Bug 104404 - Incorrect CFI_cdesc_t "type" member for assumed-type, assumed-rank complex dummy arguments

Bug 104404 - Incorrect CFI_cdesc_t "type" member for assumed-type, assumed-rank complex dummy arguments

[Damian Rouson]

For an assumed-type, assumed-rank complex dummy argument in a C interface,
gfortran 11.2.0 passes a CFI_cdesc_t object with a "type" member that does
not match the corresponding CFI_type_float_Complex and
CFI_type_double_Complex values. In the case of a complex(c_float_complex)
argument, the passed "type" member corresponds to CFI_type_double_Complex.
For a complex(c_double_complex) argument, the "type" member has a value
that I don't recognize.  Does anyone know whether this has been fixed on
the 12 branch?

% cat c_descriptor.c
#include <stdio.h>
#include <ISO_Fortran_binding.h>

void c_descriptor(CFI_cdesc_t* c)
{
  printf("a_desc->type = %d \n", c->type);
  printf("a_desc->elem_len = %d \n", c->elem_len);
  printf("CFI_type_float_Complex = %d \n", CFI_type_float_Complex);
  printf("CFI_type_double_Complex = %d \n", CFI_type_double_Complex);
}

% cat c_descriptor.c
#include <stdio.h>
#include <ISO_Fortran_binding.h>

void c_descriptor(CFI_cdesc_t* c)
{
  printf("a_desc->type = %d \n", c->type);
  printf("a_desc->elem_len = %d \n", c->elem_len);
  printf("CFI_type_float_Complex = %d \n", CFI_type_float_Complex);
  printf("CFI_type_double_Complex = %d \n", CFI_type_double_Complex);
}
(base) rouson@CLaSS adhoc % cat assumed-type.f90
module c_descriptor_m
  implicit none
contains
  module subroutine print_type_info(a)
    type(*), intent(inout), contiguous, target :: a(..)

    interface
      subroutine c_descriptor(a) bind(C)
        implicit none
        type(*) a(..)
      end subroutine
    end interface

    call c_descriptor(a)

  end subroutine
end module

  use c_descriptor_m
  use iso_c_binding
  implicit none

  complex(c_float_complex) :: z_float = (0._c_float, 0._c_float)
  complex(c_double_complex):: z_double = (0._c_double, 0._c_double)

  print*, "-----  complex(c_float_complex) --------"
  call print_type_info(z_float)
  print*, "-----  complex(c_double_complex) -------"
  call print_type_info(z_double)
end

 % gfortran c_descriptor.c assumed-type.f90

% ./a.out
 -----  complex(c_float_complex) --------
a_desc->type = 2052
a_desc->elem_len = 8
CFI_type_float_Complex = 1028
CFI_type_double_Complex = 2052
 -----  complex(c_double_complex) -------
a_desc->type = 4100
a_desc->elem_len = 16
CFI_type_float_Complex = 1028
CFI_type_double_Complex = 2052

% gfortran --version
GNU Fortran (Homebrew GCC 11.2.0_3) 11.2.0

Re: Bug 104404 - Incorrect CFI_cdesc_t "type" member for assumed-type, assumed-rank complex dummy arguments

[Paul Richard Thomas]

Hi Damian,

Tobias Burnus fixed it in 12-branch:
GNU Fortran (GCC) 12.0.1 20220203 (experimental)
./a.out
 -----  complex(c_float_complex) --------
a_desc->type = 1028
a_desc->elem_len = 8
CFI_type_float_Complex = 1028
CFI_type_double_Complex = 2052
 -----  complex(c_double_complex) -------
a_desc->type = 2052
a_desc->elem_len = 16
CFI_type_float_Complex = 1028
CFI_type_double_Complex = 2052

Cheers

Paul


On Sun, 6 Feb 2022 at 03:39, Damian Rouson via Fortran <fortran@gcc.gnu.org>
wrote:

> For an assumed-type, assumed-rank complex dummy argument in a C interface,
> gfortran 11.2.0 passes a CFI_cdesc_t object with a "type" member that does
> not match the corresponding CFI_type_float_Complex and
> CFI_type_double_Complex values. In the case of a complex(c_float_complex)
> argument, the passed "type" member corresponds to CFI_type_double_Complex.
> For a complex(c_double_complex) argument, the "type" member has a value
> that I don't recognize.  Does anyone know whether this has been fixed on
> the 12 branch?
>
> % cat c_descriptor.c
> #include <stdio.h>
> #include <ISO_Fortran_binding.h>
>
> void c_descriptor(CFI_cdesc_t* c)
> {
>   printf("a_desc->type = %d \n", c->type);
>   printf("a_desc->elem_len = %d \n", c->elem_len);
>   printf("CFI_type_float_Complex = %d \n", CFI_type_float_Complex);
>   printf("CFI_type_double_Complex = %d \n", CFI_type_double_Complex);
> }
>
> % cat c_descriptor.c
> #include <stdio.h>
> #include <ISO_Fortran_binding.h>
>
> void c_descriptor(CFI_cdesc_t* c)
> {
>   printf("a_desc->type = %d \n", c->type);
>   printf("a_desc->elem_len = %d \n", c->elem_len);
>   printf("CFI_type_float_Complex = %d \n", CFI_type_float_Complex);
>   printf("CFI_type_double_Complex = %d \n", CFI_type_double_Complex);
> }
> (base) rouson@CLaSS adhoc % cat assumed-type.f90
> module c_descriptor_m
>   implicit none
> contains
>   module subroutine print_type_info(a)
>     type(*), intent(inout), contiguous, target :: a(..)
>
>     interface
>       subroutine c_descriptor(a) bind(C)
>         implicit none
>         type(*) a(..)
>       end subroutine
>     end interface
>
>     call c_descriptor(a)
>
>   end subroutine
> end module
>
>   use c_descriptor_m
>   use iso_c_binding
>   implicit none
>
>   complex(c_float_complex) :: z_float = (0._c_float, 0._c_float)
>   complex(c_double_complex):: z_double = (0._c_double, 0._c_double)
>
>   print*, "-----  complex(c_float_complex) --------"
>   call print_type_info(z_float)
>   print*, "-----  complex(c_double_complex) -------"
>   call print_type_info(z_double)
> end
>
>  % gfortran c_descriptor.c assumed-type.f90
>
> % ./a.out
>  -----  complex(c_float_complex) --------
> a_desc->type = 2052
> a_desc->elem_len = 8
> CFI_type_float_Complex = 1028
> CFI_type_double_Complex = 2052
>  -----  complex(c_double_complex) -------
> a_desc->type = 4100
> a_desc->elem_len = 16
> CFI_type_float_Complex = 1028
> CFI_type_double_Complex = 2052
>
> % gfortran --version
> GNU Fortran (Homebrew GCC 11.2.0_3) 11.2.0
>


