Blogging Roller

... and that is just a few of the recent posts ...

The topic of Java persistence frameworks just keeps on trucking through the Java blogs.  I think that the reason the topic keeps coming up is that there is no clear concensus on the "right way" to do persistence. I certainly don't know the right way. But anyway...

I'd like to understand the available tools a little better, so I've started looking at the tools mentioned in the above posts.  I found the feature comparison on the Cayenne project's Wiki to be helpful and I also found a very nice set of Java persistence framework development scenarios on the Hibernate website.  The Hibernate folks have outlined four development scenerios for using a Java persistence framework: top-down, bottom-up, middle-out, and meet-in-the-middle.  Here is my generalized description of these four scenarios, plus a fifth "generic-object" approach that I have added to accomodate OFBiz-EE:

I think the Hibernate guys have really nailed it.  Every tool I have seen supports one or more of these scenarios.  So I decided to use these scenarios as a way to compare the available Java persistence frameworks.  I started to build a table to compare the scenarios supported by the various persistence frameworks and some other factors that I think are important interesting (thanks ceperez): support for the JDO API, support for ODMG API, development activity, and database support.  

Here is the first cut of my comparson table, I will be trying to complete it over the next couple of days:

	Abra	Castor JDO	Cayenne	Commons SQL	Hiber -nate	Jakarta OJB	OFBiz EE	Jakarta Torque	TJDO
Top-down	Y		Y	Y	Y	Y	N	N	Y
Bottom-up					Y		N	Y
Middle-out	Y		Y	Y	Y		N	Y
Meet-in-middle		Y			Y		N	N
Generic-object							Y
JDO API	N	N	N	N	N	soon	N	N	Y
ODMG3 API		subset	N	N	Y	Y	N	N
License	BSD	BSD	BSD	APL	LGPL	APL	BSD	APL	APL
Last release	0.9.6 Sept 2002	0.9.4.1 November 2002	1.0a3 Sept 2002		1.1.4-b Oct 2002		2.0-b2 July 2002		2.0-b Sept 2002
SF Activity	19.4%	-	98.3%	-	99.9%	-	98.9%	-	95.3%
SF Downloads	7	-	2698	-	775	-	86869	-	278
Developers	3		3	3	11	17	11 (1)	10 (2)	7
MySQL	Y	Y	Y		Y	Y	Y	Y
PostgreSQL	Y	Y			Y	Y	Y	Y
HSQL		Y			Y		Y	Y
Oracle	Y	Y	Y		Y	Y	Y	Y
SQL Server		Y			Y		Y	Y
DB2	Y	Y			Y	Y	Y	Y
Sybase	Y	Y	Y		Y		Y	Y

Footnotes:
Other notes:

• I took JGrinder and JRF out of he mix because they seem to be a little too intrusive.  They force your to make your persistent objects extend framework objects and they don't seem to offer any tools to help you generate DDL or persistent objects.
• I'm going to limit myself to open source frameworks for the time being, but I'd also like to consider Oracle's TopLink, CocoBase, Oracle's BC4J, and the commercial JDO implementations.  
• I'm also going to ignore EJB CMP for now.
• Blank cells still need to be filled in and hints are appreciated
• A dash "-" indicates "not applicable.
  

By the way, Roller uses the bottom-up approach.  Roller uses Castor, but not the Castor Source Generator.  Instead, Roller uses XDoclet to generate the persistent data objects, Struts form beans, and the Castor Mapping file all from the same source meta-data.  The Castor/XDoclet combo is not perfect, but I still don't know that moving to a new framework would be an improvement.  We might just be exchanging one set of idiosyncracies for another set.