My point of view is that most likely one or two people will be doing all the work, but that the separate specification of tx and security properties is a wonderful idea because you are forced to think about them as separate aspects to the application logic. IMO this results in better more maintainable designs.
david jencks On 2002.07.02 11:21:48 -0400 Dan Christopherson wrote: > Well, I think you've got it right. I beleive that the distinct and > arbitrary seperation of roles was an attempt at simplification by the > spec authors, but in reality it tends to confuse people who are more > accustomed to reality 8^}). Of course the assembler and deployer roles > also make more sense when you consider J2EE applications produced by > ISVs. This email got long on me, but my opinion on the matter is that > the transactional semantics and some other descriptor elements should be > specified in/with the code and merely be overridable in the descriptor > (see XDoclet). > > In practice (and I think this was implied in the spec, but I may be > remembering wrong) the bean author does define the original descriptor > for his beans. An application assembler (who in reality is probably the > same person as the bean developer, or at least a co-worker) then takes > the ejb-jar(s) and hooks them up with the application UI (web or J2EE > client), tweaking the original descriptor if need be. The deployer (who > is usually a co-worker of the assembler) then adjusts the security maps, > resource references, env-entries, etc. for the actual deployment > environment. > > Since most J2EE applications are developed by one organization either > for its own use or for use by a specific client, all three of these > roles are usually in a single organization, with good communication and > a lot of interaction. Often, the deployer simply follows instructions > given to him by the assembler/app developer/bean developer. In small > organizations (where production support is done by the development > staff) all of these roles (+ administrator) are often performed by one > person. > > As I said before, the roles make more sense in the case of a product > build by an ISV: the ISV will perform the roles of developer and > assembler, and their customers will (sometimes) perform the role of > deployer. > > Also, consider a perfect world where a company in need of a solution, > specifies that solution, contracts oh, say me, for instance, to perform > architecture for the system, the contracts a flaming-logo guy for the > web work and a hide-in-the-dark-and-code guy for the EJBs. Both coders > are handed specs (including the interfaces to be implemented by the > EJBs) by the architect, and write to those specs. Then someone assembles > them, using the transaction and security specified by the architect, and > someone deploys them in the environment desired by the client. Yep, > that'll work! 8^}) In reality, that would require either a ridiculously > detailed design by the architect or a high bandwidth communication > between architect, bean coder, and web dewd. Not to mention an actual > requirement spec, something most companies really don't know how to > prepare. So usually, the same set of contractors will handle > requirements gather, architecture/design, and construction. And then > help the client deploy it. > > hope this clarified more than it obscured, > danch > > Eric Kaplan wrote: > > The theory certainly seems to suggest that there is the ejb coder and > the > > ejb deployer, and that neither one should have to do much with the > other. > > If bean A calls bean B, a deployer can choose to have bean B be > Required or > > RequiresNew at deployment time. However, in reality, the choice has a > lot > > to do with how bean A is coded, doesn't it? In other words, bean B > throws > > EJBException on error, causing the transaction to roll back. If the > > deployer has set B to Required and A continues merrily it's a problem, > since > > the transaction that A was in is now rolled back. If the deployer sets > B to > > RequiresNew and A continues merrily, no problem. If this is true, > what's a > > bean coder to do if he doesn't know the intended transaction semantics > ahead > > of time? Conversely, if he does know the intended transaction > semantics, > > what's a bean deployer to do? Am I thinking about this wrong? I've > been > > grappling with this issue for a while. > > > > Thanks > > > > Eric Kaplan > > Armanta, Inc. > > 55 Madison Ave. > > Morristown, NJ 07960 > > Phone: (973) 326-9600 > > > > > > > > -- > > ========================================================================= > Disclaimer: > All statements are my own, personal opinions and have nothing to do > with the opinions of employer, wife, or cat. They may not have anything > to do with objective reality, as a matter of fact. Your Milage May Vary, > Excludes Delivery Charges, etc. > ========================================================================= > > > > ------------------------------------------------------- > This sf.net email is sponsored by:ThinkGeek > Welcome to geek heaven. > http://thinkgeek.com/sf > _______________________________________________ > JBoss-user mailing list > [EMAIL PROTECTED] > https://lists.sourceforge.net/lists/listinfo/jboss-user > > ------------------------------------------------------- This sf.net email is sponsored by:ThinkGeek Welcome to geek heaven. http://thinkgeek.com/sf _______________________________________________ JBoss-user mailing list [EMAIL PROTECTED] https://lists.sourceforge.net/lists/listinfo/jboss-user
