[Python/C++] Array constructor should not truncate floats when casting to int

[asfimport]

I would expect the following code to raise instead of truncating the float

In [4]: pa.array([1.9], type=pa.int8())
Out[4]:
<pyarrow.lib.Int8Array object at 0x113455e58>
[
  1
]

Reporter: Florian Jetter / @fjetter

_{Note: This issue was originally created as ARROW-2856. Please see the migration documentation for further details.}

[asfimport]

Antoine Pitrou / @pitrou:
This is consistent with Numpy behaviour. I would expect PyArrow to not be stricter than Numpy here:

>>> np.int8([1.1])
array([1], dtype=int8)

[asfimport]

Wes McKinney / @wesm:
I'm inclined to agree with @pitrou on this. Raising could also potentially expose the user to issues caused by small floating point errors

[asfimport]

Wes McKinney / @wesm:
NumPy does raise with NaN, which aligns with work happening in ARROW-2806

>>> np.array([np.nan], dtype='i8')
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
ValueError: cannot convert float NaN to integer

[asfimport]

Antoine Pitrou / @pitrou:
Yes, we raise on NaN as well:

>>> pa.array([float('nan')], type=pa.int8())
Traceback (most recent call last):
  File "<ipython-input-3-357aa3c7f8d3>", line 1, in <module>
    pa.array([float('nan')], type=pa.int8())
  File "pyarrow/array.pxi", line 186, in pyarrow.lib.array
    return _sequence_to_array(obj, size, type, pool, from_pandas)
  File "pyarrow/array.pxi", line 40, in pyarrow.lib._sequence_to_array
    check_status(
  File "pyarrow/error.pxi", line 81, in pyarrow.lib.check_status
    raise ArrowInvalid(message)
ArrowInvalid: ../src/arrow/python/builtin_convert.cc:920 code: AppendPySequence(seq, size, real_type, builder.get(), from_pandas)
../src/arrow/python/iterators.h:60 code: func(value)
../src/arrow/python/builtin_convert.cc:454 code: internal::CIntFromPython(obj, &value)
../src/arrow/python/helpers.cc:259 code: CheckPyError()
cannot convert float NaN to integer

[asfimport]

Antoine Pitrou / @pitrou:
I'm rejecting the issue. Please re-open if you disagree.