fix: encoder not flushing data

[infiniteregrets]

hi, thank you for your work on this library!

after upgrading async-compression, our custom protocol stopped working i.e. receiving responses like for eg, ACKs would get stuck and never reach the client, this happened on release 0.4.33 so i ran git bisect:

good: b79f66d40697ab7dadf791a416ab2ba15cccc3a3
bad: 2aa1b5f8122618004b9bbab6dc679bafca616ff2

and 6c0835eb5dd26cb6176acd0455437be13fdaadac being the one where this problem happens (#402)

i went over the changes and it seemed like the loop exited before setting the counter was set so i set:

self.state = State::Encoding(read)

and it seemed to work on that commit with my sample code, but then again broken when i applied this change on main, so i ran git bisect again etc comparing against my working version and saw some behavior was changed changing AsyncRead impl back to returning Ready whenever data was written to buffer worked

idk how correct this is, but wanted to try to fix this as it did break things with Tower's compression layer enabled for us when we upgraded deps to move from 0.4.32 -> 0.4.36

happy to work on this further and improve this. appreciate any feedback to fix this (:

[NobodyXu]

Thanks!

However when I checked the diff, it seems to be no different from the latest main branch.

We already have an early return

async-compression/crates/async-compression/src/generic/bufread/decoder.rs

Line 178 in 81084f5

let mut input = PartialBuffer::new(match reader.as_mut().poll_fill_buf(cx)? {

And the read state assignment is no-op.

[NobodyXu]

cc @infiniteregrets can u try the latest release of async-compression please?

[infiniteregrets]

cc @infiniteregrets can u try the latest release of async-compression please?

Yeah the first change might just be relevant to that broken commit, I'll re run it. But other changes seem to fix my issues so I'm not sure... will get back to you

[orium]

Didn't notice this PR existed. I've fixed the same issue #456. It also includes a flushing test.