Caught by surprise using field() without a type annotation

[kjohnsen]

I know, I know you...recommend against subclassing...but I was doing it in the hopes of reducing boilerplate. And I was really caught off guard by what happened when I mixed type annotations (which I now see the docs say you shouldn't do).

Specifically, I couldn't figure out why my child classes were using parent class default attributes. It took me a while to track it down: when I thought I was redefining the default value for an attribute for a child class, it was being ignored since I also had a field()-defined attribute without a type annotation.

MRE code

#%%
from attrs import define, field
# %%
@define
class Parent:
    x: int = 0
    y: int = 0

@define
class ChildNoAnnotNoField(Parent):
    x: int = 1
    y = 1


@define
class ChildMixedAnnot(Parent):
    # this x Attribute definition is ignored because
    # there is a field without a type annotation
    x: int = 1
    y = field(default=1)

@define
class ChildAnnotField(Parent):
    x: int = 1
    y: int = field(default=1)


print(Parent())               # x=0, y=0
print(ChildNoAnnotNoField())  # x=1, y=0
print(ChildMixedAnnot())      # x=0, y=1
print(ChildAnnotField())      # x=1, y=1

What I suggest is potentially raising a warning for cases like this to spare users from headaches. I had even read most of the docs in some detail, if not super closely, and this threw me for a loop.

Honestly, it wasn't until running into this that I realized you needed type annotations to define an Attribute either. The only place I see is on the Overview page: "Simply assign attrs.field() to the attributes instead of annotating them with types" which could be easily missed. So I'd also suggest making it a little more obvious (maybe on the 'By example' page?) that attributes are defined either by type annotations or by field.

[hynek]

I've added two warnings; one as suggested to examples, the other to the types chapter. I hope that suffices, thank you for reporting!