[news.eclipse.technology.ejb-orm] Re: JDO fetch-groups important for RCP

[michael.keith@xxxxxxxxxx (Mike Keith)]

Joerg von Frantzius wrote:

<blockquote>
I don't know much of the internals of the EJB3 expert group, but I
heard that the subject of fetch-groups had already been brought up
there. I don't know about Oracle, but my guess is that it is exactly
JBoss who had their share in turning it down, probably most of all
because Hibernate doesn't support it, and seemingly without thinking
further than that.<br>
<br>
Sorry, I don't mean to say you personally or JBoss are sabotaging
client-side Java on purpose, but the responsible persons at JBoss
should know that this is the consequence of their behaviour. <br>
</blockquote>

Joerg,

The subject of fetch groups did in fact come up in the EJB 3 expert group, 
and we (Oracle) along with others including JBoss, BEA and IBM concluded 
that fetch groups sound better in theory than they are in practice. In 
practice, fetch groups are much like lazy loading of attributes in that 
very rarely do they ever really result in better performance. The two 
cases where they can do are when there is a large number of columns in a 
table mapped to one object or when there is a single attribute that is 
very large (e.g. a LOB). All other cases will normally end up causing 
worse performance, not better. The solution to the first case is to use an 
embedded object, while the second can be lazily loaded at the attribute 
level.

To be clear, they were NOT rejected because we do not support fetch groups 
-- because we do in TopLink and I believe that Hibernate does as well, 
although I am not sure about IBM and BEA -- but because they are prone to 
misuse and a complication that we did not feel was warranted in the 
specification.

Hope this clears things up a little.

Regards,
Mike