[Patch, Fortran] PR 43199 - Handle class%component references when reading .mod files

[Janus Weil]

Hi Tobias,

> this patch prevents an ICE when handling class%component in
> find_array_spec. The problem is that with CLASS(type) one has
> effectively another layer around type, i.e. kind of
>
> TYPE class.type
>   type(type) :: data
>
> and one needs to accommodate for this extra layer. I think the attached
> patch correctly handles this.

Yes, in principle I think your patch is ok. For reasons of clarity I
would perhaps prefer the following version:

Index: gcc/fortran/resolve.c
===================================================================
--- gcc/fortran/resolve.c	(revision 157120)
+++ gcc/fortran/resolve.c	(working copy)
@@ -4005,7 +4005,11 @@ find_array_spec (gfc_expr *e)

       case REF_COMPONENT:
 	if (derived == NULL)
-	  derived = e->symtree->n.sym->ts.u.derived;
+	  {
+	    derived = e->symtree->n.sym->ts.u.derived;
+	    if (derived->attr.is_class)
+	      derived = derived->components->ts.u.derived;
+	  }

 	c = derived->components;


Moreover, a few lines further in find_array_spec, there is another
occasion where 'derived' is being set. Probably one also needs to
modify that one for CLASS components, maybe like this:

@@ -4015,6 +4019,8 @@ find_array_spec (gfc_expr *e)
 	      /* Track the sequence of component references.  */
 	      if (c->ts.type == BT_DERIVED)
 		derived = c->ts.u.derived;
+	      else if (c->ts.type == BT_CLASS)
+		derived = c->ts.u.derived->components->ts.u.derived;
 	      break;
 	    }

I haven't tried creating a test case which fails without this addition
(but maybe it would be good to include one in the patch).


> PS: I marked the test as "dg-do compile" but one can also use "run";
> what do you prefer?

dg-do compile is fine with me.

Cheers,
Janus