bytes::BytesMut::extend_from_slice or mio::net::TcpSocket produces unexpected result

[softstream-link]

Ok, so I have extracted a "minimum" amount of code to replicate it and posted it on HackMD with details of cargo toml and actual error message. This is quite a bit of code to go thought but here is the overall structure

1. This has one example file called /examples/bytes_bug.rs

2. The file has two #[test]'s so you can run each one independently

   a) test 1

   called test_emulated() and it writes a fixed size, random bytes frame but arbitrarily splitting it in "half" and then tries to reconstruct the frame but checking if enough bytes are available in the bytes::BytesMut buffer. It seems to never fail on either linux or Mac

   b) test 2

   called test_socket() and this test has two thread

   a new thread which runs server socket, it will read bytes and try to reassemble and count frames.

   the main thread has a client socket and will write N frames, N has to be set pretty high, to get a failure.

   Note: in this test, both read_frame & write_frame take place via a helper struct which will indicate whether io operation completed or socket was not ready. This is due to socket being intentionally non-blocking.

3. Post:
   bytes_bug - HackMD

It never fails on macOS

[softstream-link]

closing in favor of rust-lang/rust#115451