Indexing problem with in-place operator

[dieterv77]

Hi, thanks to all of you who spend time on this project, it is invaluable to me.
Unfortuatnely, i think I have run into an indexing-type issue, either that or my expectations are incorrect. Consider the following script:

import random
import pandas
import numpy as np

columns = list('abcdefg')
subcols = columns[1:-1]
X = pandas.DataFrame(0.0, columns=columns, index=range(100))
Z = pandas.DataFrame(np.random.randn(100,len(subcols)), columns=subcols, index=range(100))
block1 = list(subcols)
random.shuffle(block1)

X[block1] -= Z
print X.corrwith(Z)

X = pandas.DataFrame(0.0, columns=columns, index=range(100))
X[block1] -= Z[block1]
print X.corrwith(Z)

print pandas.__version__

The output i get is something like this (obviously depends on the random inputs)
a NaN
b -0.127386
c -0.073521
d -0.073521
e -0.127386
f -1.000000
g NaN
dtype: float64
a NaN
b -1
c -1
d -1
e -1
f -1
g NaN
dtype: float64
0.15.0rc1-10-g215569a

where the second result is what i would have expected.
I was surprised to find that I have to "reindex" Z by block1 to get a correct result. Is my expectation incorrect?

Many thanks

[jreback]

yep, something going on here.

[jreback]

@dieterv77 #8511 fixes this (and an odd aliasing bug).

Was a bit tricky acutally. We were defining __iadd__ == __add__ which normally is just fine.
The args are aligned properly and such. But python then does a __setitem__ with the original key (which is in a different order), and the aligning code couldn't intecept this. Adding __i*__ methods fixes this properly (and solves another aliasing issue from a while back, #5104)