Section: New Results

Software Reengineering

The Case for Non-Cohesive Packages. While the lack of cohesiveness of modules in procedural languages is a good way to identify modules with potential quality problems, we doubt that it is an adequate measure for packages in object-oriented systems. Indeed, mapping procedural metrics to object-oriented systems should take into account the building principles of object-oriented programming: inheritance and late binding. Inheritance offers the possibility to create packages by just extending classes with the necessary increment of behavior. Late binding coupled to the "Hollywood Principle" are a key to build frameworks and let the users branch their extensions in the framework. Therefore, a package extending a framework does not have to be cohesive, since it inherits the framework logic, which is encapsulated in framework packages. In such a case, the correct modularization of an extender application may imply low cohesion for some of the packages. We confirm these conjectures on various real systems (JHotdraw, Eclipse, JEdit, JFace) using or extending OO frameworks. We carry out a dependency analysis of packages to measure their relation with their framework. The results show that framework dependencies form a considerable portion of the overall package dependencies. This means that non-cohesive packages should not be considered systematically as packages of low quality. [22]

Identifying Classes in Legacy JavaScript Code. JavaScript is the most popular programming language for the Web. Although the language is prototype-based, developers can emulate class-based abstractions in JavaScript to master the increasing complexity of their applications. Identifying classes in legacy JavaScript code can support these developers at least in the following activities: (i) program comprehension; (ii) migration to the new JavaScript syntax that supports classes; and (iii) implementation of supporting tools, including IDEs with class-based views and reverse engineering tools. We propose a strategy to detect class-based abstractions in the source code of legacy JavaScript systems. We report on a large and in-depth study to understand how class emulation is employed, using a dataset of 918 JavaScript applications available on GitHub. We found that almost 70% of the JavaScript systems we study make some usage of classes. We also performed a field study with the main developers of 60 popular JavaScript systems in order to validate our findings. The overall results range from 97% to 100% for precision, from 70% to 89% for recall, and from 82% to 94% for F-score. [20]

A Critical Analysis of String APIs: The Case of Pharo. Most programming languages, besides C, provide a native abstraction for character strings, but string APIs vary widely in size, expressiveness, and subjective convenience across languages. In Pharo, while at first glance the API of the String class seems rich, it often feels cumbersome in practice; to improve its usability, we faced the challenge of assessing its design. However, we found hardly any guideline about design forces and how they structure the design space, and no comprehensive analysis of the expected string operations and their different variations. We first analyze the Pharo 4 String library, then contrast it with its Haskell, Java, Python, Ruby, and Rust counterparts. We harvest criteria to describe a string API, and reflect on features and design tensions. This analysis should help language designers in understanding the design space of strings, and will serve as a basis for a future redesign of the string library in Pharo. [19]