Jason Madden

Another useful primitive might be a thread-safe way to schedule a callback in another thread's hub. (I feel like that's come up before…) That's *very* primitive, though, given the limitations...

I'm not exactly clear what the failure is (it looks like not all the lines were covered that the test expected to be?), but I can think of one definite...

On Windows, we also use the CFFI-based libuv loop implementation now, compared to the old Cython-based libev implementation. There are [some small differences](http://www.gevent.org/loop_impls.html), but nothing that I would expect to...

I have no idea why the results would change for that loop implementation. Do they also change (get fixed) with the `libev-cffi` implementation? That would point to some sort of...

[This comment](https://github.com/nedbat/coveragepy/issues/663#issuecomment-399708957), talking about the C implementation vs the Python implementation of `gevent.queue.Queue` being the difference, still makes more sense to me than it being the loop implementation.

(EDIT: This is basically all context-swapped out of my head at this point, I was just reviewing the cython changelog.) I'm wondering if https://github.com/cython/cython/issues/2776 might be related? `gevent.queue` plays tricks...

Duplicate of #1878

Thank you all for your patience and contributions! This has been merged as-of https://github.com/gevent/gevent/commit/e29bd2ee11ca5f78cc9c9bc784d875681afb8d23

It's certainly not an entirely "dumb idea!" In fact it's been requested and discussed before. It's just that there hasn't been any kind of efficient way to implement anything like...

> Ultimately there is always one (or more) data structures which encode things that are "ready" vs those that are "not ready/waiting" plus some algo that selects one of the...