زهرا منصوري

Software testing, particularly in sensitive industrial applications, is crucial for ensuring effectiveness. Various methods have been developed to generate test data; however, none of these methods cover the entire input space because the generated test dataset may be selected from only one subdomain of the original domain and may not cover all acceptable subdomains of the program under test. Also, in many complex systems, the input domain might not fully represent the system behavior. Therefore, it is necessary to model the behavioral domain of the system considering the relationships between input parameters visible at program branches. Not using domain coverage means that many hidden program faults remain undetected. These faults, such as syntactic faults, cannot be identified and fixed by compiler tools. So far, various coverage criteria have been presented in software testing, but none of them cover the entire domain. In general, it is an open problem to identify an adequate test dataset that sufficiently captures the basic behavior of a program. The weakness of the existing coverage criteria, not only affect testing success but also cause the ineffectiveness of the main debugging step, i.e. fault localization. However, so far, no research has been presented that directly reviews the effect of domain coverage on fault localization and eva‎luates test data generation methods from this point of view. To address this gap, we introduce the EvoDomain algorithm, a dynamic method combining genetic and hill climbing algorithms for generating targeted test sets. The research aims to enhance fault localization effectiveness by developing a logic-based criterion, modified condition/decision coverage. Since fault localization is inherently linked with test data generation, in the second step, we generate a test suite based on the developed metric and eva‎luate its impact on fault localization quality. The eva‎luations demonstrate that selecting test data from the boundaries of the behavioral domain outperforms data selected based on well-known criteria, namely MC/DC and RoRG, with 74.44% and 65.06% improvements in fault detection rates, respectively. Additionally, the proposed method exhibits a 32% improvement in convergence effectiveness compared to COSMOS, a static domain analysis method.

آزمون نرم‌افزار , پوشش دامنه , آزمون مبتني بر منطق , ژنتيك , مكان‌يابي خطا

Software testing , Domain coverage , Logic-based testing , Genetic algorithms , Fault localization

Zahra Mansouri

Dr. Saeed Parsa