Precise Identification Of Problems For Structural Test Generation
Xusheng Xiao,Tao Xie,Nikolai Tillmann,Jonathan De Halleux
DOI: https://doi.org/10.1145/1985793.1985876
2011-01-01
Abstract:An important goal of software testing is to achieve at least high structural coverage. To reduce the manual efforts of producing such high-covering test inputs, testers or developers can employ tools built based on automated structural test-generation approaches. Although these tools can easily achieve high structural coverage for simple programs, when they are applied on complex programs in practice, these tools face various problems, such as (1) the external-method-call problem (EMCP), where tools cannot deal with method calls to external libraries; (2) the object-creation problem (OCP), where tools fails to generate method-call sequences to produce desirable object states. Since these tools currently could not be powerful enough to deal with these problems in testing complex programs in practice, we propose cooperative developer testing, where developers provide guidance to help tools achieve higher structural coverage. To reduce the efforts of developers in providing guidance to tools, in this paper, we propose a novel approach, called Covana, which precisely identifies and reports problems that prevent the tools from achieving high structural coverage primarily by determining whether branch statements containing not-covered branches have data dependencies on problem candidates. We provide two techniques to instantiate Covana to identify EMCPs and OCPs. Finally, we conduct evaluations on two open source projects to show the effectiveness of Covana in identifying EMCPs and OCPs.