Incorrect ParseError for type comment in unexpected location

[JelleZijlstra]

From https://travis-ci.org/python/typeshed/jobs/212532837:

Running pytype tests...
pytd error processing "stdlib/2and3/cgi.pyi":
  File: "stdlib/2and3/cgi.pyi", line 107
    def values(self, key: Any) -> Any: ...  # type: ignore
                                            ^
ParseError: syntax error, unexpected TYPECOMMENT, expecting DEF or '@'

Type comments should be allowed here.

[matthiaskramm]

You need the type: ignore for disabling mypy's check, for methods being compatible with their superclass's methods?

Maybe we need a # mypy: ignore, after all. See python/typed_ast#14

[gvanrossum]

Only # type: ignore should be allowed virtually anywhere. Type comments expressing types are only allowed in specific places (though the spec is probably a bit vague -- IIRC typed_ast just picked a set of rules that are good enough).

But I'm also fine with just working around this in typeshed by moving the ... to the next line.

[JelleZijlstra]

Moving ... to the next line still generates a parse error in pytype.

[gvanrossum]

Oh, that's unfortunate. But I do think that this is not just a mypy-specific thing. Liskov substitutability is taken pretty seriously by PEP 484. (It's the reason we have to declare variance.)

[matthiaskramm]

Fair enough. We'll add #type: ignore after function signatures to the pytype parser.

[matthiaskramm]

Fix pushed.