Exception hierarchy

[tomchristie]

Refs #949

• Decommissions the too-broad to be useful for anything HTTPError, in favour of an actually meaningful RequestError.
• Rearranges the hierarchy as follows, dropping the RedirectError which encompassed two totally different kinds of error (one a coding error, the other a request error), and introducing TransportError.
• Ensures consistently enforced attributes / __init__ parameters on all exceptions.

---

• RequestError Exposes .request
  • TransportError
    • TimeoutException
      · ConnectTimeout
      · ReadTimeout
      · WriteTimeout
      · PoolTimeout
    • NetworkError
      · ConnectError
      · ReadError
      · WriteError
      · CloseError
    • ProxyError
    • ProtocolError
  • DecodingError
  • TooManyRedirects
  • RequestBodyUnavailable
  • InvalidURL
• HTTPStatusError Exposes .request, .response
• NotRedirectResponse
• CookieConflict
• StreamError
  • StreamConsumed
  • ResponseNotRead
  • RequestNotRead
  • ResponseClosed

[florimondmanca]

Wonderful ✨

The split between "request errors" vs "developer errors" is a real improvement over other HTTP clients (eg Requests crams everything into RequestError, so genuine developer errors typically unnoticed due to catch-all except requests.RequestError).

[florimondmanca]

? (Applicable to all other decoders here)

Suggested change

	self.request = request
	super().__init__(request)

[tomchristie]

Yeah a guess a lot of other Python developers would tend always just rely on "hey I'm supposed to call super right". Personally, I think explicit is preferable to indirect here. Calling into super isn't always necessary or inherently the "right" thing to do.

[florimondmanca]

I see, well then I guess Decoder would really be better served by a Protocol, rather than a full-fledged base class… But we only have them on Py38+, which is unfortunate.

class Decoder(Protocol):
    request: Request

    def decode(self, data: bytes) -> bytes:
        ...

    def flush(self) -> bytes:
        ...

But maybe we could then also use the request: Request annotation on the Decoder interface, remove its constructor, and add an explicit constructor for IdentityDecoder (for consistency with other decoder classes)?

[cdeler]

MultiDecoder inherits the Decoder interface, but do not have this property

From my point of view, MultiDecoder, TextDecoder and LineDecoder should be renamed, since they:

1. on one hand these Decoders imitates the real decoders (having decode and flush)
2. on the other hand these decoders violates Liskov's substitution principle (I don't want to be so nerd, mypy pointed it for me), e.g. as they accept different types as args of decode