Make Request and Response picklable

[hannseman]

Refs: #1562

I took the liberty of creating a PR as I felt that there was a consensus about implementing this.

Some open questions is what state the instances should end up in after being loaded. I.e should we override is_closed, is_stream_consumed and if so to what? I also think we need to set stream to something to avoid unexpected AttributeError being raised on access. Currently set it to an empty ByteStream to make mypy happy.

It might also be cleaner to set the overrides in the dict produced in __getstate__ instead of directly in __setstate__.

An alternative to the __attrs__ pattern would be to pop invalid attributes from __dict__ in __getstate__ but being explicit about it feels safer.

[lovelydinosaur]

Okay, so thoughts here...

Firstly, I don't really like the __attrs__ thing. Although it looks like a Python built-in, it turns out that's just a bit of requests internal naming that they decided on.

I think we probably want to approach this like so:

classUnattachedStream(AsyncByteStream, SyncByteStream):
""" If a request or response is serialized using pickle, then it is no longer attached to a stream for I/O purposes. Any stream operations should result in `httpx.StreamClosed`. """def__iter__(self) ->Iterator[bytes]:
raiseStreamClosed()
asyncdef__aiter__(self) ->AsyncIterator[bytes]:
raiseStreamClosed()

classResponse:
...
def__getstate__(self):
return {
name: valueforname, valueinself.__dict__.items()
ifnamenotin ['stream', 'is_closed', '_decoder']
}
def__setstate__(self, state):
forname, valueinstate.items():
setattr(self, name, value)
self.is_closed=Trueself.stream=UnattachedStream()

classRequest:
...
def__getstate__(self):
return {
name: valueforname, valueinself.__dict__.items()
ifnamenotin ['stream']
}
def__setstate__(self, state):
forname, valueinstate.items():
setattr(self, name, value)
self.stream=UnattachedStream()

I'd consider #1584 to be a pre-requisite.

The tidying up in #1583 is also relevant here.

(Aside: We might well end up with an is_closed on the Request class too at some point in the future, but we don't really need to consider that here.)

[hannseman]

@tomchristie thanks for the review. I agree that the __attrs__ thing felt a bit awkward, much better 👍

[hannseman]

We need the yield here to avoid TypeError:

 async def aread(self) -> bytes:
"""
Read and return the request content.
"""
if not hasattr(self, "_content"):
assert isinstance(self.stream, typing.AsyncIterable)
> self._content = b"".join([part async for part in self.stream])
E TypeError: 'async for' received an object from __aiter__ that does not implement __anext__: coroutine

[lovelydinosaur]

Right, let's go with this.

Great work @hannseman!

[hannseman]

@tomchristie thanks a lot for the review and fixups! 🎉