The connection between antipatterns and maintainability in Firefox

Abstract

The notion that antipatterns have a detrimental effect on source code maintainability is widely accepted, but there is relatively little objective evidence to support it. We seek to investigate this issue by analyzing the connection between antipatterns and maintainability in an empirical study of Firefox, an open source browser application developed in C++. After extracting antipattern instances and maintainability information from 45 revisions, we looked for correlations to uncover a connection between the two concepts. We found statistically significant negative values for both Pearson and Spearman correlations, most of which were under -0.65. These values suggest there are strong, inverse relationships, thereby supporting our initial assumption that the more antipatterns the source code contains, the harder it is to maintain. Lastly, we combined these data into a table applicable for machine learning experiments, which we conducted using Weka [10] and several of its classifier algorithms. All five regression types we tried had correlation coefficients over 0.77 and used mostly negative weights for the antipattern predictors in the models we constructed. In conclusion, we can say that this empirical study is another step towards objectively demonstrating that antipatterns have an adverse effect on software maintainability.