How to type hint a dictionary with values of different types

levraininjaneer picture levraininjaneer · Jun 25, 2018 · Viewed 11.5k times · Source

When declaring a dictionary as a literal, is there a way to type-hint what value I am expecting for a specific key?

And then, for discussion: are there guiding principles around dictionary typing in Python? I'm wondering whether it is considered bad practice to mix types in dictionaries.

Here's an example:

Consider the declaration of a dictionary in a class's __init__ :

(disclaimer: I realize in the example, some of the .elements entries would probably be more appropriate as class attributes, but it's for the sake of the example).

class Rectangle:
    def __init__(self, corners: Tuple[Tuple[float, float]], **kwargs):
        self.x, self.z = corners[0][0], corners[0][1]
        self.elements = {
            'front': Line(corners[0], corners[1]),
            'left': Line(corners[0], corners[2]),
            'right': Line(corners[1], corners[3]),
            'rear': Line(corners[3], corners[2]),
            'cog': calc_cog(corners),
            'area': calc_area(corners),
            'pins': None
        }


class Line:
    def __init__(self, p1: Tuple[float, float], p2: Tuple[float, float]):
        self.p1, self.p2 = p1, p2
        self.vertical = p1[0] == p2[0]
        self.horizontal = p1[1] == p2[1]

when I type type the following

rec1 = Rectangle(rec1_corners, show=True, name='Nr1')
rec1.sides['f...

Pycharm will suggest 'front' for me. Better still, when I do

rec1.sides['front'].ver...

Pycharm will suggest .vertical

So Pycharm remembers the keys from the dictionary literal declaration in the class's __init__, and also their values' expected types. Or rather: it expects any value to have any one of the types that are in the literal declaration - probably the same as if I had done a self.elements = {} # type: Union[type1, type2] would do. Either way, I find it super helpful.

If your functions have their outputs type-hinted, the Pycharm will also take that into account.

So assuming that in the Rectangle example above, I wanted to indicate that pins is a list of Pin objects... if pins was a normal class attribute, it would be

    self.pins = None  # type: List[Pin]

(provided the necessary imports were done)

Is there a way to give the same type hint in the dictionary literal declaration?

The following does not achieve what I am looking for:

Add a Union[...] type hint at the end of the literal declaration?

            'area': calc_area(corners),
            'pins': None
        }  # type: Union[Line, Tuple[float, float], float, List[Pin]]

Adding a type-hint to every line:

            'area': calc_area(corners),  # type: float
            'pins': None  # type: List[Pin]
        }

Is there a best practice for this kind of thing?

Some more background:

I work with Python in PyCharm and I make extensive use of typing, since it helps me to predict and validate my work as I go along. When I create new classes, I also sometimes throw some less frequently used properties into a dictionary to avoid cluttering the object with too many attributes (this is helpful in debug mode).

Answer

Michael0x2a picture Michael0x2a · Jun 28, 2018

You are looking for TypedDict. It is currently only a mypy-only extension, but there are plans to make it an officially sanctioned type in the near-future. I am not sure if PyCharm supports this feature yet, though.

So, in your case, you'd do:

from mypy_extensions import TypedDict

RectangleElements = TypedDict('RectangleElements', {
    'front': Line,
    'left': Line,
    'right': Line,
    'rear': Line,
    'cog': float,
    'area': float,
    'pins': Optional[List[Pin]]
})

class Rectangle:
    def __init__(self, corners: Tuple[Tuple[float, float]], **kwargs):
        self.x, self.z = corners[0][0], corners[0][1]
        self.elements = {
            'front': Line(corners[0], corners[1]),
            'left': Line(corners[0], corners[2]),
            'right': Line(corners[1], corners[3]),
            'rear': Line(corners[3], corners[2]),
            'cog': calc_cog(corners),
            'area': calc_area(corners),
            'pins': None
        }  # type: RectangleElements

If you are using Python 3.6+, you can type this all more elegantly using the class-based syntax.

In your specific case though, I think most people would just store those pieces of data as regular fields instead of a dict. I'm sure you've already thought through the pros and cons of that approach though, so I'll skip lecturing you about it.