bug on DatatimeBlock._try_coerce_args() with pandas0.13.0

[wangshun98]

When running the following example code, it throws error in DatatimeBlock._try_coerce_args()

T1 = pd.to_datetime(pd.Series([ np.nan, datetime(2012,1,1) ]))
T2 = pd.to_datetime(pd.Series([ datetime(2013,1,1), datetime(2012,1,1) ])
T1.where(~T1.isnull(), other=T2)

---

ValueError Traceback (most recent call last)
in ()
----> 1 T1.where( ~T1.isnull(), other = T2)

C:\Python64\lib\site-packages\pandas\core\generic.pyc in where(self, cond, other, inplace, axis, level, try_cast, raise_on_error)
2887 new_data = self._data.where(other, cond, align=axis is None,
2888 raise_on_error=raise_on_error,
-> 2889 try_cast=try_cast)
2890
2891 return self._constructor(new_data).finalize(self)

C:\Python64\lib\site-packages\pandas\core\internals.pyc in where(self, _args, *_kwargs)
2279
2280 def where(self, _args, *_kwargs):
-> 2281 return self.apply('where', _args, *_kwargs)
2282
2283 def eval(self, _args, *_kwargs):

C:\Python64\lib\site-packages\pandas\core\internals.pyc in apply(self, f, _args, *_kwargs)
2265
2266 else:
-> 2267 applied = getattr(blk, f)(_args, *_kwargs)
2268
2269 if isinstance(applied, list):

C:\Python64\lib\site-packages\pandas\core\internals.pyc in where(self, other, cond, align, raise_on_error, try_cast)
1076 # see if we can operate on the entire block, or need item-by-item
1077 # or if we are a single block (ndim == 1)
-> 1078 result = func(cond, values, other)
1079 if self._can_hold_na or self.ndim == 1:
1080

C:\Python64\lib\site-packages\pandas\core\internals.pyc in func(c, v, o)
1059 return v
1060
-> 1061 v, o = self._try_coerce_args(v, o)
1062 try:
1063 return self._try_coerce_result(

C:\Python64\lib\site-packages\pandas\core\internals.pyc in _try_coerce_args(self, values, other)
1518 values is always ndarra like, other may not be """
1519 values = values.view('i8')
-> 1520 if isnull(other) or (np.isscalar(other) and other == tslib.iNaT):
1521 other = tslib.iNaT
1522 elif isinstance(other, datetime):

ValueError: The truth value of an array with more than one element is ambiguous. Use a.any() or a.all()

The issue I found is the "if" condition. When input other is array-like, isnull return and the second "or" condition won't have the same size. A potential fix could be changing line 1520 to:
if np.isscalar(other) and (isnull(other) or other == tslib.iNaT):

[jreback]

was fixed on 0.13.1, IIRC in #5458, #5689

0.13.1 is out (but announcement not yet)

In [5]: T1
Out[5]: 
0          NaT
1   2012-01-01
dtype: datetime64[ns]

In [6]: T2
Out[6]: 
0   2013-01-01
1   2012-01-01
dtype: datetime64[ns]

In [7]: T1.where(~T1.isnull(), other=T2)
Out[7]: 
0   2013-01-01
1   2012-01-01
dtype: datetime64[ns]