Robert Dewar wrote:
> 
> Yoav posted a large program and one problem mentioned is that controlled
> types can only be declared at the library level. THis is correct, GNAT
> is giving you a correct diagnostic. Controlled types are tagged types,
> and tagged types must be extended at the same level as their parents.
> 
> As for the other "problem", no idea! Please post short concise examples,
> CLA is not the place to post large chunks of code and ask "what's wrong?"
> 
> You might also want to follow the GNAT directions and send GNAT
> questions to report@gnat.com, although there too, you need to make
> an effort to provide a concise example of your question rather than
> just dumping your whole program!

I've taken Robert's reply into consideration and since he gave me the answer for the controlled problem
I now post a small spec-only example for the compile time error of gnat regarding dispatching of abstract 
routine.

in the following example , I try to define an ADT , set, which can be implemented in various ways
and a routine , convert, to convert between two such implementations.
The problem is that gnat does not know how to dispatch the 'unit' function call in 'convert' body.
any ideas ?
I guess the problem is that the signature of the routine (except the returned type) does not contain
any derivation of the set class.
But I think that the compiler can deduce which routine to call by the type expected by the unit routine
to return .

So here it is...