Dynamically Detecting Function Likely Program Invariants Based On Relational Database Theory

Along with the devolepment of computer science, softwares are widely applied in kinds of business. The quality and security of software cause more serious attention than before. In order to guarantee the quality of software, people has researched various respects such as software developing methods, management of software development process and software developing tools and made a good progress. Design by contract is a very important widely applied method of building quality software.Contracts are formal specifications that must be followed when developping software, which can be classified into four levels,grammer contract,data contract,behavior contract and service contract.Software supplier and client ensure qualify by organizing their collaboration on the basis of those contracts. Invariant is the basic fabric of data contract, which is valueable for program revolution, verification, rebuilding, testing, debugging and updating. However most software are lack of specified documents, invariants are not explicitly stated.To discovery the implicit invariants from program is essential for implementing behaviour constract and improving the program reliability.This paper focus on dynamic detecting techniques in discovering function likely invariants.the concepts and traditional approaches (daikon and Diduce) are dicussed first. After analysing the disvantages of the two techniques,a new instrument tool base on database is introduced,which is disigned and implemented byobject-oriented method.A heuristic detecting method with a high degree of flexibility,ease of use assosiated with database is proposed . In this new method, whether the expected variables are dependent with function is checked before inferential reasoning function parsing formula so that it resolves the problem of blindly detecting in tradtional method. Finally, experiments have been done to compare tradtional method and the new one which demonstrate that the latter is obviously better in detecting realibility and effeciency.