Degree Type

Dissertation

Date of Award

2010

Degree Name

Doctor of Philosophy

Department

Computer Science

First Advisor

Carl K. Chang

Abstract

Testing object-oriented software is critical because object-oriented languages have been commonly used in developing modern software systems. Many efficient test input generation techniques for object-oriented software have been proposed; however, state-of-the-art algorithms yield very low code coverage (e.g., less than 50%) on large-scale software. Therefore, one important and yet challenging problem is to generate desirable input objects for receivers and arguments that can achieve high code coverage (such as branch coverage) or help reveal bugs. Desirable objects help tests exercise the new parts of the code. However, generating desirable objects has been a significant challenge for automated test input generation tools, partly because the search space for such desirable objects is huge.

To address this significant challenge, we propose a novel approach called Capture-based Automated Test Input Generation for Objected-Oriented Unit Testing (CAPTIG). The contributions of this proposed research are the following.

First, CAPTIG enhances method-sequence generation techniques. Our approach intro-duces a set of new algorithms for guided input and method selection that increase code coverage. In addition, CAPTIG efficently reduces the amount of generated input.

Second, CAPTIG captures objects dynamically from program execution during either system testing or real use. These captured inputs can support existing automated test input generation tools, such as a random testing tool called Randoop, to achieve higher code coverage.

Third, CAPTIG statically analyzes the observed branches that had not been covered and attempts to exercise them by mutating existing inputs, based on the weakest precon-dition analysis. This technique also contributes to achieve higher code coverage.

Fourth, CAPTIG can be used to reproduce software crashes, based on crash stack trace. This feature can considerably reduce cost for analyzing and removing causes of the crashes.

In addition, each CAPTIG technique can be independently applied to leverage existing testing techniques. We anticipate our approach can achieve higher code coverage with a reduced duration of time with smaller amount of test input. To evaluate this new approach, we performed experiments with well-known large-scale open-source software and discovered our approach can help achieve higher code coverage with fewer amounts of time and test inputs.

Copyright Owner

Hojun Jaygarl

Language

en

Date Available

2012-04-30

File Format

application/pdf

File Size

150 pages

Share

COinS