Doctest failure depending on terminal width

[astrofrog]

If I run the test suite in a large terminal I get a doctest failure:

___________________________________________________ [doctest] io.rst ____________________________________________________
104 
105     >>> from astropy import units as u
106     >>> from astropy.timeseries import BinnedTimeSeries
107     >>> binned_filename = get_pkg_data_filename('data/binned.csv',
108     ...                                          package='astropy.timeseries.tests')
109     >>> ts = BinnedTimeSeries.read(binned_filename, format='ascii.csv',
110     ...                            time_bin_start_column='time_start',
111     ...                            time_bin_size_column='bin_size',
112     ...                            time_bin_size_unit=u.s)
113     >>> ts[:3]  # doctest: +ELLIPSIS
Differences (unified diff with -expected +actual):
    @@ -1,8 +1,8 @@
     <BinnedTimeSeries length=3>
    -     time_bin_start     time_bin_size ...    E       F
    -                              s       ...
    -          Time             float64    ... float64 float64
    ------------------------ ------------- ... ------- -------
    -2016-03-22T12:30:31.000           3.0 ...   28.87   63.44
    -2016-03-22T12:30:34.000           3.0 ...   27.76   59.98
    -2016-03-22T12:30:37.000           3.0 ...   27.04   59.61
    +     time_bin_start     time_bin_size         time_end           A       B       C       D       E       F   
    +                              s                                                                              
    +          Time             float64             str23          float64 float64 float64 float64 float64 float64
    +----------------------- ------------- ----------------------- ------- ------- ------- ------- ------- -------
    +2016-03-22T12:30:31.000           3.0 2016-03-22T12:30:34.000  164.93  114.73   26.27   19.21   28.87   63.44
    +2016-03-22T12:30:34.000           3.0 2016-03-22T12:30:37.000  164.89  114.75   26.22   19.07   27.76   59.98
    +2016-03-22T12:30:37.000           3.0 2016-03-22T12:30:40.000  164.63  115.04   25.78   19.01   27.04   59.61

/home/tom/Code/astropy/docs/timeseries/io.rst:113: DocTestFailure

@taldcroft - as you must have encountered this in the table docs before, do you know of a way to prevent this?

[taldcroft]

@astrofrog - I don't recall encountering that in doc tests, likely because I usually use example tables that are short and narrow.

I notice you added # doctest: +ELLIPSIS, which in a perfect world would have done the trick here. Barring that, I'd suggest using t[:3].pprint_all() to force printing the entire table. This is basically the strategy in unit tests to ensure consistency.

[astrofrog]

@taldcroft - ah yes table suffers from this too, if you make your terminal e.g. 40 characters wide, some of the doctests fail. We should probably figure out a way to make all these more robust!

[neutrinoceros]

I vaguely recall seeing this issue a couple month back, but I can't find the duplicate, so maybe it was never reported. However, it seems related to #15828

[taldcroft]

Is there a way in doctest to have some invisible Python setup that always gets run? We could configure the terminal to be big in that case.

[neutrinoceros]

I couldn't find anything of the sort in https://docs.python.org/3/library/doctest.html or pytest-doctest-plus's documentation, so I think the answer is no. Additionally, it would probably make all doctests slightly slower and slightly harder to debug so I'm not convinced it's a desirable approach.

I notice you added # doctest: +ELLIPSIS, which in a perfect world would have done the trick here.

I actually can't understand why it does not do the trick here. Does anyone know ?

[pllim]

Maybe try removing the leading whitespace before ...? 🤷‍♀️

Or maybe Tom R should use a smaller terminal. I heard 80-char width is pretty neat. 😉

[pllim]

invisible Python setup that always gets run

Yes, https://www.sphinx-doc.org/en/master/usage/extensions/doctest.html#directives (e.g., .. testsetup::).

[taldcroft]

Yes, https://www.sphinx-doc.org/en/master/usage/extensions/doctest.html#directives (e.g., .. testsetup::).

That looks promising for exactly this situation (consistent output of a table that can depend on terminal size).

[astrofrog]

Maybe we could just set things globally in conftest.py?

[taldcroft]

Right, if a global conftest.py is actually used in doc tests then I think that would be a fine option. We could just set conf.max_width and conf.max_lines both to -1 implying no limits. Then see if any existing tests break, but hopefully not many.

[pllim]

Maybe here:

astropy/astropy/conftest.py

Line 71 in 802dc27

def pytest_configure(config):

But maybe also a good practice to unset it after testing is done?

astropy/astropy/conftest.py

Line 123 in 802dc27

def pytest_unconfigure(config):

[saimn]

Reminds me #8691 :)

[pllim]

Ah, so this and #8691 are the same problem? Should we close one of them as duplicate? Which one?