什么是使用 python 数据类的没有设置器的属性获取器的 pythonic 方式？

Question

I have a situation where I need to store variables a,b, and c together in a dataclass, where c = f(a,b) and a,b can be mutated. I need c to be displayed along with a and b when printing the object, and c cannot be changed except through changing a or b . I felt the best way to do this was to have c be a value created with a property.

Minimal example of what I tried to use:

@dataclass
class Demo:
    a:int
    b:int
    c:int = field(init=False)

    @property
    def c(self) -> int:
        return self.a * self.b

However, calling this results in AttributeError: can't set attribute . I suspect the code is trying to set c to field(init=False) , and failing because c doesn't have a setter.

Options I've considered (not counting alternatives I tried that makes the code crash in a different manner):

• not having c in init
  • doesn't display c in repr, fields, etc. , which causes further issues down the line
• having a @c.setter that does nothing
  • this technically works, but involves writing boilerplate code that seemingly does nothing.
  • having a function that never gets called explicitly by any code I write but makes the program break is removed feels dangerous and unclean.
  • I also want the program to crash instead of silently doing nothing if someone tries to modify c directly.
• as above, but use sys._getframe().f_back.f_code in the setter to raise an error if the caller is not __init__ (function at endnote)
  • Is a Cpython implementation detail only.
  • Using functions that are marked as private is a bad idea, generally speaking.
• have Demo be immutable, have c be created in __post_init__ , and use dataclasses.replace to modify a and b.
  • this works in a vacuum, but other uses of Demo (well, the actual thing, not the minimal example) require it to be mutable.

endnote:

    @c.setter
    def c(self, val):
        caller = sys._getframe().f_back.f_code.co_name
        if caller != '__init__':
            raise ValueError # or some other error

Answer 1

If you want c to be a computed property, then take out the c field entirely and just leave the property:

@dataclass
class Demo:
    a:int
    b:int

    @property
    def c(self) -> int:
        return self.a * self.b

Fields are for stored data, which c should not be.

If you want c to show up in repr , you'll have to write the __repr__ method yourself. If later code needs c to show up in fields , then you probably need to rework your code at a deeper level, and possibly switch to something other than dataclasses .

Answer 2

Your method c overwrites your field c -- the original field declaration is lost, including the init=False .

Try changing the order and using the default= parameter.

from dataclasses import dataclass, field


@dataclass
class Demo:
    @property
    def c(self) -> int:
        return self.a * self.b

    a: int
    b: int
    c: int = field(init=False, default=c)


d = Demo(5, 10)
print(d)
d.a = 10
print(d)
d.c = 4  # generates an AttributeError

Output from Python 3.8.0:

Demo(a=5, b=10, c=50)
Demo(a=10, b=10, c=100)
Traceback (most recent call last):
...
  File "/home/rob/src/plot/dc.py", line 20, in <module>
    d.c = 4  # generates an AttributeError
AttributeError: can't set attribute