Using _Metadata as a typing

[paulefoe]

Hi, thanks for the great library, we are really enjoying using it!

My question is regarding _Metadata being a protected member of the module. Is there any reason for this?
Because sometimes we want to manipulate metadata inside our business logic and we send it as an argument to a function.
Then, trying to annotate our function with types we would get a warning from IDE that it is protected.
I would be glad to provide PR if you think this is okay?

Here's the small example:

async def hello_world(
        self,
        stream: Stream[HelloWorldRequest, HelloWorldResponse]
) -> None:
        request = await stream.recv_message()
        assert request is not None
        result = await say_hello_with_metadata(
              request.name,
              stream.metadata,
        )
        await stream.send_message(result)

async def say_hello_with_metadata(name: str, metadata: _Metadata) -> str:
        ...

[vmagamedov]

Hi, thanks!

I didn't wanted to expose implementation details. Metadata should be an abstract multidict-like data structure from a user's perspective, not a specific type – multidict.Multidict.

There are two possible solutions:

• if you don't expect keys with multiple values you can use the typing.Mapping type, because Multidict implements the Mapping protocol and mypy should be ok with that
• if you expect multiple values you can implement a class Multidict(typing.Protocol): protocol and use it, mypy should be also ok with that

I don't want to add this protocol into grpclib because protocols were introduced only in Python 3.8 or you should depend on typing_extensions, and I think that there should be other place for such generic protocols.

[paulefoe]

Great, thank you for the quick reply! Will do just that. Have a nice day!

I'm closing it.