Pickled Responses should include a None value for the raw attribute

[Lukasa]

I received this issue in CacheControl where a pickled response throws an error because there is no raw attribute on the response object. The __setstate__ function should set this explicitly to an empty urllib3.response.HTTPResponse.

[Lukasa]

Yep, this is a good catch. We should fix this up.

[Lukasa]

I don't think we should add a HTTPResponse object because it violates Kenneth's condition about being too tied to urllib3. It also makes an incorrect assumption that urllib3 is always being used to provide a raw response, which is simply untrue.

I'd argue it should be present and set to None.

[Lukasa]

Can I get review from @sigmavirus24 and @ionrock?

[ionrock]

This seems to fix things in CacheControl. Thank you @Lukasa!

My reasoning for suggesting the HTTPResponse was because then someone doesn't have to check whether or not a response was cached or created outside a normal request. For example, if you used resp.raw.seek(0) you'd get an error as None obviously doesn't have a seek method.

With that said, seek doesn't work anyway! No blood, no foul.

Thanks for fixing this so quickly. I'll add some docs in CacheControl to help communicate that a cached response's raw attribute will be None.

[sigmavirus24]

Not that you needed it, but 👍. Also we should document this for others relying on pickling and using libraries like multiprocessing (which pickle objects).