Hi Oliver, Sure I gave it a shot and I stashed the code changes for what I came up with at [1] if you wish to review. Most of the effort was to create a simple tool to scan all the possible features and find out what bundles are included in each of those and then scan each of those bundles to find out what packages/capabilities are provided and match up what was missing with the smallest feature that contained those things.
After building the project from the fork, I launched the feature to confirm it starts with this command: java -jar ~/.m2/repository/org/apache/sling/org.apache.sling.feature.launcher/1.1.16/org.apache.sling.feature.launcher-1.1.16.jar -f target/slingfeature-tmp/feature-it-testing.json 1. https://github.com/enapps-enorman/sling-org-apache-sling-jcr-maintenance/commit/a713ad93496f5d58a246c84c7207f6e714c79036 How does that look to you? Regards, -Eric On Tue, Mar 2, 2021 at 3:05 AM Oliver Lietz <[email protected]> wrote: > On Tuesday, 2 March 2021 01:04:06 CET Eric Norman wrote: > > Hi Oliver, > > Hi Eric, > > > I'm not sure about how a "feature model" feature would express > dependencies > > on other features or whether that is even necessary. > > > > But as an experiment, I just tried to create a simple aggregate feature > > that contains just the jcr maintenance feature and nothing else. > > > > During the build the analyse-features goal processes that incomplete > > aggregate feature and I get a nice report back (see below) about what > > packages and capabilities are missing and then I can go look for the > > features or bundles that would provide those and add them, and repeat > that > > cycle until there are no errors. I guess I'm just saying that it is a > > nicer error report and fails faster than trying to parse the log files > > generated during a paxexam test to figure out why it didn't work. And > if I > > go through all the work to create a working aggregate feature model it > > seems like it would be nice to be able to use that directly in my > > paxexam integration tests rather than trying to recreate the same > > configuration in java code. > > > > For example, the mvn output may look something like this: > > ========================================= > > [INFO] Starting analyzing feature > > 'experiment:starter:slingosgifeature:jcr_maintenance:1.0.0-SNAPSHOT'... > > [INFO] - Executing APIs jar properties check [apis-jar]... > > [INFO] - Executing Bundle Import/Export Check [bundle-packages]... > > [INFO] - Executing Repoinit Check [repoinit]... > > [INFO] - Executing Requirements Capabilities check > > [requirements-capabilities]... > > [INFO] Analyzing feature > > 'experiment:starter:slingosgifeature:jcr_maintenance:1.0.0-SNAPSHOT' > > finished : 0 warnings, 2 errors. > > [ERROR] org.apache.sling:org.apache.sling.jcr.maintenance:1.0.0: is > > importing package(s) [org.apache.jackrabbit.oak.api.jmx, > > org.apache.jackrabbit.oak.commons.jmx, org.slf4j, > org.apache.felix.hc.api, > > javax.jcr.version, org.apache.sling.api.resource, javax.jcr] in start > level > > 20 but no bundle is exporting these for that start level. > > [ERROR] org.apache.sling:org.apache.sling.jcr.maintenance:1.0.0: Artifact > > org.apache.sling:org.apache.sling.jcr.maintenance:1.0.0 requires > > [org.apache.sling.jcr.maintenance/1.0.0] osgi.extender; > > > filter:="(&(osgi.extender=osgi.component)(version>=1.4.0)(!(version>=2.0.0)) > > )" in start level 20 but no artifact is providing a matching capability > in > > this start level. > > [ERROR] Analyser detected errors on feature > > 'experiment:starter:slingosgifeature:jcr_maintenance:1.0.0-SNAPSHOT'. See > > log output for error messages. > > Can you proceed with this approach, collect all bundles and configurations > and > let me know how much effort it is in total to have a working Sling > instance to > test Sling JCR Maintenance? (There are no ITs currently for the module so > effort is not wasted.) > > Thanks, > O. > > > > Regards, > > -Eric > > > > On Mon, Mar 1, 2021 at 2:57 PM Oliver Lietz <[email protected]> > wrote: > > > On Monday, 1 March 2021 18:51:42 CET Eric Norman wrote: > > > > Hi Oliver, > > > > > > Hi Eric, > > > > > > > The problem I frequently run into is that the default configurations > > > > from > > > > SlingOptions are almost always providing versions of bundles that are > > > > too > > > > old to use for testing new features. And then at that point it > requires > > > > > > a > > > > > > > lot of research and trial and error to find a compatible set of > bundles > > > > > > to > > > > > > > get it to start up properly. This is why I would think the feature > > > > > > model + > > > > > > > the build time "analyse-features" goals may be useful in identifying > > > > configuration troubles earlier and with better error messages when > > > > things > > > > go wrong. > > > > > > Feel free to create options for