Writing R extensions

Overview

• R makes it easy to use legacy C and Fortran code
• comes with beautiful graphics and powerful statistical tools

• 10000 packages on CRAN

• packages follow high standards for documentation
• think of R as python with built-in matplotlib and pandas

Take this simple Fortran subroutine:

      subroutine facto(n, answer)
C simple routine to compute factorial
C return result by reference in variable answer
      integer n, answer, i

      answer = 1
      do 100 i = 2,n
         answer = answer * i
 100  continue
      end

Build a shared library:

$ R CMD SHARED facto.f
$ ls
facto.f facto.o facto.so

We can now use this R-interface to call our routine:

dyn.load("facto.so")
facto <- function(n) {
  foo <- .Fortran("facto", n=as.integer(n), answer=integer(1))
  # returns a list with components n and answer
  foo$answer
}
m <- facto(5)
print(m)

or if you would hide this in a package called facto:

library(facto)
m <- facto(5)
print(m)

Example: RADEX

During the tech talk I'll be demonstrating how to build an R-interface to RADEX, a publically available Fortran code for radiative transfer. The legacy code is here.

Idea:

• replace main function with R wrapper
• remove all I/O from original program
• provide input via R variables
• return result as an R data frame

Extra:

• use R routines for warnings, errors, debug statements
• optional: replace linear algebra code with R built-in routines
• turn into R package
• add an interface to read a molecular data file needed by RADEX, returning e.g. the molecular levels and transitions as R data frames.

The final package is available on github