Deviation from the Faraday adapter specification

[ebababi]

Faraday adapters are expected to represent empty body response as blank string, but async-http-faraday deviates from this specification.

[ioquatix]

I am okay if you want to submit a PR to make this gem more consistent with the Faraday spec. However, there can be some difference between a response with an empty body (content-length: 0) and no body (e.g. HTTP/2 headers frame with stream close flag).

cc @iMacTia wdyt?

[ioquatix]

Released in v0.20.0.

[iMacTia]

There can be some difference between a response with an empty body (content-length: 0) and no body (e.g. HTTP/2 headers frame with stream close flag).

I'm open to exploring what it would mean in Faraday to support the nil body case, but I fear it may have important repercussions.
One possible trick we could use is to allow adapters to return nil, and then Faraday turns that into ''.
This way the behaviour is preserved, but if one day we decide to treat those scenarios in a different way, we just need to change Faraday without having to change all adapters as well