Decorator API with auto_attribs=True doesn't work.

[ckutlu]

When I try to use attribute decorators via auto attributes style, I get an error saying that the name doesn't exist. This doesn't happen if I initiate attributes with attr.ib(). Is this expected behavior?

Here is an example that fails:

@attr.s(auto_attribs=True)
class A:
    x: float

    @x.validator
    def _check_x(self, attrib, value):
        print("Validator called.")
        print(self, attrib, value)
a = A(5)

gives

Traceback (most recent call last):
  File "test.py", line 5, in first
    @attr.s(auto_attribs=True)
  File "test.py", line 9, in A
    @x.validator
NameError: name 'x' is not defined

[hynek]

Yes, unfortunately this is to be expected. There is no way for us to cheat x into the class namespace. It's just type declared but doesn't exist until @attr.s runs over it.

This should be documented in http://www.attrs.org/en/stable/types.html

[ckutlu]

I understand, thanks for the prompt response. That's a bummer, but I guess using attr.ib() for only attributes with validators isn't that much of an ugly workaround.

[AndydeCleyre]

This should be mentioned in the Validators section of the examples page, and in my opinion the existing notice in the Validators section of the init page is insufficiently clear, and should explicitly mention that it won't work when using the @attrs(auto_attribs=True... decorator.

[jimcushing-natgeo]

It might help to note that this still works:

@attr.s(auto_attribs=True)
class A:
    x: float = attr.ib()

So you still can use type annotations, validators, and auto_attribs.

[alexchandel]

attrs should provide a workaround for this. Like

@attr.s(auto_attribs=True)
class A:
    x: float

    @attr.validator
    def _check_x(self, attrib, value):
        print("Validator x called.")
        print(self, attrib, value)

    y: float
    z: float

    @attr.validator_for('y')
    def _check_y(self, attrib, value):
        print("Validator y called.")
        print(self, attrib, value)


a = A(5, 6, 7)

I think the first case requires python >= 3.7 dicts remembering order insertion.