This commit contains a few refactorings, but the major goal is to remove the
`Arc` that's stored inside of each `ReadinessStream` and `Scheduled` slot in the
event loop. The original purpose of this `Arc` was to share the I/O object among
the concrete handle itself and the event loop. The event loop would then change
how the socket is registered over time and then deregister it when it gets a
"shutdown request".
Nowadays, however, once an I/O object is registered with the event loop it's
never updated. Additionally, we don't actually need to call `deregister` but can
rather just instead close the I/O object itself and let the kernel/event loop
take care of the cleanup. All we need to do on deregistering is free up the slab
entry.
The major result of this commit is that I/O objects no longer need to be `Sync`
(as they're not stored in an `Arc`). Instead they just need to be `Send +
'static` as one might otherwise expect.
Along the way this also refactors a few pieces here and there to make more sense
in this new scheme. The `ReadinessStream` type now has a type parameter
indicating an owned reference to the I/O object it wraps. This can be accessed
via the `get_ref` and `get_mut` methods. Additionally I/O tokens on the event
loop are now a full-fledged `IoToken` type which we can change in the future if
we need to.
* Auto-register interest whenever we see WouldBlock
* Remove implementations of `Stream<Item=Ready>`, no longer needed
* Add explicit `poll_{read,write}` methods, if needed
* Remove all I/O streams, libstd ones suffice
* Update all I/O futures
A more appealing model is actually just automatically inferring what needs to be
scheduled based on what actions are done during poll. For example if during a
poll you check a oneshot channel, then the current task is registered for being
woken up if it's not ready. Similarly this will apply to I/O where if I/O is
attempted but we see EAGAIN then we'll schedule the task to get notified when
it's ready.
This may also have performance benefits in some niche situations because you
don't need to recompute where you are in the state machine both during poll and
during schedule. Instead, it now happens all at once.
Alex Crichton