change in MultiIndex.has_duplicates behavior from 0.15.0 -> 0.15.2

[lebedov]

The following code prints False when run with Pandas 0.15.0, but True with 0.15.2; I'm using Python 2.7.6 on Ubuntu 14.04.1 in both cases. Is this change in behavior expected? (I was using has_duplicates to detect duplicate tuples in the index.)

import pandas as pd
t = [(u'x', u'out', u'z', 5, u'y', u'in', u'z', 169),
     (u'x', u'out', u'z', 7, u'y', u'in', u'z', 119),
     (u'x', u'out', u'z', 9, u'y', u'in', u'z', 135),
     (u'x', u'out', u'z', 13, u'y', u'in', u'z', 145),
     (u'x', u'out', u'z', 14, u'y', u'in', u'z', 158),
     (u'x', u'out', u'z', 16, u'y', u'in', u'z', 122),
     (u'x', u'out', u'z', 17, u'y', u'in', u'z', 160),
     (u'x', u'out', u'z', 18, u'y', u'in', u'z', 180),
     (u'x', u'out', u'z', 20, u'y', u'in', u'z', 143),
     (u'x', u'out', u'z', 21, u'y', u'in', u'z', 128),
     (u'x', u'out', u'z', 22, u'y', u'in', u'z', 129),
     (u'x', u'out', u'z', 25, u'y', u'in', u'z', 111),
     (u'x', u'out', u'z', 28, u'y', u'in', u'z', 114),
     (u'x', u'out', u'z', 29, u'y', u'in', u'z', 121),
     (u'x', u'out', u'z', 31, u'y', u'in', u'z', 126),
     (u'x', u'out', u'z', 32, u'y', u'in', u'z', 155),
     (u'x', u'out', u'z', 33, u'y', u'in', u'z', 123),
     (u'x', u'out', u'z', 12, u'y', u'in', u'z', 144)]
idx = pd.MultiIndex.from_tuples(t)
print idx.has_duplicates

Reference on StackOverflow here.

[jreback]

a bug caused by the indexer overflowing the smaller dtypes. fixed in #9077. This only had one test.

That said, the drop_duplicates/duplicated methods exist on Index objects (also as of 0.15.0.)
Another way to do this is idx.duplicated().any() which in this case is faster, though it depends on the length of the index versus the level depth.