Hey, I've been looking through the runtime code today and noticed that the algorithm to "check module, then layer, then used layers" is replicated like umpteen times for various purposes. That sucks. It is also implemented in the instance level rather than the model level, even though many of the things being done should be on the model level. That sucks too.
One thing that really sucks is that service dependency injection binding *should* be done once in the model, and then just "executed" at runtime. This will be faster, and will also help the visualizer, since it could then trivially show all service references that have been bound to a particular @Service injection. So I attempted a solution with visitors, which have helped us already in other places. It turned out to work very well, which means that the "check module, then layer, then used layers" algorithm is in *one* place in the code (in the model), and then visitor implementations decide what needs to be done for the various cases. Much much nicer and easier to read. Apart from also pushing many of the algorithms down to the model layer, it makes many of the results cacheable, as the visitor after usage contains the result of the computation. Just put that in a cache, and subsequent calls will be superfast. I hope to be able to finish this tomorrow. /Rickard _______________________________________________ qi4j-dev mailing list [email protected] http://lists.ops4j.org/mailman/listinfo/qi4j-dev
