Re: Possible changes to abstract types proposal

[Stewart Greenhill <greenhil@murdoch.edu.au>]

At 07:29 PM 6/2/99 +0100, hartmut Goebel wrote:
>Stewart Greenhill (greenhil@murdoch.edu.au) schrieb:
>
>
> >- Use of the NEW attribute to distinguish newly introduced methods from
> >re-defined methods. This makes it more difficult to accidentally introduce
> >a new method where you meant to override an existing one (eg. due to a typo
> >in a method name).
>
>unnecessary to introduce an new keyword for this: such errors are easily
>found in the very beginning of code writing.

True, but one might not actually find the error until run-time, when the
code does something unexpected. The advantage of the NEW tagging is that
most of these errors can be caught at compile time.

> >- Object finalisation. This helps to reduce the possibility of resource
> >leakage when garbage-collecting objects that encapsulate resources other
> >than memory (eg. OS handles).
>
>No Language extension is necessary for this! Simply use a baseclass e.g.
>"Finalized" and enter each object into a queue. See the Oberon System for
>implementation. Okay, not that nice, but can easily be hitten in a module.

Yes, that is another way of doing it. I'm not familiar with the Oberon
System. If the objects are in a queue, how does the system decide when to
call their finalisation procedures?

In BlackBox, ANYREC is an implicit base type for all record types. It has
built in declarations equivalent to:

ANYPTR = POINTER TO ANYREC;

PROCEDURE (a: ANYPTR) FINALIZE-, NEW, EMPTY;

This means that the default FINALIZE method does nothing, and that any type
can override it. However, it is exported as implement-only, so FINALIZE can
never be called by users' code, only by the system. The garbage collector
calls FINALIZE prior to deallocating an object. 

- Stewart