-- 
"If you can't explain it simply, you don't understand it well enough" -
Albert Einstein

Re: Bug 104404 - Incorrect CFI_cdesc_t "type" member for assumed-type, assumed-rank complex dummy arguments

[Damian Rouson]

That's great news.  Thanks for letting me know.

Damian

On Sun, Feb 6, 2022 at 12:48 AM Paul Richard Thomas <
paul.richard.thomas@gmail.com> wrote:

> Hi Damian,
>
> Tobias Burnus fixed it in 12-branch:
> GNU Fortran (GCC) 12.0.1 20220203 (experimental)
> ./a.out
>  -----  complex(c_float_complex) --------
> a_desc->type = 1028
> a_desc->elem_len = 8
> CFI_type_float_Complex = 1028
> CFI_type_double_Complex = 2052
>  -----  complex(c_double_complex) -------
> a_desc->type = 2052
> a_desc->elem_len = 16
> CFI_type_float_Complex = 1028
> CFI_type_double_Complex = 2052
>
> Cheers
>
> Paul
>
>
> On Sun, 6 Feb 2022 at 03:39, Damian Rouson via Fortran <
> fortran@gcc.gnu.org> wrote:
>
>> For an assumed-type, assumed-rank complex dummy argument in a C interface,
>> gfortran 11.2.0 passes a CFI_cdesc_t object with a "type" member that does
>> not match the corresponding CFI_type_float_Complex and
>> CFI_type_double_Complex values. In the case of a complex(c_float_complex)
>> argument, the passed "type" member corresponds to CFI_type_double_Complex.
>> For a complex(c_double_complex) argument, the "type" member has a value
>> that I don't recognize.  Does anyone know whether this has been fixed on
>> the 12 branch?
>>
>> % cat c_descriptor.c
>> #include <stdio.h>
>> #include <ISO_Fortran_binding.h>
>>
>> void c_descriptor(CFI_cdesc_t* c)
>> {
>>   printf("a_desc->type = %d \n", c->type);
>>   printf("a_desc->elem_len = %d \n", c->elem_len);
>>   printf("CFI_type_float_Complex = %d \n", CFI_type_float_Complex);
>>   printf("CFI_type_double_Complex = %d \n", CFI_type_double_Complex);
>> }
>>
>> % cat c_descriptor.c
>> #include <stdio.h>
>> #include <ISO_Fortran_binding.h>
>>
>> void c_descriptor(CFI_cdesc_t* c)
>> {
>>   printf("a_desc->type = %d \n", c->type);
>>   printf("a_desc->elem_len = %d \n", c->elem_len);
>>   printf("CFI_type_float_Complex = %d \n", CFI_type_float_Complex);
>>   printf("CFI_type_double_Complex = %d \n", CFI_type_double_Complex);
>> }
>> (base) rouson@CLaSS adhoc % cat assumed-type.f90
>> module c_descriptor_m
>>   implicit none
>> contains
>>   module subroutine print_type_info(a)
>>     type(*), intent(inout), contiguous, target :: a(..)
>>
>>     interface
>>       subroutine c_descriptor(a) bind(C)
>>         implicit none
>>         type(*) a(..)
>>       end subroutine
>>     end interface
>>
>>     call c_descriptor(a)
>>
>>   end subroutine
>> end module
>>
>>   use c_descriptor_m
>>   use iso_c_binding
>>   implicit none
>>
>>   complex(c_float_complex) :: z_float = (0._c_float, 0._c_float)
>>   complex(c_double_complex):: z_double = (0._c_double, 0._c_double)
>>
>>   print*, "-----  complex(c_float_complex) --------"
>>   call print_type_info(z_float)
>>   print*, "-----  complex(c_double_complex) -------"
>>   call print_type_info(z_double)
>> end
>>
>>  % gfortran c_descriptor.c assumed-type.f90
>>
>> % ./a.out
>>  -----  complex(c_float_complex) --------
>> a_desc->type = 2052
>> a_desc->elem_len = 8
>> CFI_type_float_Complex = 1028
>> CFI_type_double_Complex = 2052
>>  -----  complex(c_double_complex) -------
>> a_desc->type = 4100
>> a_desc->elem_len = 16
>> CFI_type_float_Complex = 1028
>> CFI_type_double_Complex = 2052
>>
>> % gfortran --version
>> GNU Fortran (Homebrew GCC 11.2.0_3) 11.2.0
>>
>
>
> --
> "If you can't explain it simply, you don't understand it well enough" -
> Albert Einstein
>