Reconciling Dataclasses And Properties In Python - Florimond Manca

[utterances-bot]

Reconciling Dataclasses And Properties In Python - Florimond Manca

I love Python dataclasses, but combining them with properties is not obvious. This is a problem solving report — and a practical introduction to dataclasses!

https://florimond.dev/en/posts/2018/10/reconciling-dataclasses-and-properties-in-python/

[sujittiwari]

Appreciate your effort to bring this detailed analysis, It really help us to take deep dive.
But still, have a question about @wheels.setter if we provide a setter to a protected/private variable then didn't we lose the intent of having one(protected/private )?

[ruslaniv]

Great article!
Could you maybe extend this article and write about setting a default value to an attribute using dataclasses and properties (because of this: python/cpython#94067)? Right now I'm setting a default value in a parent class and then subclassing it but maybe there is a more elegant solution?

[florimondmanca]

@ruslaniv It seems you can set _wheels: int = field(..., default=...).

from dataclasses import dataclass, field

@dataclass
class Vehicle:
    wheels: int
    _wheels: int = field(init=False, repr=False, default=4)

    @property
    def wheels(self) -> int:
        print("getting wheels")
        return self._wheels

    @wheels.setter
    def wheels(self, wheels: int) -> None:
        print("setting wheels to", wheels)
        self._wheels = wheels

But this wouldn't work: the wheels setter receives the wheels property on init, instead of 4.

>>> Vehicle()
setting wheels to <property object at 0x1045ff100>
getting wheels
Vehicle(wheels=<property object at 0x1045ff100>)

After some digging into the source code of the dataclasses module, this is because of these lines:

# dataclasses.py, L730:732
    # If the default value isn't derived from Field, then it's only a
    # normal default value.  Convert it to a Field().
    default = getattr(cls, a_name, MISSING)

When processing the class, getattr(Vehicle, "wheels", MISSING) returns the wheels property. This value is later used as the default and passed to the setter via self.wheels = _wheels_dflt in the generated __init__.

We can't set a default using wheels: int = 4, because the later definition of the property would override it.

One solution that would work is ignoring that initial property value...

    @wheels.setter
    def wheels(self, wheels: int) -> None:
        if isinstance(wheels, property):
            return
        print("setting wheels to", wheels)
        self._wheels = wheels

Output:

>>> Vehicle()
getting wheels
Vehicle(wheels=4)

This could be moved to a minimal dataclass_property implementation, like so...

class dataclass_property(property):
    def __set__(self, obj, value):
        if isinstance(value, property):
            # dataclasses tries to set a default and uses the
            # getattr(cls, name). But the real default will come
            # from: `_attr = field(..., default=...)`.
            return
        super().__set__(obj, value)

Usage would be to replace @property with @dataclass_property.

I won't comment on the level of hackiness, but any usage of the approach described in this blog post was already in the "NO WARRANTY PROVIDED" zone. :-)

[hasii2011]

Thanks for this walk through and the short and sweet "recipe" at the end