guess you are thinking about 1.3.1 already :-) >> Cheers, Sergey >> >>> Cheers, >>> >>> David >> >>
