[
https://issues.apache.org/jira/browse/SLING-8104?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16686972#comment-16686972
]
David Bosschaert commented on SLING-8104:
-----------------------------------------
Maybe we could have a mechanism as follows: if more than one feature provides a
certain artifact (bundle) then, if both are _not_ exactly the same some
configuration needs to be present to do the resolution. Otherwise the merge
will fail.
For example maybe some configuration in the creation of the aggregate in the
slingfeature-maven-plugin, like this:
{code:java}
...
<aggregate>
...
<artifact-clashes>
<org.foo.my.bundle>1.2.3</org.foo.my.bundle> <!-- use a specific version -->
<org.foo.my.bundle>HIGHEST(1.*)</org.foo.my.bundle> <!-- use the highest
version available with a major version of 1.x -->
<org.foo.my.bundle>HIGHEST</org.foo.my.bundle> <!-- use the highest version
(what it does now) -->
<org.foo.my.bundle>ALL</org.foo.my.bundle> <!-- keep all versions -->
</artifact-clashes>
</aggregate>
{code}
I think the mechanism should work on artifact gav's rather than bsn/version
since the merger only knows about artifacts. This also makes the mechanism more
generic.
Just an initial proposal, would be more than happy with improvement
suggestions. Note that the launcher will also need a mechanism to specify this.
> Avoid magic when merging features
> ---------------------------------
>
> Key: SLING-8104
> URL: https://issues.apache.org/jira/browse/SLING-8104
> Project: Sling
> Issue Type: Improvement
> Components: Feature Model
> Reporter: Carsten Ziegeler
> Priority: Blocker
> Fix For: slingfeature-maven-plugin 1.0.0, Feature Model 0.2.2
>
>
> Currently when features are merged a simple algorithm is applied which just
> picks the highest version based on the bundle version. However this version
> might not have no meaning at all and might not really reflect what has
> changed inside the bundle.
> Especially when there is a major version change, this approach seems to be
> clearly wrong
> But in the end, picking a single version is magic.
> While the problem could probably be solved by using something like a resolver
> and figure out if just one version is enough or if both versions are needed,
> without a resolver there is no way to figure this out.
> Therefore we should provide a similar way as we do for variables at the
> moment: if there is a clash the caller needs to provide context on what to
> choose.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
