Monday, May 26, 2008

Eclipse Monkey - Core design notes

I am currently working on the new Eclipse monkey, starting this week. You know it is amazing how much time unemployment can free up for you or, er, contract sabbatical. I will make available a temporary public repository when I have something to show for this. I am always open to comments. I am writing this blog entry in order to state what are the design goals for this version of monkey, in particular the core plugin.

In many ways it is very very tempting to just take Groovy Monkey, rename a few packages and classes and then voila, here is the new Eclipse Monkey. There would be nothing wrong with that, I am after all proud of Groovy Monkey, its simple and it works. Still I would be remiss if I did not take advantage of this natural break to consider some refactoring and or redesign.

The main consideration I wish to take into account for the new version of Eclipse Monkey is to make it more modular. What I mean by that, is that I wish it to be possible to use monkey across the full spectrum of the Eclipse platform. Now what does that mean? I think I would be best able to explain it in terms of goals. It is a goal to allow people developing using just an OSGi container to be able to leverage some useful subset of Eclipse Monkey functionality in their applications. Allowing a bundle developer to be able to run scripts embedded in a bundle, specified via a URI/URL or specified directly in a string. Another goal would be to allow that core functionality plus some additional support to be used by a developer of Eclipse RCP applications. At the RCP level, there is now the notion of a workspace, SWT/JFace UI API and Eclipse jobs, so the monkey functionality to support that should be available. The same idea would apply to Eclipse Platform, JDT and finally the PDE (which would constitute the full Eclipse SDK).

There are some consequences of these decisions. There will be more bundles/fragments than would be otherwise, but at least those components should be simpler and smaller in size. I will focus on those that deal with the new org.eclipse.monkey.core component, which will run in the base OSGi container only.

We will have to consider OSGi services (perhaps declarative services), since we are missing Eclipse extension points for one. This will be decidedly different from Groovy Monkey where language extensions and DOM(s) are specified via extension points. Hopefully this will be possible without too much difficulty, also consideration will be made to allow parallel extension points, since the PDE support for extension points is much further along than with OSGi services. Of course allowing for parallel definitions, we will hopefully not have too many headaches with monkey bundle developers trying to define their extensions as both OSGi services and Eclipse defined extension points.

Another consequence of the OSGI only approach is that there will be no IResources API, so there will need to be support for allowing scripts to be embedded in a bundle, located via URI/URL or even to allow script text to be taken in verbatim. This will be a significant difference from Groovy Monkey where everything is assumed to be a resource in the workspace.

Finally, another major impact of this choice, will be the fact that support for threading through the Eclipse Jobs API will not be available. This in and of itself also makes me want to abandon the OSGi only approach. This means that use of IWorkspaceRunnable and Eclipse Jobs will commence only when the org.eclipse.core.runtime bundle is available.

There are two other things that this line of thinking has led me to consider. One, the metadata keywords and definitions should be pluggable themselves. As of right now in Groovy Monkey the script metadata is pretty statically defined in the code for the core plugin. If the layering support is to work, they have to be plugged by different plugins/fragments depending on the framework that monkey is being run in. Second, probably need a way to define which bundles are required or which platform the script is meant to run against. For instance, putting in a Job definition for a script running in an OSGi container is nonsense, however once in the RCP level, it is required. I think that there should be probably hardcoded values: OSGi, RSP, RCP, Platform, JDT and PDE. RSP would be the OSGi container plus the org.eclipse.core.runtime and org.eclipse.core.resources plugins, but no UI support.

Finally, all this work has to be done with the idea of backwards compatibility in mind. Hey I have written lots of monkey scripts I don't wanna throw away either.

No comments: