When testing Eclipse RCP plugins, it is good practice to have all test classes and resources live in a dedicated (test) fragment.
This way all classes in the host plugin can be accessed, no matter if those packages are actually exported or not.
But what if I have a common test infrastructure, that is used by many plugins?
I cannot put this into a fragment, since I cannot reuse classes inside a fragment somewhere else.
A fragment does not exist at runtime. It is merged with its Fragment-Host.
You either need to place the code in a regular bundle/plug-in, or use a poxy bundle that exposes the classes of the fragment to be accessible from other fragments or bundles. For example:
Host
Bundle-SymbolicName: org.example.test.util
Eclipse-ExtensibleAPI: true
Fragment
Bundle-SymbolicName: org.example.test.util.impl
Export-Package: org.example.test.util.impl
Fragment-Host: org.example.test.util
However, the latter approach uses the Eclipse-specific Extensible-API header that is only understood by the Equinox OSGi implementation.
Therefore, my recommendation would be to leave the actual test classes in a fragment that corresponds to the bundle-under-test and put the reusable test helpers in a dedicated bundle.
Related
I'd like to include a resource file (e.g. some xml config file) in my bundle and make it visible to all other bundles in the container. Is it possible without using the Fragment-Host manifest header? I'd like this resource file to always be visible in the classpath of all bundles running alongside my bundle, even those that do not exist yet, but will potentially be added in future.
EDIT:
To clarify - that resource must be available passively, i.e. the other bundles should be able to find it in their classpath, and not by refering to any special API or service of my bundle.
Some more background - my environment is a bit messy but I have no control over it and cannot change its existing bundles. The only way I can modify it is by adding my own bundles. That environment includes several copies of the ch.qos.logback.classic bundle. When logback starts up, it looks for specific XML config files in the classpath. If it doesn't find any of them, then its default behaviour is to print everything to stdout with debug level. This environment was previously used to host a GUI application so it didn't matter that much before, but now I am trying to adapt it so I can use some of its functionality in headless mode. So now it becomes important to me to be able to configure it in such a way that only warning and errors are printed to the console.
In general, no you cannot do this. Class-space isolation is at the heart of OSGi, but you want to put a resource in the class loader of one bundle and make it visible to all other bundles. That's not OSGi, it's the global application classpath.
The only thing you can do to add to the internal classpath of a specific bundle is to write a fragment which attaches to that bundle. A fragment can attach to multiple host bundles, but only if those hosts have the same symbolic name, i.e. because they are different versions of the same bundle. See OSGi R6 Core Specification, section 3.14.
You did however state that the bundles you want to attach are all copies of ch.qos.logback.classic. If that means they all have that exact symbolic name then perhaps a fragment will work after all.
You can not change the classpath of other bundles this way.
What you can do is retrieve the classloader of your bundle from your bundleContext. You can give this classloader to another bundle to retrieve your resource.
ClassLoader cl = context.getBundle().adapt(BundleWiring.class).getClassLoader();
Another option is to give the other bundle the URL of the resource.
As long as the resource is on the classpath, any bundle can access the resource if it can get hold of the class loader of the bundle that contains the resource.
For example:
ClassLoader classLoaderOfBundleWithResource = ...
classLoaderOfBundleWithResource.getResourceAsStream("org/example/resource.xml");
From a maintenance and API point of view, I would not recommend exposing a resource that way. Java types are much better suited therefore. Instead, let the resource bundle export a class that gives clients access to the contents of the resource.
For example:
public class XmlDocumentProvider {
public InputStream openDocument() {
return getClass().getResourceAsStream("resource.xml");
}
}
Assuming that both the resource.xml and the XmlDocumentProvider reside in the same package, openDocument will return the resource content just like in the first example.
I am using Anypont Studio 5.3.0 and server runtime 3.7.0. I want to invoke a main() method from my component. Application is developed using Maven, SpringBoot and JPA. It sits in the jar file and have the following structure.
com
package
Application.class (with main method)
another package
Other classes
lib
other jars
META-INF
persistance.xml
MANIFEST.MF
Org
springframework boot loader and other spring classes.
when file arrives with file pattern that I detect with mule polling component I would like to invoke Java component in mule flow that has main class and all the supporting classes.
Thanks,
David
did you mavenize your Application? If yes, you can add that as a dependency in your mule project pom, which is also mavenize. But you need to make sure that the jars are added in your maven repository i.e. execute first "mvn clean install" to your java application. Otherwise, add the jars in you build path. When you are able to do those, you can create a spring bean or create a java component in mule where they could call your class with main() method.
