Bad smell detection using quality metrics and refactoring opportunities

Abstract

Bad smells are bad practices in developing software. These poor solutions significantly influence the understandability and maintainability of source code. Therefore, bad smell detection plays a vital role in the refactoring, maintaining, and measuring the quality of large and complex software systems. Researchers believe that bad smells should be precisely identified and addressed. However, bad smell detection is complicated by issues such as informal and inconsistent specifications of bad smells and high false positive rates in the detection process, all of which affect the success rate in detection. In this paper, we present a new method to detect bad smells in code by addressing the aforementioned issues. Our proposed method is a multi‐step process using software quality metrics and refactoring opportunities. In this method, after obtaining the bad smell formal specifications based on software metrics, we utilize them to achieve a set of candidates for each bad smell. Afterwards, each of the instances will be examined and compared with the corresponding refactoring situations specified for that bad smell. This examination strikes out the false positives created in the previous step. The evaluation of this method on four open‐source systems demonstrates the improved effectiveness of bad smell detection in code.

Towards a systematic approach to manual annotation of code smells

“…The cyclomatic complexity structural metric counts branches, loops, and several other code constructs to determine a method's complexity. This metric is used in several smell detectors [17] to determine the Long Method smell, where a complexity above 5 signals the presence of this smell.…”

“…These rules can vary greatly depending on what the average LOC for the project is. For projects with inexperienced engineers, the average can quickly go above the conventional recommendations from the industry and other research, where functions with LOC above 30 [3][4] [5] and classes with LOC above 100 [17] might be affected by a smell. Furthermore, even functions with LOC above 10 can impose a readability issue if the code is sufficiently complex [4] [5].…”

“…Furthermore, the studies either purposefully avoid training the annotators or list a short workshop. This preparation might be problematic, as manual code smell detection is subjective [15], produces high disagreement among experienced software engineers [3] [9], and significantly differs between the scientific literature [17] and industry best practices [4] [5]. Another common limitation is the lack of a usable and easily accessible dataset.…”

“…• Studies of automated smell detection. We examined studies gathered in the systematic literature review on the topic [17] to extract the domain knowledge researchers strived to encode in their algorithms. • Studies that examine engineer perception of code smells.…”

“…Likewise, engineers use the length of the class to determine the presence of the Large Class smell [3] [21]. A recent literature review [17] examined rule-engines and other code smell detectors. For Large Class, they found that the most relaxed rule (with the highest number of true and false positives) checked if the class had more than 100 lines of code.…”