BUG: series.fillna() fails with quoted numbers

[jlowin]

Here's a simple example:

import pandas as pd
import numpy as np

x = pd.Series(range(5))
x.iloc[1] = np.nan
x.fillna(0) # works
x.fillna('foo') # works
x.fillna('0') # crash

in the function _putmask_smart(), two numpy arrays are compared to determine if they have compatible dtypes. However if one array contains quoted numbers, then the test does NOT resolve as element wise compare, but rather as a "normal" python comparison. This means the result is False, rather than an array, and the attempt to get False.all() obviously fails.

Not sure exactly why the arrays don't compare as expected -- perhaps there are other criteria that would trigger it -- but an easy fix is to add AttributeError to the try/except clause as a "valid" error.

[jreback]

hmm, that is odd. Pls add a test for this.

[jreback]

@jlowin this needs a test and release note.

[jreback]

closing pls reopen if/when updated