[xwiki-users] How Plugable is Data Model?

This one's for the hard-core XWiki gurus... My group is interested in going heavily into using XWiki to manage (semi) structured information that will be partially shared with other applications (non-wikis, possibly desktop apps). We figured since XWiki has an extensible object-model and accompanying persistence layer, we might let it be the authoritative keeper of all data in the distributed system. After playing with XWiki for a couple of weeks and prototyping some pages/objects plus a component to package and exchange data with another Java application, we're starting to wonder if there couldn't be a more efficient way to access/distribute object data, and even to define a common data model across applications and wiki: - Regarding access/distribution: As an example, we're prototyping a graphical tool that would want to get a whole graph of objects at start-up to build a node/link picture of some situation (where each node or link would also be a page/object in the wiki). Given the existing DB schema, we feel like we're struggling at this point to imagine a DB query that would pull all relevant object data for a given graph at one go, and even if we could pull the data efficiently, we're not sure how best to package it up for transmission. - Regarding more efficient object class definition: We're imagining a system with maybe 25 classes (and growing), and it would be nice to define them once and have them exist both as Java POJOs (for the other application bits) and XWiki classes (for XWiki to generate page views). We've been using the Eclipse Modeling Framework (EMF) quite a bit lately to do that kind of thing--e.g. to define a model in a UML-ish way, and generate different artifacts, including Java class code. I drag EMF into this because I like the idea of having one modeling point for all parts of the system, and because its native generated POJO object implementations are pretty efficient (it also provides freebie object-graph serialization to XML, which we've used successfully for the "distribution" issue above in other systems, and it even allows for dynamic runtime class definitions). One thought we had was to create modified EMF generator templates to generate code that would invoke XWiki APIs to build needed XWiki object class structures, as well as packer/unpacker routines to convert between XWiki objects and EMF-POJO objects. This approach has some challenges: (1) I don't know if there's an exposed XWiki API for defining object classes, (2) if there is, I don't know how tricky it might be to use it correctly, and (3) hacking the EMF generator would be a pretty daunting task too. Another thought we had was to totally replace the underlying XWiki object model. This approach is likely even more challenging, but could produce a better (more efficient) overall implementation. We checked out the XWiki source from the public SVN and on first glance it appear com/xpn/xwiki/objects and com/xpn/xwiki/objects/classes are the packages to look at. It seems like it might be possible to use EMF-POJOs as our data model, by either wrapping or extending them to satisfy ObjectInterface and ClassInterface. If we could do that, it might also make sense to also replace the persistence layer (com/xpn/xwiki/store ?) by using EMF's ability to generate DB-persistence through Teneo and Hibernate (if we do our own persistence, we can perhaps set up schema/indexing that better supports key high-volume queries in our application). So here's the question part of this question: 1- Do the overall goals of using XWiki in this kind of distributed/multi-application system make sense and fit within the envisioned envelope of what XWiki is reasonably intended to support? 2- Has anyone worked on pulling large sets of objects from the XWiki DB in one go, and packaging them for transmission over the wire to some other Java application? 3- Is there a programmatic/API way to drive XWiki to define new object classes, and if so, is there any documentation on how to use it? 4- Is replacing the data and/or persistence models underlying XWiki something that is supposed to be doable? How hard/complicated should we expect it to be? Is that the kind of thing we could get help with on this list? 5- Do you have any other comments or advice on how to think about this kind of problem? Hope this strikes someone as interesting and worthy of comment. Thanks for any guidance you can offer. --Eric ========================================================================= Eric Domeshek Phone: 617-902-2223 AI Project Manager Fax: 617-902-2225 Stottler Henke Associates, Inc. EMail: domeshek(a)stottlerhenke.com 86 Sherman St., Cambridge, MA 02140 Web: www.stottlerhenke.com =========================================================================