Java Magazine - May/June 2012

//java architect /

Conclusion

We conducted an empirical study to investigate the use of covariant arrays in real-world Java programs. Our corpus was composed of 64 open source programs from the Qualitas Corpus, ranging from games to system software and libraries. As we have shown with the help of an example, covariant arrays in Java can lead to type errors that are not caught by the compiler but cause runtime exceptions.

In our empirical study, we found relatively few uses of covariance given the large size of our corpus: 324 covariant array uses in more than 5 million lines of code. More importantly, almost 75 percent of these uses were caused by calls to Java library methods. This demonstrates that covariant arrays are used very infrequently by Java developers in practice.

Since Java SE 5, generics and wildcards provide a type-safe alternative to covariant arrays. All of the uses of covariant arrays we found in our corpus could be transformed and made completely type safe using these Java features.

The existence of generics as a better and safer solution and the infrequent use of covariant arrays in Java lead us to suggest that covariant arrays should not be included in future programming languages and should be avoided whenever possible in Java development.

We intend to extend our work by investigating the use of wildcards in

Figure 3

DefaultTableModel Arrays.toString JTable() JList.setListData Arrays.equals Arrays.binarySearch JList() JComboBox() Arrays.sort Vector.copyInto DefaultComboBoxMode JOptionPane (Vector/List).toArray 34 17 15 8 6 4 3 3 2 2 1 1

146

COMMUNITY JAVA IN ACTION

Java programs. Wildcards and generics provide a better alternative to covariant arrays, and we are interested to see how frequently they are used in practice. </article>

Register Now

We would like to thank Alan Mycroft and Luke Church for their encouragement and useful advice and suggestions during our work. We would like to thank our wider research group, the Cambridge Programming Research Group, as well as attendees of our talk at the FOSDEM conference for valuable feedback and suggestions. Further, Raoul-Gabriel Urma wishes to thank Qualcomm and Janina Voigt wishes to thank Rutherford Foundation of the Royal Society of New Zealand for supporting this work through their studentships.