Re: [PING^2] ada: Add limitations on the acceptable prefixes of 'Old

[Robert Dewar <dewar at adacore dot com>]

Samuel Tardieu wrote:
> On 11/05, Robert Dewar wrote:
>
> > the deep check is quite easy to write with our standard traversal
> > routines, and this results in a simple rule to state.
>
> Here is an updated patch.

Seems fine, thanks for the patience in reaching a good solution here.
I suggest minor style changes to the sem_attr patch, as well as
including the entity name in the msg:

> --- sem_attr.adb        (revision 130883)
> +++ sem_attr.adb        (working copy)
> @@ -3509,6 +3509,68 @@
>                ("?attribute Old applied to constant has no effect", P);
>           end if;
>
> +         --  Check that the expression does not refer to local entities
> +
> +         Check_Local : declare
> +            Subp : Entity_Id := Current_Subprogram;
> +
> +            function Process (N : Node_Id) return Traverse_Result;
> +            --  Check that N does not contain references to local variables
> +            --  or other local entities of Subp.
> +
> +            -------------
> +            -- Process --
> +            -------------
> +
> +            function Process (N : Node_Id) return Traverse_Result is
> +            begin
> +               if Is_Entity_Name (N)
> +                 and then not Is_Formal (Entity (N))
> +                 and then Enclosing_Subprogram (Entity (N)) = Subp
> +               then
> +                  Error_Msg_Node_1 := Entity (N);
> +                  Error_Attr
> +                    ("attribute % cannot refer to local variable&", N);
> +               end if;
> +
> +               return OK;
> +            end Process;
> +
> +            procedure Check_No_Local is new Traverse_Proc;
> +
> +         --  Start of processing for Check_Local
> +
> +         begin
> +            Check_No_Local (P);
> +
> +            if In_Parameter_Specification (P) then
> +
> +               --  We have additional restrictions on using 'Old in parameter
> +               --  specifications.
> +
> +               if Present (Enclosing_Subprogram (Current_Subprogram)) then
> +
> +                  --  Check that there is no reference to the enclosing
> +                  --  subprogram local variables. Otherwise, we might end
> +                  --  up being called from the enclosing subprogram and thus
> +                  --  using 'Old on a local variable which is not defined
> +                  --  at entry time.
> +
> +                  Subp := Enclosing_Subprogram (Current_Subprogram);
> +                  Check_No_Local (P);
> +
> +               else
> +                  --  We must prevent default expression of library-level
> +                  --  subprogram from using 'Old, as the subprogram may be
> +                  --  used in elaboration code for which there is no enclosing
> +                  --  subprogram.
> +
> +                  Error_Attr
> +                    ("attribute % can only appear within subprogram", N);
> +               end if;
> +            end if;
> +         end Check_Local;
> +
>        ------------
>        -- Output --
>        ------------

and some very minor style changes in the sem_util patch:

> --- sem_util.adb        (revision 130946)
> +++ sem_util.adb        (working copy)
> @@ -5029,6 +5029,26 @@
>        return False;
>     end In_Package_Body;
>
> +   --------------------------------
> +   -- In_Parameter_Specification --
> +   --------------------------------
> +
> +   function In_Parameter_Specification (N : Node_Id) return Boolean is
> +      PN : Node_Id;
> +
> +   begin
> +      PN := Parent (N);
> +      while Present (PN) loop
> +         if Nkind (PN) = N_Parameter_Specification then
> +            return True;
> +         end if;
> +
> +         PN := Parent (PN);
> +      end loop;
> +
> +      return False;
> +   end In_Parameter_Specification;
> +
>     --------------------------------------
>     -- In_Subprogram_Or_Concurrent_Unit --
>     --------------------------------------

OK to commit this patch!