A replicated and refined empirical study of the use of friends in C++ software

The friend mechanism is widely used in C++ software even though the potential benefits of its use are disputed and little is known about when, where and why it is employed in practice. Furthermore, there is limited empirical analysis of its impact in object-oriented software, with only one study (Counsell and Newson, 2000) reported at journal level. This paper aims to add to the empirical evidence of friendship's impact by replicating Counsell and Newson (2000)'s original study. The study's design is refined to improve the construct validity of the evaluation and a larger cohort of systems is used to improve the generalisability of the results. The findings suggest that classes involved in friendship are coupling hotspots and that there is no link between inheritance and friendship, contrary to the findings presented in Counsell and Newson (2000). The findings also suggest that the use of friends in a class is independent of the number of hidden members in a class.