I'm proposing to extend enum.Flag member functionality so it is iterable in a manner similar to enum.Flag subclasses.

from enum import Flag, auto


class FlagIter(Flag):
    def __iter__(self):
        for memeber in self._member_map_.values():
            if member in self:
                yield member


class Colour(FlagIter):
    RED = auto()
    GREEN = auto()
    BLUE = auto()
    YELLOW = RED | GREEN

cyan = Colour.GREEN | Colour.Blue

print(*Colour)  # Colour.RED Colour.GREEN Colour.BLUE Colour.YELLOW

# Without the enhancement, 'not iterable' is thrown for these
print(*cyan)  # Colour.GREEN Colour.BLUE
print(*Colour.YELLOW)  # Colour.RED Colour.GREEN Colour.YELLOW
print(*~Colour.RED)  # Colour.GREEN Colour.BLUE

This functionality is now in the third-party aenum* library.

Are there any strong use-cases for this behavior such that it should be in the stdlib?


* as of version 2.0.10, available on PyPI, and I am its author

Problem:  What to do when the Flag has compound members?

    class Color(Flag):
        BLACK = 0
        RED = 1
        GREEN = 2
        BLUE = 4
        PURPLE = RED|BLUE
        WHITE = RED|GREEN|BLUE

I think the answer is: only return the single members.  So

  --> list(Color.WHITE)
  [Color.BLUE, Color.GREEN, Color.RED]

What does aenum do and has there been any feedback on it?

To me I would see what you suggest as surprising but I don't use enums often (I should use them more!) so take that with a grain of salt, and also surprising != wrong/not good.

Final outcome:

`Flag` has been redesigned such that any flag comprised of a single bit is canonical; flags comprised of multiple bits are considered aliases.  During iteration only canonical flags are returned.