Section: New Results
Our advances in Model-based testing are related to language modelisation and test generation with properties.
Test Generation from Behavioral Models
Participants : Fabrice Bouquet, Thibaud Brocard, Pierre-Christophe Bué, Kalou Cabrera, Jean-Franois Couchot, Frédéric Dadeau, Stéphane Debricon, Alain Giorgetti, Adrien de Kermadec, Jonathan Lasalle, Vincent Pretre.
We have introduced an original model-based testing approach that takes a UML behavioural view of the system under test and automatically generates test cases and executable test scripts according to model coverage criteria  . In parallel, we are working on the improvement of the test generation technique, by combining constraint solving and theorem proving, in order to detect inconsistencies in the behaviors extracted from the model, and to find a relevant instantiation of the initial test data.
A rebuild of the architecture of the BZ-Testing-Tools engine will start in december 2008, with the help of an "ingénieur jeune diplomé". It aims at integrating the latest works on constraint solving and theorem proving, in a modular architecture dedicated to the analysis and exploitation of formal behavioral models for test generation purposes.
Test Generation from Scenarii
In the context of the RNTL POSE project (http://www.rntl-pose.info ), the team has developed and experimented a language describing test scenarii. Basically, a scenario is a regular expression describing sequences of operations calls (without specifying their possible parameters) along with intermediate states that have to be reached. Each scenario is unfolded and played using a symbolic animation engine, that instantiates the sequence. This approach has been experimented on the IAS case study of Gemalto, and also applied on a model of the POSIX standard  .
In addition, we have defined conformance relationships dedicated to establishing a verdict when testing the correct implementation of security policies (namely access control policies) in smart cards applications  ,  . These conformance relationships are variants of input-output conformance and are based on the inclusion of traces of the implementation w.r.t. traces computed on a security-dedicated model, involving possible mappings between the values of these two levels.
We are also beginning experiments on the combination of random- and model-based testing. A first attempt has been done to automatically produce LTL formula using uniform random test generation. More recently, an approach has considered the automated generation of automata in order to evaluate various FSM-based test generation algorithms. A major result is the highlighting of an error in a widely-spread implementation of the chinese postman algorithm. We also proposed a test generation technique, driven by a final number of test cases, and combining random testing and model-based testing. It consists in arbitrarily augmenting a FSM in order to reach a given number of test cases when selected FSM-based test generation algorithms are applied. A realistic experiment has illustrated the efficiency of this approach. These works are summarized in  .