Yes I did! Thanks and well spotted...
Import-Package: !org.example.hostpkg, *
Rgds
Neil
On Thursday, 15 March 2012 at 14:16, Richard S. Hall wrote:
> On 3/15/12 09:32 , Neil Bartlett wrote:
> > Bnd doesn't do any special processing for fragments, it will still analyse
> > the contents of the fragment bundle and generate an Import-Package
> > statement. The reason for this is that Bnd analyses bundles in isolation,
> > and in general there is no way of knowing what packages might be already
> > present in the host.
> >
> > Of course in runtime terms, the Import-Package statement of the fragment
> > will be appended to the Import-Package statement of the host, so the
> > combined host+fragment might end up trying to import packages that it
> > already contains. On the other hand I would expect this to lead to OSGi
> > resolution errors rather than CNFEs....
> >
> > When building fragments it's probably best to use a bang pattern to exclude
> > host packages from Import-Package, for example:
> >
> > Import-Package: org.example.hostpkg, *
>
> Did you mean to include a "!" in there?
>
> -> richard
>
> >
> > Where "org.example.hostpkg" is a package that is known to be contained
> > within the host. Note the trailing catch-all asterisk in this instruction,
> > since you DO want to include imports that are not going to be satisfied by
> > the host.
> >
> > Rgds
> > Neil
> >
> >
> >
> > On Thursday, 15 March 2012 at 13:20, Richard S. Hall wrote:
> >
> > > Well, given your last message I was going to suggest that this was an
> > > issue with your imports in your fragment, but it seems that you
> > > discovered this below. Yeah, I don't know how bnd supports creating
> > > fragments, but you definitely shouldn't have to include what is on the
> > > class path of the host bundle, because you get that by default.
> > >
> > > -> richard
> > >
> > > On 3/15/12 05:14 , Christopher BROWN wrote:
> > > > I've just found a "hack" solution that avoids having to export the
> > > > "freemarker.template.*" package from the main bundle's bundle
> > > > classpath.
> > > >
> > > > I tweaked the Import-Package directive in the fragment bundle (the
> > > > ".bnd" file, to be specific) so that it imports
> > > > "freemarker.template.*;resolution:=optional" (adding optional
> > > > resolution). It works, but it just seems like a hack. I'm guessing
> > > > there should be some way to tell ".bnd" to accept the bundle classpath
> > > > without requiring that it be duplicated into the fragment. Maybe I'm
> > > > wrong, maybe it doesn't exist, or maybe it does (and I just don't know
> > > > where to find documentation for that feature).
> > > >
> > > > I tried:
> > > > http://www.aqute.biz/Blog/2007-02-19
> > > > ...but it discussed adding to the bundle-classpath with fragments, not
> > > > referring to classes on the host's bundle classpath.
> > > >
> > > > Thanks,
> > > > Christopher
> > > >
> > > >
> > > >
> > > > On 15 March 2012 09:58, Christopher BROWN<br...@reflexe.fr
> > > > (mailto:br...@reflexe.fr)> wrote:
> > > > > Hello,
> > > > >
> > > > > So, I have the bundle "B" which contains lib/freemarker.jar ("F" for
> > > > > short) in the Bundle-Classpath, and a fragment for testing "T" which
> > > > > declares "B" as the fragment host.
> > > > >
> > > > > When I don't have the export directive in the manifest of "B", when I
> > > > > try to USE (compiling and BND usage is OK) it, I get this output for
> > > > > my JUnit test runner:
> > > > > ________________________________________
> > > > >
> > > > > Failed to load "viewer.system.themes.freemarker.TestFM_T001" from
> > > > > bundle system.themes-freemarker [11]
> > > > > (java.lang.ClassNotFoundException: *** Class
> > > > > 'viewer.system.themes.freemarker.TestFM_T001' was not found. Bundle
> > > > > system.themes-freemarker [11] does not import package
> > > > > 'viewer.system.themes.freemarker', nor is the package exported by any
> > > > > other bundle or available from the system class loader. ***).
