simplefortran - demo package for calling C code with .C()

simplefortran is a small demo package showing how Fortran code could be included in a package and called with .Fortran()

This is one of a series of small demo packages for
calling other languages from R:

• {simplec} - calling C with .C()
• {simplecall} - calling C with .Call()
• {simplercpp} - calling C++ with {Rcpp}
• {simplefortran} - calling Fortran with .Fortran()

Rough comparison of .C(), .Call(), {Rcpp} (and .Fortran())

	.C()	.Call()	Rcpp
Overview	No real understanding of R objects	Need to understand SEXP macros & internals	C++ classes hide the complexity of the SEXP internals
What code?	Basic C code. Numeric calcs.	Complex C code. Can manipulate R objects from C	Complex C and C++ code involving numerics and R objects
Pros	Simple to understand and use	Simple. No unnecessary copying.	Great documentation. Wrapping of R objects very intuitive.
Cons	Too simple for most interesting things	Need to understand SEXP & R internals
Cons	Performs copying of data to call functions
Demo R package	{simplec}	{simplecall}	{simplercpp}
Compiled size	17 kB	17 kB	92 kB (stripping can bring this down: see issue1)

	.Fortran()
Overview	No real understanding of R objects
What code?	Basic Fortran code. Numeric calcs.
Pros	Simple to understand and use
Cons	Too simple for most interesting things
Cons	Performs copying of data to call functions
Cons	Need to know Fortran!
Demo R package	{simplefortran}
Compiled size	17 kB

Installation

You can install from GitHub with:

# install.package('remotes')
remotes::install_github('coolbutuseless/simplefortran)

What’s in the box?

Package contains 2 Fortran functions, and 2 functions in R which call the Fortran functions using .Fortran().

Fortran function	R function
subroutine add_(x, y, answer)	add_Fortran(x, y)
subroutine mul_(x, y, answer)	mul_Fortran(x, y)

What’s in the R functions?

• A call using .Fortran()
• First argument is the Fortran function name e.g. add_
• x and y arguments are passed through from the R function
• the third argument to the function is space in which to store the result i.e. numeric(1)
• .Fortran() returns as many arguments as you gave it originally
• We only want to return the actual result, which is the third argument, hence [[3]]
• @useDynLib [packagename] [Fortran function name] is used to generate the NAMESPACE entry useDynLib(simplefortran, add_)

#~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
#' Add two numbers
#'
#' @param x,y numbers to add
#'
#' @useDynLib simplefortran add_
#' @export
#~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
add_Fortran <- function(x, y) {
  .Fortran(add_, x, y, numeric(1))[[3]]
}

What’s in the Fortran functions?

• Actual result must be stored in one of the arguments

        subroutine add_(x, y, answer)
c
c add 2 numbers
c
        double precision x, y, answer

        answer = x + y

        end

How do R variables map to Fortran variables?

• See R manual on writing Extensions for more info

R	Fortran
logical	INTEGER
integer	INTEGER
double	DOUBLE PRECISION
complex	DOUBLE COMPLEX
character	CHARACTER(255)
raw	(none)

What does the Fortran function look like in R?

• An object of class NativeSymbolInfo
• Holds an externalptr to the loaded function

simplefortran:::add_
#> $name
#> [1] "add_"
#> 
#> $address
#> <pointer: 0x1055e3ec0>
#> attr(,"class")
#> [1] "NativeSymbol"
#> 
#> $dll
#> NULL
#> 
#> attr(,"class")
#> [1] "NativeSymbolInfo"

Resources

• Peter Dalgaard’s Keynote from UseR 2004 discusses R’s language interfaces
• Using R — Calling C Code ‘Hello World’
• Hadley’s Advanced R book chapter - ‘Rs C interface’
• Rcpp

Acknowledgements

• R Core for developing and maintaining such a wonderful language.
• CRAN maintainers, for patiently shepherding packages onto CRAN and maintaining the repository