An Empirical Study of Bugs in Software Build System

抄録

A build system converts source code, libraries and other data into executable programs by orchestrating the execution of compilers and other tools. The whole building process is managed by a software build system, such as Make, Ant, CMake, Maven, Scons, and QMake. Many studies have investigated bugs and fixes in several systems, but to our best knowledge, none focused on bugs in build systems. One significant feature of software build systems is that they should work on various platforms, i.e., various operating systems (e.g., Windows, Linux), various development environments (e.g., Eclipse, Visual Studio), and various programming languages (e.g., C, C++, Java, C#), so the study of software build systems deserves special consideration. In this paper, we perform an empirical study on bugs in software build systems. We analyze four software build systems, Ant, Maven, CMake and QMake, which are four typical and widely-used software build systems, and can be used to build Java, C, C++ systems. We investigate their bug database and code repositories, randomly sample a set of bug reports and their fixes (800 bugs reports totally, and 199, 250, 200, and 151 bug reports for Ant, Maven, CMake and QMake, respectively), and manually assign them into various categories. We find that 21.35% of the bugs belong to the external interface category, 18.23% of the bugs belong to the logic category, and 12.86% of the bugs belong to the configuration category. We also investigate the relationship between bug categories and bug severities, bug fixing time, and number of bug comments.