INTRINSIC statement in module causes an error

[ur]

I have to comment out the INTRINSIC statement in the code below with ifort/ifx, but not with gfortran/nvfortran.  Should the INTRINSIC statement be allowed?

module M_overload
! allow SIGN(3f) to take a single argument
use, intrinsic :: iso_fortran_env, only : int32
implicit none
private
public :: sign
!intrinsic :: sign ! ifort 2023 bug(?)
interface sign ; module procedure sign_int32;   end interface

contains

elemental function sign_int32(value)
integer(kind=int32),intent(in) :: value
integer(kind=int32)            :: sign_int32
   sign_int32=sign(1_int32,value)
end function sign_int32

end module M_overload

program tryit
use M_overload, only : sign
implicit none
   write(*,*) merge('sign works','sign fails', sign(10).eq.1 .and. sign(-10).eq.-1 )
end program tryit

[ur]

The message I get with ifx 2023 and ifort 2023 is

xx.f90(23): error #6505: This intrinsic procedure reference contains too few arguments. [SIGN]
write(*,*) merge('sign works','sign fails', sign(10).eq.1 .and. sign(-10).eq.-1 )
-----------------------------------------------^
xx.f90(23): error #6505: This intrinsic procedure reference contains too few arguments. [SIGN]
write(*,*) merge('sign works','sign fails', sign(10).eq.1 .and. sign(-10).eq.-1 )
```
-------------------------------------------------------------------^
compilation aborted for xx.f90 (code 1)

[FortranFan]

It's a compiler bug with being unable to resolve a generic interface.  While it gets addressed, but a clearer instruction per the standard that all the processors will likely also prefer will be to move the `intrinsic` to the scope of the overriding procedure:

contains
   elemental function sign_int32(value)
      integer(kind=int32),intent(in) :: value
      integer(kind=int32)            :: sign_int32
      intrinsic :: sign ! move the intrinsic to this subscope
      sign_int32=sign(1_int32,value)
   end function sign_int32

[ur]

That does make is less ambigious as just the intrinsic is needed at that point, but the overlaid procedure is in scope, which makes for an

recursive situation.

[Barbara_P_Intel · Accepted solution]

Good news! I ran this with the coming release of ifort 2021.9.0 and ifx 2023.1.0 and it compiles and runs!

$ ifx --version
ifx (IFX) 2023.1.0 20230216
Copyright (C) 1985-2023 Intel Corporation. All rights reserved.

$ ifx tryit.f90
$ a.out
 sign works

[FortranFan]

@Barbara_P_Intel ,

In your tryit.f90 test, was the line 7 uncommented?  Assuming your test was the code in the original post with line 7 as "!intrinsic :: sign ! ifort 2023 bug(?)"   OP had the line commented out.

[Barbara_P_Intel]

@FortranFan, I missed that subtlety on line 7. I get the same error message with the coming release.

[Barbara_P_Intel]

Bug filed, CMPLRLLVM-45073.

[Barbara_P_Intel]

The erroneous error message is removed now in the Fortran compilers that were released last week as part of oneAPI HPC Toolkit 2023.2.

Please try it!