WIP: Cleanup pandas tests

[znicholls]

An attempt to clean the pandas tests up a bit. @hgrecco I'm still a bit unsure whether to use HAS_PANDAS or set pd = None if there's no Pandas available. It's not that clear which convention to use where and it seems like pd = None implies there is no pandas available (like np = None implies no numpy available).

[hgrecco]

Do you need help?

[znicholls]

Yep. The easy question: how are e.g. HAS_NUMPY and pd.compat.np related?

The harder question, is it possible to clean up the pandas related tests. I wanted to get rid of this massive if clause that we currently have but can't work out how. I've arrived at trying to use SkipTest, but that's failing for Python2.7 and Python3.3 and I have no idea why/how to fix it. Any ideas are appreciated but it may just be overkill.

[hgrecco]

The idea is that for a feature X there is an HAS_X boolean that tell us that the feature is available. Additionally (and certainly redundant) we provide all the classes / modules equal to None just to allow any rogue import to work. But the ideas is that everybody test agains HAS_X

A way out of this problen (annoying but that would work) is to migrate to UnitTest. The conditional testing there works flawlesly and you can use the current NumPy test in pint as example.

[znicholls]

A way out of this problen (annoying but that would work) is to migrate to UnitTest

Sadly I can't do that as all the underlying pandas tests are written with pytest.

What do you think about moving all the pandas stuff out into a separate, wrapper repo in the style of cyberpandas https://github.com/ContinuumIO/cyberpandas? It would mean that Pint stays clean and dependency free and all the nasty dependency handling can go elsewhere? I would set things up so that the tests are run daily to tell us fairly quickly if any Pint changes break the pandas interface.

[hgrecco]

Sounda good to me. Maybe we can create pint associates packages like pint-pandas Hernán

…

On Thu, Dec 13, 2018, 4:01 AM Zeb Nicholls ***@***.*** wrote: A way out of this problen (annoying but that would work) is to migrate to UnitTest Sadly I can't do that as all the underlying pandas tests are written with pytest. What do you think about moving all the pandas stuff out into a separate, wrapper repo in the style of cyberpandas https://github.com/ContinuumIO/cyberpandas? It would mean that Pint stays clean and dependency free and all the nasty dependency handling can go elsewhere? — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#741 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AARAJjCSBJ_ydqNUIhF8YqBXuVmzCaXzks5u4ftmgaJpZM4Y80s5> .

[andrewgsavage]

Give us a shout when you do that as I've made a branch that's passing pandas tests again.

[hgrecco]

I created a stub repo in https://github.com/hgrecco/pint-pandas

[hgrecco]

I think we can close this, right?

[znicholls]

Yes if you're happy to go down the route of a new repository

…

On Sun, Dec 30, 2018 at 3:30 PM Hernan Grecco ***@***.***> wrote: I think we can close this, right? — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#741 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AWh-myBK29yR5psXctzbETkPTBREX-c_ks5u-EF0gaJpZM4Y80s5> .