Stack overflow on applying numpy functions to DataFrame with duplicated column entries.

[skycaptain]

Applying a numpy function, such as np.round, to a DataFrame with duplicated column indicies can cause an unrecoverable stack overflow error (Fatal Python error: Cannot recover from stack overflow.), which causes e.g. a ipython kernel to crash. E.g. take the following example, where python crashes at line 4:

x = pd.DataFrame(np.random.randn(3,3))
y = pd.DataFrame(np.random.randn(3,3))
z = pd.concat((x, y), axis=1)
print(np.round(z))

However, removing the duplicate column entries, works as expected:

...
z = pd.concat((x, y), axis=1, ignore_index=True)
print(np.round(z))

---

python 3.5.0, numpy 1.10.1, pandas 0.17.0

[jreback]

this is specifically with np.round, which ends up calling DataFrame.round, which does not handle duplicates properly.

iteration needs to use .iteritems() which correctly handles duplicate iteration, rather than column selection

pull-requests to fix are welcome

[jreback]

closed by #11618