Hello everyone.
This message is particularly important for anyone who embeds Felix into
their own projects (i.e., directly creates an instance of Felix).
Two weeks ago I was working on simplifying the Felix shutdown process,
since it is somewhat circular and confusing. During that process it
dawned on me that there is a better way to handle the entire way we
create Felix instances and start/stop the framework.
In summary, the idea is to have the Felix class implement the Bundle
interface; thus, when you create an instance of Felix what you get in
return is the actual system bundle.
In this scenario, the Bundle.start()/Bundle.stop() methods will be used
to start/stop the framework, respectively. All other methods will work
the way that the current system bundle methods work (generally as
defined by the spec). Following this approach will not expose any
methods that are not already currently exposed by the system bundle.
Further, it will actually leverage the standard Bundle API to provide
access to what was previously framework-specific API (e.g., typically
when embedding Felix, the host application needs access to the system
bundle context, now it can get the system bundle context via the
standard Bundle.getBundleContext() method).
The current approach will further be modified so that the configuration
property resolver and the system bundle activator list will be passed
into the Felix constructor, instead of the Felix.start() method.
I almost have these changes ready to commit (hopefully next week), so
this message serves as advanced warning that this will break embedding
applications; however, converting to the new API should be very simple.
For those using the standard Felix launcher, this will not impact you at
all.
When I commit these changes I will also update the "launching and
embedding" documentation. I will make an announcement when all of this
happens.
Feedback and input is welcome.
-> richard
