BUG: MultiIndex needs to validate length of levels on setting.

[jtratner]

related #4687

Bad errors and/or not checking:

>>> ind = tm.makeCustomIndex(5, 5)
>>> ind.set_levels([[1, 2]]) # Wrong Error
Traceback
    ...
ValueError: Length of names (5) must be same as level (1)
>>> ind = tm.makeCustomIndex(5, 5)
>>> ind.set_levels([[1, 2]], inplace=True) # Wrong Error
Traceback
    ...
ValueError: Length of names (5) must be same as level (1)
>>> ind = tm.makeCustomIndex(5, 5)
>>> ind.levels = [[1, 2]] # Wrong Error
Traceback
    ...
ValueError: Length of names (5) must be same as level (1)

[jreback]

What's wrong with these errors? they are informative

[jtratner]

Well, it's more just that it's not validating levels, so you can end up
with the wrong level length accidentally. These examples might be wrong,
but the overall point that labels but not levels are checked is valid (and,
in particular, needs to check that, for each level/label pair, level.max()
<= len(label)). One upside of refactoring Index to not be ndarray (and use
a _data attribute) is that we can skip these checks on slicing or with
fastpath=True.

[jreback]

ok.....leaving open in 0.13 then

[jtratner]

I looked at this again and the reason its wrong is because names are (generally) irrelevant. Instead it's the fact that there is a level/label mismatch.

[jreback]

@jtratner in 0.13?

[jtratner]

Yes, I'm adding to my near-term todo list now. :)