I don't know if we necessarily want to support type annotation prefixes that MyPy considers invalid. It might be better to just nudge a user in the right direction, so they fix their prefixes to work with mypy.

This regex is a little confusing, won't it also match # type:?

it does, but had there been # type: we wouldn't have even run this regex due to this check : if len(type_lines) == 0:

it's valid to have multiple type lines, so it's not enough to check that if there is one valid match. Can we instead filter all lines that include type, check for the exact match, and if there isn't an exact match throw if the regex matches?

In this particular case, we would still get Number of type annotations (1) did not match the number of function parameters (2):. We could tweak that error message to advise our user to check their type annotation prefixes.

What about

def foo(
   a, # type: int
   b #     type: bad type annotation
)
# type: (...) -> int

As long as there's at least one valid type annotation prefix, we should get an error.
I just double checked that this exact example also gives the same error.

@torch.jit.script
def add_two(a,# type: int
            b #     type: int
           ):
    # type: (...) -> int
    return a+b

This should either use the actual warnings module or be an error. I'm leaning towards it being an error since people should be fixing this if it comes up.

let's turn this into an error then.

Also let's not refer to MyPy directly and make people think it's secretly running under the hood

on the other hand, wouldn't we annoy the heck out of our user if we don't explain her why exactly we don't just support this syntax and raise an error instead? I don't mind removing the mypy comment if you think that's the best course of action here.

Yea i would rather not reference MyPy directly, there are other python type checkers as well. It's a python standard not a mypy standard

You can refer to the PEP or something lol