Re: RFC: (PR 45586) Pointer, target, and alias analysis of derived types (RESTRICT qualifier)

[Michael Matz <matz at suse dot de>]

Hi,

On Tue, 26 Jul 2011, Tobias Burnus wrote:

> a) Fortran question: My understanding is that any POINTER/TARGET attribute
> propagates through such that the last-ref can/is a pointer target. Example:
> 
>   TYPE realspace_grid_type
>      REAL(KIND=dp), DIMENSION ( :, :, : ), ALLOCATABLE :: r
>   END TYPE realspace_grid_type
> 
> Here, "r" cannot be a pointer target, unless, one has, e.g.,
> 
>   TYPE realspace_grid_p_type
>      TYPE(realspace_grid_type), POINTER :: rs_grid
>   END TYPE realspace_grid_p_type
> 
> such that "rs_grid"'s pointer target has TARGET attribute such that also "r"
> has the target attirbute. Is this correct?

It seems so.  realspace_grid_type.r can have restrict set, but 
realspace_grid_p_type.rs_grid.r cannot.  Hence the type of the FIELD_DECL 
corresponding to rs_grid needs to have a nonrestrict type, exactly because 
of the pointer modifier in that type definition.  Therefore if something 
needs to change it is in actually building the RECORD_TYPE from the 
frontend representation, and hence ...

> b) GCC/gfortran code question: Is the following patch correct and sufficient?
> It does not ICE, but I am not sure whether that's sufficient. (Before, it
> ICEed with -flto, cf. comment 53.)
> 
> --- a/gcc/fortran/trans-types.c
> +++ b/gcc/fortran/trans-types.c
> @@ -2421,6 +2421,9 @@ gfc_get_derived_type (gfc_symbol * derived)
>                &&  !c->attr.proc_pointer)
>         field_type = build_pointer_type (field_type);
> 
> +      if (c->attr.pointer)
> +       field_type = gfc_nonrestricted_type (field_type);
> +
>        /* vtype fields can point to different types to the base type.  */
>        if (c->ts.type == BT_DERIVED&&  c->ts.u.derived->attr.vtype)
>           field_type = build_pointer_type_for_mode (TREE_TYPE (field_type),

... this seems to be the right thing.  But it probably also needs to do 
the same for c->attr.target, right?  E.g. for:

   TYPE murks
      TYPE(realspace_grid_type), TARGET :: rs_grid
   END TYPE murks

I'm not sure if murks.rs_grid.r needs to have a nonrestrict type.  The 
fortran standard seems to imply this, doesn't it?  (As in, rs_grid.r also 
has the target attribute implicitely).


Ciao,
Michael.