> > > > > Failed to load "viewer.system.themes.freemarker.TestFM_T002" from
> > > > > bundle system.themes-freemarker [11]
> > > > > (java.lang.ClassNotFoundException: *** Class
> > > > > 'viewer.system.themes.freemarker.TestFM_T002' was not found. Bundle
> > > > > system.themes-freemarker [11] does not import package
> > > > > 'viewer.system.themes.freemarker', nor is the package exported by any
> > > > > other bundle or available from the system class loader. ***).
> > > > > Failed to load "viewer.system.themes.freemarker.TestFM_T003" from
> > > > > bundle system.themes-freemarker [11]
> > > > > (java.lang.ClassNotFoundException: *** Class
> > > > > 'viewer.system.themes.freemarker.TestFM_T003' was not found. Bundle
> > > > > system.themes-freemarker [11] does not import package
> > > > > 'viewer.system.themes.freemarker', nor is the package exported by any
> > > > > other bundle or available from the system class loader. ***).
> > > > > ________________________________________
> > > > >
> > > > > As I said, I use BND (version 0.0.384) -- incidentally, I'm using
> > > > > Felix 3.2.2 as I'm deferring the upgrade to 4.x until I have time to
> > > > > get the hang of compiling from source using Java 6 -- and for
> > > > > information, here's the ".bnd" files".
> > > > >
> > > > > First, for bundle "B" (note the commented-out #Export directive which
> > > > > I enable to make it work) ; also the Bundle-Classpath.
> > > > > ________________________________________
> > > > >
> > > > > Bundle-Version: 0.6.0
> > > > > Bundle-RequiredExecutionEnvironment: JavaSE-1.6
> > > > > Bundle-Activator: viewer.system.themes.freemarker.FmBundleActivator
> > > > > Private-Package: viewer.system.themes.freemarker.*
> > > > > Import-Package:
> > > > > org.osgi.framework;version="1.5",org.osgi.*,org.slf4j,fr.reflexe.viewer.api.*;version="[0.6,2.0)",javax.el.*;resolution:=optional,javax.servlet.*;resolution:=optional,javax.swing.*;resolution:=optional,javax.xml.*;resolution:=optional,com.sun.org.apache.*;resolution:=optional,org.apache.commons.logging.*;resolution:=optional,org.apache.log.*;resolution:=optional,org.apache.log4j.*;resolution:=optional,org.apache.tools.ant.*;resolution:=optional,org.apache.xml.*;resolution:=optional,org.apache.xpath.*;resolution:=optional,org.dom4j.*;resolution:=optional,org.jaxen.*;resolution:=optional,org.jdom.*;resolution:=optional,org.mozilla.*;resolution:=optional,org.python.*;resolution:=optional,org.slf4j.spi.*;resolution:=optional,org.w3c.dom.*;resolution:=optional,org.xml.*;resolution:=optional,org.zeroturnaround.javarebel.*;resolution:=optional
> > > > > #Export-Package: freemarker.*;version=2.3.19
> > > > > Bundle-ClassPath: ., lib/freemarker.jar
> > > > > Include-Resource: lib=lib
> > > > > -removeheaders: Include-Resource, TODAY, EXEC_LOG_LEVEL,
> > > > > TEST_LOG_LEVEL
> > > > > ________________________________________
> > > > >
> > > > >
> > > > > Then, for fragment "T" which declares "B" as its host:
> > > > > ________________________________________
> > > > >
> > > > > Bundle-Version: 0.6.0
> > > > > Bundle-RequiredExecutionEnvironment: JavaSE-1.6
> > > > > Private-Package: viewer.system.themes.freemarker.*
> > > > > Import-Package:
> > > > > org.osgi.framework;version="1.5",org.osgi.*,org.slf4j,fr.reflexe.viewer.api.*;version="[0.6,2.0)",javax.el.*;resolution:=optional,javax.servlet.*;resolution:=optional,javax.swing.*;resolution:=optional,javax.xml.*;resolution:=optional,com.sun.org.apache.*;resolution:=optional,org.apache.commons.logging.*;resolution:=optional,org.apache.log.*;resolution:=optional,org.apache.log4j.*;resolution:=optional,org.apache.tools.ant.*;resolution:=optional,org.apache.xml.*;resolution:=optional,org.apache.xpath.*;resolution:=optional,org.dom4j.*;resolution:=optional,org.jaxen.*;resolution:=optional,org.jdom.*;resolution:=optional,org.mozilla.*;resolution:=optional,org.python.*;resolution:=optional,org.slf4j.spi.*;resolution:=optional,org.w3c.dom.*;resolution:=optional,org.xml.*;resolution:=optional,org.zeroturnaround.javarebel.*;resolution:=optional,freemarker.template.*,junit.framework,viewer.system.test.*;version="[0.6,2.0)"
> > > > > -removeheaders: Include-Resource, TODAY, EXEC_LOG_LEVEL,
> > > > > TEST_LOG_LEVEL
> > > > > Fragment-Host: system.themes-freemarker; bundle-version="0.6.0"
> > > > > ________________________________________
> > > > >
> > > > > If it's any use, I can provide the resulting MANIFEST.MF files (but
> > > > > they're a bit big, due to all the packages in freemarker.jar, when I
> > > > > enable the export directive).
