I have been putting off writing this email for a while... mostly because I am lazy.
Any ways, I am concerned about the current trend for packaging which is being employed in JBoss. By packaging I mean the way in which we organize our classes into packages. I believe the current "package by type" scheme is a poor choice for large (and rapidly growing) projects. By "package by type", I mean that we are organizing different functional components by the type of "thing" it is, rather by the functional group in which it belongs. How about an example, that always seems to explain this issue. At one company I used to work at (which I will leave unnamed to protect the innocent... or rather the stupid... or both), they were building a webapp based on servlets. These servlets talked to databases via objects (written poorly and completely non-j2ee like). The app was big and had lots of different functions, but there staff of "senior" engineers (degrees and all) chose to organize *every* servlet and *every* db object into two packages: some.dumb.company.servlets and some.dumb.company.db. Needless to say this approach is not very manageable. Imagine you are a new employee trying to get up to speed fast on one function of the system. There isn't any good documentation or class naming scheme to tell which classes are for what and all of your coworkers are too busy making sense out of the spaghetti they have written (poorly) to be of much help. It would have been much better if they had packaged by function, such that all of the login servlets got a package, and then all of the user management servlets got another and so on. * * * I would like to avoid this for JBoss. Granted we are not in the same situation as the folks I mentioned above (we actually know how to write Java code), but I think we have overlooked some of the organization required to make a large software project truly effective and easy to manage. One example of this is the org.jboss.metadata package. There is metadata classes from all aspects of the system here, rather than closer to the functional groups in which they are used. This package only really makes sense if there are helper, abstract or type classes which are used by cocreate metadata implementations. Another is the org.jboss.deployment package. This guy has j2ee stuff, sar stuff and more. Why not put those in the packages which deal with those areas? Leaving this again for abstract impls and such. Lastly take a look at the ejb proxy's. This is a different case, but still a little backwards (IMHO). org.jboss.proxy.ejb, contains the ejb container proxies. This is a sub-package inside of the package by type methodology. It would make more sense to put them into org.jboss.ejb.proxy, and leave org.jboss.proxy for abstracts and the like. Any ways, I wanted to get this off my chest. I hope that some (or most) of you understand what I am trying to get at... avoiding the spaghetti (tastes good, but I don't want to clean it up off the floor). If we spend a little effort in optimizing the organization of our codebase, it will be easier to manage, maintain and much easier for newbies/learners to get up to speed faster to help make the system even better. --jason _______________________________________________ Jboss-development mailing list [EMAIL PROTECTED] https://lists.sourceforge.net/lists/listinfo/jboss-development
