TST: make _skip_if into pytest decorators

[jreback]

• _skip_if_32bit (Converted windows / 32bit skips into decorators #18693)
• _skip_if_no_mpl (TST: Skipif decorator for matplotlib #18190 #18427)
• _skip_if_mpl_1_5 (Added mpl_15 decorator #18682)
• _skip_if_no_scipy (Skipif no scipy #18794)
• _skip_if_no_lzma (Skipif no lzma ne #18820)
• _skip_if_no_xarray (Removed old xarr and ipython decorators #18814)
• _skip_if_windows_python_3 (Converted windows / 32bit skips into decorators #18693)
• _skip_if_windows (Converted windows / 32bit skips into decorators #18693)
• _skip_if_no_pathlib (TST: Skip if Decorators for Localpath and Pathlib #18765)
• _skip_if_no_localpath (TST: Skip if Decorators for Localpath and Pathlib #18765)
• skip_if_no_ne (Skipif no lzma ne #18820)
• _skip_if_has_locale (Created decorators for skip_if locale functions #18745)
• _skip_if_not_us_locale (Created decorators for skip_if locale functions #18745)
• _skip_if_no_mock
• _skip_if_no_ipython (Removed old xarr and ipython decorators #18814)
• skip_if_no_package

we should move the _skip_if_* functions out of pandas.util.testing to another (private module)

then we can add skipif decorators

and use like this

@skip_if_windows_py3
def test_.......():

rather than calling tm._skip_if_windows_py390 in the body of the function (sometimes you also need to do that, so we leave the functions themselves as well).

this makes much more idiomatic and readable pytest code and removes the need to roll your own when using the decorator.

[WillAyd]

I could take a stab at this. What private module do you want the decorators to be in? Something like _test_decorators.py to enable from pandas.util._test_decorators import *?

[jreback]

i wouldn’t import * them but the name is ok

[WillAyd]

For the skipif decorators, we need to provide a function that evaluates to True to get it to skip whatever is being wrapped. A lot of the pandas-internal skip functions use pytest's importorskip, which raises a Skipped when the required module cannot be imported.

I would propose here to catch a Skipped in the decorator function and return True if that occurs. Sample code is provided below to illustrate for mpl.

Before I go too far - do you agree with this direction? I'm somewhat hesitant to use it given the Skipped exception needs to be imported from _pytest. However, the alternative would be to refactor the existing functions to not use importorskip, which has its own downsides. FWIW if we wanted to go with the try...except methodology I'd create a wrapper function internal to the _test_decorators module to handle that rather than coding that block for each decorator that we want to create

import pytest
from _pytest.outcomes import Skipped

def _skip_if_no_mpl():
    try:
        mpl = pytest.importorskip("matplotlib")
        mpl.use("Agg", warn=False)
    except Skipped:
        return True

skip_if_no_mpl = pytest.mark.skipif(_skip_if_no_mpl(), reason="No matplotlib")

[jreback]

I would refactor like your 2nd method, IOW have a safe importer that returns a boolean, then you could actually auto-create the functions & skip decorators.

[jreback]

certainly if you want a PR for a simple version would be great to see before you convert everything.

key is that has to be a separate module that is only imported when tests are run and not in pandas import (so deps are not imported until then)

[WillAyd]

EDIT - moved checklist to top of thread

[WillAyd]

@jreback as far as the max_version argument goes in the current testing module's skip_if_no_package function I only since one usage in the following location:

pandas/pandas/tests/sparse/test_frame.py

Line 1234 in 7a0ee19

tm.skip_if_no_package('scipy', max_version='0.19.0')

I believe that was introduced by 56b5a30 and I looked through the related comments in #15668 but am not clear on why that particular test needs the v <= 0.19.0 criteria. Any insight as to if that is needed or not?

FWIW that test runs on my machine with SciPy 1.0, save the dok_matrix fixture which may have an unrelated issue.

[jreback]

@WillAyd I updated the top list, but I think I missed a couple.

[max-sixty]

I'm looking at how we should skip tests in xarray.

Why did you do this rather than pytest's importorskip?

[jreback]

importerskip is a hammer
we need fine grain skipping

[WillAyd]

Some other things to consider:

• The mark option is easy to apply to functions or classes interchangeably
• If using importorskip for a class (in the setup) you lose visibility to how many tests get skipped
• The mark option gives you metadata about your tests you miss out on with importorskip