Don't show @_ in the error message

[asukaminato0721]

Describe the Bug

  Attribute `__iter__` has incompatible types: `Generator[LLMResultChunkWithStructuredOutput, None, None] | LLMResultWithStructuredOutput.__iter__` has type `BoundMethod[LLMResultWithStructuredOutput, (self: LLMResultWithStructuredOutput) -> Generator[tuple[str, Any], None, None]]`, which is not assignable to `BoundMethod[Generator[LLMResultChunkWithStructuredOutput, None, None] | LLMResultWithStructuredOutput, (self: Generator[LLMResultChunkWithStructuredOutput, None, None] | LLMResultWithStructuredOutput) -> Generator[@_, @_, @_]]`, the type of `Generator.__iter__`

it comes from

   --> core/plugin/backwards_invocation/model.py:124:33
    |
124 |         if isinstance(response, Generator):
    |                                 ^^^^^^^^^
    |

in https://github.com/langgenius/dify, in case someone needed.

Sandbox Link

No response

(Only applicable for extension issues) IDE Information

No response

[yangdanny97]

It's an unpinned type variable in partial type inference

Here's an example, if you have isinstance(x, Iterable) and x is a list[int], the implementation of isinstance checks that list[int] is a subtype of Iterable[@_], which pins the var to int. Then we are able to know that x is Iterable[int] inside the if body

Same thing applies to generator here

IDK the specifics of that error or whether it's a false positive or not tho

[asukaminato0721]

Then I think the error message can be improved a bit :P

[yangdanny97]

Yes, the fact that it shows up in the error message is leaking an implementation detail.

[yangdanny97]

Substituting it with Any isn't more clear IMO, so we should come up with something other than that