[
https://issues.apache.org/jira/browse/FELIX-4190?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13910322#comment-13910322
]
Guillaume Nodet commented on FELIX-4190:
----------------------------------------
So I come back to my earlier proposal which is to duplicate the loop at the end
after having invalidted the registration.
I.e.
{code}
// Now forcibly unget the service object for all stubborn clients.
Bundle[] clients = getUsingBundles(reg.getReference());
for (int i = 0; (clients != null) && (i < clients.length); i++)
{
while (ungetService(clients[i], reg.getReference()))
; // Keep removing until it is no longer possible
}
((ServiceRegistrationImpl) reg).invalidate();
for (int i = 0; (clients != null) && (i < clients.length); i++)
{
while (ungetService(clients[i], reg.getReference()))
; // Keep removing until it is no longer possible
}
{code}
The call to getUsingBundles does not need to be synchronized, as the method
itself synchronizes on this.
At this point, no new bundles can acquire a service reference, so we can keep
the list of bundles and iterator on it safely.
Then, we do a first pass on ungetService which should remove most of the
references, except if a bundle is re-acquiring the service.
We then invalidate the registration, which will prevent any further call to
getService.
We iterate another time on the bundles in case a bundle has acquired the
service during the first loop.
> The framework should not hold any lock while calling ServiceFactory#unget
> -------------------------------------------------------------------------
>
> Key: FELIX-4190
> URL: https://issues.apache.org/jira/browse/FELIX-4190
> Project: Felix
> Issue Type: Bug
> Components: Framework
> Reporter: Guillaume Nodet
> Assignee: Guillaume Nodet
> Fix For: framework-4.4.0
>
>
--
This message was sent by Atlassian JIRA
(v6.1.5#6160)
