TypeError in renaming.py

[The-Compiler]

I don't have time to dig into this right now I'm afraid, but it looks like the newest release broke pytest's build:

[...]
  File "/home/travis/build/pytest-dev/pytest/.tox/py35/lib/python3.5/site-packages/_pytest/assertion/rewrite.py", line 213, in load_module
    py.builtin.exec_(co, mod.__dict__)
  File "/home/travis/build/pytest-dev/pytest/.tox/py35/lib/python3.5/site-packages/hypothesis/strategies.py", line 1232, in <module>
    min_datetime=None, max_datetime=None,
  File "/home/travis/build/pytest-dev/pytest/.tox/py35/lib/python3.5/site-packages/hypothesis/internal/renaming.py", line 52, in accept
    'in a future version of Hypothesis.'
TypeError: unsupported operand type(s) for +=: 'NoneType' and 'str'

Travis build: https://travis-ci.org/pytest-dev/pytest/jobs/268491937#L653

Affected code which I don't really understand right now:

https://github.com/HypothesisWorks/hypothesis-python/blob/ea8f3a125adf98d90b7a9de1809f596b41880f8b/src/hypothesis/internal/renaming.py#L42-L54

[alexwlchan]

This occurs if with_name_check is a function without a docstring. I can reproduce this with the following snippet, Pythons 2 and 3:

@renamed_arguments(y='x')
def f(x):
    print(x)

So I think we just need to check that the docstring is not-None, and use the empty string if not.

[alexwlchan]

I’m surprised this wasn’t caught by the Hypothesis build, because anything we’ve decorated with that should be tested – but I wonder if something inside pytest is throwing away docstrings?

[alexwlchan]

Okay, patch available, this was an easy spot upon code read.

@The-Compiler For your benefit: renamed_arguments is a decorator we added in 3.20.0 to rename some arguments on the datetime-related strategies. Here’s an example of it in use:

https://github.com/HypothesisWorks/hypothesis-python/blob/master/src/hypothesis/strategies.py#L1233-L1242

Because this is an internal function, and only gets applied to strategies defined in the Hypothesis codebase, there’s an implicit assumption that any function it decorates will have a non-empty docstring (as we use them to generate the strategy documentation).

I haven’t seen any similar issues reported elsewhere, so it seems likely that pytest CI is stripping docstrings from our strategies. Does that sound plausible?

(Also, has this only just started failing CI? 3.20.0 is 6 days old, and I’d expect this to start causing CI failures as soon as it was released. Has something changed in pytest recently?)

[The-Compiler]

Ah, I can confirm it failed earlier, I just never noticed (I'm currently mostly ignoring mails because of exams going on). This is in a test which sets PYTHONOPTIMIZE=2 which throws away docstrings.