Angle.to_string() returns a 0-d array with dtype=object - #1374

I've run into this before and I'm +1 on changing this to return a unicode object.

I commented on this already here: #1006 (comment)

Practically speaking, what is the downside here. Arrays of objects is, unfortunately, the "standard" way to represent variable length strings in Numpy, and Numpy provides no good way to do otherwise.

When I ask for something as a string, then I expect string operations to work:

In [10]: a = Angle('45d')
In [12]: out = a.to_string()
In [13]: out[:2]  # get the first two characters.
ERROR: ValueError: cannot slice a 0-d array [IPython.core.interactiveshell]
---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-13-d16174cd7ed4> in <module>()
----> 1 out[:2]

ValueError: cannot slice a 0-d array

What would be the downside of a final check for out.ndim == 0 in to_string() and then returning unicode(out) instead (assuming out is the result of the final format_ufunc(values) in to_string().

I see. That makes sense. It seems that in the case of an array (not scalar), this requirement already holds, so there's nothing to do there:

>>> import numpy as np
>>> x = np.array(['foo', 'bar', 'baz'], dtype=object)
>>> x[0][:2]
'fo'