Leveraging text analytics to improve software testing

软件测试是确保软件正确性并在软件发布前发现尽可能多的错误的常见做法。软件测试的一个基本挑战是执行路径的数量随着程序大小的增加呈指数级增长，并且可以是无限的。对于大多数现实世界的程序，测试所有执行路径是不可能的。在实践中，人们需要使用搜索启发式来确定执行路径的优先顺序，通过组合路径来折衷准确性，并使用其他技术来测试程序。有效的程序输入通常需要遵循某些约束。关注有效或接近有效的输入(对于边界情况)可以帮助测试程序的核心功能，从而提高测试覆盖率和有效性。传统的软件测试技术不知道输入约束，因此不知道哪些路径会导致有效输入的处理。该项目将通过使用文本分析技术自动从文档中提取输入约束。然后，它将使用这些输入约束来支持在软件测试期间执行程序核心功能的执行路径。使用无效输入进行测试也是非常重要的，例如，检查错误处理代码或发现由于输入格式错误而导致的缺陷。因此，还将探索有效地否定测试错误处理代码的约束的策略。无论关注什么输入(有效或无效)，如果不知道哪些输入是有效的，哪些是无效的，就不能这样做。所提出的方法通过自动从软件文档中提取输入约束来实现这种选择。此外，还将设计新的方法来分析其他软件文本，以指导软件测试将重点放在更有可能包含缺陷的代码区域。这些方法将增强自动测试生成和错误检测的有效性。

Software testing is the common practice to ensure software correctness and discover as many bugs before software release. A fundamental challenge of software testing is that the number of execution paths grow exponentially as the program size increases and can be infinite. For most real-world programs, it is impossible to test all execution paths. In practice, one needs to use search heuristics to prioritize execution paths, compromise accuracy by combining paths, and use other techniques to test programs. Valid program inputs typically need to follow certain constraints. Focusing on valid or close-to-valid inputs (for boundary cases) can help test the core functionalities of the program, which improves testing coverage and effectiveness. Traditional software testing techniques are unaware of input constraints, and thus unaware of which paths lead to the processing of valid inputs. This project will automatically extract input constraints from documents by using text analytics techniques. It will then use these input constraints to favour execution paths that execute a program's core functionalities during software testing. Testing with invalid inputs is also very important, e.g., to check error-handling code or find defects due to malformed inputs. Therefore, strategies to negate the constraints efficiently for testing error-handling code will also be explored. Regardless of what inputs (valid or invalid) to focus on, one cannot do so without knowing what inputs are valid and what are not. The proposed approach enables this choice by extracting input constraints from software documents automatically. In addition, new approaches will be designed to analyze other software text to guide software testing to focus on code regions that are more likely to contain defects. These approaches will enhance the effectiveness of automatic test generation and bug detection.