Deduplicate generic::bufread::Decoder impl of tokio/futures-io

[NobodyXu]

Ref #384

[NobodyXu]

Some explanation on my design/API choice

[NobodyXu]

cc @robjtede I made this design choice since

• for async I/O inlining usually doesn't help with performance, as it is usually the I/O that is bottleneck and inlining doesn't eliminate it
• it would avoid monomophization

[NobodyXu]

I would also say that even for in-memory objects or AsyncBufRead with (de)compression it wouldn't matter, because the AsyncBufRead::poll_fill_buf returns a &[u8], AFAIK this is something inline cannot optimize out since it involves a buffer and our de/encoder impl often involves complex stuff.

The on,y case it could be bottleneck is when poll_fill_buf returns very small slice or the (de)compression only consumes a very small part of it, like literally a few bytes, but that would usually mean the user is using a very terrible BufReader implementation

[NobodyXu]

I've think of an alternative design which does not need to pass the reader at all, it also removes the variable first within do_poll_read, making it truly stateless:

impl Decoder {
    fn do_poll_read(
        &mut self,
        output: &mut PartialBuffer<&mut [u8]>,
        decoder: &mut impl Decode,
        input: &mut PartialBuffer<&[u8]>,
        first: bool,
    ) -> ControlFlow<Result<NextAction>> {
        let mut first = true;
        loop {
            match self.state {
                Decode => {
                    // logic unchanged ...

                    if res? {
                        self.state = State::Flushing;
                    } else {
                        return ControlFlow::Continue(()); // Ask caller to call `reader.consume` and `reader.poll_fill_buf`
                    }
                }
            }
        }
    }
}

Caller of this funciton will then do:

let mut input = PartialBuffer(&[][]);
let mut first = true;
loop {
    match this.inner.do_poll_read(output, this.decoder, &mut input) {
        Continue(_) => {
            if !first {
                this.reader.consume(input.written.len());
            }
            first = false;
            input = PartialBuffer(ready!(this.reader.poll_fill_buf(cx))?);
        },
       Break(res) => break res,
    }
}

this design is much cleaner

[NobodyXu]

@robjtede I've updated the code with this new design, and use macro_rules! to further reduce duplicate code.

I quite like the current iteration, I think it'd make maintenance much, much simpler

[NobodyXu]

cc @robjtede I decided to keep the state as simple as possible, anything else can be passed by parameter

[NobodyXu]

If there's no objection, I'll merge it this evening

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 0.00%. Comparing base (fcf91fd) to head (2c24af2).
⚠️ Report is 2 commits behind head on main.

Additional details and impacted files

@@     Coverage Diff     @@
##   main   #391   +/-   ##
===========================
===========================

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[NobodyXu]

I will ping you again when releasing this change @robjtede , if you think of something you'd like me to change I can open a separate pr