How Fast and Effectively Can Code Change History Enrich Stack Overflow?

Abstract

Stack Overflow (SO) is one of the most popular Q&A sites for not only providing valuable information to software developers but also encouraging the sharing of knowledge and problem solving. Unfortunately, the information provided by SO is not always sufficient for developers. In this paper, we empirically show how fast and effectively historical code changes can substitute for missing or unanswered SO articles. Developers in all around the world encounter many problems daily and their trial-and-error experiences to resolve the problems are accumulated in the code change history. The extracted source code differences are expected to provide valuable information to developers before the questions and answers are posted on SO. In our study, we focus on the usage of APIs as the topic of SO articles, because many developers are interested in API programming and suffer from the problems related to API usage. We extracted 4,780 code differences from 713 repositories of Android applications (F-Droid). As a result, we found that 64% of SO articles on Android APIs are related to code differences, whereas 44% of code differences are related to SO articles. Not a few code differences appear before the corresponding SO articles are actually posted. The median of time lag between the first appearance of code changes and the first actual SO postings is 22 months.