Usually in model based testing a "Model" is generated from specification. We do not include any source code information. What if I need some information to include in the model which can be generate from source code? Is it possible to include source code information in a model? If so, then how to do it?
It depends on what you meant by a "model". If you use the UML (Unified Modeling Language) model, you can include source code constructs in UML's Activity Diagram, Sequence Diagram, or Communication Diagram. Also, please search for any reference on refactoring or reverse engineering where source code is used to generate models. Hope this helps.
Thank you Cyril. Its a great help. I am planning to use UML model. I will look through your recommendation.
I have a question here. You said that i can include source code constructs in UML's Activity Diagram, Sequence Diagram, or Communication Diagram. What if source code is erroneous? That mean the model will be erroneous too. How to address that issue?
Well, wrong model will generate the wrong code and vice versa. Automatic programming is far from perfect. If C = C + B should be C = C - B, there is no way the machine will know unless you build in some constraints on the data (e.g., after the operation, the value of C should be less if C and B are positive). Bottom line - logical error is the hardest to detect in programming.
I think that model should be as abstract and simple as possible. So you can use OCL (Object Constraint Langage ) in UML diagrams to express the code information that you want to include in your model.
Actually, you can add any program code in any language to a UML state machine, activity, or action which allow the inclusion of something called "OpaqueBehavior". This is a catch-all for model elements that are unknown to UML and, hence, not verifiable against the UML standard. In other words, there is indeed a danger that such element could include code that violates the semantics of UML, which would render the model erroneous.
In practice it is quite common to embed programing language fragments (e.g., C, C++, Java) in UML models, when such models are used to automatically generate code. There are literally thousands of industrial applications that do that. And, many of them have hit the problem of such program code being inconsistent with the semantics of UML. To get around that, the OMG has now standardized a language called Alf, which is based on UML semantics and which can be used with UML in place of such "foreign" fragments. Several UML tools now support Alf, including IBM's RSA (where it is called UAL) and the open-source Papyrus UML tool.
As to whether models should or should not include such detail, it is a question of the purpose of a model. In one case, models are primarily for analysis and communication. Such models, sometimes referred to as Descriptive models, should be as simple aand as clear as possible. The other case are Prescriptive models, which are used as programs and they need to include whatever detail is necessary to produce an implementation. Both kinds of models are useful and used. One nice thing about using |UML is that you can seamlessly move from descriptive to prescriptive models.
Finally, I should correct an earlier post on this thread: you CANNOT directly embed program code fragments (i.e., opaque behaviors) with sequence diagrams. However, you can have execution occurrence specifications (those blocks overlapping lifelines), which point to such fragments that are defined in either state machines, activities, or actions.
I should clarify my comment on including code constructs in sequence diagram. For example, generic sequence diagram provide representation of conditional (if) statements (enclose each condition for an if statement in square brackets). I meant to say that these constructs (e.g., if statements) reflect the code skeleton.
Thank you Bran Selic and Cyril S. Ku for your valuable comments.
I probably cannot help you in how to put information using your UML tool :) But assuming you can just stick any information in some XML describing your model, the technical problem is more in collecting the information you need. Generally, you would need a parser to extract information from your source code; they you need to figure out, to which element of the model this information belongs to. The latter is probably not trivial. In any case, this probably means a lot of programming of your own custom tooling.
- Badges
- Science topic