But for things like large transaction apps that absolutely have to work for absolutely everyone, and one screwed transaction can mean millions of renminbi (financial sector type stuff, meaning Citibank, not FQ), EJB3 looks nice. Of course they’re definitely not first movers and probably are still looking at EJB1.

Java 1.5 is much better than 1.4. Going back to 1.4 temporarily recently has been painful since it keeps bitching at me because I’ve gotten used to autoboxing and generics.

Presentation and navigation stuff stays up in the web layer and anything business related happens in a session bean or below (DAO, etc). The reason I like them is that you get transactions for free and they are independent of the web layer. This is good for reuse.

With EJB3 they’ve basically peeled all the crap away that made EJB’s a pain to use. Now you just code a Java class and add maybe one or two annotations. For that small amount of work, you get container managed transactions, dependency injection, the ability to expose your bean as a web service (one more annotation), etc. They look pretty good to me. The only downside is (still) that the code must be run inside of a J2EE container. This can be a bit painful for testing but there’s Cactus to address that. If you don’t like this aspect, then maybe Spring is the way to go. I think the new EJB looks extremely good - better by leaps and bounds than previous versions.

As for annotations, just think of them simply as source code. Anything you would hardcode with source, annotations are appropriate for. Anything you would not hardcode, use a database instead - just like you always did.

Anyhow, I’ll just be glad to get on to Java 5. I would kill for auto-unboxing alone.