Series.where return unmodified series in upcasting corner case

[dalejung]

import pandas as pd
import numpy as np
s = pd.Series(range(10))
s.where(s > 4, np.array([np.nan]))

Out[4]:
0    0
1    1
2    2
3    3
4    4
5    5
6    6
7    7
8    8
9    9
dtype: int64

The culprit is that if upcast is required, ser doesn't get changed in the com call. Since we're making a copy of self anyways, shouldn't we always pass in change?

        change = ser if inplace else None
        com._maybe_upcast_putmask(ser,~cond,other,change=change)

I didn't hit this organically and I'm assuming it's a fairly contrived example. Was just reading code and found this.

[jreback]

this is related to this (sort of): #3217

[jreback]

fyi, this is equivalent to doing:

s[s>4] = np.nan

[dalejung]

yea, it's not something I'd likely run across. I was actually planning to use com._maybe_upcast_putmask, saw its behavior, went to see how it was used, and then couldn't figure out how the quoted lines form Series.where would return the correct results when upcasting and inplace=False.

[jreback]

closing as fixed in master