Why does pyright deduce Mapping[Unknown, Unknown] in this case?

[Holt59]

I have the following function that I would like to use to convert either a mapping or a float to a mapping (with defaultdict):

def asdict(value: Mapping[_K, _V] | _V) -> Mapping[_K, _V]:
    if not isinstance(value, Mapping):
        return defaultdict(lambda: value)
    return value

I get a warning from pyright because in the second return, the type of value is deduced as Mapping[_K@asdict, _V@asdict] | Mapping[Unknown, Unknown]*.

Why is the type deduce like that? Is there a way to have it deduce Mapping[_K, _V]? If no, can this be safely ignored?

[erictraut]

I presume that you have strict mode enabled because pyright does not generate any type errors for this code in standard mode.

I'm guessing that _K and _V are type variables with no explicit upper bound, in which case they have implicit upper bounds of object. I can't tell from your code snippet wither _K and _V are scoped to a class. I'm guessing that they're not.

A simplified version of the function should make it more clear why pyright narrows _V to Mapping[Unknown, Unknown] in this case.

def func[_V](value: _V):
    if isinstance(value, Mapping):
        reveal_type(value) # Mapping[Unknown, Unknown]

Here value could be of any type including a Mapping. The type arguments are unknown, so pyright must assume that the type is Mapping[Unknown, Unknown].

[Holt59]

Thanks for the clarification, and your hypothesis are right (_K and _V have no bounds and are no scoped).

If I summarize, the issue is because _V could be a mapping type itself?