Difference between revisions of "Equinox Planning - 3.2"
|Line 118:||Line 118:|
* drag 'n drop install
* drag 'n drop install
Latest revision as of 16:19, 13 February 2007
- Refactoring/refining of the framework
- two different packagings, one with all the extra stuff in extensions (for maximum flexibility) and another with everything lumped together the way it is now (for maximum convenience)
- Look for ways of reducing footprint
- Look for ways to simplify the framework
- Update the supplement bundle. It is supposed to be everything in org.eclipse.osgi that is not standard OSGi. Locations, NLS, ... We don't have a good build story around this so just copy the code for now.
- Look at eliminating the use of our special library lookup techniques and investigate the standard OSGi mechanisms (e.g., Bundle-NativeCode)
- Update the quickstart guide and general how tos on the web
- Application model: In 3.2 we investigated the MEG application model but ran out of time. We should reopen that investigation and see where we want to go for 3.3. The primary goal here should be to get more flexibility and control wrt the applications. Starting and stopping, querying, running several, ...
- Server-side OSGi: The server side incubator has several parts that should be reviewed with an eye to graduating them out of the incubator and into the Bundles component. These have to be reviewed carefully in all aspects.
- rework startup.jar: Pascal's favorite topic. This little JAR is hugely complicated and subtle. A nightmare to maintain and test. There are many different scenarios in which is it used and it does a number of things. It also turns out that much of this has to be done in situations such as WebStart, the Servlet Bridge and launching nested frameworks.
- Investigate making startup.jar a bundle (both in form/name and in location and reality). This gets it out of the rootfile set and gets it versioned. As such it is more easily updated and managed. Further, if someone wants to launch a nested framework, they need this code (or a friend of it)
- The goal here is to have all the complicated stuff in one place and then that is reused in the different scenarios (e.g., imagine launching a nested framework while doing selfhosted development)
- improve the mechanism for discovering osgi.framework.extensions
- Service performance
- Registry scaleability
- ServiceTracker laziness and short circuit when no customizer
- Security: there has been alot of work in the incubator over the past year. We should get up to date on it and what it all means. Primarily this means the Java 2 permissions and the code modifications. Need to investigate SWORD4J and get familiar with the process of securing bundles.
- start all by default in simple configurator
- file manager packaging
- runtime clean-up
- x-internal/friends in refactored runtime
- run in strict mode/test cases
- equinox builds/features/downloads
- patching fragments
- simple configurator - consolidate platform.cfg files from multiple install/updaters
- component framework
- be good OSGi citizens
- credential store
- pervasive use of "uses"
 Declarative Services
- implement optimizations
- investigate how to generate componentized JavaDoc. Maven may have a story for this but in any event, we should understand how we can build JavaDoc for each bundle as we build the bundle.
- create finer grained features (help, update, ...)
- jar signing
 Core tools and JMX
- Get JMX code released
- update the core tools in terms of hooks, options, and implementation
- expose the core tools "management objects" as mbeans
- for example, assuming the options are turned on, under a bundle there should an "activation" child that, if selected, gets the stack trace of when that bundle was activated. Similarly there should be a "classes" link
 PDE UI
- remove use of "plugins dir"
- secondary classpath
- import management
- populate a target via update
- tools for version management
- Binary compatibility checker.
- Version checker to see if the version evolved properly.
- API inclusion checker (to ensure that all the APIs called are available when running with the lower range).
- Filtered code assist based on the value in the @since tag and the version specified in the manifest.mf.
- Warm fuzzy feeling checker.
- named targets
- service constraints
- splash info gathered in product and fanned into config.ini
- "Add to target" button
- tooling around "uses" directives
 PDE Build
- look at how Maven does incremental building
- Do a prototype of building Eclipse things using Maven
- use the Maven OSGi plugin and see how it works, what kind of metadata we have to create (do it by had first perhaps)
- start with the Equinox bundles
- then try RCP (this brings in SWT (always a problem child) and the launchers as well as packaging issues like zipping, links, permissions, etc.
- explore some of the other Maven plugins. There is one that packages as RPMs, OS services, installers, ...
- Wagon: this is the transport layer under Maven's repo client
- Can this be used to populate update sites?
- Can this be used to populate targets?
- Repo indexing: Maven has a repo indexer that could, for example, be used to generate packed content or the update site digest
- investigate Maven's configuability wrt repo structures etc.
- Mangen: Rob Walker recently contributed mangen (manifest generator) to Felix. This does class file analysis to see what packages etc are used by a piece of code. We should look at this and see what it can do. Is this something that can/should be integrated into our development model.
- parallel X build
- predict size of downloads
- populate target
- alien configs - be able to modify configurations which aren't the currently running one (e.g. remove references to PlatformConfiguration#getDefault)
- performance and robustness
- pluggable transports
- role-based provisioning
- OBR - OSGi Bundle Repository
- marketing #
- simple configurator use
- minimize dependanices for update configurator
- update deltas
- signed JARs
- add to site
- small update application
- warm fuzzy tool
- import package requirements in features
- maven back-end
- drag 'n drop install