CXXR: An Ideas Hatchery for Future R Development

• Activity Number: 101
• Type: Invited
• Date/Time: Monday, August 1, 2011 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistical Computing
• Author(s): Andrew Runnalls*+
• Companies: University of Kent
• Keywords: R ; C++ ; refactoring ; computing ; object-oriented

Abstract

The continued growth of CRAN is testament to the increasing number of developers engaged in R development. But far fewer researchers have experimented with the R interpreter itself. The code of the interpreter, written for the most part in C, is structured in a way that will be foreign to students brought up with object-oriented programming, and the available documentation, though giving a general understanding of how the interpreter works, does not really enable a newcomer to start modifying the code with any confidence.

The CXXR project is progressively refactoring the interpreter into C++, whilst all the time preserving existing functionality: by restructuring the code into tightly encapsulated and carefully documented classes, CXXR aims to open up the interpreter to more ready experimentation by statistical computing researchers.

This paper focusses on two example tasks: (a) providing, as a package, a new type of data vector, and (b) adding the capability to track the provenance of R objects. The paper shows how CXXR greatly facilitates these tasks by internal changes to the structure of the interpreter, and by offering a higher-level interface for packages to exploit.