Re: [Patch, Fortran] SELECT TYPE via ASSOCIATE

[Daniel Kraft <d at domob dot eu>]

Hi again,

thanks for the test!

Salvatore Filippone wrote:
> Here you go; I've been able to tseal 10 mins (while waiting for a
> colleague). 
> It's a variation of the original test case, and produces: 
> [sfilippo@donald bug20]$ ./bug20_dx 
>  Dynamic type on entry: XBASE   
>  Not implemented yet 
>  Dynamic type CLASS DEFAULT clause: DBASE   
>
> which is pure nonsense :-) 

I agree :)  Looking at the tree-dump (and with some modification to your 
code), I get:

{
    struct class$d_base_sparse_mat_a * aa;
    struct d_base_sparse_mat * tmp$type$d_base_sparse_mat;

    aa = &a->a;
    tmp$type$d_base_sparse_mat = aa->$data;
    switch (aa->$vptr->$hash)
      {
        case 65139841:;
        goto L.1;
        default:;
        {
          character(kind=1) str.1[5];
          struct class$d_base_sparse_mat class.0;

          class.0.$vptr = (struct vtype$d_base_sparse_mat * {ref-all}) 
&vtab$d_base_sparse_mat;
          class.0.$data = tmp$type$d_base_sparse_mat;
          d_base_get_fmt ((character(kind=1)[1:5] *) &str.1, 5, &class.0);
          __builtin_memmove ((void *) &string, (void *) &str.1, 5);
          __builtin_memset ((void *) &string + 5, 32, 3);
        }
        {
          struct __st_parameter_dt dt_parm.2;

          dt_parm.2.common.filename = &"test.f03"[1]{lb: 1 sz: 1};
          dt_parm.2.common.line = 78;
          dt_parm.2.common.flags = 128;
          dt_parm.2.common.unit = 0;
          _gfortran_st_write (&dt_parm.2);
          _gfortran_transfer_character (&dt_parm.2, &"Dynamic type 
CLASS DEFAULT clause: "[1]{lb: 1 sz: 1}, 35);
          _gfortran_transfer_character (&dt_parm.2, &string, 8);
          _gfortran_st_write_done (&dt_parm.2);
        }
        goto L.1;
      }
    L.1:;
  }

I don't think the dynamic type gets clobbered, but rather the CLASS 
DEFAULT clause seems to use tmp$type$d_base_sparse_mat in case of "aa" 
-- this probably is a problem with the select type selector stack; at 
least this is the first that comes to mind.

Tobias suggested to check-in my patch anyways, and afterwards I (or 
Janus, if he has an idea?) can try to fix this.

Thanks a lot for all your testing of OOP ;)

Daniel

--
http://www.pro-vegan.info/
--
Done:  Arc-Bar-Cav-Kni-Ran-Rog-Sam-Tou-Val-Wiz
To go: Hea-Mon-Pri