Calling __new__ via super

[ltworf]

I just encountered this bug again: #1053

I checked this code, to make a subclass of tuple:

class A(tuple):
    def __new__(cls, fields):
        return super(A, cls).__new__(cls, tuple(fields))

With those settings:

python_version=3.5
warn_unused_ignores=True
warn_redundant_casts=True
strict_optional=True
scripts_are_modules=True
check_untyped_defs=True

Using mypy 0.590

And it results in

/tmp/test.py:3: error: Too many arguments for "__new__" of "object"

[ilevkivskyi]

I think this is actually a typeshed problem. I just looked through some stubs and noticed that there are some places where __new__ is missing. Namely, tuple defines __init__ instead on both Python 2 and 3 which doesn't really makes sense, since it is immutable, tuple should actually define __new__ with the same signature. I don't know if other projects depend on this "fake" tuple.__init__, but I propose to close this issue and move to typeshed tracker. @JelleZijlstra what do you think?

[ltworf]

Sorry if it's misplaced. I opened it here because I thought it might be a regression from the old bug.

[JelleZijlstra]

@ilevkivskyi that makes sense; I hadn't realized before that it makes a difference whether stubs define __new__ or __init__.

[ilevkivskyi]

@ltworf Could you please close this issue and open a corresponding one on the typeshed tracker?

[ltworf]

python/typeshed#2091