Update bytes to v0.6

[olix0r]

Depends on:

• Upgrade to bytes 0.6 h2#497
• Bump bytes crate to 0.6 http-body#29
• Bump bytes crate to 0.6 http#440

[Urhengulas]

This doesn't currently compile due to the tokio::ReadBuf API.

@olix0r: Thank you for taking ownership for this :) Can you please elaborate what's holding the PR back?

[olix0r]

@Urhengulas Sorry, that comment is outdated (addressed in 34e60f7). It looks like there are some remaining changes needed in examples that I can try to fix. I won't be offended if someone else takes this branch and gets it over the line, though ;)

[olix0r]

Also, this is blocked on a few other PRs:

• Update to bytes v0.6 h2#498 -- not currently passing CI
• Update bytes to v0.6 http-body#30
• Update bytes to v0.6 http#441

[paolobarbolini]

My h2 PR does pass CI hyperium/h2#497

[olix0r]

@paolobarbolini whoops! didn't mean to double up on those PRs, I just didn't look before branching. Closed my open PRs in favor of yours.

[paolobarbolini]

It looks like there are some remaining changes needed in examples that I can try to fix

It probably just needs bytes::buf::BufExt to be changed into bytes::Buf

[olix0r]

OK, build passes. gitcop won't tell me what it doesn't like(?!), so i'm going to rebase and force push.

[Urhengulas]

That line looks threatening..

Can you please add a comment saying what it is doing?

[olix0r]

I don't actually know what it's doing ;)

I borrowed this from https://github.com/tokio-rs/tokio/blob/d78655337a68bded305782a8a8b4ac7be42aa6a7/tokio/src/io/util/read_buf.rs#L53-L55, which was the only updated use of ReadBuf::uninit I could find.

I would appreciate feedback from someone who understands these APIs better than I do.

[Urhengulas]

@seanmonstar ?

[Urhengulas]

But let's see who introduced it 🙂

[Urhengulas]

tokio-rs/tokio#3121

[olix0r]

Thanks, added a comment summarizing that discussion.

[Urhengulas]

Perfect 😄

[seanmonstar]

Don't worry about gitcop (it's not "required"), it's just a reminder to me to fix the commit message format.

[hawkw]

An alternative to all of this (which avoids the scary-looking pointer cast that @Urhengulas was concerned about) is to just use tokio_util::io::poll_read_buf from the latest tokio-util here:

Suggested change

	let dst = self.read_buf.bytes_mut();
	// Safety: `dst` may include uninitialized memory, but `ReadBuf` will
	// never uninitialize memory that has already been initialized.
	let dst = unsafe { &mut *(dst as *mut _ as *mut [MaybeUninit<u8>]) };
	let mut buf = ReadBuf::uninit(dst);
	match Pin::new(&mut self.io).poll_read(cx, &mut buf) {
	match tokio_util::io::poll_read_buf(Pin::new(&mut self.io), cx, &mut self.read_buf) {

That would mean adding a tokio-util dependency, but it's already a transitive dependency via h2.

[Urhengulas]

That seems like a good solution!

And this is a useful function which can't be found in the docs... (code: https://github.com/tokio-rs/tokio/blob/master/tokio-util/src/lib.rs#L67-L134).

[hawkw]

I believe #2339 closes this.