Calling .Fortran from R with helper functions

Question

I have a two-part question that might brush on preferred habits, and whether I'm unnecessarily complicating a problem. I've been using .C and learned .Call to do the heavy lifting from R. I wanted to do a comparison with .Fortran, and so I've been learning it as I go. Therefore, my issue may be due to my limited Fortran knowledge.

There are several functions implemented in R that I'd like to translate to Fortran that are related to each other. Rather than hash out the details of the problem I've created a minimum synonymous example that's impractical but a good illustration.

I have two functions and two related subroutines to find the magnitude and squared-magnitude, which make the entirety of the file:

! Actual functions
  real function sumofsquaresfun (x,y)
    implicit none
    real :: x,y
    sumofsquaresfun=x*x+y*y
  end function sumofsquaresfun


  real function magnitudefun(x,y)
    implicit none
    real :: x,y,tmp
    tmp=sumofsquaresfun(x,y)
    magnitudefun=sqrt(tmp)
  end function magnitudefun

! Wrappers for R accessibility since R requires subroutines
  subroutine sumofsquares(x,y,answer)
    implicit none
    real :: x,y
    real answer
    answer=sumofsquaresfun(x,y)
  end subroutine sumofsquares

  subroutine magnitude(x,y,answer)
    implicit none
    real :: x,y
    real answer
    answer=magnitudefun(x,y)
  end subroutine magnitude

Suppose that both subroutines are useful when I'm in R, so keeping the two separate is important.

When I try t compile using

R CMD SHLIB fortrancode.f95

I have several errors the same two types of errors:

Error: Return type mismatch of function 'sumofsquaresfun' at (1) (UNKNOWN/REAL(4))

Error: Function 'sumofsquaresfun' at (1) has no IMPLICIT type

etc.

My questions in order of importance:

1. Even if feasible, is this the preferred approach to such a problem? Should I be using modules instead? I've seen this discussion (using a Fortran module in R?) and the related one by the same OP but don't know if involving modules is necessary/optimal.
2. If what I am trying to do is the correct way, what causes these embarrassing errors?
3. I like to use functions because that's what they're meant for, but am I being too much of a purist by creating the subroutine wrappers? Is it better style to just do away with them so I only have the two subroutines?

Many thanks!

Answer 1

This is more of a Fortran question than an R question. You haven't declared the functions in your subroutines. Do it like this and it should compile.

! Actual functions
 real function sumofsquaresfun (x,y)
   implicit none
   real :: x,y
 sumofsquaresfun=x*x+y*y
 end function sumofsquaresfun


real function magnitudefun(x,y)
  implicit none
  real :: x,y,tmp
  real sumofsquaresfun
  tmp=sumofsquaresfun(x,y)
  magnitudefun=sqrt(tmp)
end function magnitudefun

! Wrappers for R accessibility since R requires subroutines
subroutine sumofsquares(x,y,answer)
  implicit none
  real :: x,y
  real answer
  real sumofsquaresfun
  answer=sumofsquaresfun(x,y)
end subroutine sumofsquares

subroutine magnitude(x,y,answer)
  implicit none
  real :: x,y
  real answer
  real magnitudefun
  answer=magnitudefun(x,y)
end subroutine magnitude

Warning: You are using real but don't forget that R uses double precision. real is single precision. You should use something like real*8 or the Fortran95 equivalent of that.