Can't tell apart untyped from typed class variable assignment

[euresti]

While working on #2088 I came across an interesting problem. Take this code

class A:
   x: int
   y: ClassVar[int] = 67
   z = 67

It's clear that x is an instance variable and y is a class variable. However z is a bit vague. Technically, as far as python is concerned it's a class variable. Both dataclasses and attr.s(auto_attribs=True) will ignore z when creating the __init__. (Because they only look in cls.__annotations__) (Full disclosure, I haven't run dataclasses yet but looking at the code it looks that way) However I can't figure out a way to tell the difference in mypy code.

Basically I'm walking the class body looking for AssignmentStmts. And I thought "Oh hey. node.is_classvar" should help me here. It return False for x, True for y, but sadly it also returns False for z. To make it harder, if you look at visit_assignment_stmt it does analyze_simple_literal_type to infer the type of z and then store_declared_types sets is_inferred_def = False so I can't tell apart z = 67 from z: int = 67

[ilevkivskyi]

I think PEP 526 is clear on this, everything that is not typed with ClassVar is an instance variable. In particular:

class C:
    z = 42

introduces an instance variable with a default value. As for how to check this in mypy, there are several options and the simplest one is probably to re-use stmt.new_syntax, see semanal.check_namedtuple_classdef.

[euresti]

Ooh. new_syntax is exactly what I need. Thanks!