An Empirical Analysis of the Impact of Comment Statements on Fault-Proneness of Small-Size Module

Abstract

Code size metrics are commonly useful in predicting fault-prone modules, and the larger module tends to be more faulty. In other words, small-size modules are considered to have lower risks of fault. However, since the majority of modules in a software are often small-size, many ``small but faulty'' modules have been found in the real world. Hence, another fault-prone module prediction method, intended for small-size module, is also required. Such a new method for small-size module should use metrics other than code size since all modules are small size. This paper focuses on ``comments'' written in the source code from a novel perspective of size-independent metrics, comments have not been drawn much attention in the field of fault-prone module prediction. The empirical study collects 11,512 small-size modules, whose LOC are less than the median, from three major open source software, and analyzes the relationship between the lines of comments and the fault-proneness in the set of small-size modules. The empirical results show the followings: 1) A module in which some comments are written is more likely to be faulty than non-commented ones, the fault rate of commented modules is about 1.8-3.5 times higher than that of non-commented ones. 2) Writing one to four lines of comments would be thresholds of the above tendency.