Alexander Burger wrote:
> Let me correct a wrong statement:
> On Thu, Dec 10, 2009 at 08:32:38AM +0100, Alexander Burger wrote:
>> The reason is that all inheritance mechanisms (message passing,
>> lookup with 'meta', 'var:', etc) start searching one level above the
>> current object.
> This is only true for 'meta' and derived functions, which search for a
> property in the classes and superclasses of a symbol.
> In case of messages, however, the search for a method starts in the
> object, and then continues in the classes and superclasses. Methods are
> not stored in properties, but in an assoc list in the object's or
> class's value cell.
> This was a deliberate design decision. It should be possible for an
> object to override a method privately. But "class variables" (the
> properties in classes and superclasses) should
>    1. really reside in classes, and
>    2. not conflict with local attributes of objects
> Therefore the term (or (: name) (var: name)) must be used explicitly if
> both places should be searched.

Perhaps we should propose a naming convention, that class variables
should begin with an uppercase letter, whereas instance variables should
begin with a lowercase letter. So we could avoid a possible glitch where
a class define an instance variable (let's say) 'id' thereby shadowing a
class variable 'id'. With your proposed term above, the instance
variable would silently be taken. So adhering to such a naming
convention could prevent hassle ... :-)



Reply via email to