Remove Chunk in favor of using Bytes directly

[abonander]

Compared to Bytes, Chunk is pointless and kinda frustrating:

• It's a trivial wrapper around Bytes but Bytes is exposed via the From/Into impl anyways so it's not like it's keeping it out of the public API
• Chunk adds 0 new functionality over Bytes, all of its methods forwards to Bytes anyway
• There's no way to split a Chunk without round-tripping through Bytes
• Any third-party crate that wants to build on Stream<Item = Bytes> needs special adapters to work with hyper::Body

That last one is a real problem for me because I'm finally updating multipart-async and trying to maximize interop by generifying the API as much as possible. Because hyper::Body doesn't implement AsyncRead I've instead standardized around a Stream<Item = impl BodyChunk> API where BodyChunk covers two operations: splitting and dereffing to &[u8]. My final constraint is that I want to be able to return the same impl BodyChunk type, mostly to make it easy to drop multipart-async into existing request handlers.

Without implementing BodyChunk specifically for hyper::Chunk while still having an applicable impl for it, I would need an impl that's generic over T: From<Bytes>, Bytes: From<T>, T: AsRef<[u8]>. My hangup on this is that I now cannot implement BodyChunk for &[u8] directly, which complicates testing as I have to convert all bytestring literals to Bytes. (I also want to avoid a generic impl based on Into<Bytes> as that operation may perform implicit copies and I'm trying to be near-zero-copy.)

I realize Chunk exists because it used to be a custom implementation of Bytes but at this point it can be changed to just be a type alias and basically nothing will break (except usages of Chunk::into_bytes(), ironically).

[davidbarsky]

I'm a firm +1 to this suggestion. I'd also be okay with maybe replacing Bytes usage with just Vec<u8>, but I feel less strongly about the latter point.

[abonander]

I haven't checked if BytesMut is reused internally once all strong refs are gone. I assumed that was the major reason for using Bytes in the first place.

[davidbarsky]

I think it might be, but I’m not qualified to give an answer on that at the moment. Disregard my comment about Bytes!

[seanmonstar]

This could work out. I'd just hope all the same convenient constructors match up (I think in bytes 0.4, Bytes::from(static_str) will make a copy, but that's fixed in 0.5.x).

[davidbarsky]

When you say "same convenient constructors match up", are you referring to the constructors on Bytes in 0.5.x, or something else?

[seanmonstar]

Yea, I think my concerns are fixed in 0.5.x, so I'd probably wait until that is available before making the change in hyper.

[bstrie]

@seanmonstar , I interpreted this issue as implying that you wanted to make Chunk into a type alias, but it's also possible that you intended that Chunk should be removed entirely. I've just submitted a PR for the former.