Refs #1207 and #1209.

[skip ci]

[skip ci]

Run leakchecks on Python 3.7.

Fix ABCs in the MRO of gevent.local. Fixes #1201

Try to make it take less than the 200s timeout.

Update config.guess/sub to latest versions.

Fixes #1197

Switch to objgraph to handle the measurement for us. That cleared up a
few of the obscure issues with references to
functions/getset_descriptors and the like. (Possibly because it keys
by string names and we were keeping type objects alive.)

Many of the real failures were due to re-using exception instances,
which is bad because of chaining.

Most of the @ignore_leakcheck are for performance, only one is for a
real issue---and that test was skipped already on CI anyway for being
too flaky.

[skip ci]

[skip ci]

Sigh, the timer resolution is really flaky on CI. Patch test__core_timer.py:test_resolution to use the mechanism to handle this.

As suggested by https://packaging.python.org/guides/making-a-pypi-friendly-readme/ and https://github.com/pypa/warehouse/issues/3884#issuecomment-386188844, this may now be necessary on warehouse.

Adding it to setup.py results in 'distutils/dist.py:274: UserWarning: Unknown distribution option: 'long_description_content_type'' even with setuptools 39, but setup.cfg is equivalent. I verified that PKG-INFO has the appropriate 'Description-Content-Type: text/x-rst' field.

Queue all libuv callbacks and run them at the end of the iteration

Invalid FDs were getting through because of the order in which we ran loop callbacks.

This solves the timer delay. Add a specific test for this.

We're still sensitive to exactly when these run, for reasons that
aren't clear (comment inline). I want to spend some more time
investigating that.

This disables user-access to prepare watchers on libuv (otherwise
we're right back where we started). It could be worked around if
needed without *too* much trouble.

This should eliminate the need to patch and the need to avoid 0 duration timers.

This works more like libev. We could theoretically implement priorities using this system.

lock: fix typo [skip ci]

It just bothers me so much whenever I hit it, sorry

Compile the monitor greenlet tracer with Cython

This makes things 54% faster. In fact, the monitor tracer is now faster than a trivial tracer implemented in python (settrace(lambda e, a: None)).

+-------------------+-----------------+-----------------------------+
| Benchmark         | 37_bench_tracer | 37_bench_tracer_cython_opt3 |
+===================+=================+=============================+
| monitor tracer    | 1.62 us         | 739 ns: 2.20x faster (-54%) |
+-------------------+-----------------+-----------------------------+
| max switch tracer | 3.06 us         | 874 ns: 3.50x faster (-71%) |
+-------------------+-----------------+-----------------------------+
| hub switch tracer | 2.16 us         | 815 ns: 2.66x faster (-62%) |
+-------------------+-----------------+-----------------------------+

Not significant (2): no tracer; trivial tracer

Unoptimized still makes them 25% faster:

+-------------------+-----------------+------------------------------+
| Benchmark         | 37_bench_tracer | 37_bench_tracer_first_cython |
+===================+=================+==============================+
| trivial tracer    | 792 ns          | 786 ns: 1.01x faster (-1%)   |
+-------------------+-----------------+------------------------------+
| monitor tracer    | 1.62 us         | 1.24 us: 1.31x faster (-24%) |
+-------------------+-----------------+------------------------------+
| max switch tracer | 3.06 us         | 1.89 us: 1.62x faster (-38%) |
+-------------------+-----------------+------------------------------+
| hub switch tracer | 2.16 us         | 1.68 us: 1.29x faster (-22%) |
+-------------------+-----------------+------------------------------+

Current numbers on 3.7b3:

no tracer: Mean +- std dev: 414 ns +- 10 ns
trivial tracer: Mean +- std dev: 792 ns +- 16 ns
monitor tracer: Mean +- std dev: 1.62 us +- 0.12 us
max switch tracer: Mean +- std dev: 3.06 us +- 0.12 us
hub switch tracer: Mean +- std dev: 2.16 us +- 0.04 us

Fix a typo [skip ci]

Make the libuv run QUEUE part of the loop.

Benchmarking (link in the email) showed that malloc/free had substantial and widely varying overhead.

I didn't really see much of a difference in the gevent benchmarks, but I didn't run them all. However, if any patch gets upstreamed, it will probably be something like this.

The link referenced in the email contains the discussion on the libuv mailing list.