I have no idea what the issue is when running on Felix, but they appear to be a bugs in the framework itself. I remember Equinox used to have a problem with updating a host bundle to a fragment and vise-versa. But that was fixed in Luna.
Did you try the same thing on Equinox? When I proposed and designed extension activators for OSGi core R6 the main motivation was the region usecase. Before there were serious timing issues because 'normal' bundles could get resolved and active in the service layer before the region digraph had a chance to establish its hooks to enforce the region connections. Now that it is a system bundle fragment it can ensure the hooks are available before any normal bundles are resolved and it can ensure it is enforcing the region connections all the way until the framework is shutdown. Tom From: David Bosschaert <[email protected]> To: "[email protected]" <[email protected]> Date: 10/09/2015 09:10 AM Subject: Re: Recent performance improvements in equinox regions (used by aries subsystems) Looking closer at the latest region digraph bundle I see that it has been converted to a system bundle fragment. Previously it was just an ordinary bundle. It seems to cause issues when updating this bundle to the new version of the bundle in a running framework, although I haven't been able to figure out yet why. I am seeing that the RegionDigraph service does not get registered in such a scenario, even after refreshing the framework. Additionally, stopping the framework (Felix 5.2.0) I get an exception [1]... I have also seen the exception [2] appear when then new fragment bundle is installed. Why was is necessary to convert it to a system bundle fragment? Any ideas on why these problems in the extension activator occur? Thanks, David [1] WARNING: Bundle org.eclipse.equinox.region [21] Unable to stop Extension Activator (java.lang.NoClassDefFoundError: org/eclipse/equinox/internal/region/StandardRegionDigraphPersistence) [2] WARNING: Bundle org.eclipse.equinox.region [25] Unable to start Extension Activator (java.lang.ClassCastException: org.eclipse.equinox.internal.region.management.StandardManageableRegionDigraph$1 cannot be cast to org.eclipse.equinox.internal.region.RegionLifecycleListener) On 2 October 2015 at 17:13, Thomas Watson <[email protected]> wrote: > Yes the following performance bugs have been fixed in Mars.1 > > https://bugs.eclipse.org/bugs/show_bug.cgi?id=476074 > https://bugs.eclipse.org/bugs/show_bug.cgi?id=476136 > > There is an additional performance fix only in master for Neon > > https://bugs.eclipse.org/bugs/show_bug.cgi?id=476640 > > I'm unsure about publishing of the artifacts to maven central. There is > ongoing requests for Eclipse to get their act together and publish all > their jars to maven central, but the solution to that is ongoing. I think > the one you see in maven central was put there by John Ross. > > Tom > > > > > > From: David Bosschaert <[email protected]> > To: "[email protected]" <[email protected]> > Date: 10/02/2015 10:06 AM > Subject: Re: Recent performance improvements in equinox regions > (used by aries subsystems) > > > > Hi Tom, > > I see that Mars SR1 has been released and that the Equinox Regions > bundle org.eclipse.equinox.region_1.2.101.v20150831-1342.jar is part > of this release, so I assume that's the version that contains these > fixes... > > I was wondering, is there a Maven repository somewhere where this > artifact will be published? The one in Maven Central is from a couple > of years ago. > > Cheers, > > David > > On 31 August 2015 at 19:37, Thomas Watson <[email protected]> wrote: >>>Thanks for the note, Tom! >>> >>>Would it be possible to send a note again when the updated Equinox >>>Regions bundle is available? >>> >>>Cheers, >>> >>>David >>> >> >> Sure, and if I forget, there will be milestone build for Mars.1 > available on the Equinox download page on Wednesday, Sept 2. >> >> Tom >> > > > >
