- Converting Tycho Projects to maven-bundle-plugin, Initial Phase
- Winter Project #2: Maven P2 Repository Resolver
- OpenNTF Fork of p2-maven-plugin
- The Intricate Work of OSGi Dependencies on Domino
It's been one of my long-running goals to reduce my use of Tycho for my work. While Tycho does what it says on the tin, the way PDE works in Eclipse means it's an ongoing nightmare to deal with when I want to do simple things like add a new dependency. This isn't really Tycho's fault as such, and the project itself is making major steps to alleviate some issues, but it's the nature of the surrounding tooling. Even beyond that, the shaky support in IntelliJ and total lack of support in Visual Studio Code and similar editors makes it a real thorn in my side.
Still, though, it brings a lot to the table, particularly when dealing with Domino-targeted projects. Because Domino's OSGi layout is... fiddly, it's often safest to use the "Manifest-first" approach for dependencies, and it's definitely important to still be able to do feature projects and p2 repositories for importing into Designer and Domino.
But I've still been trying to whittle away at the constraints over time, and I got fed up enough yesterday to make some major strides.
The Original Project
One of the major tools in my toolbelt for years has been the
p2-maven-plugin, which does a lot of heavy lifting when it comes to taking non-Tycho or non-OSGi-focused projects and making them palatable for an OSGi environment. Even when I don't use it as the backbone of a project, I tend to use it to gather third-party dependencies and process them to make them Domino-friendly.
It has its limitations, though, that have kept me from using it to replace the final steps of a Tycho build, and those are the ones that I set out to improve. Yesterday, I forked the project and got to work. Most of my work centered around letting it pull more information out of existing p2 repositories. While it already has some knowledge of such repos, it was still geared heavily towards only using them to pick up a bundle here or there. The big annoyance for me there was that I wanted to bring in entire existing p2-housed features into the final update site.
For example, one of my big projects consumes and redistributes a bunch of upstream projects, such as ODA and the XPages Jakarta EE support. While the
p2-maven-plugin made it possible to reference those projects as Maven artifacts or individual bundles, I couldn't do what I wanted and just say "bring X and Y features in, including all their bundles".
I also went in and added a few other niceties needed for Domino: generation of the antiquated "site.xml" file for the NSF Update Site, archiving of the final site for distribution, and so forth.
With my changes, I was able to delete all of the feature projects in the tree, which lowers the mental complexity a bit. That also means that the only parts "controlled" by Tycho now are the actual bundle projects, and those have a clear path to de-Tycho-ization. Though doing that will make it a little more difficult to know when dependencies are Domino-suitable ahead of time, the conversion should save a ton of hassle overall.
So now, I have a toolchain that should be able to work together to replace Tycho while still working with the Equinox-heavy target:
maven-bundle-pluginto generate the OSGi metadata in
META-INF/MANIFEST.MF. I could also use
bnd-maven-plugindirectly for this and bndtools in Eclipse, but I'm not sure that it'd gain me much in practice
generate-domino-update-siteto create p2 repositories from post-9.0.1 Domino releases' XPages framework, which remains damnably non-Mavenized
p2-layout-providerto resolve p2-housed artifacts like those from above and OpenNTF projects and make them available as normal-enough Maven dependencies on the fly
- The forked
p2-maven-pluginto generate features and update sites, as well as to repackage existing bundles to be more Domino-friendly
What's missing now is an ability to run compile-time test suites in a true Equinox environment. I'm hemming and hawing on how important that really is, though. The tests I write only rarely expect the presence of OSGi - the main way it comes into play is for extensions, which are papered over by IBM Commons anyway. I've had a delightful time lately running tests of JAX-RS resources with Liberty's dev mode, and I'm pretty sure I saw some examples somewhere of building up and tearing down a scaffolding to run them during compilation, so maybe I'll switch to that anyway.
In any event, just having a tool to do this stuff is a huge weight off my back, and now the goal of a fully-normal-enough Maven project tree is tantalizingly in sight.