@JBO sure I'm on IRC Mon-Fri ;)
I'm not going to force my idea. The reasoning behind this is my experience when trying to align feature/bundle versions from CXF, Camel, ActiveMQ, hawtio, fabric8, switchyard, PAX (Web, CDI, JDBC, JMS), JClouds, Hibernate.... In most cases, "<bundle depdendency="true">" helps, but it's not always a case. Having 6 different scala-library bundles or shipping 8 different guava bundles in a distro wasn't nice... This is not going to be "user friendly" tool, it's rather aimed for distro creators, where you want to control more aspects of versions without having to fork. best regards Grzegorz Grzybek 2017-10-31 12:37 GMT+01:00 Jean-Baptiste Onofré <j...@nanthrax.net>: > Hi > > From a quick glance, it sounds good to me. However I would like to take a > deeper look and discuss with you about that. > > Regards > JB > > On Oct 30, 2017, 15:09, at 15:09, Grzegorz Grzybek <gr.grzy...@gmail.com> > wrote: > >Hello > > > >Continuing my investigation of Processor mechanism for feature > >definitions > >(a.k.a. "better overrides")[1], I want to give you insight to what I > >plan > >to change in Karaf's FeatureService. > > > >Currently we have two separate mechanisms: > > > >"blacklisting": > > - may remove bundles from feature at features XML load time > > - may remove features from repository at features XML load time > >- may skip adding/analyzing features XML at karaf-maven-plugin:assembly > >invocation time > > - doesn't prevent given features XML to be added later > >- clears runtime information about blacklisted features/bundles - we > >can't > >query for blacklisted features or see (in `feature:list`) which > >features > >were blacklisted > > > >"overrides": > >- may replace G:A:V of some bundle with another G:A:V2 when downloading > >bundles of a feature > > - doesn't allow to change G:A:V into different G2:A2:V2 (e.g., > >"mvn:org.eclipse.jetty.orbit/javax.servlet/3.0.0.v201112011016" → > >"mvn:org.apache.geronimo.specs/geronimo-servlet_3.0_spec/1.0") > > - doesn't allow to change "dependency" flag on given bundle > > - doesn't allow to add bundles to a feature (assuming we "know better" > >than original feature's author - but it's a problem more often than > >we'd > >want) > > > >What I'm working on (till you tell me we don't need it), is better > >separation of concerns. I assume that feature XML files are loaded in > >one > >place (JAXB) and then may be further processed (which involves > >blacklisting, overriding and altering of entire features) before > >they're > >actually used by FeaturesService. > > > >Currently org.apache.karaf.features.internal.service.RepositoryCache > >class > >holds JAXB model and uses > >org.apache.karaf.features.internal.service.Blacklist class to alter (to > >some degree) it. > >I want to integrate "Blacklist" class into more generic > >org.apache.karaf.features.internal.service.FeaturesProcessor interface. > > > >Repository, Feature and BundleInfo classes will have "isBlacklisted()" > >method for diagnostic purposes. Blacklisted items are not simply > >removed, > >but can't be used to alter runtime (State). > > > >I want to be as much consistent between FeaturesService (dynamic > >aspect) > >and karaf-maven-plugin:assembly (static aspect) as possible. Maven > >goal's > >configuration will be used to generate special (new?) file in etc/ of > >the > >distro and then used at runtime. > > > >As the area I'm playing with is very fragile, it's slower (than I want) > >process. I'll continue my work, but I welcome any comments if I attempt > >something silly. > > > >regards > >Grzegorz Grzybek > >=== > >[1]: https://issues.apache.org/jira/browse/KARAF-5376 >
