>From the specs (section 3.15.1, https://github.com/apache/felix/blob/trunk/resolver/src/main/java/org/apache/felix/resolver/ResolverImpl.java#L2027-L2028), it seems that the fact that there are 2 wires on the fragment is expected. However, such a wire should definitely not cause a refresh of the framework if a fragment is refreshed. So I'm planning to add an additional test around the following line: https://github.com/apache/felix/blob/trunk/framework/src/main/java/org/apache/felix/framework/BundleRevisionDependencies.java#L173
if (HostNamespace.HOST_NAMESPACE.equals(bw.getCapability().getNamespace())) { result.add(bw.getProvider().getBundle()); } 2017-06-06 16:12 GMT+02:00 Guillaume Nodet <gno...@apache.org>: > Actually, the bug (if it's a bug) is much older, as I can observe the same > behavior with framework 5.0.0. > I can't easily try with 4.x lines though... > > 2017-06-06 15:47 GMT+02:00 Guillaume Nodet <gno...@apache.org>: > >> I just hit a problem where refreshing a fragment cause the framework to >> be wrongly refreshed. >> This is caused by the fragment's wiring having 2 wires: one to its host, >> and another one for an osgi.ee requirement. >> My understanding is that fragments should only be wired to their hosts. >> This is a regression between 5.6.1 and 5.6.2, which may have been cause >> by the upgrade to resolver 1.12.0. >> >> At first glance, it could be related to the following commit: >> https://github.com/apache/felix/commit/b844e1ebd944bbbb5d1 >> 8c8a3d37f38f5e43e1a13 >> >> Any hint appreciated... >> Btw, i've raised FELIX-5649 >> <https://issues.apache.org/jira/browse/FELIX-5649> for that. >> >> -- >> ------------------------ >> Guillaume Nodet >> >> > > > -- > ------------------------ > Guillaume Nodet > > -- ------------------------ Guillaume Nodet