I think my example is a syntactically correct instance of procedure
pointer initialization but both gfortran and an old version of ifort
error out on it:

module funcs
implicit none
interface
subroutine a() bind(C,name='a1')
implicit none
end subroutine a
subroutine b() bind(C,name='a2')
implicit none
end subroutine b
end interface
end module funcs

program start
use funcs
implicit none
type has_fun
procedure(a), pointer, nopass :: fun
end type has_fun
type(has_fun) :: lots(2) = [has_fun(a),has_fun(b)]
procedure(a), pointer :: stuff => b
end program start

subroutine c() bind(C,name='a1')
implicit none
end subroutine c

subroutine d() bind(C,name='a2')
implicit none
end subroutine d

Hi James,
indeed, the code looks standard-conforming to me. The problem is the
initialization of lots, using of array of objects built by structure
constructors. gfortran and ifort complain that they don't know what
a is, so they can't resolve the interfaces from module funcs, nagfor
compiles without complaints. If you use
type(has_fun), dimension(2) :: lots
[...]
lots = [has_fun(a), has_fun(b)]

gfortran compiles this as well, while ifort 2021.5.0 produces an ICE.
You should report this to gfortran and Intel.
Cheers,
JRR

Am 22.05.22 um 09:54 schrieb James Van Buskirk:
> I think my example is a syntactically correct instance of procedure
> pointer initialization but both gfortran and an old version of ifort
> error out on it:
>
> module funcs
>   implicit none
>   interface
>      subroutine a() bind(C,name='a1')
>         implicit none
>      end subroutine a
>      subroutine b() bind(C,name='a2')
>         implicit none
>      end subroutine b
>   end interface
> end module funcs
>
> program start
>   use funcs
>   implicit none
>   type has_fun
>      procedure(a), pointer, nopass :: fun
>   end type has_fun
>   type(has_fun) :: lots(2) = [has_fun(a),has_fun(b)]
>   procedure(a), pointer :: stuff => b
> end program start
>
> subroutine c() bind(C,name='a1')
>   implicit none
> end subroutine c
>
> subroutine d() bind(C,name='a2')
>   implicit none
> end subroutine d
>

--
Juergen Reuter
Theoretical Particle Physics
Deutsches Elektronen-Synchrotron (DESY)
Hamburg, Germany