An Empirical Study to Evaluate Resource Usage Caused by Android Code Smells
Percom-2018

To analyze the positive or negative impacts from applying the refactoring technique on mobile devices, we use a Moto G XT1032 smartphone. To minimize the risk of biasing in our study, we tested each application five times continuously, and we use the arithmetic mean of the achieved results to evaluate the impact from refactoring technique. Five steps were performed by our study to analyze the adopted refactoring techniques. These steps are illustrated in figure below and described the follows.

1) Code smells detections - We run the JDeodorant tool to detect three types of code smells: God Class, God Method, and Feature Envy. The nine applications analyzed were obtained from our data set.

2) Running applications with code smells in Android - From the data obtained in the previous step, we run each application individually, five times on the Moto G model XT1032 smartphone. With the termination of the five tests per application, we obtained the results of memory use and CPU consumption through the average of the five executions in each application

3) Refactoring using JDeodorant- From the conclusion of the previous steps, we can obtain the consumption of the features of the smartphone, such as memory and CPU. Therefore, the objective of this paper is to evaluate the positive or negative impacts of adopting the refactoring technique in mobile applications.

4) Run of the Apps after applying the refactoring technique - After us applying the automated refactoring technique through the JDeodorant tool was possible to identify if improved or deteriorated the consumption of memory and CPU.

5) Result analysis - At the end of all the previous steps, we obtained the data regarding the consumption of resources in Android that was submitted to refactoring technique. With these results, it is possible to compare the feasibility of this technique in mobile devices

To investigate the impacts caused by Android smells and refactorings technique, we chose only apps from the context-aware domain for several reasons: First, these apps, in general, are intuitive and easy to evaluate; Second, there is a lot of apps available for download on GitHub. Third, these apps are strongly used since context-aware is a well-defined domain, the authors of this work believed that it would be easy to find code smells from this domain application; Fourth, one of the characteristics of this type of app is the need to use various hardware resources to identify the context, for instance, GPS and movement sensors.