Project : vertecs
Section: New Results
Test generation on enumerative and symbolic models
Test generation based on coverage directives
Our test generation techniques were previously based on a selection by test purposes. However, this approach necessitates to specify those test purposes. Users sometimes want more automatic ways to generate test cases, from more general selection mechanisms. In the context of the Agedis european project (see 7.1), we have defined more general selection mechanisms, called test selection directives. They allow to describe both coverage directives (on states, transitions and more generally expressions on variables), test purposes (extended to more general observers) and constraints on data values, and to combine them. Taking into account these test directives involved a deep modification in some test generation algorithms. We also designed algorithms that generate test cases randomly, without any test directive. All these algorithms are incremental in the sense that they produce test cases when these are computed, without waiting the end of the process, thus allowing users to interupt the process with a partial result. These results are still unpublished.
From Safety Verification to Safety Testing
In this work, we define a methodology that combines verification and conformance testing for validating safety requirements of reactive systems. The safety requirements, specified as observers of visible behaviors, are first automatically verified on the system's specification. Then, test cases are automatically derived from the specification and the requirements, and executed on a black-box implementation of the system. This allows to check whether the requirements hold on the implementation as well. It is shown that an implementation conforms to its specification (for the conformance relation of Tretmans) if and only if it satisfies all the relevant safety requirements that are satisfied by the specification. The main differences with our previous works on test generation from test purposes is that test purposes express reachability properties, while requirements express safety properties, which are the most used type of property used in verification, and our methodology establishes a direct link between verification and test generation as what is tested on the implementation is exactly what is verified on the specification. This work will be presented in Testcom in 2004.
Symbolic test generation
The PhD. thesis of Elena Zinovieva describes the integration of the approximated reachability and co-reachability algorithms embodied into the NBac tool, into the general symbolic test generation algorithm of the STG tool. The new reachability/co-reachability algorithms allow for a precise handling of data, and as a result, the new version of the tool is able to generate better test cases (i.e., having less Inconclusive verdicts).