> > > > >
> > > > > If I drop (for the .bnd file for the test fragment) the Import-Package
> > > > > value "freemarker.template.*", BND chokes with this message:
> > > > > [bnd] Unresolved references to [freemarker.template] by class(es)
> > > > > on the Bundle-Classpath[Jar:dot]:
> > > > > [viewer/system/themes/freemarker/MockStringModel.class]
> > > > > [bnd]
> > > > > /Users/cbr/dev/build/temp/Viewer/0.6.0/bnd-tmp/themes-freemarker/fragment/system.themes-freemarker-test-fragment.bnd:
> > > > > bnd failed
> > > > >
> > > > > That's why I added the "import" in the fragment, but I can't see how
> > > > > to avoid this (in other words, tell BND that it's OK, it'll magically
> > > > > find the freemarker packages when the fragment is merged with the host
> > > > > which has all that stuff in private packages available from the bundle
> > > > > classpath).
> > > > >
> > > > > For completeness, here's the source code for the class that BND
> > > > > complains about in the above error message:
> > > > > ________________________________________
> > > > >
> > > > > package viewer.system.themes.freemarker;
> > > > >
> > > > > import java.util.ArrayList;
> > > > > import java.util.List;
> > > > > import freemarker.template.TemplateModelException;
> > > > > import freemarker.template.TemplateScalarModel;
> > > > >
> > > > > public class MockStringModel implements TemplateScalarModel
> > > > > {
> > > > > private final String _value;
> > > > >
> > > > > static List<TemplateScalarModel> args(String... args)
> > > > > {
> > > > > List<TemplateScalarModel> list = new
> > > > > ArrayList<TemplateScalarModel>(args.length);
> > > > > for (String arg : args)
> > > > > {
> > > > > list.add(new MockStringModel(arg));
> > > > > }
> > > > > return list;
> > > > > }
> > > > >
> > > > > MockStringModel(String value)
> > > > > {
> > > > > assert value != null;
> > > > > _value = value;
> > > > > }
> > > > >
> > > > > @Override
> > > > > public String getAsString() throws TemplateModelException
> > > > > {
> > > > > return _value;
> > > > > }
> > > > > }
> > > > > ________________________________________
> > > > >
> > > > > As you can see, the class (from my test fragment bundle) DIRECTLY
> > > > > refers to classes from FreeMarker in order to create a mock object
> > > > > required by a test case, specifically:
> > > > >
> > > > > ________________________________________
> > > > >
> > > > > public void test001_PlainFunctionAdapter() throws Exception
> > > > > {
> > > > > MockFunction tf = new MockFunction("fn");
> > > > > MockDataContext td = new MockDataContext();
> > > > > FunctionAdapter fa = new FunctionAdapter((ITemplateFunction)tf, td);
> > > > >
> > > > > assertEquals(tf.getName(TEMPLATE_SYSTEM_NAME), fa.getName());
> > > > > assertEquals(0, fa.getUsageCount());
> > > > > assertEquals("fn( [foo, bar] )", fa.exec(MockStringModel.args("foo",
> > > > > "bar")));
> > > > > assertEquals(1, fa.getUsageCount());
> > > > > }
> > > > > ________________________________________
> > > > >
> > > > > If I drop the use of "MockStringModel" and comment out the direct
> > > > > references to FreeMarker, it works. I use a similar approach for
> > > > > other bundles, with bundle classpaths and so on; the problem only
> > > > > occurs here because of the mock object's direct use of classes on the
> > > > > bundle classes (I need it here to build callback-style test cases).
