To: J3                                                     J3/24-154
From: Malcolm Cohen
Subject: Recommendation on constant expressions for generic resolution.
Date: 2024-June-26
References: 24-103.


This paper presents the /DATA subgroup recommendations on the proposed
constant expressions for generic resolution, paper 24-103.

This paper suggests some intriguing possibilities for extending generic
resolution in a manner that follows how we have done some of the intrinsic
functions, e.g. SIZE.

Subgroup considers that those intrinsic functions are not an appropriate
model for user-defined generic procedures.

The first example in the paper can be done using generic subroutines
instead of generic functions, for example,

    Generic Simple Subroutine get_size(c, cs)
      Use Iso_Fortran_Env
      Class(container),Intent(In) :: c
      Integer(int_kinds),Intent(Out) :: cs
      cs = Size(c%vals,Kind(cs))
    End Subroutine

Subgroup further notes that some intrinsic functions use a MOLD argument
rather than a KIND argument, to specify the return type and kind. Thus, if
functions are preferred over subroutines, one can already write

    Generic Integer Function size(c) Result(r)
      Class(container),Intent(In) :: c
      r = Size(c%vals)
    End Function
    Generic Function size(c, mold) Result(r)
      Class(container),Intent(In) :: c
      Integer(*),Intent(In) :: mold
      r = Size(c%vals,Size(mold))
    End Function
    ...
    Print *,size(c)
    Print *,size(c,0_int32)
    Print *,size(c,0_int64)

The paper further suggests using character string values in generic
resolution, for example
    call report_info("weather")
    call report_info("time-of-day")

This suggestion fills subgroup with horror.
Those procedures can already be written as
    call report_info_weather
    call report_info_time_of_day
which needs less typing and has been standard-conforming since Fortran 90.

Therefore, subgroup recommends that the proposal in 24-103 not be included
in Fortran 202y.

===END===