5.4 Summary

It has not been an objective of this chapter to identify all possible metric primitives for program modules. The metrics defined herein are those that we have found of maximum utility for measuring the activity for program modules. These metrics represent a first step in the development of a software measurement program.

We have found that there are very many program attributes that can easily be counted, such as Comments and LOC. Enumeration of these attributes for a program will produce data. The problem with these two values is that their information content is negligible. We can learn nothing really useful from them. They are not very useful in predicting software quality attributes. On the other hand, there are other measures of program size, such as total executable statement count, Exec, that are good predictors of software quality and are not nearly so dependent on the uncontrolled source of variation of style of individual programmers.

An overriding objective of this book is to focus on the software measurement process. We would expect that this measurement process itself should come under close measurement scrutiny. It should be continually measured and enhanced. The metrics presented in this chapter are intended to be an initial working metric set for the institution of a measurement process. Any measurement process should have at its heart a process for measurement improvement. With this basic set of working metrics, we know that we can account for at least 80 percent of the total variation in software faults, for example. That is quite a bit more information than the vast majority of software managers have at their current disposal. Now we can begin the process of refinement. Careful analysis of software fault data will now tell us what we are missing in the 20 percent of variance for which we cannot account.