Re: [oscar-devel] Another OBR issue

[Richard S. H. <he...@un...>]

Eric Swindell wrote:

>If the motivation for bundle uniqueness is driven by the ability to uniquely 
>identify bundles for their replication from server to locale repositories, 
>the bundle name and version number could form a unique key.  You could add 
>this as a separate manifest entry, but then again, it could easily be 
>constructed from Bundle-Name and Bundle-Version.  
>  
>

I hadn't thought of this, but yes, that could be one way to get a unique 
ID. Of course, this still requires that Bundle-Name is unique too. For 
example, I might release a Wire Admin implementation and so might you. 
We can't both call our bundles "Wire Admin", otherwise we could suffer 
name clashes. I would have to call mine "Richard's Wire Admin" and you 
would have to call yours "Eric's Wire Admin" or we could use something 
less dorky.  :-)

If we had a separate unique ID, then both could be called Wire Admin. Of 
course, we could just ignore this issue, since we don't have many 
duplicate bundles yet.

>Such keys could also be used to provide limited bundle version dependency 
>management, where one bundle could list it's dependent bundle keys for OBR 
>purposes.  OSGi provides package versioning in the import/export entries, but 
>these assume backward compatibility.  Chances are, you're going to test your 
>bundles against specific dependent bundles so that these dependent bundle 
>implementations are sort of 'certified' as interoperable without ill 
>consequences anyway.
>  
>

I don't think I want to open this can of worms just yet.

>However, isn't OSGi's versioning based 
>on the 'specification version', and excludes any consideration of an 
>'implementation version' or 'service provider'?  Sorry, I'm off topic here :)
>  
>

You are precisely correct on this issue. This topic actually came up at 
the OSGi meeting last week.

-> richard

p.s. The OSGi meeting was really, really interesting by the way...