Installing features that are not matching (e.g. ECF), probably need a special UI
Memory consumption, need to be able to install Ganymede without OOME
Showing in the UI the plug-ins that have been installed
Is the filtering of IU on filter desirable and can't it be potential source of problems?
Why can't we watch the plug-ins folder in self-contained setups?
Do we have a p2-ized version of the platform build?
There was a lot of excitement about p2 at EclipseCon. People are eager to get the new functionality offered by p2: simpler workflows, the ability to uninstall, management of non-bundle artifacts, touchpoint actions, etc.
There are many people jumping on p2 and ready to use it in their products: NASA/JPL, Wind River, Genuitec (Pulse), Cognos, and others.
The p2 talk was well attended - 200+ people, standing room only. p2 BOF was also well attended. Another enterprise provisioning BOF was heavy on p2 content.
We need to focus on responding to bugs from early adopters and ease the pain of transition and adoption.
The main problem area is that the dropins folder is strict about resolving dependencies. If any plug-in in the dropins folder has unsatisfied dependencies, we fail to install anything from dropins. We need a "best-effort" approach where we install as much of the dropins content as possible.
The other big pain point is replacing bundles that are already present in the install. This is more problematic and we'll address this early in M7.
Another case that occurs is when multiple features have conflicting includes on the same bundles. For example the org.eclipse.rcp feature now includes some bundles from ECF that are also available in ECF features. Users want to be able to install the ECF feature even if it conflicts with existing features in the configuration. This problem likely requires non-trivial changes to p2 metadata and we don't have time to address this for M6
Memory consumption problems when installing all of Ganymede
Slow performance installing and updating even simple IUs
We need to investigate to find out more details about where the time/space is going
Unless there are some simple/safe optimizations we'll likely need to defer further performance work to M7