Traits, Generics, and modern-day OO for Fortran

[FortranFan]

@certik

in OO-based generics you would do:

implements (INumeric,IPrintable) :: real
  procedure :: add, cast, output
end implements real

contains

function add(self,other) result(res)
real, intent(in) :: self, other
real :: res
res = self + other
end function add
...

    s_i = simple_sum(a_i)

@kkifonidis , @certik et al. and all the practitioners of Fortran,

Being able to author and consume a generic procedure like SUM intrinsic in as close to a manner as the intrinsic shall be a starting, minimum requirement for any Generics enhancement to the Fortran language.

This should require minimal “gymnastics” on part of the program author. Fortran language has achieved this in an entirely another context with coarrays i.e., PGAS-based SPMD parallel computing. Syntactically [ … ] yielded the magic for the array language that is Fortran, the rest is mostly semantics for the processor implementation.

Simplicity like with coarrays is what is deserved by the practitioners of Fortran.

This is the ideal in any work on Generics in Fortran that the language developers should seek inspiration and imagination. The current pursuit with templates by J3 Generics subgroup fails in this respect and which is why a pause or a change in direction is being petitioned with WG5.

Traits with OO too comes across as way too much “gymnastics”, plus it does not solve the basic use case for a practitioner of Fortran i.e., to be able to author a generic procedure similar to SUM, FINDLOC, etc.

Hence the strong recommendation to go back to the drawing board.