> > > > >
> > > > > Thanks,
> > > > > Christopher
> > > > >
> > > > >
> > > > >
> > > > > On 14 March 2012 14:06, Richard S. Hall<he...@ungoverned.org
> > > > > (mailto:he...@ungoverned.org)> wrote:
> > > > > > On 3/14/12 05:00 , Neil Bartlett wrote:
> > > > > > > It certainly is possible to use fragments this way, it fact it's
> > > > > > > almost
> > > > > > > the sole purpose of fragments. So there must be something else
> > > > > > > going on.
> > > > > > > Please show what kind of error messages you're seeing.
> > > > > > >
> > > > > >
> > > > > >
> > > > > > Agree with Neil. Fragment classes are loaded by the same class
> > > > > > loader as the
> > > > > > classes in the host to which they are attached, so they should have
> > > > > > the same
> > > > > > visibility to other classes.
> > > > > >
> > > > > > -> richard
> > > > > >
> > > > > >
> > > > > > > Rgds Neil
> > > > > > >
> > > > > > > Sent from my iPhone
> > > > > > >
> > > > > > > On 14 Mar 2012, at 08:35, Christopher BROWN<br...@reflexe.fr
> > > > > > > (mailto:br...@reflexe.fr)> wrote:
> > > > > > >
> > > > > > > > Hello,
> > > > > > > >
> > > > > > > > The software I'm creating uses Felix to separate one API bundle
> > > > > > > > (a set
> > > > > > > > of "official" interfaces) from various implementation bundles.
> > > > > > > > Generally, to discourage customers (who aren't OSGi experts)
> > > > > > > > from
> > > > > > > > creating dependencies on JARs upon which we require, as we want
> > > > > > > > to be
> > > > > > > > able to keep them as hidden implementation details that we're
> > > > > > > > free to
> > > > > > > > change over time. There's only a few JARs we've " promoted" or
> > > > > > > > shared
> > > > > > > > as proper bundles.
> > > > > > > >
> > > > > > > > Our testing strategy involves binding fragment bundles with
> > > > > > > > tests, and
> > > > > > > > I hit a problem, given the approach outlined above. In short,
> > > > > > > > although
> > > > > > > > the fragment bundle is in the same package, it doesn't seem
> > > > > > > > possible
> > > > > > > > to share access to the Bundle-Classpath without exporting from
> > > > > > > > the
> > > > > > > > host and importing it from the Fragment. It's therefore
> > > > > > > > available to
> > > > > > > > world+dog as an unwanted side effect, and yet I thought
> > > > > > > > Fragments were
> > > > > > > > supposed to share classloader access with the host..?
> > > > > > > >
> > > > > > > > Is there another way to do this without having to export
> > > > > > > > "private
> > > > > > > > dependency JARs"?
> > > > > > > >
> > > > > > > > Thanks,
> > > > > > > > Christopher
> > > > > > > >
> > > > > > >
> > > > > > >
> > > > > >
> > > > > >
> > > > >
> > > > >
> > > >
> > > > ---------------------------------------------------------------------
> > > > To unsubscribe, e-mail: users-unsubscr...@felix.apache.org
> > > > (mailto:users-unsubscr...@felix.apache.org)
> > > > For additional commands, e-mail: users-h...@felix.apache.org
> > > > (mailto:users-h...@felix.apache.org)
> > > >
> > >
> > >
> > >
> > > ---------------------------------------------------------------------
> > > To unsubscribe, e-mail: users-unsubscr...@felix.apache.org
> > > (mailto:users-unsubscr...@felix.apache.org)
> > > For additional commands, e-mail: users-h...@felix.apache.org
> > > (mailto:users-h...@felix.apache.org)
> > >
> >
> >
>
>
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: users-unsubscr...@felix.apache.org
> (mailto:users-unsubscr...@felix.apache.org)
> For additional commands, e-mail: users-h...@felix.apache.org
> (mailto:users-h...@felix.apache.org)
>
>
