Fix #12564 for Categorical: consistent result if comparing as DataFrame

[facaiy]

• closes Categorical equality check raises ValueError in DataFrame #12564
• tests added / passed
• passes git diff upstream/master | flake8 --diff
• whatsnew entry

[jreback]

I don't think you need all of these extra checks which are block specific. does it break when you take them out?

[facaiy]

1. x.ndim < self.ndim
   if x is a normal object, _block_shape will increase x.ndim so that ValueError: Wrong number of dimensions raises. Thus I check x.ndim < self.ndim.
2. getattr(self, 'is_categorical', None)
   Because we expect that x.ndim < self.ndim should only occur in NonConsolidatableMixIn class, so if it is not the case, I believe that it might be a bug.

[jreback]

these are too specific here for this type of general routine. you might want to do this in _try_coerce_results for only Categoricals.

[facaiy]

Ok, I will take a look at _try_coerce_results. Thanks for your advice.

[jreback]

pls add a whatsnew note in Bug Fixes as well (for 0.18.1)

[facaiy]

ok, I'll add a whatsnew note by tomorrow.

[jreback]

you don't need this

[facaiy]

Thanks for your advice, @jreback.
Is it OK? and do I need to rebase for resolving conflicts?

[TomAugspurger]

@ningchi if you could that'd be great. It's probably just a conflict in the whatsnew file. And if your comfortable you can squash your commits down to a nice single message too.

[jreback]

yes, pls rebase

[facaiy]

many thanks! I have rebased these commits.

[jreback]

thanks @ningchi