Revisiting measurement of software complexity

Abstract

Software complexity measures are often proposed as suitable indicators of different software quality attributes. The paper presents a study of some complexity measures and their correlation with the number of failure reports, and a number of problems are identified. The measures are such poor predictors that one may as well use a very simple measure. The proposed measure is supposed to be ironic to stress the need for a more scientific approach to software measurement. The objective is primarily to encourage discussions concerning methods to estimate different quality attributes. It is concluded that either completely new methods are needed to predict software quality attributes or a new view on predictions from complexity measures is needed. This is particularly crucial if the software industry is to use software metrics successfully on a broad basis.