Adrian Trenaman-2 wrote: > > Just to pitch in on this: I really like to write my own feature files for > Karaf. Generally, I want to create one features file that pulls together > artifacts from a whole set of projects, in a way that creates features as > meaningful 'applications' that an ops guy can install/uninstall. While I > appreciate that we could auto-generate the features file from a Maven POM, > I prefer to *design* my features so that they're ergonomic. And, as has > been mentioned earlier on this trail, hand-crafted features allow me to > add config, documentation, etc. Any tooling we can write that helps the > developer do this in an easy, no fuss way (and maybe spots any missing > dependencies) should be preferred over a 'generate-from-Maven' approach, > IMHO. > > Cheers, > Ade > > ----- Original Message ----- > From: Andreas Pieber [mailto:[email protected]] > Sent: Wednesday, March 02, 2011 02:33 AM > To: [email protected] > Subject: Re: KAR feature not doing what the docs say it should > > Hey David > > On Wed, Mar 2, 2011 at 8:09 AM, David Jencks wrote: >> I'm trying to understand why you want to maintain a features.xml by hand >> so that the versions in it will differ from those in your maven project >> rather than generating the features.xml from your maven project so the >> versions will match up. > > I'm not sure why they should? You can easily wire them together by > filter the features.xml and use something like ${xyz.version} in your > features.xml. In addition you do not always want to use the same > artifacts as you reference in your pom. E.g. if you wrap Apache > Wicket, you typically have one package with all deps, but in the src > you really want to reference the single packages to help maven finding > sources and jdoc. > >> I agree with KARAF-459 to the extent that if we keep the archive-kar goal >> it should use more info from the supplied features.xml. I am arguing >> that we should not keep it. Why is it a good idea to encourage people to >> get their dependencies out of sync? > > I don't think that we encourage them to do so, but using the > features.xml it is much easier to add config files, configurations, > required features, other repositories... All of that have to be > configured otherwise anyhow in the maven plugin and I think this will > make it much more complex. > >> If you convince me this is a good idea :-) then I think making the kar >> packaging so it can start with a (possibly partial) features.xml, add >> maven dependencies to it, and put all the resulting dependencies into the >> generated kar would be a good idea. I think this would solve KARAF-459? > > I've tried ;) > > Kind regards, > Andreas > >> >> I don't know what the add-features-to-repo goal does yet so I'm not sure >> if I think it's useful :-) >> >> thanks >> david jencks >> >> On Mar 1, 2011, at 10:34 PM, Jean-Baptiste Onofré wrote: >> >>> I'm not sure to follow you. >>> >>> The kar goal is exactly as the add-features-to-repo goal: you start from >>> a features descriptor (that you wrote by hand) and the goal package the >>> descriptor and the bundles/dependencies into a repo (kar or local). >>> >>> Regards >>> JB >>> >>> On 03/02/2011 07:34 AM, David Jencks wrote: >>>> OK, but you are in a maven environment. You've now disconnected the >>>> versions in the features.xml which you are presumably maintaining by >>>> hand from those in your maven poms. I consider that a non-starter. >>>> >>>> My point is that you want to construct the features.xml from maven >>>> dependencies in the first place. At the same time you can construct >>>> the kar, including (some of) the dependencies. >>>> >>>> happy to be convinced otherwise... >>>> thanks >>>> david jencks >>>> >>>> On Mar 1, 2011, at 10:05 PM, Jean-Baptiste Onofré wrote: >>>> >>>>> The main advantage is that it starts from the features descriptor. So >>>>> you simply define the features what you want to embed in the Kar and >>>>> the plugin is responsible to download and embed all bundle >>>>> dependencies. >>>>> >>>>> For instance, in place of having: >>>>> >>>>> >>>>> >>>>> >>>>> >>>>> >>>>> >>>>> >>>>> >>>>> >>>>> >>>>> >>>>> >>>>> >>>>> you simple have in the plugin >>>>> >>>>> myfeature >>>>> >>>>> >>>>> So the POM is light, the version is defined in the features descriptor >>>>> and it manages transitive dependencies to others features. >>>>> >>>>> Regards >>>>> JB >>>>> >>>>> On 03/02/2011 07:00 AM, David Jencks wrote: >>>>>> I might understand what the archive-kar goal does now, from the jira >>>>>> issue. >>>>>> >>>>>> I would like to suggest that we eliminate this goal and just use the >>>>>> kar packaging which generates both the features.xml and the kar from >>>>>> the maven dependencies. >>>>>> >>>>>> When would the archive-kar goal be useful compared to the kar >>>>>> packaging? >>>>>> >>>>>> thanks >>>>>> david jencks >>>>>> >>>>>> On Mar 1, 2011, at 9:47 PM, Jean-Baptiste Onofré wrote: >>>>>> >>>>>>> Hi guys, >>>>>>> >>>>>>> The purpose of the kar goal is to take a features descriptor and >>>>>>> package the features descriptor and the related bundle into a kar >>>>>>> archive (that's it's a goal of the features maven plugin). >>>>>>> The kar deployer create a repo for these bundles. >>>>>>> I raised KARAF-459 about that. At least, the kar goals should take >>>>>>> an argument to define if the bundle are embedded in the kar or not. >>>>>>> But, if the kar doesn't embed the bundle, what's the advantage of >>>>>>> using a kar more than directly drop the features descriptor into the >>>>>>> deploy directory :) >>>>>>> >>>>>>> Regards >>>>>>> JB >>>>>>> >>>>>>> On 03/01/2011 11:40 PM, David Jencks wrote: >>>>>>>> I couldn't quite understand what the docs expected. What I think >>>>>>>> is usable is the (undocumented) kar packaging which ought to look >>>>>>>> something like this: >>>>>>>> >>>>>>>> >>>>>> >>>>>>>> xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"; >>>>>>>> xsi:schemaLocation="http://maven.apache.org/POM/4.0.0 >>>>>>>> >>>>>>>> http://maven.apache.org/xsd/maven-4.0.0.xsd";> >>>>>>>> >>>>>>>> 4.0.0 >>>>>>>> >>>>>>>> hibernate-osgi >>>>>>>> hibernate-osgi >>>>>>>> 0.0.1-SNAPSHOT >>>>>>>> kar >>>>>>>> hibernate-osgi >>>>>>>> >>>>>>>> >>>>>>>> >>>>>>>> >>>>>>>> >>>>>>>> >>>>>>>> >>>>>>>> >>>>>>>> org.apache.karaf.tooling >>>>>>>> features-maven-plugin >>>>>>>> 2.99.99-SNAPSHOT >>>>>>>> true >>>>>>>> >>>>>>>> >>>>>>>> >>>>>>>> >>>>>>>> >>>>>>>> >>>>>>>> This should generate a features.xml file inside the kar and include >>>>>>>> the bundles you mentioned as entries in the feature.xml and copied >>>>>>>> into the kar. >>>>>>>> >>>>>>>> thanks >>>>>>>> david jencks >>>>>>>> >>>>>>>> On Mar 1, 2011, at 2:15 PM, karafman wrote: >>>>>>>> >>>>>>>>> To test the KAR feature, I compiled the trunk and executed the >>>>>>>>> following >>>>>>>>> pom.xml file: >>>>>>>>> >>>>>>> >>>>>>>>> xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"; >>>>>>>>> >>>>>>>>> xsi:schemaLocation="http://maven.apache.org/POM/4.0.0 >>>>>>>>> >>>>>>>>> http://maven.apache.org/xsd/maven-4.0.0.xsd";> >>>>>>>>> >>>>>>>>> 4.0.0 >>>>>>>>> >>>>>>>>> hibernate-osgi >>>>>>>>> hibernate-osgi >>>>>>>>> 0.0.1-SNAPSHOT >>>>>>>>> pom >>>>>>>>> hibernate-osgi >>>>>>>>> >>>>>>>>> >>>>>>>>> >>>>>>>>> >>>>>>>>> org.apache.karaf.tooling >>>>>>>>> features-maven-plugin >>>>>>>>> 2.99.99-SNAPSHOT >>>>>>>>> >>>>>>>>> >>>>>>>>> archive-kar >>>>>>>>> >>>>>>>>> archive-kar >>>>>>>>> >>>>>>>>> >>>>>>>>> src/main/resources/features.xml >>>>>>>>> >>>>>>>>> >>>>>>>>> >>>>>>>>> >>>>>>>>> >>>>>>>>> >>>>>>>>> >>>>>>>>> >>>>>>>>> >>>>>>>>> Using this features.xml file: >>>>>>>>> >>>>>>>>> <?xml version="1.0" encoding="UTF-8"?> >>>>>>>>> >>>>>>>>> >>>>>>>>> >>>>>>>>> mvn:javax.xml.stream/com.springsource.javax.xml.stream/1.0.1 >>>>>>>>> mvn:org.dom4j/com.springsource.org.dom4j/1.6.1 >>>>>>>>> >>>>>>>>> mvn:org.jboss.javassist/com.springsource.javassist/3.9.0.GA >>>>>>>>> >>>>>>>>> mvn:javax.persistence/com.springsource.javax.persistence/1.0.0 >>>>>>>>> mvn:org.antlr/com.springsource.antlr/2.7.7 >>>>>>>>> >>>>>>>>> mvn:net.sourceforge.cglib/com.springsource.net.sf.cglib/2.2.0 >>>>>>>>> >>>>>>>>> mvn:org.apache.commons/com.springsource.org.apache.commons.collections/3.2.1 >>>>>>>>> >>>>>>>>> mvn:org.apache.commons/com.springsource.org.apache.commons.logging/1.1.1 >>>>>>>>> >>>>>>>>> mvn:org.objectweb.asm/com.springsource.org.objectweb.asm/1.5.3 >>>>>>>>> >>>>>>>>> mvn:org.objectweb.asm/com.springsource.org.objectweb.asm.attrs/1.5.3 >>>>>>>>> >>>>>>>>> mvn:org.hibernate/com.springsource.org.hibernate/3.3.2.GA >>>>>>>>> >>>>>>>>> mvn:org.hibernate/com.springsource.org.hibernate.annotations/3.3.1.ga >>>>>>>>> >>>>>>>>> mvn:org.hibernate/com.springsource.org.hibernate.annotations.common/3.3.0.ga >>>>>>>>> >>>>>>>>> mvn:org.hibernate/com.springsource.org.hibernate.ejb/3.3.2.GA >>>>>>>>> >>>>>>>>> >>>>>>>>> >>>>>>>>> The .kar file created didn't contain any of the bundles, just the >>>>>>>>> features.xml file. The expected behavior is to (according to >>>>>>>>> http://karaf.apache.org/manual/2.2.1-SNAPSHOT/users-guide/kar.html): >>>>>>>>> The kar-archive goal: >>>>>>>>> 1. Reads all features specified in the features descriptor. >>>>>>>>> 2. For each feature, it resolves the bundles defined in the >>>>>>>>> feature. >>>>>>>>> 3. All bundles are packaged into the kar archive. >>>>>>>>> >>>>>>>>> So, it appears the KAR feature is not doing what is stated in the >>>>>>>>> docs. I >>>>>>>>> suggest we either change the documentation, or the archive-kar >>>>>>>>> goal. >>>>>>>>> >>>>>>>>> ----- >>>>>>>>> Karafman >>>>>>>>> Slayer of the JEE >>>>>>>>> Pounder of the Perl Programmer >>>>>>>>> >>>>>>>>> -- >>>>>>>>> View this message in context: >>>>>>>>> http://karaf.922171.n3.nabble.com/KAR-feature-not-doing-what-the-docs-say-it-should-tp2606973p2606973.html >>>>>>>>> Sent from the Karaf - Dev mailing list archive at Nabble.com. >>>>>>>> >>>>>>>> >>>>>> >>>> >> >> >
I propose we look at this from a different point of view, the implementer with basic knowledge of Karaf. With that in mind, whatever we decide to do with the plugin should result in the following use-cases being satisfied: 1) Provide a strong "starting point" for a features.xml file based on what's listed in the pom, leveraging maven. In my program, we spend 1 - 2 FTE (full time equivalents, basically8 to 16 hours of development time) per bundle (here I mean a bundle that we write, not that we use) to create a good features.xml file. We then aggregate these into a featues.xml file for our overall application. Being able to generate a good starting point for our features.xml documents would shave about three-quarters of this time off of the process. 2) Provide a good starting point for a deployable KAR file based on the pom and leveraging maven. When we finally get to the point where the KAR feature is usable in Karaf, I imagine it will take some time to create a deployable KAR. Implementers of smaller applications would be able to use this functionality to create thier final KAR file, while implementers of larger applications would be able to use this KAR as a starting point and then refine it. 3) Provide the ability to use a completed features.xml file to create a final KAR file for an application. This means that the plugin would read-in the features.xml, resolve all of the bundles, and then package the bundles with the features.xml file in a deployable KAR file. This is similar to how the KAR feature currently works. The overarching goal for the community should be to provide a tool that will ease the implementation of a new or existing Java application into Karaf. I'd also like to take this opportunity to comment on some of the discussion in this thread and others about how much a developer should know about OSGi before being able to successfully use it. Consider the lessons learned when we started moving our legacy enterprise applications (CORBA and home-grown) into the J2EE in the early 2000's. There were numerous developers of widely ranging skillsets who attempted to migrate thier applications, and early-on many were not successful, writing off the J2EE as too hard or complex to be viable. Most of this was caused by the learning curve of the new way-of-thinking required by the J2EE, and a lack of tools to do hand-holding for the developers. We can expect the same thing to occur with OSGi and Karaf. One major thing we can do to reduce this is to create tools that will do a lot of the heavy-lifting of OSGi for the developers. So, while from a knowledgable OSGi developer's point of view, it would be nice for all of our users to know everything there is to know about Karaf before they use it, that will not be the case for 80% of the users. If the features.xml file produced by the plugin isn't perfect, or needs tweaking, so be it. At least we've gotten the users most of the way to thier final solution. ----- Karafman Slayer of the JEE Pounder of the Perl Programmer -- View this message in context: http://karaf.922171.n3.nabble.com/KAR-feature-not-doing-what-the-docs-say-it-should-tp2606973p2614569.html Sent from the Karaf - Dev mailing list archive at Nabble.com.
