Hi,
I have a hard time seeing actual use of modeling with the run-time model library, but without any specific target/action language selected. As I already indicated in my little sketch, the run-time model library (if placed in the "Common Runtime" part) gets included by selected a specific target/action language feature (C++ or Java in the sketch).
As soon as you want to model run-time stuff, e.g. add an SAP based on any of the system protocols in the run-time model library, then you are a very small step away from needing a target language and actually reference that SAP in your action code. Keep in mind that two of the three system protocol actually do not have any protocol messages, since they are accessed directly as methods on the C++ representation of the port. So the use for adding SAPs based on empty protocol, without writing action code is pretty limited.
Sure, there could be some special cases where you are doing some "sketchy" modeling and still want to add SAP ports (based on "empty" protocols), but not write any action code. But I have a feeling that this is such an uncommon case, and if this is needed, then I would like to propose that you have to select a action/target language to be able to do that. Just to reduce the number of choices and alternatives.
So I would like to simplify things to cover two cases: Structure modeling (optionally with high level state machine modeling), but no code-generation/run-time support (and no run-time model library) and full-fledged UML-RT modeling with code-generation/run-time (including the run-time model library).
/Peter Cigéhn