Confusion about __call__ and attribute lookup

[Kent Johnson]

I am learning about metaclasses and there is something that confuses me.

I understand that if I define a __call__ method for a class, then instances of the class become callable using function syntax:
... def __call__(self):
... print 'Called Foo'
...Called Foo

To create a class instance, you call the class. This made me think that the class' class must define __call__, and indeed it does, and calling it as an unbound method also creates a class instance:

>>> dir(type) [..., '__call__', ...]
>>> f=type.__call__(Foo)
>>> f

<__main__.Foo object at 0x00A35EB0>

But why doesn't Foo.__call__ shadow type.__call__? Normally an instance attribute takes precedence over a class attribute. Is it something special about how function call syntax is handled internally, or do all special methods work this way, or is there something else going on?

PS Is there any place in the standard Python docs where the details of attribute lookup are spelled out?

Thanks,
Kent

 

[Leif K-Brooks]

But why doesn't Foo.__call__ shadow type.__call__? Normally an instance
attribute takes precedence over a class attribute. Is it something
special about how function call syntax is handled internally, or do all
special methods work this way, or is there something else going on?

New-style classes look up special methods on the class, not on the instance:
... def __invert__(self):
... return 'foo'
... Traceback (most recent call last):
'foo'

 

[Kent Johnson]

New-style classes look up special methods on the class, not on the instance:

For my future reference, is this documented somewhere in the standard docs?

Thanks,
Kent

 

[John J. Lee]

For my future reference, is this documented somewhere in the standard docs?

Maybe somewhere in here :-(

http://www.python.org/doc/newstyle.html


John

 

[Serge Orlov]

For my future reference, is this documented somewhere in the standard docs?

Looks like it's the most detailed explanation on the net:

http://mail.python.org/pipermail/python-dev/2003-May/035732.html

 

[Kent Johnson]

Maybe somewhere in here :-(

http://www.python.org/doc/newstyle.html

I have never found it there. I think something like the writeup Serge referenced should be in the language reference. I just sent the suggestion to (e-mail address removed)

Thanks to both of you,
Kent