I think we all agree that using native OSGi apis for all of karaf is no
good idea.
I only intended to use the native APIs for the feature core bundle. The
setup is definately harder than with DS but it is doable and
allows to have a core ( I think I would even name it bootstrap or
similar) that does not depend on any framework. One advantage would be
that this core would
run with any version of DS and blueprint. If we use one of these
internally we kind of dictate the version as we ship it.
Ideally I would like to see a karaf core or bootstrap that just consists
of two small bundles:
1. Support for mvn urls
2. features core or perhaps in the future subsystems
The system dir of the net distro then only would need to contain these
two bundles + the framework of course.
Of course we are quite far from that but I think it would make sense to
move a bit in the direction of such a simple setup.
Christian
Am 20.01.2014 13:19, schrieb Ioannis Canellos:
I had the impression to read an e-mail from Peter Kriens ;)
LOL
I know DS interests. My point was more DS or Blueprint vs OSGi "native". It
was not between DS and Blueprint.
Ooops. I think that I misunderstood your question.
The problem with the plain OSGi approach, is that its extremely hard
to get it done right and really hard to maintain it.
In the simplest case, you have a service that depends on an other
service and using a plain service tracker is pretty straight forward.
When a service has more than one service dependencies the complexity
starts to increase. You'll need to be extra careful to prevent
concurrency issues and the resulting code will be hard to deal with
(tons of boilerplate all over the place).
And of course, transparency and control wise we wouldn't even be as
near as DS can get us ;-)
--
Christian Schneider
http://www.liquid-reality.de
Open Source Architect
Talend Application Integration Division http://www.talend.com
