+2007-09-21 Olivier Hainque <hainque@adacore.com>
+
+ * utils.c (type_for_nonaliased_component_p): Return false for
+ all AGGREGATE_TYPEs.
+
2007-09-17 Eric Botcazou <ebotcazou@adacore.com>
* decl.c (gnat_to_gnu_entity) <object>: Make again the type of an
if (must_pass_by_ref (gnu_type) || default_pass_by_ref (gnu_type))
return false;
- /* We might need the address for any array type, even if normally
- passed by copy, to construct a fat pointer if the component is
- used as an actual for an unconstrained formal. */
- if (TREE_CODE (gnu_type) == ARRAY_TYPE)
+ /* We used to say that any component of aggregate type is aliased
+ because the front-end may take 'Reference of it. The front-end
+ has been enhanced in the meantime so as to use a renaming instead
+ in most cases, but the back-end can probably take the address of
+ such a component too so we go for the conservative stance.
+
+ For instance, we might need the address of any array type, even
+ if normally passed by copy, to construct a fat pointer if the
+ component is used as an actual for an unconstrained formal.
+
+ Likewise for record types: even if a specific record subtype is
+ passed by copy, the parent type might be passed by ref (e.g. if
+ it's of variable size) and we might take the address of a child
+ component to pass to a parent formal. We have no way to check
+ for such conditions here. */
+ if (AGGREGATE_TYPE_P (gnu_type))
return false;
return true;
+2007-09-21 Olivier Hainque <hainque@adacore.com>
+
+ * gnat.dg/nested_subtype_byref.ad[bs]: Part of new test.
+ * gnat.dg/test_nested_subtype_byref.adb: New test.
+
2007-09-21 Jakub Jelinek <jakub@redhat.com>
PR fortran/33439
--- /dev/null
+
+package body Nested_Subtype_Byref is
+
+ type Data (Stamped : Boolean) is record
+ case Stamped is
+ when True => Valid : Boolean;
+ when others => null;
+ end case;
+ end record;
+
+ type Message is record
+ F : Integer := 1;
+ D : Data (Stamped => True);
+ end record;
+
+ procedure Check is
+ M : Message;
+ begin
+ M.D.Valid := True;
+ end;
+
+end;
+