The Scent of a Smell: A Comparison Between Structural and Textual Code Smells

Code smells are symptoms of poor design and implementation choices. Several studies demonstrate their negative effects on maintenance cost. Despite this, developers tend to not refactor code smells. Almost all the current code smell detectors and refactoring tools are based on structural information. In this paper, we conjecture that textual information, as opposed to structural information, is better aligned with developers’ perception of code smells. In this paper, we systematically investigate how textually or structurally detected code smells are treated and perceived by developers. To this aim, we first conduct a mining study to obtain a quantitative perspective. Secondly, we conduct an industrial user study with 5 quality experts in order to qualitatively analyze how developers perceive code smells identified using different sources of information and whether textually or structurally detected code smells are easier to refactor. Results clearly indicate that textually detected smells are easier to perceive and subject to more refactoring operations.