Re: Type naming conventions: Any_Foo

["Randy Brukardt" <randy@rrsoftware.com>]

"Jeffrey R. Carter" <spam.jrcarter.not@spam.not.acm.org> wrote in message 
news:qsipel$t3k$1@dont-email.me...
> On 12/8/19 12:24 AM, Jere wrote:
>>
>> You are correct that you can write a high level abstraction wrapper
>> around a GUI without exposing access types, but to Dmitry's point, the
>> actual code that creates the GUI components has to use access types
>> or some sort of referential data type.  Your example GUI illustrates even
>> this.  When you look at ada_gui.adb, you will see multiple uses of
>> access types:  accesses to procedure callbacks, accesses to GUI 
>> components,
>> uses of 'Access, etc.
>
> The implementation of Ada_GUI is a hack to let me test the feasibility of 
> the idea. I think a real implementation could be done without needing 
> access types, but have not looked into it in enough detail to be sure. 
> Even if it cannot be implemented without them, this discussion was about 
> what the pkg requires from its clients, so the implementation is 
> irrelevant.

To me, this is the key point. Claw uses very few access types externally; 
the user view is generally one of objects, which can be aggregated in any 
way the user desires (records, arrays, containers, or allocation/access 
types). There were a few places where that model didn't work, although I 
think that was more because of the underlying system (Win32) and wanting to 
make all of the facilities available in some way, rather than because it was 
strictly necessary.

Note that most Claw types are non-limited (thus allowing "cloning" - a copy 
being a "clone" of the original, and more importantly, return from 
functions) and task safe in the same way as Ada (so long as a single task 
only operates on a single object, everything will work -- tasks can work on 
different clones of the same underlying object).

                             Randy.