SmellDSL: A domain-specific language to assist developers in specifying code smell patterns

Context:

The current literature has widely investigated code smell patterns over the years, which describe specific source code characteristics that indicate potential problems or areas for improvements. Empirical studies suggest that (i) metric-based strategies for code smell detection are not effective and overload the developers with false positives; (ii) code smell specifications are informal, ambiguous, and not supported by traditional IDEs like Eclipse platform; and (iii) the identification of code smells depends on the perception of software development teams.

Objective:

This article, therefore, proposes SmellDSL, a tool-supported domain-specific language to assist developers when specifying code smell patterns. SmellDSL benefits developers by introducing Eclipse built-in constructs that enable the specification of team-sensitive code smell patterns. Developers can write rules to specify single or composite architectural problems (e.g., Misplaced Concerns) and suggest code refactorings regarding severe architectural degradation symptoms.

Method:

We conducted an empirical study with 35 developers who specified eight code smells using SmellDSL, generating 280 evaluation scenarios.

Results:

The main results, supported by statistical tests, suggest that SmellDSL requires low effort to specify code smell patterns and promotes a high rate of correctly code smell specifications.

Conclusion:

We contribute with a domain-specific language for the specification of code smell patterns, empirical evidence on its usefulness, and draw worth-investigating research challenges by the research community.