DOC: Update list of functions that lose units - #10848

@mhvk So you say to add a new section like this one with the title being the one you mentioned?

And the explicit example you also mention, is the one given by the author of #10836 ?

@rpsfonseca - Yes to both!

Since the issue that this PR fixed was labeled as Hacktoberfest, I applied "hacktoberfest-accepted" label here. I am still not fully understanding how the new fest rules work... cc @astrojuanlu

Thank you, all!

thanks @pllim ! I was about to do it myself 😊

For future reference, please do not skip CI when there is real code in the docs. This caused real failures, which I am patching it over at #10874 . FYI.

Thank you for the help!
And sorry about skipping CI, I checked the CONTRIBUTING guide, and it mentioned that on documentation we could skip CI. But I will keep that in mind for future contributions.

@rpsfonseca - no worries, when I reviewed I noticed the few checks but that readthedocs was done, and I also thought that was OK. Having been around, I should have known better...

Question: would you be willing to adjust the CONTRIBUTING guide, adding a line about not skipping if there are doctests? (And for that one we can actually skip CI!)

Perhaps this line in CONTRIBUTING.md confusing: If your commit makes substantial changes to the documentation but no code changes (... the key is "no code changes" but changing code in doc is basically code changes because we test it.)

I also added something in https://docs.astropy.org/en/latest/development/docguide.html#adding-a-git-commit .

If any of those could be less confusing, please feel free to propose changes as PR. Thanks!

@mhvk I'd love to help on that for sure!
@pllim Yes, that was the line the lead me into the "mistake".

I will look into that and propose a change and check your opinion.

@mhvk , @eteq and I just found out that the UnitConversionError doesn't happen in Numpy 1.16.x, as with in Numpy 1.20.x. Is this expected? We dropped Numpy 1.16 on master but still tests against it on older branches.

>>> import numpy as np
>>> import astropy
>>> from astropy import units as u
>>> np.__version__
'1.16.5'
>>> astropy.__version__
'4.0.3'
>>> x = 1. * u.m
>>> np.full(10, x)
array([1., 1., 1., 1., 1., 1., 1., 1., 1., 1.])

@pllim - I think 1.16 just completely ignored the unit, without even realizing it.
But I guess that means the text of the docs is still not quite right...

that means the text of the docs is still not quite right

We dropped numpy 1.16 in #10664 , so it is non-issue for master but for the sake of making the release out, Erik and I decided to just skip that doctest in 4.0.x and 4.1.x branches.

Sorry for taking this long, but I opened a PR to address the confusing tips on CONTRIBUTING.md, as we talked about.