FrozenNDArray cannot repr scalars

[behzadnouri]

on master:

>>> from pandas.core.base import FrozenNDArray
>>> a = FrozenNDArray(1)
>>> a
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/usr/lib/python3.4/site-packages/pandas-0.15.1_192_g7bd1b24-py3.4-linux-x86_64.egg/pandas/core/base.py", line 65, in __repr__
    return str(self)
  File "/usr/lib/python3.4/site-packages/pandas-0.15.1_192_g7bd1b24-py3.4-linux-x86_64.egg/pandas/core/base.py", line 44, in __str__
    return self.__unicode__()
  File "/usr/lib/python3.4/site-packages/pandas-0.15.1_192_g7bd1b24-py3.4-linux-x86_64.egg/pandas/core/base.py", line 268, in __unicode__
...
...
RuntimeError: maximum recursion depth exceeded while calling a Python object

keeps going back and forth between __str__ and __unicode__. a workaround would be to change self on this line to self if self.ndim else self.item().

I ran into this bug while taking .max on the array:

>>> xs
FrozenNDArray([1, 2, 3], dtype='int64')
>>> xs.max()
RuntimeError: maximum recursion depth exceeded while calling a Python object

it is kind of unusual to me that here .max returns a zero dimensional array, whereas numpy returns a scalar:

>>> xs
FrozenNDArray([1, 2, 3], dtype='int64')
>>> type(xs.max())
<class 'pandas.core.base.FrozenNDArray'>
>>> type(np.array([1, 2, 3]).max())
<class 'numpy.int64'>

in particular, np.isscalar returns differently for the two:

>>> np.isscalar(np.array([1, 2, 3]).max())
True
>>> np.isscalar(FrozenNDArray([1, 2, 3]).max())
False

[jreback]

this whole notion of a FrozenNDarray can be taken out actually
it's only used in 1 place and u can just substitute a ndarray there

this was originally was needed to make index immutable but that's handled elsewhere now

[dsm054]

Can we yank FrozenNDArray now, then? I guess we'd need to make sure that pickles would still work, but apart from that, as you note, all the references seem very shallow now.

[jreback]

@dsm054 yes as I said a while back its not necessary. pickle compat is 'easy', just follow the pattern in pandas.compat.pickle_compat.py (might need a slight modification as this is not a rename but shouldn't be too hard).

[jreback]

xref #15477 and this #15506 (last was reverted, something odd changed in making the doc build take way longer).