I don't have much experience with the reference: url, so I'm not
really sure of the drawbacks.  What happens if the file is deleted and
overwritten because aether downloaded a new snapshot instead of the
one that was installed ?

On Wed, Apr 20, 2011 at 22:45, David Jencks <[email protected]> wrote:
> I don't think that's a good idea.
>
> No one has addressed my other concern that directly using mvn: urls for stuff 
> in system means all feature bundles get copied info the framework.
>
> I would like to consider installing bundles in system using reference: urls 
> rather than mvn urls, even if the bundle is specified with a mvn url in a 
> feature descriptor.
>
> I'd also consider optionally using reference:urls for files that aether has 
> downloaded into a local maven repo, which would further reduce copying but 
> might introduce other problems.  On the other hand it might solve the 
> BundleWatcher problem that currently requires pax-url-aether to be embedded 
> in shell.dev.
>
> thanks
> david jencks
>
>
> On Apr 20, 2011, at 1:29 PM, Achim Nierbeck wrote:
>
>> oh, I forgot about the minimal one :(
>> the standard one is using it. So do you suggest that we revert to the 
>> standard pax-url-mvn bundle
>> to get this snapshot issue resolved?
>>
>> regards, Achim
>>
>>
>>> Ah, right, aether behaves differently and does not have this notion of
>>> default repositories.  I think we may want to fix pax-aether in order
>>> to support that.  I think it's just about having a  custom local repo
>>> pointing to system and checking if the artifact can be resolved in
>>> that one first.
>>>
>>> Note that the snapshot problem isn't really trivial.  If we use
>>> pax-aether and the system dir as a local repository, we can't really
>>> use it as a 'default' repository, else snapshots won't ever be
>>> downloaded again.
>>> At the same time, we want people to be able to update snapshots easily
>>> (similar to dev:watch but using remote repositories).
>>>
>>> Where is aether installed ? The startup properties still list the mvn
>>> handler and not aether:
>>>   
>>> http://svn.apache.org/repos/asf/karaf/trunk/assemblies/apache-karaf/src/main/filtered-resources/minimal/startup.properties
>>>
>>> On Wed, Apr 20, 2011 at 18:59, Achim Nierbeck<[email protected]>  
>>> wrote:
>>>> maybe we need to ask toni if he changed something on pax-url-aether
>>>> that changed this behavior
>>>> since with 3.0 we use pax-url-aether instead of pax-url-mvn to resolve
>>>> the dependencies.
>>>>
>>>> regards, achim
>>>>
>>>> 2011/4/20 Guillaume Nodet<[email protected]>:
>>>>> Yes, that would definitely be a problem.
>>>>> However, i'm not sure why it happens.  The mvn url handler is
>>>>> configured with system as a default repository which should override
>>>>> any other repository, including the default m2 local repository (and
>>>>> obviously any remote repository).   I did that a while ago to solve
>>>>> this exact problem.
>>>>>
>>>>> On Wed, Apr 20, 2011 at 18:50, David Jencks<[email protected]>  
>>>>> wrote:
>>>>>> I discovered that features can pull in snapshots from the apache 
>>>>>> snapshot repo rather than the ones you carefully installed into system 
>>>>>> if someone does a deploy of a snapshot between when you assembled the 
>>>>>> server and started it.
>>>>>>
>>>>>> I found this behavior very disconcerting and I'm not sure it's what we 
>>>>>> want.
>>>>>>
>>>>>> One way to change this and also fix the "we're copying all the bundles 
>>>>>> into the framework" problem might be to examine each feature bundle and 
>>>>>> if its in system use a reference: url instead of the supplied mvn url.
>>>>>>
>>>>>> The situation in more detail:
>>>>>>
>>>>>> build a snapshot bundle X locally with local modifications.
>>>>>> assemble a server X in the system repo and a feature using X in boot 
>>>>>> features.
>>>>>>
>>>>>> Someone else deploys a different X snapshot to say apache snapshot repo
>>>>>>
>>>>>> Start the server you assembled..... the feature starts and 
>>>>>> pax-url-aether fetches the X from apache snapshot repo instead of the 
>>>>>> one in system.
>>>>>>
>>>>>> thoughts?
>>>>>>
>>>>>> thanks
>>>>>> david jencks
>>>>>
>>>>>
>>>>> --
>>>>> Cheers,
>>>>> Guillaume Nodet
>>>>> ------------------------
>>>>> Blog: http://gnodet.blogspot.com/
>>>>> ------------------------
>>>>> Open Source SOA
>>>>> http://fusesource.com
>>>>>
>>>>> Connect at CamelOne May 24-26
>>>>> The Open Source Integration Conference
>>>>> http://camelone.com/
>>>>>
>>>
>>>
>>
>
>



-- 
Cheers,
Guillaume Nodet
------------------------
Blog: http://gnodet.blogspot.com/
------------------------
Open Source SOA
http://fusesource.com

Connect at CamelOne May 24-26
The Open Source Integration Conference
http://camelone.com/

Reply via email to