Treat annotations more opaquely (proposal)

[valtron]

Python allows annotations to be any expression, but if you use MyPy on your project, you can't use annotations for anything other than types (MyPy will error "Invalid type ...").

This issue proposes the following solution:

• provide class typing.Annot (name for the purposes of this discussion):

class Annot:
    def __init__(self, type_=None, **kwargs):
        self.type = type_
        self.kwargs = kwargs

• during typechecking, MyPy treats annotations that are instances of Annot as if they were their type property

Benefits/Use cases

MyPy/other tools (including runtime inspection) also have extra info in kwargs available, if they know how to use it. For example, dataclasses (and attrs, and this concept) store extra field information in the value of the attribute:

@dataclass
class C:
    z: int = field(repr=False, default=10)

...which is a workaround. With general-purpose annotations, it could look like this:

@dataclass
class C:
    z: Annot(int, repr=False, default=10)

This could also subsume class attribute annotations and Final annotations:

class A:
    x: Annot(int, classvar=True) = 0
    y: Annot(final=True) = ''

But the main benefit is that with this in place, it will be possible to add annotations as well as, or instead of, plain type info, without foregoing MyPy, paving the way for more tooling.

[Michael0x2a]

You may be interested in this proposal in the "typing" github repo, which proposes an idea similar to yours.

[ilevkivskyi]

I think this is essentially a duplicate of python/typing#600, we can continue discussing details in that issue.