[
https://issues.apache.org/jira/browse/FELIX-4497?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Thomas Watson updated FELIX-4497:
---------------------------------
Attachment: org.apache.felix.resolver.patch
This patch takes an approach that uses two steps:
1) During the prepare phase (end of populate phase) establish a table of
Capability->Requirement that maps each capability to the requirement that
potentially could be used to substitute the capability
2) For each permutation: check for substitutes. If substitutes are found then
A) create a new permutation that selects the next candidate for the
substitution requirement and B) for the existing permutation remove the
substituted capability as the candidate
> Invalid wirings may result from substitutable exports
> -----------------------------------------------------
>
> Key: FELIX-4497
> URL: https://issues.apache.org/jira/browse/FELIX-4497
> Project: Felix
> Issue Type: Bug
> Reporter: Thomas Watson
> Attachments: org.apache.felix.resolver.patch
>
>
> FELIX-737 is a framework issue but the same issue also exists in the
> standalone resolver implementation. I am opening a separate bug to
> specifically address the issue in the standalone resolver implementation.
> The felix framework resolve has a known issue with substitutable exports
> FELIX-737. The scenario is pretty complex. It involves bundles multiple
> bundles that export and import the same package X at different versions and
> an importer that imports the package with a range that requires a specific
> package. If that specific package gets substituted there are cases where the
> resolver will return wires to the substituted export.
> The ideal solution would do something smart with selection of substituted
> exports so that different combinations or selections can be made if the first
> substitution did not provide good results. This potentially will explode the
> permutations. For now we should at least change the current permutation to
> drop the substituted exports so that invalid wires do not get created. That
> could drop some good valid combinations, but that is better than the current
> state where completely invalid wires get created.
--
This message was sent by Atlassian JIRA
(v6.2#6252)
