__name__ inconsistently applied in class definition

[eukreign]

BPO	14092
Nosy	@terryjreedy, @benjaminp

^{Note: these values reflect the state of the issue at the time it was migrated and might not reflect the current state.}

Show more details

GitHub fields:

assignee = None
closed_at = <Date 2012-02-25.01:15:32.065>
created_at = <Date 2012-02-22.23:29:07.646>
labels = ['interpreter-core', 'type-feature']
title = '__name__ inconsistently applied in class definition'
updated_at = <Date 2012-02-25.01:15:32.063>
user = 'https://bugs.python.org/eukreign'

bugs.python.org fields:

activity = <Date 2012-02-25.01:15:32.063>
actor = 'terry.reedy'
assignee = 'none'
closed = True
closed_date = <Date 2012-02-25.01:15:32.065>
closer = 'terry.reedy'
components = ['Interpreter Core']
creation = <Date 2012-02-22.23:29:07.646>
creator = 'eukreign'
dependencies = []
files = []
hgrepos = []
issue_num = 14092
keywords = []
message_count = 8.0
messages = ['154013', '154027', '154032', '154033', '154034', '154035', '154040', '154178']
nosy_count = 3.0
nosy_names = ['terry.reedy', 'benjamin.peterson', 'eukreign']
pr_nums = []
priority = 'normal'
resolution = 'rejected'
stage = None
status = 'closed'
superseder = None
type = 'enhancement'
url = 'https://bugs.python.org/issue14092'
versions = ['Python 3.3']

[eukreign]

The following behavior doesn't make sense, especially since it works correctly for other special attributes:

>>> class F:
	__name__ = "Foo"

>>> F.__name__
'F'
>>> F().__name__
'Foo'
>>> F.__name__ = 'Foo'
>>> F.__name__
'Foo'

Works fine for __module__ and others:

>>> class F:
	__module__ = "mod"
	
>>> F.__module__
'mod'
>>> F
<class mod.F at 0x0212B360>

[benjaminp]

I don't think this is actually incorrect. Basically setting something in the class body is not equivalent to setting it as an attribute on the class.

This happens with other attributes. Consider

>>> class X:
...     __class__ = list
... 
>>> X.__class__
<class 'type'>
>>> X().__class__
<class 'list'>

[eukreign]

I think for __class__ it might make sense but for __name__ it seems not intuitive.

[benjaminp]

Well, what do you expect the name of the class to be? The one you assign __name__ to or the one you pass to type()?

[eukreign]

The one in __name__ since logically that happens after the class declaration ('class X' line) and should overwrite the name in the class declaration.

[benjaminp]

It can be implemented but I'm skeptical that it's correct. You might try convincing Python-dev.

[eukreign]

I don't particularly need this functionality. It was just something that seemed counter intuitive to me.

I discovered this while working on a python to javascript compiler. I'll probably implement the compiler to allow overriding with __name__ as it seems more intuitive to me that way.

Hopefully cpython will be changed to work that way too.

[terryjreedy]

This is not a bug report, as Python works as documented.
Double underscore names are defined as *reserved* for the interpreter, with the ones actually in use having defined meanings.

type.__new__ sets several internally used attributes on new classes. The attribute look-up mechanism for classes looks at them first before looking in __dict__, which is for attributes of both the class and its instances. Here is another example similar to yours.

>> class C: __dict__ = 1

>>> C.__dict__
dict_proxy({'__dict__': 1, '__module__': '__main__', '__weakref__': <attribute '__weakref__' of 'C' objects>, '__doc__': None})
>>> C().__dict__
1

__dict__ is not writable, but __class__ is. You can already rename a class if you really want:

>>> C.__name__ = 'bizarre'
>>> C.__name__
'bizarre'

Conceptually, this seems the right way as one normally would not want the name of the class to be the default name for every instance.

>>> C().__name__
Traceback (most recent call last):
  File "<pyshell#12>", line 1, in <module>
    C().__name__
AttributeError: 'bizarre' object has no attribute '__name__'

If you really want instances to have that also, then also do as you did.

There are other class-only, not for instances, attributes:
__mro__ and __subclasses__ and perhaps others.