Series of unsigned dtype can't have diff() without upcast

[alvorithm]

This is not a Pandas issue but a numpy one.
see also #5230
However, it is highly confusing and it could be appropriate to fix in Pandas by promoting uint(2^x) dtypes to int(2^(x+1)) in Series.diff().

[jreback]

yep....can be fixed

In [93]: Series(np.array([3,2,0], dtype=np.uint32))
Out[93]: 
0    3
1    2
2    0
dtype: uint32

In [94]: Series(np.array([3,2,0], dtype=np.uint32)).diff()
Out[94]: 
0           NaN
1    4294967295
2    4294967294
dtype: float64

In [95]: Series(np.array([3,2,0], dtype=np.uint16)).diff()
Out[95]: 
0      NaN
1    65535
2    65534
dtype: float64

[jtratner]

The problem with upcasting is what you do with something like this:

Series(np.array([2 ** 63 + 10, 2 ** 63 + 5, 5], dtype='uint64')).diff()

[jreback]

yep...that should really be object even before the diff....

[OmerJog]

was this fixed?

[TomAugspurger]

No, it's still open.