Re: Access type to member procedure of instance (Object Oriented programming in Ada)

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

"Adam Beneschan" <adam@irvine.com> wrote in message 
news:87af1bfd-41d2-45c7-953a-c1954d4f5ceb@googlegroups.com...
>On Tuesday, November 20, 2012 1:38:38 PM UTC-8, Randy Brukardt wrote:
>> "Adam Beneschan" wrote in message
>>
>> >If Message had type Message_Type'Class, I'd assume that at some point,
>> >Message.Print'Access would need to determine which Print subprogram 
>> >would
>> >be
>> >called, and figure out the address of it.  I don't think it matters 
>> >whether
>> >it's done when the
>> >'Access value is computed or when the call is made, since the tag of an
>> >object can't be
>> >changed during the object's lifetime.  I could be wrong, though; maybe 
>> >I've
>> >missed a case
>> >where it matters.  (But I'm not concerned about the possibility of an
>> >object's tag being
>> >changed through extralegal means like Unchecked_Conversion.)
>>
>> I don't think this is quite true; if the object is a parameter (for
>> instance), it can have different tags for different calls. But I don't 
>> see
>> that happening for something like Param.Print'Access -- access values are
>> never dispatching.
>>
>> Anyway, since something like Param.Print'Access is illegal in any case, 
>> we
>> don't have to decide what this means or how code would be generated for 
>> it.
>
>I probably confused things.  Yes, Param.Print'Access is illegal, but I was
>thinking in terms of "what if" we added a language feature to make it 
>legal--probably
>by adding a new kind of access type that contained references both to the 
>prefix
>object and, somehow, to the subprogram.  In that case, I'd want it to be
>dispatching, if Param's type were classwide.

*IF* we made it legal, we'd simply change the convention to Ada. There is no 
reason for the complication of a new kind of access type, because creating a 
wrapper is simple. Moreover, both Ada 2005 and Ada 2012 require creating 
wrappers in certain circumstances, so it's not like this is something a 
compiler never has done before. (And there are at least 4 other places where 
Janus/Ada uses wrappers, it's just not a big deal. And one of them is for 
'Access when the  subprogram was declared inside a generic while the access 
type was not.)

Note that GNAT is apparently using a wrapper to implement the renames of a 
prefixed view (and mistakenly allowing 'Access to be taken of the result). 
Doing exactly that for 'Access is not likely to be a big deal.

Whether or not that wrapper ought to dispatch is a separate issue and 
probably doesn't really make any difference to how the 'Access is supported.

                                                        Randy.