atom feed19 messages in[jr3] Plugin architecture
FromSent OnAttachments
Jukka ZittingFeb 17, 2010 8:06 am 
Carsten ZiegelerFeb 17, 2010 8:46 am 
Adam FoltzerFeb 17, 2010 5:59 pm 
Justin EdelsonFeb 17, 2010 8:04 pm 
Thomas MüllerFeb 18, 2010 12:20 am 
Stefan GuggisbergFeb 18, 2010 1:37 am 
Stefan GuggisbergFeb 18, 2010 1:45 am 
Carsten ZiegelerFeb 18, 2010 1:53 am 
Bertrand DelacretazFeb 18, 2010 2:02 am 
Guo DuFeb 18, 2010 3:54 am 
Alexander KlimetschekFeb 18, 2010 4:05 am 
Alexander KlimetschekFeb 18, 2010 4:07 am 
Felix MeschbergerFeb 18, 2010 4:34 am 
Guo DuFeb 18, 2010 4:53 am 
Alexander KlimetschekFeb 18, 2010 5:00 am 
Justin EdelsonFeb 18, 2010 5:15 am 
Thomas MüllerFeb 18, 2010 6:34 am 
Marcel ReuteggerFeb 19, 2010 6:40 am 
Michael DürigFeb 20, 2010 6:11 am 
Subject:[jr3] Plugin architecture
From:Jukka Zitting (
Date:Feb 17, 2010 8:06:00 am


Regardless of whether we go with a microkernel approach as discussed in the other thread, making Jackrabbit more modular and extensible would be quite useful. Besides the design benefits, plugin architectures typically also increase community participation as they make it easier to customize or extend the product.

The current architecture already provides a number of extension points via various core interfaces and configuration points, but they are a bit cumbersome to use and not always in the points where you'd want them to be. Here's what I have in mind for what we could do about this:

* dynamic configuration: The current XML-based configuration mechanism needs to be updated whenever new extension points are introduced and makes it difficult to support dynamic plugin environments like OSGi or IoC containers.

* higher level extension points: Most of our extension points are currently deep down at the bottom of the Jackrabbit architecture (PersistenceManager, Journal, FileSystem, SearchIndex, etc.). It would be useful to offer also higher level constructs like Repository and Session lifecycle listeners or transaction boundary checkers to be injected into the system.

* whiteboard pattern: Instead of the custom context objects (PMContext, ClusterContext, etc.) and related init calls that statically wire our components together we should look at implementing more dynamic whiteboards from where all components could do on-demand lookups for the things they need.

WDYT? Any other ideas on what we should do in this are?