Equinox OSGi Project
A Transition Proposal
The existing Equinox Technology project is to transition to be an Eclipse Project.
Download
the transition review slides
In mid-2003, while recovering from releasing Eclipse 2.1, the Eclipse Runtime team started to think of how to make Eclipse more dynamic. One of the drivers for this was the community push for Eclipse as a Rich Client Platform. A further driver was broadening the appeal of the Eclipse runtime infrastructure and leveraging existing standards and facilities. The team wanted to engage the community in the search for and investigation of new runtime technology for Eclipse. To support this, the Equinox Technology project was created.
Equinox was set up as an incubator -- the first such project in Eclipse. As a separate project under the Technology PMC, the investigations were isolated from the main code and the entry barriers for interested parties lower. The output of the project was to either graduate into the main Eclipse code base or wither due to lack of interest. Having setup the project, the team proceeded to investigate various runtime component models (e.g., Avalon, OSGi, ...) and eventually settled on the OSGi framework as the most promising. In parallel, other team members looked at areas such as dynamic registry behaviour.
By the end of 2003, the team had completed implemenations of a dynamic registry and an OSGi R3 framework implementation (plus various extensions). As of Eclipse 3.0 M6, the original Eclipse Runtime was seamlessly replaced with a fully OSGi-based runtime. In June of 2004, Eclipse 3.0 shipped and thousands of people began running and shipping OSGi in their development environments and products!
The Equinox work did not end there. The team could have closed down the project as a raging success. Instead, "Phase 2" was started with a focus on security, framework layering and the like. Some of this work was subsequently absorbed into the main Eclipse OSGi framework and some continues today.
Similarly, the OSGi work did not stop. Several of the original Equinox committers became Platform Core committers and continued to work on the code. The team also worked closely with the OSGi Core Platform Experts Group (CPEG) to standardize the framework extensions (e.g., fragments, Require-Bundle) added to the Eclipse implementation in support of our expanded usecases. Much of this work is now exposed in the recent OSGi R4 framework specification draft and the OSGi Alliance has elected to use the Eclipse 3.1 OSGi implementation as their R4 specification reference implementation. The Eclipse team continues to be committed to OSGi as the base for its runtime technology.
To go to the next level with OSGi, we need a community around OSGi and the framework implementation. The need for an independent OSGi implementation project at Eclipse was highlighted at the 2004 OSGi World Congress. It became apparent that the lack of a clear and obvious community rallying point was an inhibitor to people using our implementation and in fact, joining the community. Simple things such as the lack of an independent download, obvious mailing list or bug home were sending a signal that the Eclipse OSGi implementation was not standalone. In many ways, this reflected reality but in others, the community and opportunities were there but not well advertised.
The need for such a community is further supported by the recent interest in OSGi, the specificiation and the technology. The desire to use OSGi, and more generally a Java component model, is growing as evidenced by the emergence of groups such as the Apache Felix incubator project and JSR-277 (Java Module System). People are still asking why Eclipse has not setup a community around its OSGi implementation. This proposal addresses that concern.
Concretely we propose to transition the Equinox Technology project to be an Eclipse project. The new Equinox Eclipse project will be the home of the OSGi community at Eclipse.org. As a peer of the Platform, JDT and PDE projects, the OSGi code will continue to be managed by the Eclipse PMC and ship with the Eclipse project major releases.
This setup provides the needed infrastructure (mailing lists etc), retains the linkage to the Platform (as OSGi is a key element there) and provides a plausible, visible rallying point for an OSGi community at Eclipse. Creating a top-level project is another option but this option seems like additional overhead for no apparent gain (it is unclear that people readily distinguish between projects and subprojects).
Why move Equinox? Why not create a new subproject? There are several reasons.
What of the other work in Equinox? Equinox was the first technology incubator project. It was put in place before we had the ability to invite others to join existing projects for speculative work. The need to bring in others but not give them full commit rights was one of the key motivators in spawning a Technology project. Since then we have several cases where people have been invited to join ongoing work with less than full commit rights. This is the vision for the new Equinox. The current team working in the various workareas will move to an incubator area of the rehosted Equinox. Nothing will change for them other than the repo location, mailing list name, etc.
The new Equinox will be open to:
As such, the OSGi framework implementation previously developed in Equinox and now hosted in the Platform project will move back to Equinox for further development and maintenance.
The new Equinox will not undertake development of the Eclipse Runtime function such as Jobs, Preferences and Content Types. This will remain part of the Platform project. This furthers the vision of Eclipse as a landscape of OSGi bundles rather than calling out particular bundles as "OSGi" and others as Eclipse. That is, a large (and increasing) number of Eclipse plug-ins can be run on any OSGi framework -- there is no particular value in cluttering Equinox with unrelated content.
The project will be structured into the following components:
Each of these components will have a developer mailing list, a team website, a bugzilla component and other infrastructure as appropriate. Equinox will have just one newsgroup to host the user community. For now, separate newsgroups would fragment the discussion too much. This point can be revisited if the message volume gets unwieldy.
The Equinox repository will be structured as follows:
The current Platform Runtime committers will be grandfathered as full Equinox committers. Current Equinox committers will be grandfathered as committers in the new Equinox Incubator area. The project as a whole will continue to be led by Jeff McAffer, IBM.
The submitters of this proposal welcome interested parties to post to the eclipse.equinox newsgroup and ask to be added to the list as interested parties or to suggest changes to this document.