Re: status of Ada STL?

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

Mark Biggar wrote:

> "Dmitry A. Kazakov" wrote:
>> 
>> procedure '':=" (Target : in out A; Source : A); -- (*)
>> 
>> Note I do not demand that the copy constructor would be derived from
>> ":=". It is IMO a wrong idea to attempt to derive one from another.
>> C++ does it simplier. What is actually wrong in C++ is that the result
>> of "=" is not void.
>> 
>> >Wouldn't it also fall apart if we started including "+:="?
>> 
>> No, they should be just user-defined procedures. Whether they are
>> primitive or not should only depend on the freezing rules and whether
>> particular parameters are class-wide.
>> 
>> (*)  There is a minor problem with unconstrained subtypes passed as
>> inout and that discriminants cannot be changed.
> 
> There are other problems as well.  This was looked at in detail during
> the Ada95 revision.  Allowing users to redefine ":=" using a procedure
> definition like that above was determine to be way too course grained.
> Do you expect your new assignment operation to compose for record or
> array types? What about variable initialization? 

This should be performed by a copy-constructor. The user may influence it 
by overriding Adjust which is a part of it. Compiler should never use ":=" 
implicitly.

> Or, the assign out
> during procedure exit for out parameters?

Out parameters are now allowed to be read, so it is no more a problem. A 
similar problem (though with all predefined operations) is that people 
might want to get rid of predefined ":=" and have

procedure '':=" (Target : in out A; Source : A'Class);

This is unsolvable, because there is no way to disallow a primitive 
operation in Ada.

> This is way ada95 controlled types are tagged and break the assignment
> process down into finer parts with Initialize, Adjust and Finalize
> as user definable hooks into a larger assignment process.  This way
> use defines the type specific parts of data movement and the compiler
> is free to use them in various ways to move data around in ways the
> guarantees that the type's data abstractions are preserved.

This is IMO bad. Assignment should be just a user-defined procedure. It 
could be well defined as Finalize + Copy-constructor, but no longer user 
overrides it.

-- 
Regards,
Dmitry Kazakov
www.dmitry-kazakov.de