yes, that is correct

Current coverage is 62.10% (diff: 50.00%)

Merging #7854 into master will increase coverage by <.01%

@@             master      #7854   diff @@
==========================================
  Files           174        174          
  Lines         56065      56067     +2   
  Methods           0          0          
  Messages          0          0          
  Branches          0          0          
==========================================
+ Hits          34820      34823     +3   
+ Misses        21245      21244     -1   
  Partials          0          0          

Powered by Codecov. Last update fa95b58...c189bf9

Should I add a new test for this code change?

Yes, a test would be good. Maybe in test_pickle.py?

I struggle a bit with creation of test for this case without pandas
do you know how I can make it easily?
not sure that adding pandas dependency is a correct approach here...

Can you create a mock object whose API is close to the one from pandas?

(pandas should not be added as a dependency).

I don't think anything should depend on Pandas, or needs to here. If the problem is pickling an rrulelocator, it should be possible to test it very directly by analogy with some other tests in the test_pickle.py module.

I agree--mocking Pandas seems a bit heavy-handed, if not wrong. If the problem is pickling, test pickling.

We have some tests that run with pandas, you can just use it. Follow the pattern from

You could, but it's not really necessary to use pandas to reproduce the problem. Something like this is enough to trigger the recursion:

>>> import matplotlib.dates
>>> import pickle
>>> r = matplotlib.dates.rrulewrapper(2)
>>> r
<matplotlib.dates.rrulewrapper object at 0x7f3c9de18470>
>>> pickle.loads(pickle.dumps(r))
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  ...
  File "/home/elliott/code/matplotlib/lib/matplotlib/dates.py", line 716, in __getattr__
    return getattr(self._rrule, name)
RecursionError: maximum recursion depth exceeded while calling a Python object

It just needs to be wrapped up in the test functions and given better names and such.

awesome, QuLogic, thank you for the example!
I will add a test now

added the test
4e08061

does anybody know how a new test could lead to failure of those checks?
it's a bit strange that code coverage decreased..

Part of the failure is our PEP8 checker, You have a blank line containing whitespaces. I suggest fixing that and let the tests rerun. I think the other failure is a random fluke

Thanks for the patch!

👍 to backporting, but it needs a bit more work as this 'fixes' the problem via

In [27]: r.__getstate__
---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
<ipython-input-27-9cc7b4e1bb10> in <module>()
----> 1 r.__getstate__

/home/tcaswell/src/p/matplotlib/lib/matplotlib/dates.py in __getattr__(self, name)
    713         self._construct.update(kwargs)
    714         self._rrule = rrule(**self._construct)
--> 715 
    716     def __getattr__(self, name):
    717         if name in self.__dict__:

AttributeError: type object 'object' has no attribute '__getattr__'