Re: Augusta: An open source Ada 2012 compiler (someday?)

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

On Thu, 27 Mar 2014 13:38:16 -0700 (PDT), Lucretia wrote:

> i.e. 1) add exception records
> 
> type E is exception
>    record
>       My_Data : Integer;
>    end record;

You will have a problem catching such exceptions because in Ada exceptions
are "values" not "types", like in C++.

That they are "value" among other allows enumeration like:

   when Error : Constraint_Error | Data_Error =>

If Constraint_Error and Data_Error were "types" what would be the type of
Error? You could move to classes of exceptions but that still would not
allow simple enumerations when instances are not siblings.

A related problem is renaming of exceptions. A renaming produce a new
object, but you still can catch it under original name because the result
has same "value."

In general if exceptions form a tree of related types you have a problem of
choosing between sets, possibly nested sets of types in exception handlers.
C++ does this by linear matching exceptions in the order they appear. This
is a mess.

> 2) User defined types having 'Image and 'Value, which call a subprogram
> which must be defined if these attributes are to be used - much like the
> stream operations do.

They should be primitive operations instead.

But the actual problem here is MD. It works with stream attributes through
cascaded dispatch *only* because streams enforce Stream_Element on all
implementations which decompose objects into Stream_Element_Arrays. This is
the reason why a stream attribute can *first* dispatch on the object type
and *then* on the stream. For 'Image and 'Value it would not work in
presence of Character, Wide_Character, Wide_Wide_Character. That is full
MD, or else you have to restrict it to single character type.
 
> 3) Replace tagged with just class, so:
> 
> type C is
>    class
>       ...
>    end class;

Class is not a type. Class is a set of types. A class can have a
"representative" type, which is already named in Ada as T'Class.

In short, you need not to specify that some type may be member of a class.
All types may. Why not?

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