Wrong result in assigning this_image() to a complex coarray

[Steve Kargl]

On Fri, Jun 26, 2015 at 07:18:55AM -0700, Damian Rouson wrote:

(Courtesy wrap to less than 80 characters)

> I just submitted the code below in bug report  66681
> (https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66681).  The problem
> disappears if the variable declaration is for an integer or real or
> if it contains the allocatable attribute and a subsequent allocation
> happens.  This appears to be a front-end problem because it even
> occurs when compiling without OpenCoarrays.
> 
> Damian
> 
> $ cat assign_this_image_to_complex.f90 
> complex a[*]
> a = this_image()
> print *,this_image(),a 
> end
> 
> $ gfortran assign_this_image_to_complex.f90 -fcoarray=single
> 
> $ ./a.out
>            1 (  0.00000000    ,  0.00000000    )
> 

The relevant part of -fdump-tree-original for Damain's
testcase is

{
  complex(kind=4) a;

  SAVE_EXPR <a> = __complex__ (1.0e+0, 0.0);
  {
    _gfortran_transfer_complex_write (&dt_parm.0, &a, 4);
  }
}


Changing COMPLEX to INTEGER in the testcase yields

{
  integer(kind=4) a;

  (integer(kind=4)) a = 1;
  {
    _gfortran_transfer_integer_write (&dt_parm.0, &a, 4);
  }
}

For the simple program 

complex a
a = 1
print *, a
end

I get 
{
  complex(kind=4) a;

  a = __complex__ (1.0e+0, 0.0);
  {
    _gfortran_transfer_complex_write (&dt_parm.0, &a, 4);
  }
}

which suggest that the SAVE_EXPR <a> is missing an (gfc_)evaluate(_now)
step. I'm not sure where to look to fix this.  Tobias?

-- 
Steve