Abstract

Functional programming appears to be enjoying a renaissance of interest for developing practical, ―real-world ‖ applications. Proponents have long maintained that the functional style is a better way to modularize programs and reduce  complexity. What is new in this paper is we test this claim by studying the complexity of open source codes written in Scala, a modern language that unifies functional and object programming. We downloaded from GitHub, Inc., a portfolio of mostly ―trending‖ Scala repositories that included the Scala compiler and standard library, much of them written in Scala;the Twitter, Inc., server and its support libraries; and many other repositories, several of them production-oriented and commercially inspired. In total we investigated approximately 22,000 source files with 2 millions lines of code and 223,000 methods written by hundreds of programmers. To analyze these sources, we developed a novel compiler kit that measures lines of code and adaptively learns to estimate the cyclomatic complexity of functional-object codes. The data show, first, lines of code and cyclomatic complexity are positively correlated as we expected but only weakly which we did not expect with Kendall’s τ=0.258–
0.274. Second, 75% of the Scala methods are straight-line, that is, they have the lowest possible cyclomatic complexity. Third, nearly 70% of methods have three or fewer lines. Fourth, the distributions of lines of code and cyclomatic complexity are both non-Gaussian (P<0.01), which is as surprising as it is interesting. These data may offer new insights into software complexity and the large-scale structure of applications including but not necessarily limited to Scala.