[C++/Python] Casting float NaN to int should raise an error on safe cast

[asfimport]

Currently the following code works:

pa.array([np.nan, 1.]).cast(pa.int64())

It produces:

<pyarrow.lib.Int64Array object at 0x111ca6778>
[
  -9223372036854775808,
  1
]

The expected behaviour should be that it raises.

Reporter: Uwe Korn / @xhochy
Assignee: Krisztian Szucs / @kszucs

PRs and other links:

• GitHub Pull Request #2612

_{Note: This issue was originally created as ARROW-2854. Please see the migration documentation for further details.}

[asfimport]

Wes McKinney / @wesm:
Agreed. Closely related to ARROW-488

[asfimport]

Antoine Pitrou / @pitrou:
Retargetting to 0.11.0 as we'll probably want to merge the fix for ARROW-488 first.

[asfimport]

Wes McKinney / @wesm:
Should we push this off or try to tackle for 0.11?

[asfimport]

Krisztian Szucs / @kszucs:
ARROW-3158: [C++] Handle float truncation during casting has resolved this case:

>>> pa.array([np.nan, 1.]).cast(pa.int64())
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "pyarrow/array.pxi", line 413, in pyarrow.lib.Array.cast
    check_status(Cast(_context(), self.ap[0], type.sp_type,
  File "pyarrow/error.pxi", line 81, in pyarrow.lib.check_status
    raise ArrowInvalid(message)
pyarrow.lib.ArrowInvalid: Floating point value truncated

>>> pa.array([np.nan, 1.]).cast(pa.int64(), safe=False)
<pyarrow.lib.Int64Array object at 0x116155778>
[
  -9223372036854775808,
  1
]

Of course the error message could be better.

@xhochy can We close this?

[asfimport]

Wes McKinney / @wesm:
Do we have a unit test for this specifically?

[asfimport]

Uwe Korn / @xhochy:
@kszucs We can close this when we have an unit test for this.

[asfimport]

Wes McKinney / @wesm:
Unit test in cc8d636