PHP code smells in web apps: Evolution, survival and anomalies

Abstract

Context:Code smells are symptoms of poor design, leading to future problems, such as reduced maintainability. Therefore, it becomes necessary to understand their evolution and how long they stay in code. This paper presents a longitudinal study on the evolution and survival of code smells (CS) for web apps built with PHP, the most widely used server-side programming language in web development and seldom studied. Objectives:We aimed to discover how CS evolve and what is their survival/lifespan in typical PHP web apps. Does CS survival depend on their scope or app life period? Are there sudden variations (anomalies) in the density of CS through the evolution of web apps? Method:We analyzed the evolution of 18 CS in 12 PHP web applications and compared it with changes in app and team size. We characterized the distribution of CS and used survival analysis techniques to study CS’ lifespan. We specialized the survival studies into localized (specific location) and scattered CS (spanning multiple classes/methods) categories. We further split the observations for each web app into two consecutive time frames. As for the CS evolution anomalies, we standardized their detection criteria. Results:The CS density trend along the evolution of PHP web apps is mostly stable, with variations, and correlates with the developer’s numbers. We identified the smells that survived the most. CS live an average of about 37% of the life of the applications, almost 4 years on average in our study; around 61% of CS introduced are removed. Most applications have different survival times for localized and scattered CS, and localized CS have a shorter life. The CS survival time is shorter and more CS are introduced and removed in the first half of the life of the applications. We found anomalies in the evolution of 5 apps and show how a graphical representation of sudden variations found in the evolution of CS unveils the story of a development project. Conclusion:CS stay a long time in code. The removal rate is low and did not change substantially in recent years. An effort should be made to avoid this bad behavior and change the CS density trend to decrease.