BUG: After reindex(), astype() raises exception for categorical series

[junjiecai]

Hi

I meet an unexpected behavior for categorical series

After reindex() a categorical series, astype() will raise Exception

For example, code below raise ValueError: Wrong number of dimensions

s = Series([1, 2, 3], index = [2, 0, 1])
(s.astype('category')
 .reindex([0, 1, 2])
 .astype(str))

However, if I reindex() first then set the series categorical, astype() works as expected.
Code below works fine

s = Series([1, 2, 3], index = [2, 0, 1])
(s.reindex([0, 1, 2])
 .astype('category')
 .astype(str))

I think both code should behave the same way

[jreback]

hmm this is a bug, not sure exactly, but I think something is getting created in an odd way.

[alanbato]

I'm looking into it, and it seems that the bug is due to its CategoricalBlock step having ndim evaluating to 2, whereas its values.ndim evaluate to 1. I think it has to do with how a Block object has an ndim of 2 by default.
Maybe we aren't sending the ndim parameter and we should, or is this supposed to be correct?

[mroeschke]

Looks to work on master. Could use a test.

In [30]: s = Series([1, 2, 3], index = [2, 0, 1])
    ...: (s.astype('category')
    ...:  .reindex([0, 1, 2])
    ...:  .astype(str))
Out[30]:
0    2
1    3
2    1
dtype: object

In [31]: pd.__version__
Out[31]: '0.26.0.dev0+565.g8c5941cd5'