| Description |
In computer science, functional software testing is a method of ensuring that software gives expected output on specific inputs. Software testing is conducted to ensure desired levels of quality in light of uncertainty resulting from the complexity of software. Most of today's software is written by people and software development is a creative activity. However, due to the complexity of computer systems and software development processes, this activity leads to a mismatch between the expected software functionality and the implemented one. If not addressed in a timely and proper manner, this mismatch can cause serious consequences to users of the software, such as security and privacy breaches, financial loss, and adversarial human health issues. Because of manual effort, software testing is costly. Software testing that is performed without human intervention is automatic software testing and it is one way of addressing the issue. In this work, we build upon and extend several techniques for automatic software testing. The techniques do not require any guidance from the user. Goals that are achieved with the techniques are checking for yet unknown errors, automatically testing object-oriented software, and detecting malicious software. To meet these goals, we explored several techniques and related challenges: automatic test case generation, runtime verification, dynamic symbolic execution, and the type and size of test inputs for efficient detection of malicious software via machine learning. Our work targets software written in the Java programming language, though the techniques are general and applicable to other languages. We performed an extensive evaluation on freely available Java software projects, a flight collision avoidance system, and thousands of applications for the Android operating system. Evaluation results show to what extent dynamic symbolic execution is applicable in testing object-oriented software, they show correctness of the flight system on millions of automatically customized and generated test cases, and they show that simple and relatively small inputs in random testing can lead to effective malicious software detection. |
