WakerRegistrar mentioned in comment is not present

[psumbera]

https://github.com/tokio-rs/mio/blame/f6a20da1c81c2d56a78bc6f6832b9904b9215914/src/sys/unix/selector/poll.rs#L267

                        // Remove the interest which just got triggered
                        // the IoSourceState/WakerRegistrar used with this selector will add back
                        // the interest using reregister.
                        poll_fd.events &= !poll_fd.revents;

when I use mio within https://github.com/rust-lang/rustup on Solaris it does not reregister (and rustup doesn't download stuff)

Looking at IoSourceState (mentioned in the comment as well) it does reregistration just in case of error!?

https://github.com/tokio-rs/mio/blob/master/src/sys/unix/selector/poll.rs#L663

@jasta, where is WakerRegistrar and how is reregistration supposed to work?

[Thomasdezeeuw]

https://github.com/tokio-rs/mio/blame/f6a20da1c81c2d56a78bc6f6832b9904b9215914/src/sys/unix/selector/poll.rs#L267

                        // Remove the interest which just got triggered
                        // the IoSourceState/WakerRegistrar used with this selector will add back
                        // the interest using reregister.
                        poll_fd.events &= !poll_fd.revents;

when I use mio within https://github.com/rust-lang/rustup on Solaris it does not reregister (and rustup doesn't download stuff)

Looking at IoSourceState (mentioned in the comment as well) it does reregistration just in case of error!?

https://github.com/tokio-rs/mio/blob/master/src/sys/unix/selector/poll.rs#L663

This is to emulate edge triggers. We remove the interest when it's triggered. For example if we get a readable event we won't poll for it again until it's reregistered, which is what happens in IoSourceState.

@jasta, where is WakerRegistrar and how is reregistration supposed to work?

I believe it was renamed to WakerInternal:

mio/src/sys/unix/waker.rs

Line 224 in f6a20da

pub struct WakerInternal {

[jasta]

Indeed, the comment must not have been updated after the big refactor. The logic that this is referring to is the call to reregister in the same file under IoSourceState::do_io. This happens through a wrapper around all I/O that checks for EWOULDBLOCK, which actually breaks in tokio for certain cases due to tokio-rs/tokio#5866.

I'll send up a PR to fix the comments though.