8.6 Limitations

The following DCharts features are not supported. They will be studied in future research:

• Currently, actions and guards are not supported for target-languages Java and C#. If Python is chosen as the target-language, actions and guards are optionally included in the synthesized code (if parameter --ext is given on the SCC command-line). The behavior of this Python code with actions and guards is the same as the simulation in SVM. If C++ is chosen instead, the actions and guards may also be included (with the same --ext parameter). The code must then be linked to a Python run-time library. In an execution, the binary code automatically loads the Python library, and executes the actions and evaluates the guards. With this, the behavior of the model is also preserved.

  There are many other choices for the implementation of actions and guards. One possibility is to use languages that are independent of specific target-languages, such as action semantics [46] [47] and Modelica [48] [49]. Action semantics is not yet standardized. There is no mature library for it until now. Modelica is a powerful language capable of specifying non-causal equation sets. Actions ``a=b+c, d=a/2'' can thus be written as ``a=2*d, a-b-c=0''. The Modelica compiler symbolically and automatically determines the unknown variables and sorts the equations in an order in which all the equations can be solved sequentially. For example, suppose a and d are unknown before the actions are executed. Modelica changes the order of the equations and symbolically transforms them. As a result, a is solved with `` a=b+c'' first, and then d is solved with ``d=a/2''. Though, there is no non-commercial Modelica solver until now, its has a bright future as both an action language and a constraint language.

• Transition parameters are supported only for Python and C++. The --ext parameter must be explicitly given on the command-line to invoke SCC.

• Timed transitions are supported only for Python and C++. The --ext parameter must be explicitly given on the command-line to invoke SCC. Scheduling events in an execution requires extra threads. This limits the portability and predictability of the model.

• Macros are statically substituted with their values by SCC. The generated code does not contain macros any more. This also implies that macro redefinition is no longer allowed in the synthesized code. When a model imports a submodel, it instantiates the submodel class, which cannot be modified at run-time. Because of this, the behavior of a model is fixed when code is generated.

• Distributed simulation is not supported. Ports and connections defined in a model description are simply ignored by SCC. This feature is left as future work. It is important and meaningful for SCC to automatically generate distributed systems, where different components (objects in the target language) communicate via a network. If they conform to the same communication protocols as PYPVM and SVMDNS, those components may coexist in a system with some DCharts components simulated by SVM. This makes the system extremely flexible.

• As discussed a previous section, the implementation of history is not efficient.