Raoul-Gabriel Urma

This talk aims to present current ideas about enabling the Java developer community to directly participate in the evolution of their favourite programming language.

First, we briefly introduce 4 controversial Java features as seen by the research community: 1) Generics 2) Arrays 3) Wildcards 4) Overloading. We describe their intended purpose as seen by the language designers. Next, we describe what programmers actually do with them by looking at recent empirical studies.

Second, we present a new platform (Wiggle) developed at the University of Cambridge that intends to democratise studies of language features in Java. We describe the technical architecture of the platform, the research and design decisions behind it. The audience will learn about Java source code query languages, graph databases and interesting recent development in the Java compiler. In addition, we will demonstrate a prototype in live.

Finally, we present a recent idea about classifying Java language features based on different dimensions such as typical applications and conceptual complexity. We describe how such a classification could be useful both for collaborative development, software comprehension and teaching.

The Java type system is unsafe since it allows covariant use of arrays. For example, an array of Bananas can be used where an array of Fruit is expected, given that Banana is a subclass of Fruit. This allows the following unsafe code: Banana[] bananas = new Banana[5]; Fruit[] fruits = bananas; fruits[0] = new Apple(); // runtime exception bananas[0]; // Apple? In the above example, the programmer gets back an Apple where he was expecting Bananas, causing an undesired runtime exception. We investigate the covariant use of arrays in open source Java programs to see whether programmers use this problematic feature of Java. We make use of the freely available OpenJDK, modifying its Java compiler to disallow covariant use of arrays, thus reporting any such uses as compile errors. We apply this compiler to a code corpus of 106 open source programs, including large and well-known programs such as Eclipse, JUnit and Ant. Initial results of our investigation show that covariance is rarely used outside the context of legacy code and Java libraries. This leads us to question its relevance in the Java language. In our talk, we also discuss solutions to covariance taken by other languages such as Scala and Google Dart, explain our modifications to the OpenJDK Java compiler and report the full results of our study.

Other people: Janina Voigt