Strange tab completion oddity with enums?

[Peter Otten]

Chris Angelico wrote:

> I'm not sure what's going on here, and it's probably not actually
> enum-specific, but that's where I saw it.
> 
> If you create a plain class and have an attribute with an annotation,
> you can see that:
> 
>>>> class Foo:
> ...     spam: "ham" = 1
> ...
>>>> Foo.__a
> Foo.__abstractmethods__  Foo.__annotations__
>>>> Foo.__annotations__
> {'spam': 'ham'}
> 
> Note that __annotations__ shows up when tab-completing "__a".
> 
> Now consider an enumeration:
> 
>>>> from enum import Flag, auto
>>>> class Bar(Flag):
> ...     quux: "asdf" = auto()
> ...
>>>> Bar.__
> Bar.__abstractmethods__  Bar.__getattr__(         Bar.__ne__(
> Bar.__base__(            Bar.__getattribute__(    Bar.__new__(
> Bar.__bases__            Bar.__getitem__(         Bar.__prepare__(
> Bar.__basicsize__        Bar.__gt__(              Bar.__qualname__
> Bar.__bool__(            Bar.__hash__(            Bar.__reduce__(
> Bar.__call__(            Bar.__init__(            Bar.__reduce_ex__(
> Bar.__class__(           Bar.__init_subclass__(   Bar.__repr__(
> Bar.__contains__(        Bar.__instancecheck__(   Bar.__reversed__(
> Bar.__delattr__(         Bar.__itemsize__         Bar.__setattr__(
> Bar.__dict__             Bar.__iter__(            Bar.__sizeof__(
> Bar.__dictoffset__       Bar.__le__(              Bar.__str__(
> Bar.__dir__(             Bar.__len__(             Bar.__subclasscheck__(
> Bar.__doc__              Bar.__lt__(              Bar.__subclasses__(
> Bar.__eq__(              Bar.__members__          Bar.__subclasshook__(
> Bar.__flags__            Bar.__module__           Bar.__text_signature__
> Bar.__format__(          Bar.__mro__              Bar.__weakrefoffset__
> Bar.__ge__(              Bar.__name__
>>>> Bar.__annotations__
> {'quux': 'asdf'}
> 
> Double-tabbing "__" shows everything but, and double-tabbing "__ann"
> has nothing... but the attribute is most definitely there.
> 
> Perhaps notable is dir():
> 
>>>> dir(Foo)
> ['__annotations__', '__class__', '__delattr__', '__dict__', '__dir__',
> '__doc__', '__eq__', '__format__', '__ge__', '__getattribute__',
> '__gt__', '__hash__', '__init__', '__init_subclass__', '__le__',
> '__lt__', '__module__', '__ne__', '__new__', '__reduce__',
> '__reduce_ex__', '__repr__', '__setattr__', '__sizeof__', '__str__',
> '__subclasshook__', '__weakref__', 'spam']
>>>> dir(Bar)
> ['__class__', '__doc__', '__members__', '__module__', 'quux']
> 
> But that's not the whole story, since tab completing "Bar.__" will
> still show "__class__" and "__init__" that aren't in dir().
> 
> Tested with the default REPL CPython 3.6, 3.7, 3.8, and 3.9. Tested
> also in IDLE on 3.9 but tab completion of dunders behaves differently
> there (it ONLY seems to want to tab complete __class__, for some
> reason) so it's not comparable.
> 
> What's actually going on here?

With Python 3.7:

>>> from enum import Flag
>>> import inspect
>>> print(inspect.getsource(Flag.__dir__))
    def __dir__(self):
        added_behavior = [
                m
                for cls in self.__class__.mro()
                for m in cls.__dict__
                if m[0] != '_' and m not in self._member_map_
                ]
        return (['__class__', '__doc__', '__module__'] + added_behavior)

Looks like everything starting with an underscore (except class, doc, and 
module) is suppressed, probably to suppress some noise...