Metrics-Based Code Smell Detection in Highly Configurable Software Systems

A code smell is a symptom in computer programming that may indicate design flaws or code decay within a software system. As such, much research has been conducted regarding their detection and impact on understandability and changeability of source code. Current methods, however, can not be applied to highly configurable software systems, that is, variable systems that can be configured to fit a wide range of requirements or platforms. This variability is often based on conditional compilation implemented using C preprocessor annotations (#ifdef). These annotations directly interact with the host language (e.g., C) and therefore may have negative effects on understandability and changeability of the source code. In this case, we refer to them as variability-aware code smells. In this thesis, we propose 1) a concept of how to detect such variability-aware code smells with a metrics-based approach, and 2) a code smell detector, called Skunk, that employs this concept. In our evaluation, we use the detector on 7 open-source system of medium size to find potential variability-aware code smells samples, on which we perform a manual inspection on 20 samples of each subject system. Our results show an average precision of 40.7% (Annotation Bundle) and 62.1% (Annotation File) for detecting actual code smells. For the Large Feature we used a statistical approach to decide at which point features implement an excessive amount of functionality. The results show that features with more than 1122 lines of code are an uncommonly large. Additionally, we proposed severity ratings for ranking potential code smells, which proved to be a good indicator for the smelliness of samples. Higher-rated samples are more likely to have a negative effect on understandability and changeability than low-rated ones. For each code smell, we examined frequently recurring patterns across all systems.