[C++] power_checked incorrectly returns NaN

[rohanjain101]

Describe the bug, including details regarding any error messages, version, and platform.

>>> a = pa.array([-117], type=pa.int8())
>>> b = pa.array([7], type=pa.decimal128(38,18))
>>> pa.compute.power_checked(a, b)
<pyarrow.lib.DoubleArray object at 0x000001F398AF9F00>
[
  nan
]
>>>

I would expect this to return -300124211606973, not NaN. If B is re-scaled, then it returns correct result:

>>> b = pa.array([7], type=pa.decimal128(3,2))
>>> pa.compute.power_checked(a, b)
<pyarrow.lib.DoubleArray object at 0x000001F398AF8DC0>
[
  -3.00124211606973e+14
]
>>>

Component(s)

Python

[js8544]

It was due to an inaccurate cast from decimal to float.

>>> b = pa.array([7], type=pa.decimal128(38,18))
>>> pc.cast(b, pa.float64())
<pyarrow.lib.DoubleArray object at 0x7fbb78518d00>
[
  7.000000000000001
]

Powering a negative number to a non-integer number results in NaN.
The casting issue is solved by #35997, so this error won't happen any more starting from release 13.0.

[rohanjain101]

Thanks for checking, do you know when release 13.0 will be available?

[mapleFU]

Would coming soon, 13.0.0 code freeze seems done yet, I guess it could be release about this month

[js8544]

Thanks for checking, do you know when release 13.0 will be available?

Actually, I was wrong. The PR I mentioned solves Float->Decimal, not Decimal->Float in your case. This is another issue #35942 which hasn't been resolved yet. I'll take a look at this issue today.

[js8544]

@rohanjain101 Hi, this was fixed in #36667, but will unfortunately only be available since pyarrow 14, which should be released in a few months. If you need this fix now, you can use the nightly pyarrow package.