M. Halstead's Software Science is the origin of some program metrics. Among them are well-known as well as new measures. Their evaluation is commonly used a counting strategy, i.e. a definition of Halstead's tokens of a programming language. We study the differences between the two strategies for Ada programming language, one base on compiler oriented approach (1987) and the other base on human oriented approach (2007). Analytical consideration and testing both shows the first strategy gives higher values of program vocabulary (excess is usually moderate). It is more difficult to compare the respective estimations of program length. Under certain conditions, they approximately equal. Despite the different approaches, the recent counting strategy for Ada 95 (2004-07) seems sufficiently coherent to the previously tested strategy for Ada 83 (1987).
