Ideas for adding more features/optimizing utimeq module

[pfalcon]

First of all, currently entries lack a unique identifier, so it's not even possible for an entry, because there's no way to identify it. But actually there's a unique id stored, it's counter. So, we should switch from ordering entries by "time" to ordering them by this id, and return it on insert operation. "Time" then just becomes a payload in entry. That's not going to work, LOL.

Then, we can store pointers to entries in the heap, not the entries themselves. This will allow us to have stable addresses for entries, so we can refer to them. Of course, that takes extra memory word. We could try to compensate that by removing existing entry member. A candidate is "args" field. It's not needed for coros, only for callbacks. But we can say "callbacks with params should create a closure and store that instead".

But the reason of this ticket is trying to optimize timeout handling in uasyncio, and timeouts are handled by callbacks taking a coro as an argument. So, we would need to optimize that too. As it doesn't make to store integers as objects in utimeq, we can say that an object with low bit set is an "implicit" timeout object for coro which is stored in the object field. So, event loop handles such specially and cancels the coro on processing such an entry.

We now just need to allow a coro to cancel such implicit timeout object in its turn. For this we use "stable address" feature introduced previously. So, when timeout object is allocated, we get its stable handle, and can store that in the wait_for wrapper coro, and then call special utimeq function to "cancel" it (e.g. nullize object field).

The last optimization would be to get rid of wait_for wrapper coro, and instead store timeout object handle in the coro itself. That's of course not easy, "normal" way would be to introduce timed await on the language syntax level and make compiler spawn coro objects with extra field. Or go for compromise and recognize await wait_for(...) as a special syntax form with the expected imitations. Or can just stuff extra field into any coro %).

[pfalcon]

And if we really care about removing (timeout) entries from the queue (instead of marking them as removed), then could use beap data structure, but that has O(N^1/2) insertion complexity (but such a search complexity too, except there's nothing to search for, per the strike-thru above).