Google Groups no longer supports new Usenet posts or subscriptions. Historical content remains viewable.
Dismiss

a new object definition

1 view
Skip to first unread message

Sylvain Ferriol

unread,
Sep 1, 2006, 4:28:41 AM9/1/06
to
hello everybody,

i want to talk with you about a question i have in mind and i do not
find a answer. it 's simple:
why do we not have a beatiful syntax for object definition as we have
for class definition ?

we can define a class in python in 2 ways:
1. by using the metaclass constructor
my_class = MyMetaClass(....)
2. by using the classic definition syntax:
class my_class(object):
__metaclass__ = MyMetaClass

if i want to instanciate an object, i only have one way to define it:
my_obj = my_class(....)

why not a better syntax like class syntax ?

example:
>> instance my_obj:
>> __class__ = my_class


with this syntax, we are coherent between objects and classes.
why do we have a specific syntax for the class object definition and not
for objects of different type?

so if i want to define a new class object (my_class for example) with a
new object syntax:
>> instance my_class:
>> __class__ = object
>> __metaclass__ = MyMetaClass
>> ....

thanks

Sylvain Ferriol
Ingénieur de recherche
Laboratoire TIMC/IMAG
http://www-timc.imag.fr/

Michele Simionato

unread,
Sep 1, 2006, 4:55:26 AM9/1/06
to
Sylvain Ferriol wrote:
> hello everybody,
>
> i want to talk with you about a question i have in mind and i do not
> find a answer. it 's simple:
> why do we not have a beatiful syntax for object definition as we have
> for class definition ?

See http://www.python.org/dev/peps/pep-0359 (already rejected by
Guido).

Michele Simionato

Bruno Desthuilliers

unread,
Sep 1, 2006, 4:58:31 AM9/1/06
to
Sylvain Ferriol wrote:
> hello everybody,
>
> i want to talk with you about a question i have in mind and i do not
> find a answer. it 's simple:
> why do we not have a beatiful syntax for object definition as we have
> for class definition ?

Python's classes are objects too - instances of their metaclass.

> we can define a class in python in 2 ways:
> 1. by using the metaclass constructor
> my_class = MyMetaClass(....)
> 2. by using the classic definition syntax:
> class my_class(object):
> __metaclass__ = MyMetaClass
>
> if i want to instanciate an object, i only have one way to define it:
> my_obj = my_class(....)
>
> why not a better syntax like class syntax ?
>
> example:
>>> instance my_obj:
>>> __class__ = my_class

I fail to see how it's "better", nor what would be the use case. But
anyway I think it could be possible by using the metaclass as class and
the class as instance. Just make sure the metaclass wraps all methods
into classmethods and you should be done.

>
> with this syntax, we are coherent between objects and classes.
> why do we have a specific syntax for the class object definition and not
> for objects of different type?
>
> so if i want to define a new class object (my_class for example) with a
> new object syntax:
>>> instance my_class:
>>> __class__ = object
>>> __metaclass__ = MyMetaClass
>>> ....
>
> thanks
>
> Sylvain Ferriol
> Ingénieur de recherche
> Laboratoire TIMC/IMAG
> http://www-timc.imag.fr/
>


--
bruno desthuilliers
python -c "print '@'.join(['.'.join([w[::-1] for w in p.split('.')]) for
p in 'on...@xiludom.gro'.split('@')])"

Sylvain Ferriol

unread,
Sep 1, 2006, 6:44:52 AM9/1/06
to
Michele Simionato a écrit :
i do not understand the withdrawal note, what do "different level" mean ?
do you have an example or is it python core implemantation problem ?
> Michele Simionato
>

Michele Simionato

unread,
Sep 1, 2006, 7:09:11 AM9/1/06
to
Sylvain Ferriol wrote:
> Michele Simionato a écrit :

> >
> > See http://www.python.org/dev/peps/pep-0359 (already rejected by
> > Guido).
> >
> i do not understand the withdrawal note, what do "different level" mean ?
> do you have an example or is it python core implemantation problem ?

I asked Guido in person at EuroPython. He said the syntax didn't look
"right" to him. It is as simple as that.

Michele Simionato

Sylvain Ferriol

unread,
Sep 1, 2006, 8:53:19 AM9/1/06
to
Michele Simionato a écrit :
note that we can not use the "make syntax" for a function definition
because the block is not executed until the function call, while the
block for a class is executed before the class instanciation
> Michele Simionato
>

Steven Bethard

unread,
Sep 1, 2006, 11:58:23 AM9/1/06
to
Sylvain Ferriol wrote:
> hello everybody,
>
> i want to talk with you about a question i have in mind and i do not
> find a answer. it 's simple:
> why do we not have a beatiful syntax for object definition as we have
> for class definition ?
>
> we can define a class in python in 2 ways:
> 1. by using the metaclass constructor
> my_class = MyMetaClass(....)
> 2. by using the classic definition syntax:
> class my_class(object):
> __metaclass__ = MyMetaClass
>
> if i want to instanciate an object, i only have one way to define it:
> my_obj = my_class(....)
>
> why not a better syntax like class syntax ?
>
> example:
>>> instance my_obj:
>>> __class__ = my_class

Michele Simionato already pointed you to `PEP 359`_. One of the reasons
that I withdrew it was that people seemed to feel that you could get
most of what you want now by defining appropriate metaclasses. In your
case, for example, the appropriate metaclass and its usage might look like::

>>> def instance(name, bases, dict):
... cls = dict.pop('__class__')
... dict.pop('__metaclass__')
... dict.pop('__module__') # silently added by class statement
... return cls(**dict)
...
>>> class C(object):
... def __init__(self, a, b):
... self.a = a
... self.b = b
...
>>> class c:
... __metaclass__ = instance
... __class__ = C
... a = 'foo'
... b = 'bar'
...
>>> c.a, c.b
('foo', 'bar')

Sure, it's misleading to use a class statement when you're not actually
creating a class, but I guess people felt that wanting to do this was
uncommon enough that they weren't worried about it.

.. _PEP 359: http://www.python.org/dev/peps/pep-0359/

STeVe

Sylvain Ferriol

unread,
Sep 4, 2006, 6:28:41 AM9/4/06
to Steven Bethard
i know that there is always a solution. But this problem show that
python and others are not coherent in the syntax (contrary to lisp for
example).

with the 'make' syntax, it will be really easy to translate a program or
a data structure defined in XML format into python syntax.

i do not know how many use-cases we need for changing a PEP status :)

another advantage is that we have the same syntax in all definition
levels: metaclass, class, instance.
and if we just want to use objects and do a sort of 'prototype
programming', we can with this syntax.
example:
instance my_obj(prototype_obj):
...
... object specialisation
...

sylvain

Steven Bethard

unread,
Sep 6, 2006, 3:31:59 PM9/6/06
to
Sylvain Ferriol wrote:
> with the 'make' syntax, it will be really easy to translate a program or
> a data structure defined in XML format into python syntax.

Only if there are no ordering constraints and no need for multiple
elements with the same name. The make statement was built to mirror the
the class statement, so the body is just executed in a regular Python
dict. Hence, you can only have one value for each name, and the order
in which the names appear is not maintained. There's some discussion of
workarounds_ for this in the PEP, but they're pretty hackish.

.. _workarounds:
http://www.python.org/dev/peps/pep-0359/#customizing-the-dict-in-which-the-block-is-executed

STeVe

0 new messages