Re: Please help a newbie with Booch/Ada.

[Simon Wright <simon@pogner.demon.co.uk>]

Stephen Leake <Stephen.Leake@gsfc.nasa.gov> writes:

> Lengyel Sandor <hunnia@pacbell.net> writes:
> 
> > I am fooling around with the Booch components, since I think
> > "Collections" are very usefull.
> 
> good idea.
> 
> >    PACKAGE String_References IS
> >       TYPE String_ptr IS ACCESS ALL String;
> >       TYPE Const_String_Ptr IS ACCESS CONSTANT String;
> >    END String_References;
> > 
> > 
> > Then I created a root_string_container.ads as:
> > 
> >    with Bc.Containers;
> >    with String_References;
> >    package Root_String_Container is new Bc.Containers
> >       (String,String_References.String_ptr);
> > 
> > 
> > (Coping the format of root_container.ads. :-) )
> > 
> > When I run a semanric check on this later code, I get the following
> > message.
> > 
> > root_string_container.ads:4:08: actual for "Item" must be a definite
> > subtype.
> 
> See RM 3.3 (23). Basically, an indefinite type is any unconstrained
> type (in this case, String is an unconstrained array). A definite type
> is a constrained type.
> 
> The problem is that the Ada 95 Booch components are modeled after the
> C++ Booch components. Since there is no such thing as an "indefinite
> type" in C++, the Ada 95 Booch components restrict you to definite
> types. This is one of the advantages of Ada 95 over C++; you can give
> more information about the type. On the other hand, a careless library
> author can shoot you down.

:-)

> As a work around, you can use Ada.Strings.Unbounded.Unbounded_String
> instead of String.

This was my recommendation to another user.

> Better would be to fix the Ada 95 Booch components to allow indefinite
> types. Find the declaration of Bc.Containers, and add "(<>)" as a
> discriminant to the object type. Be aware this might break lots of
> stuff; I haven't tried it.

This is a possibility, but since the components require the generic
type Item to be copyable I don't think it will work?

A fix for Lengyel would be to declare

  type String_ptr_ptr ias access String_ptr;

and instantiate with

  package Root_String_Container is new Bc.Containers
     (String_References.String_ptr, String_References.String_ptr_ptr);

I have in mind that this instantiation with 2 pointers is confusing
and tedious, as far as I can see at the moment it's there to support
iterators where the iterating procedure can modify the item;

  function Current_Item (Obj : Iterator) return Item_Ptr;

and

  generic
    with procedure Apply (Elem_Ref : in Item_Ptr; OK : out Boolean);
  function Modify (Obj : access Passive_Iterator) return Boolean;

I think both of these would be better Ada if they used in out
parameters, eg

  generic
    with procedure Apply (Elem : in out Item; OK : out Boolean);
  function Modify (Obj : access Passive_Iterator) return Boolean;