Re: Ada 'hello world' for Android; success!

["Dmitry A. Kazakov" <mailbox@dmitry-kazakov.de>]

On Wed, 02 Nov 2011 11:59:46 -0400, Stephen Leake wrote:

> "Dmitry A. Kazakov" <mailbox@dmitry-kazakov.de> writes:
> 
>> On Sun, 30 Oct 2011 07:32:59 -0400, Stephen Leake wrote:
>>
>>> In brief, _any_ framework that uses dispatching on class-wide
>>> objects needs reference semantics.
>>
>> Wrong. Reference semantics is required to maintain object's identity.
> 
> Well, yes. What good is a system that does not maintain object identity?

Why should it? Most of the objects Ada creates lack run-time identity. Ada
also prevents the programmer from identifying such objects. E.g. by
X'Access for X without "aliased".

> Please rewrite my access_vs_object example, without using reference
> semantics. Note that I am explicitly including container cursors as a
> "reference type".

That is OK, cursor is a referential type, which identities the object it
points to. If the cursor is a publicly access type, that is fine to me.
Moreover I wished that all cursor types implemented "abstract access
interface".

> Or show me another real system that has dispatching on collections of
> heterogeneous objects without reference semantics.

Note that Ada does not have MD. So you just cannot dispatch on the
collection *and* the object in the collection. You probably meant a
different thing: to dispatch on an element of some collection. The word
"dispatch" is irrelevant to identity here, because identity comes from the
collection: the element must stay in the collection while the operation is
performed on it. Is it so?

Or did you mean that it would be impossible to implement a collection of
class-wide objects without pointers? This is true, but it does not imply
referential semantics. E.g. the objects still could be copied in and out
when accessed.

-- 
Regards,
Dmitry A. Kazakov
http://www.dmitry-kazakov.de