Feature request: Add validation option to auto_attribs

[qpwo]

This is just an idea. Some users of a library I work on use static analysis and other users don't, so we use both type hinting and validation on the attributes. We have several classes like this:

from attr.validators import instance_of
@attrs
class C:
    x: str = attrib(validator=instance_of(str))
    y: str = attrib(validator=instance_of(str))
    z: int = attrib(validator=instance_of(int))

And in this search you can see that most users of instance_of stick it on every attribute when they use it at all.

It would be nice if attrs had a kwarg like auto_attribs_with_validation that would convert the below class to the above class:

@attrs(auto_attribs_with_validation=True)
class C:
    x: str
    y: str
    z: int

Then of course if you executed c = C("foo", "bar", "baz") you would get a TypeError.

I'm aware that it may not be worth adding another kwarg to attrs.

If this is a good idea, then I can probably write the PR.

[qpwo]

@rgabbard pointed out to me that some types cannot be fully instance-checked, e.g. all subscripted generics from typing, like List[int]. So auto_attribs_with_validation could do a few things:

Option A: Error out at class definition if there are any un-checkable types

Option B: Strip the subscripts off the generics, so

@attrs(auto_attribs_with_validation=True)
class C:
    nums: List[int]

would become

@attrs
class C:
    nums: List[int] = attrib(validator=instance_of(List)) # no [int]

and other sorts of un-checkable types would still error out.

Option C: Do nothing and let the user discover that their types are un-checkable.

[gabbard]

@qpwo : This appears to basically be #301 updated for the existence of auto_attribs and type annotations.

[wsanchez]

Sounds like we agree this is a dup. I'm going to close this out as such.