your tests from the feature model. > > > > > > > On the other hand with Sling Feature Model there is a big drawback. > > > > > > There is > > > > > > > > only a full blown Sling, no fine grained features. > > > > > > > > I don't think this is true, you can assemble things in any way you > want, > > > > and in fact the starter already has many feature descriptors > published > > > > under different classifiers that you could reference independently. > > > > > > > > But I would agree that the sling starter isn't the best place to > define > > > > > > the > > > > > > > features. The starter should really just be aggregating features > > > > defined > > > > elsewhere and each feature should be released on its own schedule. > The > > > > recently released sling-org-apache-sling-jcr-maintenance project is a > > > > > > good > > > > > > > example of doing something like this as the starter is just > referencing > > > > > > the > > > > > > > released feature instead of defining everything inside the starter. > > > > > > AFAIK by design there is no way to express dependencies in OSGi/Sling > > > Features. > > > > > > You cannot tell from the feature of Sling JCR Maintenance what is > required > > > to > > > get it up and running. Is Jackrabbit required or even Oak? Which > version? > > > > > > I haven't looked into the "analyse-features" goal. Does it take > > > Capabilities/ > > > Requirements into account? How is it working around the fact that for > > > several > > > bundles the meta data is missing? > > > > > > If you tell me there is a way to declare and maintain features > > > automatically – > > > great! Until now a lot of manual work was required for Sling's Karaf > > > Features. > > > > > > Regards, > > > O. > > > > > > > Regards > > > > -Eric > > > > > > > > On Mon, Mar 1, 2021 at 2:22 AM Oliver Lietz <[email protected]> > > > > > > wrote: > > > > > On Friday, 26 February 2021 18:54:28 CET Bertrand Delacretaz wrote: > > > > > > Hi, > > > > > > > > > > > > On Fri, Feb 26, 2021 at 6:31 PM Eric Norman <[email protected]> > > > > > > wrote: > > > > > > > ...Someday, maybe the sling feature model could be used to > derive > > > > > > the > > > > > > > > > > paxexam configuration so it would be less mysterious and > easier to > > > > > > > debug?.. > > > > > > > > > > This can be implemented quite easily but you would need regular > > > > > > "releases" > > > > > > > > of > > > > > Sling Starter or you would have to deal with a moving target. > > > > > > > > > > On the other hand with Sling Feature Model there is a big drawback. > > > > > > There > > > > > > > > is > > > > > only a full blown Sling, no fine grained features. > > > > > > > > > > The purpose of Sling Testing PaxExam (extending OPS4J Pax *Exam* > for > > > > > > use > > > > > > > > in > > > > > > > > Sling project) is described here: > > > > https://sling.apache.org/documentation/development/testing-paxexam.html#fe > > > > > > > > atures > > > > > > > > > > You get well tested sets of bundles (2-3 times) via SlingOptions > and > > > > > > some > > > > > > > > helpers in TestSupport. > > > > > > > > > > Tested bundle sets: > > > > > 1. (all) > > > > > > > https://github.com/apache/sling-org-apache-sling-karaf-integration-tests/t > > > > > > > > ree/master/src/test/java/org/apache/sling/karaf/tests/bootstrap 2. > > > > > > (all) > > > > > > > > https://github.com/apache/sling-org-apache-sling-testing-paxexam/ > > > > > tree/master/src/test/java/org/apache/sling/testing/paxexam/it/tests > > > > > < > > > > > > > https://github.com/apache/sling-org-apache-sling-testing-paxexam/tree/ma > > > > > > > > ster/src/test/java/org/apache/sling/testing/paxexam/it/tests> 3. > > > > > ("launchpad") > > > > > > > https://github.com/apache/sling-org-apache-sling-karaf-launchpad-oak-tar-i > > > > > > > > ntegration-tests/blob/master/src/test/java/org/apache/ > > > > > > sling/karaf/tests/configuration/SlingQuickstartOakTarConfiguration.jav > > > > > a > > > > > < > > > > > > > https://github.com/apache/sling-org-apache-sling-karaf-launchpad-oak-tar > > > > > > > -integration-tests/blob/master/src/test/java/org/apache/sling/karaf/tests/ > > > > > > > > configuration/SlingQuickstartOakTarConfiguration.java> > > > > > > > > > > Let me know if the documentation is unclear and needs to be fixed. > > > > > > > > > > O. > > > > > > > > > > > That would be great, and in the meantime I suppose the various > > > > > > versions of the lists of bundles at [1] can help find out about > the > > > > > > right combinations. Assuming those are not defined in [2] > already. > > > > > > > > > > > > -Bertrand > > > > > > > > > > > > [1] > > > > > > > https://github.com/apache/sling-org-apache-sling-starter/tree/master/src/m > > > > > > > > a > > > > > > > > > > > in/features [2] > > > > > > https://github.com/apache/sling-org-apache-sling-testing-paxexam > > > > >
