namedtuple should support "module" parameter added in Python 3.6

[gvanrossum]

See http://bugs.python.org/issue17941

[elazarg]

The current definition is wrong. From collections.__init__.pyi:

# namedtuple is special-cased in the type checker; the initializer is ignored.
namedtuple = object()

Well, no. namedtuple is special-cased in some type checkers, but if I write my own, I might defer this specialization indefinitely, and in the meantime treat it as a simple function call, that returns type or Type or something.

Even when it is special cased by the type checker, for uniformity it might want to use the actual signature; currently this is done ad-hoc and incorrectly in mypy:

$ cat > tmp.py
from collections import namedtuple
T = namedtuple('T', 'a b', verbose=True, rename=False)
$ mypy tmp.py
tmp.py:2: error: Too many arguments for namedtuple()
tmp.py:2: error: "object" not callable

The last error message does not happen with correct signature in collections.__init__.pyi, and mistakes result in the usual, clear error messages.

(It might even be helpful to add a pseudo-implementation, so that type checkers can simply inline and continue, getting rough approximation of the actual type. This is not like adding implementation of arbitrary functions, as this is a type constructor with trivial control flow)

[gvanrossum]

I think this would be reasonable to put in the stub:

# namedtuple is special-cased in the type checker; the initializer is ignored.
def namedtuple(typename: str, field_names: Union[str, Iterable[Any]], *,
               verbose: bool = False, rename: bool = False, module: bool = None) -> Type[tuple]: ...

but we also need to add support for the keyword args to mypy. Can you file a separate bug for that? I'm happy to take a PR for the above.

[elazarg]

I assume you mean module: str? And why put the initializers instead of ellipsis?

---

While we're at it, perhaps we should add

def cast(tp: Type[_T], obj: Any) -> _T: ...

def NewType(name: str, tp: Type[_T]) -> Type[_T]: ...

But I don't know if that's valid. cast[T](obj) and NewType[T](name) would have made more sense :(

[gvanrossum]

Just send a PR already. :-)

On Tue, Sep 13, 2016 at 9:51 AM, Elazar [email protected] wrote:

I assume you mean module: str? And why put the initializers instead of

ellipsis?

While we're at it, perhaps we should add

def cast(tp: Type[_T], obj: Any) -> _T: ...

def NewType(name: str, tp: Type[_T]) -> Type[_T]: ...

But I don't know if that's valid. castT and NewTypeT would
have made more sense :(

—
You are receiving this because you authored the thread.
Reply to this email directly, view it on GitHub
#533 (comment),
or mute the thread
https://github.com/notifications/unsubscribe-auth/ACwrMoVBWs_aqhHCH7UBhfmHysn9kC1bks5qptR5gaJpZM4J6qFv
.

--Guido van Rossum (python.org/~guido)

[gvanrossum]

Fixed by #541.

(If you want to do cast() and NewType() too, just submit another PR.)