Sounds reasonable. I added a mypy issue for this: python/mypy#562

I see three different categories of modules that don't import typing: (1) modules that don't import it and don't have any annotations, (2) modules that don't import it and have type annotations, either PEP 484 function annotations or # type: comments, (3) modules that don't import typing but have non-standard annotations. Should these all be different?

Mypy currently complains about code like this if there is no annotation, even for dynamically typed modules, since module top levels are implicitly statically typed by default:

x = []   # Can't infer the type of x

I'd like to change this so that mypy could process at least many existing modules and still support some level of type checking for code that imports those modules (e.g., checking that function argument counts are compatible).

Maybe category (1) modules should default to Any as the type of top-level and class variables, but mypy would try to infer these types for category (2) modules. Thus full type checking would be signalled by the addition of any type annotation or typing import. Most existing modules don't have any annotations, so a type checker could process them without generating errors, but marking a module for type checking wouldn't need to involve importing typing. Category (3) modules would need either a # type: OFF comment or telling the type checker to not to check that module.

Moved to README.rst.