Analyzing Measurements of the R Statistical Open Source Software

Abstract

Software quality is one of the main goals of effective programming. Although it has a quite ambiguous meaning, quality can be measured by several metrics, which have been appropriately formulated through the years. Software measurement is a particularly important procedure, as it provides meaningful information about the software artifact. This procedure is even more emerging when we refer to open source software, where the need for shared knowledge is crucial for the maintenance and evolution of the code. A paradigm of open source project where code quality is especially important is the scientific language R. This paper aims to perform measurements on the R statistical open source software, examine the relationships among the observed metrics and special attributes of the R software and search for certain characteristics that define its behavior and structure. For this purpose, a random sample of 508 R packages has been downloaded from the CRAN repository of R and has been measured, using the SourceMonitor metrics tool. The resulted measurements, along with a significant number of specific attributes of the R packages, were examined and analyzed, leading to interesting conclusions such as the validity of a power law distribution regarding the majority of the sample's metrics and the absence of specific patterns due to the interdependencies among packages. Finally, the effects of the number of developers and the number of dependencies are investigated, in order to understand their impact on the metrics of the sample packages.