[Tutor] object structure

[Eike Welk]

Hello Denis!


On Wednesday May 12 2010 14:37:54 spir ☣ wrote:
> class Integer(int): pass
> i = Integer(1)
> i.name = "one"
> print i,i.name	# --> "1 one"
> 
> An instance of Integer has both a value 1 and a dict. Actually, one can do
>  more sophisticated things (below __new__ is needed because of some
>  limitation in float):

I think it's the other way round: It is quite odd that "Integer" works without 
"__new__".

> class Constant(float):
>     def __new__(cls, name, value):
>         obj = float.__new__(cls, value)
>         obj.name = name
>         return obj
>     def __str__(self):
>         return "%s:%s" %(self.name, float.__str__(self))
> PI = Constant("pi", 3.14)
> print PI      # --> "pi:3.14"
> 
> In either case, the actual numerical value is transparently stored and
>  inaccessible: reason why I need to use float.__str__ to print it out. But
>  there also is a dict available to store attributes.
> 
> So, a question arises: does the structure representing a Constant object
>  like PI have one more field? Namely for type, value *and* dict?

I think you are right. The instances of class "Constant" have a "__dict__" 
while the "float" instance don't have it. And obviously the "float" instances 
contain some hidden data to encode the value.

However this mechanism (or a very similar one) is available to Python 
programmers too: classes with a "__slots__" declaration.
http://docs.python.org/reference/datamodel.html#slots

Instances of these classes have no "__dict__" attribute. The attributes 
declared with slots are also computed attributes. This means instead of 
accessing the attribute directly, the Python interpreter calls a function, 
which returns the attribute. (The function, really the descriptor, is stored 
in the class.) 

As you can inherit from classes with a "__slots__" declaration, I think all 
built in classes could be implemented elegantly with this underlying 
mechanism: 
- It is a mechanism for conserving memory (no "__dict__"), important for 
numbers.
- It is also a mechanism for storing hidden data somewhere in the object. The 
implementation available to the Python programmers is not very exciting, but 
it could be used for everything. 



Eike.