Allow class-based NamedTuple syntax in semantic analyzer for stubs in pre-3.6 mode

[gvanrossum]

See python/typeshed#1336 (comment)

Not very important, but would be nice if stubs could use

class A(NamedTuple):
    x: int
    y: str

instead if

A = NamedTuple('A', [('x', int), ('y', str)])

[srittau]

Had this come up in python/typeshed#3278.

[JukkaL]

This seems to be the relevant code (in mypy/semanal_namedtuple.py):

    def check_namedtuple_classdef(
            self, defn: ClassDef) -> Optional[Tuple[List[str], List[Type], Dict[str, Expression]]]:
        """Parse and validate fields in named tuple class definition.

        Return a three tuple:
          * field names
          * field types
          * field default values
        or None, if any of the types are not ready.
        """
        if self.options.python_version < (3, 6):
            self.fail('NamedTuple class syntax is only supported in Python 3.6', defn)
            return [], [], {}
...

Also, there is is_typeshed_stub_file() in SemanticAnalyzerInterface which isn't quite what we need. A similar property could be added for is_stub_file(), and then the rest should be easy.

Anybody want to write a PR?

[srittau]

PR in #7646.

[gvanrossum]

Do we know if this is supported by type checkers other than mypy?

[srittau]

We checked this in python/typeshed#3321 and pytype, pyre, and pyright, PyCharm, and the Microsoft Language Server for Python all support this.