https://gcc.gnu.org/bugzilla/show_bug.cgi?id=125761
Mikael Morin <mikael at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |mikael at gcc dot gnu.org
--- Comment #4 from Mikael Morin <mikael at gcc dot gnu.org> ---
> diff --git a/gcc/fortran/trans.cc b/gcc/fortran/trans.cc
> index adf392cec6f..c10484ff636 100644
> --- a/gcc/fortran/trans.cc
> +++ b/gcc/fortran/trans.cc
> @@ -524,8 +524,13 @@ gfc_build_array_ref (tree base, tree offset, tree decl,
> span = gfc_vptr_size_get (vptr);
>
> /* Check if this is an unlimited polymorphic object carrying a
> character
> - payload. In this case, the 'len' field is non-zero. */
> - if (decl && GFC_CLASS_TYPE_P (TREE_TYPE (decl)))
> + payload. In this case, the 'len' field is non-zero. The class
> + container type does not always carry GFC_CLASS_TYPE_P, but its
> + canonical type does. */
> + if (decl
> + && (GFC_CLASS_TYPE_P (TREE_TYPE (decl))
> + || (TYPE_CANONICAL (TREE_TYPE (decl))
> + && GFC_CLASS_TYPE_P (TYPE_CANONICAL (TREE_TYPE (decl))))))
> span = gfc_resize_class_size_with_len (NULL, decl, span);
> }
> else if (decl)
We should try to have the GFC_CLASS_TYPE_P set on the type directly, otherwise
we'll have to duplicate this condition everywhere the flag is used.
gfc_typenode_for_spec has:
case BT_DERIVED:
case BT_CLASS:
basetype = gfc_get_derived_type (spec->u.derived, codim);
if (spec->type == BT_CLASS)
GFC_CLASS_TYPE_P (basetype) = 1;
The generation of class descriptors as additional symbols has its downsides,
but here it could help us as it makes polymorphic derived types (i.e.
class(...)) use a separate symbol different from regular derived types. So
gfc_get_derived_type in the snippet above could return a type with the
GFC_CLASS_TYPE_P flag already set. I think it would be set basically when the
derived type symbol has the is_class attribute.
