OSError.__init__ should do nothing

x = OSError()
x.__init__('abc')
assert x.args is ()

This is because in CPython all the work is done in OSError.__new__. For subclasses of OSError, if subclass.__init__ != OSError.__init__ and subclass.__new__ == OSError.__new__ then the work is done in OSError.__init__ instead of new:

class Test(OSError): pass

x = Test()
OSError.__init__(x, 1) # __init__ does nothing
assert x.args == ()

class Test2(OSError):
    def __init__(self, *args): pass

x = Test2()
OSError.__init__(x, 1)
assert x.args == (1,)

class Test3(OSError):
    def __new__(cls, *args):
        return super().__new__(cls, *args)

x = Test3()
OSError.__init__(x, 1) # __init__ does nothing
assert x.args == ()

class Test4(OSError):
    def __new__(cls, *args):
        return super().__new__(cls, *args)

x = Test4()
OSError.__init__(x, 1) # __init__ does nothing
assert x.args == ()

This causes failures in CPython.test_pep3151 as well as IronPython.test_exceptions.

Start with the CPython.test_pep3151 and IronPython.test_exceptions failures, then trace OSError.new and OSError.init in IronPython. Done means the shown subclass cases preserve their expected args behavior and both reported test failures pass.