False positive N805 on overriden __new__ method

[popotam]

Description

Overriden new methods are always static methods and their first parameter should not be named self, but pep8-naming prints an error when it is not self.

How to replicate:

$ cat new_override.py 

class NewOverride(object):
    def __new__(cls):
        return object.__new__(NewOverride)

$ flake8 new_override.py 
new_override.py:2:9: N805 first argument of a method should be named 'self'

[sigmavirus24]

I suspect this also happens with classmethod and staticmethod decorated methods?

[stephan-hof]

This is a valid issuse,

So in case of __new__ the naming checker should only allow 'cls' ?
I will have a look at this on the weekend and will open a pull-request for it.

I suspect this also happens with classmethod and staticmethod decorated methods?

I don't think so, the checker tries to detect classmethod and staticmethod and does not force self there

[florentx]

For __new__, it seems reasonable to allow only cls.

There's also __prepare__ which is a special method for metaclasses.
For this one, it seems there's no clear consensus.
http://docs.python.org/3.4/reference/datamodel.html#metaclass-example

And a quick search on Google: https://www.google.com/search?q=%22def+__prepare__%22&nfpr=1
The first argument is one of meta, metacls, cls, mcl.

[stephan-hof]

Looking at __prepare__ it seems to me the classmethod decorator makes the first argument to appear. If you omit the classmethod decorator the signature looks like this def __prepare__(name, bases, **kwds)
Hence I would say the classmethod rule applies here too => cls

[abraithwaite]

@stephan-hof Any plans for a release of the changes reflected in this PR?