Prevent double wrapping of inherited Hypothesis tests

[seifertm]

When looking at #231 I thought that something is wrong with the setup and teardown of the event_loop fixture. Running pytest with --show-setup proved that this is not the case, though:

tests/hypothesis/test_inherited_test.py 
        SETUP    F event_loop
        tests/hypothesis/test_inherited_test.py::TestOne::test_hypothesis (fixtures used: event_loop).
        TEARDOWN F event_loop
        SETUP    F event_loop
        tests/hypothesis/test_inherited_test.py::TestTwo::test_hypothesis (fixtures used: event_loop).
        TEARDOWN F event_loop

I dug further and noticed that regular async tests are executed just fine. Only tests that were using Hypothesis raised errors when they were executed in a subclass. Eventually, it turned out that subclassing caused multiple synchronous function wrappers to be layered around inherited Hypothesis tests. Here's why:

Pytest-asyncio identifies Hypothesis test cases by their is_hypothesis_test attribute. When setting up an async Hypothesis test pytest-asyncio replaces the function's hypothesis.inner_test attribute. The the top level function never changes.

When a Hypothesis test case is defined in a base class and inherited by subclasses, the test is collected in each subclass. Since the top-level Hypothesis test never changes, its inner test will be wrapped multiple times.

Double wrapping leads to execution errors caused by stale (closed) event loops in all test executions after the first. This also explains why a module scoped fixture could serve as a workaround for the OP in #231.

This change adds an original_test_function attribute to the async function wrapper, in order to keep track of the original Hypothesis test. When re-wrapping would occur in subclasses pytest-asyncio wraps the original test function rather than the wrapper function.

Closes #231

[codecov-commenter]

Codecov Report

Merging #239 (dc74a20) into master (57ccbc7) will not change coverage.
The diff coverage is 100.00%.

@@           Coverage Diff           @@
##           master     #239   +/-   ##
=======================================
  Coverage   95.83%   95.83%           
=======================================
  Files           2        2           
  Lines         144      144           
=======================================
  Hits          138      138           
  Misses          6        6           

Impacted Files	Coverage Δ
pytest_asyncio/plugin.py	95.80% <100.00%> (ø)

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 57ccbc7...dc74a20. Read the comment docs.

[seifertm]

Rebased to master. #220 addresses the same issue in a slightly different way. I adjusted the PR accordingly.

The PR now:

• Fixes a linting issue
• Links the fixes issues in the changelog
• Mentions flaky as an added test dependency in the changelog. This is an important information for downstream packagers.

[asvetlov]

LGTM