Re: [e4-dev] Nullpointerexception when launching a clean e4application p

[Simon Chemouil <eclipse@xxxxxxxxxxxxxx>]

Toedter, Kai a écrit :
> Tom wrote:
> > Isn't this contradicting the whole DS-System
>
> Why do you find contradicting? If you use DS you need an implementation. And besides, many e4 bundles (like org.eclipse.ui.workbench) have hard dependencies to Equinox bundles...
>
> Kai

I agree with Tom that there should be no dependency on DS.

What you can do in PDE is add implicit dependencies on equinox.ds and
equinox.util so that they will be added automatically when you hit "add
required bundles" even if they're not in a transitive dependency.

As I understand it, one aspect of the rationale of E4 is also to get 
closer to OSGi best practices (no singletons, OSGi services in the 
injection, etc), so that eventually one day someone brave enough should 
be able to swap implementations (if not the whole container, at least 
OSGi extensions like DS or blueprint).

I would find it sad to forego that goal simply for the comfort of
developers who learn very quick they have to include DS anyway: my team
here learnt it the hard way but they got it and it also taught them
something important when using OSGi...

By the way, the "real" solution is in the OSGi specs draft for r4.3
where they propose a new manifest header to say that a bundle provide a
functionality, so that target platform resolvers like PDE or p2 can find
which bundles provide a service (or a lower level feature like DS) and
select the most appropriate (or ask the user which ones he wants). But
we'll have to wait a few years before it goes through the process of
OSGi approval, Equinox implementation, PDE-resolver implementation, PDE
tooling and finally in E4 !

Cheers,

Simon