I never came across this kind of production scenario where there is a need to call main method of java class in enterprise application. Are you sure you have only main method to access other classes, it should have initialize, spring way of injection etc. Simple answer to you question, create a mule java component and override onCall method to call Application(class).main. I will never do this kind of stuff [for sure it will give more problems based on what is being written in main method]. In general we will use main method invocation in desktop application. if possible work on (or let the application team to work on) jar file to have better initializing options
This is for Struts 1.x (I'm using 1.3.10).
I've noticed that Struts is unable to pick up resource bundles in the ApplicationResources.properties file if it is not placed somewhere in the default classpath (e.g., com.abc.SomePackage).
For instance, if I put the ApplicationResources.properties file in a custom folder /WEB-INF/strutsResources and configure the struts-config.xml thus:
<message-resources parameter="/WEB-INF/strutsResources/ApplicationResources"/>
I've read that the resources need to be on the classpath so I've also tried adding the /WEB-INF/strutsResources folder to the classpath. It still does not pick up the resource keys.
I've double-checked that the strutsResources folder is actually deployed to the server (I'm using Glassfish v3), so the file is there, it's just not being parsed.
P.S.
If you're wondering why I'm trying to do this, I just wanted to organize my code a little better ("better," IMO). Since the ApplicationResources.properties file is not really a class, I wanted to place it in a resources folder by itself.
I've checked that placing the ApplicationResources file in a package in the src directory works just fine.
Ultimately, the answer is yes. You can play some interesting games by configuring a custom className and/or factory and get messages however you want (including from a database) and so on. This allows you to customize whatever you want*.
I agree the resources aren't a class, but putting them on the classpath is a common practice, and allows resources to be loaded as a resource, e.g., from inside a jar. I'm sympathetic, but I'd leave it as-is.
*Like reversing all the text; a fun prank to play on your co-workers and QA department.
Its best leave it on the classpath.
It's stadard practise to include properties files on the classpath, especially if you're planning on packaging it up in your WAR/EAR. You're keeping it under WEB-INF so you gain no benefit from moving it off the classpath, and you'll just confuse other developers who have to work on the project and you've had to put a hack in to make this work.
If you want to keep your files external to your deployable WAR/EAR then that's a valid reason for not using the classpath. Typically this will require some configuration as part of your deployment to specify where the file is to reside.
For example specify the location using
a JVM argument (e.g. -Dprops.file=/config/myapp.properites)
lookup from a JNDI resource
use a PropertiesFactoryBean if you're using the Spring framework (I
use Spring's ApplicationContext with Struts 1 MVC)
read properties from a database writing your own
ApplicationPropertiesDAO class that initialises itself durnig your
applications bootstrap process (e.g. Spring application contact,
Servlet in web.xml, Listener in web.xml, etc)
I have a main plugin project which does not depend on any of Eclipse APIs. But I do want to use Eclipse API's in one of its fragment plug-in. Will it cause any problem for the main plugin?
If you add a dependency (to a plug-in or a package) in a fragment, then you effectively modify the class path of the host project as well. Whether that will cause any changes to the semantics of the host project or any other fragment for the same host project, depends on the specific use in the project.
Having said all that, the normal answer is: no, it should not cause problems, unless you have code that depends on the class path - e.g. if you are using Class.forName(...) or similar...
One final note: when you test this, use -clean argument in the launch configuration to force OSGi to accept the changed dependency. Otherwise, it will be ignored.
I'm happily using the Maven bundle-plugin to create OSGi manifest headers for my modules. However, when there are configuration files that pull in classes which aren't referenced directly in the code, the plugin can't tell which packages it's going to need.
One example is a bundle with domain models that constitute a Persistence Unit for JPA. The driver class is part of the PU configuration and either set in an XML file or at runtime when the EntityManager is instantiated. I have to manually add an Import-Package header for the driver class that I want to load, or I get CNF errors.
Another example is a Struts war, where the web.xml pulls in the Struts dispatcher that's otherwise not found anywhere in the code and has to be manually added to the headers.
How can I avoid this?
I tried adding the required packages as dependencies with a provided scope, but that didn't help.
In the plug-in section of the bnd configuration you can specify plug-ins to analyze these files and contribute to the import-package header. For spring it looks like this:
<_plugin>aQute.lib.spring.SpringComponent</_plugin>
I am not sure, what descriptors are supported on top of spring. Just take a look at the source (it's in the Apache Felix SVN) and see for yourself. In the worst case you have to write your own plug-in, but at least it is possible! Also peter kriens site about the bnd explains the usage and some internals.
Other then that I am not aware of any simple solution.