Dict constructor typing is incorrect (too restrictive)

[beamerblvd]

The dict type has several supported constructor signatures. The one in question is this:

def __init__(self, iterable: Iterable[Tuple[_KT, _VT]], **kwargs: _VT) -> None: ...

I believe this signature is improperly restrictive. Consider this code:

parameters = dict(
    map(str.strip, p.split('=', 1)) if '=' in p else (p.strip(), None)
    for p in encoded_parameters.split(';')
) if encoded_parameters else {}  # type: Dict

This is perfectly valid and works well. The code in the dict constructor doesn't actually require the iterable argument to contain two-tuples. It merely requires the iterable argument to contain values that can be unpacked each into exactly two values. (It's for k, v in iterable: ...). It can be a tuple, a list, or any iterable. However, my static type checker (mypy) complains that the types are incompatible (says it must be an iterable of Tuple[Any, Any]).

I'm not, however, certain what the ideal signature would be. Unfortunately, there's no Unpackable type. Some possible options:

def __init__(self, iterable: Iterable[Iterable[Union[_KT, _VT]]], **kwargs: _VT) -> None: ...
def __init__(self, iterable: Iterable[Iterable[Any]], **kwargs: _VT) -> None: ...

Thoughts?

[srittau]

Unfortunately I don't think that there is anything that typeshed can do at the moment. Type checkers can not assert that map() returns an iterable with at least two items of the correct type. I have opened discussion about this in python/typing#592.