Named argument feature

I find named argument a great feature of Fortran. It really makes a difference when working with functions/subroutines with many arguments. By contrast, C and C++ don’t have it. Surprisingly, one of my friends who has used Fortran for more than 30 years doesn’t know this feature.


I tweeted about positional arguments.

You can force the user to use positional arguments by making the first argument optional with a type that is private, as shown below. I wonder if I should tweet about this or if it’s overly tricky and bad style.

module m
implicit none
public :: sd
type :: phony
end type phony
function sd(phony_arg,x,mean) result(xsd)
type(phony), intent(in), optional :: phony_arg
real       , intent(in)           :: x(:)
real       , intent(in)           :: mean
real                              :: xsd
integer                           :: n
n = size(x)
xsd = sqrt(sum((x-mean)**2)/(n-1))
end function sd
end module m
program main
use m
implicit none
integer, parameter :: n = 10**6
real :: x(n)
call random_number(x)
end program main

The issue with the phony argument is that it might not be immediately evident what the caller did wrong from the compiler error message.

Recently, I bumped into an issue where I had to use named arguments. Consider the subroutine

    subroutine dl_optimize_dense(optimum, p, nmeas, f, fparams, ctx)
        real(wp), intent(out) :: optimum
        real(wp), intent(inout), contiguous :: p(:)
        integer, intent(in) :: nmeas
        procedure(dl_callback_dense) :: f
        class(*), intent(in), target, optional :: fparams
        type(dl_solver_context_t), intent(out), optional :: ctx

Since fparams is an unlimited polymorphic object you can pass anything in its place. It’s easy to mistakenly pass the ctx object in place of fparams. The solution is to use a named argument:

call dl_optimize_dense(opt, p, nmeas, f, &
    ctx = context)

I also like named arguments very much :drooling_face: Re C++, there seem to be many proposals up to now but rejected for various reasons…

These pages show some “workaround(s)”, but it seems a bit awkward to me… (compared to usual named arguments)

Recent discussion in C++ community: