A Brief History of Software Complexity Metrics

Early complexity metrics were based almost entirely on the amount of code produced. Therefore, a function which contained 10 lines of code was considered more complex than one which contained only 5. However, since lines of code often varies due to stylistic differences, Halstead developed a set of measures based on the number of operators and operands present within the code (Kearney et ah, 1986). 

E is the number of edges, N is the number of nodes, and T is the number of terminating nodes (entry and exit points - usually 2). 

Cyclomatic complexity is extremely useful in determining how to test software. The cyclomatic complexity of a program is also the minimum number of tests needed to cover every control flow branch of a program. If the cyclomatic complexity of a program is 5, then at least 5 tests must be devised to test every branch of code. Such tests do not guarantee total coverage of all possible test conditions, but they will verify that every statement in the program will contribute to at least one test. 

The ABC metric is a simplified metric combining aspects of both lines of code and cyclomatic complexity. It works by simply cormting the number of assignment statements, the number of branches (direct flow control shifts - i.e., function calls), and the number of conditionals (Fitzpatrick, 1997, pp. 2-3). These can then be analyzed on a whole program, per-module, or per-function basis, to give an overview of complexity and size of a software program. 

---