This has new enough kernel headers that libev builds the io_uring
backend, but testing it fails:

  ======================================================================
220  ERROR: test_linux_iouring_0 (gevent.tests.test__close_backend_fd.Test)
221  ----------------------------------------------------------------------
222  Traceback (most recent call last):
223    File "/home/travis/.runtimes/snakepit/2.7.d/lib/python2.7/site-packages/gevent/tests/test__close_backend_fd.py", line 77, in test
224      self._check_backend(backend)
225    File "/home/travis/.runtimes/snakepit/2.7.d/lib/python2.7/site-packages/gevent/tests/test__close_backend_fd.py", line 45, in _check_backend
226      hub = Hub(backend, default=False)
227    File "/home/travis/.runtimes/snakepit/2.7.d/lib/python2.7/site-packages/gevent/hub.py", line 441, in __init__
228      self.loop = self.loop_class(flags=loop, default=default) # pylint:disable=not-callable
229    File "src/gevent/libev/corecext.pyx", line 444, in gevent.libev.corecext.loop.__cinit__
230      raise SystemError("ev_loop_new(%s) failed" % (c_flags, ))
231  SystemError: ev_loop_new(50331776) failed

There are also different SSL errors (of course) on different versions.
gevent.tests.test__makefile_ref.TestSSL commonly fails to find
expected open files on most versions, and some versions don't see the
cipher suites they expect, etc.

https://ci.appveyor.com/project/denik/gevent/builds/33014124/job/lug3fye0ko2gbe7q

checking whether make supports nested variables... (cached) yes
1067./configure: line 12686: syntax error near unexpected token AX_PTHREAD('

https://travis-ci.org/github/gevent/gevent/jobs/689397295

Update to Cython 3.0a5. Fixes #1599

Fix issue 1437 by detecting cross-thread lock usage and using an async watcher

Tests for semaphore are passing in both pure and compiled code. gevent.tests.test__lock.TestLockMultiThread still exposes a race condition though.

Rename C accelerator modules to avoid clashes.

Fixes #1480.

Fixes #1600

[skip ci]

Fix calling gevent.wait() on a set Event twice

The detection of when to de-queue an existing notifier was wrong, leading us to de-queue it while it was running and then promptly re-queue one.

Fixed in https://github.com/rthalley/dnspython/commit/0d018c3b94a3aed79508d7aa2bd8bd61a248b5d7

Make waiters on Events that arrive while the event is ready and notif…

Make waiters on Events that arrive while the event is ready and notifying greenlets that had to block run after they do.

Fixes #1520

Update PyPy on Travis

Fixes #1569

Properly clean up native resources in gevent.signal_handler() 

Fixes #1606

Also fix reporting errors from threadpool tasks, and make waiting for threadpool tasks a bit cheaper.

Let the greenlets be freed when destroying a hub