COMPAT: breakage with just merged: https://github.com/numpy/numpy/pull/7325

[jreback]

see here
xref #12603
#12644

numpy issue 7325

cc @gfyoung
cc @njsmith

[jreback]

@gfyoung I think we just need to also accept **kwargs (and validate)

[gfyoung]

@jreback : The test needs to be modified to check which version of numpy we are using. That exception will only be thrown if we're not using numpy-dev (on numpy-dev, no exceptions will be thrown because the DataFrame or Series gets converted to a numpy array). What's the best the way to check numpy version?

[jreback]

I don't think its a test failure, rather need to accept **kwargs AS WELL as *args (similar to how we do in the stat functions). The signature is consistent across all versions of numpy. I don't want to see specific version tests, except in very specific cases.

In [1]: pd.Series.round?
Signature: pd.Series.round(self, decimals=0, *args)

[gfyoung]

Fair enough. I'll tackle this issue in a PR that will also fix #12644.

[jreback]

@gfyoung if you have a chance on this one. annoying when master fails :< (even though its the allowed failures, that's just to make it finish faster :)

[gfyoung]

@jreback : No worries. As I said above, I'll fix it in a PR that will address all of the fromnumeric.py compat issues.

[jreback]

@gfyoung ok, great.

[jreback]

@gfyoung if you are looking for something to do.....I want to fix the failing build on travis (numpy compat build) for this (can leave open for more general things).

[gfyoung]

@jreback : Alas, the fix is more complicated than just adding **kwargs. I didn't consider in #7325 the fact that a function can raise TypeError that isn't necessarily a signature compatibility issue, like when you try to take the sum of a Series with Categorical data.

xref #7491 in numpy (this PR will fix the issue on the numpy end)

[gfyoung]

My massive PR #12810 will address this breakage on the pandas side now. The combination of my numpy PR and this one should fully fix the problem.

[charris]

Does this fix the pandas breakage without any work on the pandas side? I'm somewhat annoyed that there was a problem.

OOPS: wrong PR ;)