GH-48457: [Python] Overflow check in datetime conversion in pa.array

[HyukjinKwon]

Rationale for this change

arrow/python/pyarrow/src/arrow/python/numpy_to_arrow.cc

Lines 499 to 500 in 7a36fcc

	// TODO(wesm): How pedantic do we really want to be about checking for int32
	// overflow here?

Should check and work together with safe paramter.

What changes are included in this PR?

This PR proposes to implement overflow check in date and integer conversions

Are these changes tested?

Yes, manually tested, and also added a unittest:

pytest pyarrow/tests/test_array.py -k "test_array_from_numpy_datetime_overflow" -v

Are there any user-facing changes?

Yes.

import numpy as np
import pyarrow as pa
pa.array(np.array([np.int64(-3000000000)], dtype='datetime64[D]'), safe=True)

Before:

>>> pa.array(np.array([np.int64(-3000000000)], dtype='datetime64[D]'), safe=True)
<pyarrow.lib.Date32Array object at 0x1207f85e0>
[
  <value out of range: 1294967296>
]

After:

pyarrow.lib.ArrowInvalid: Integer value -3000000000 out of bounds for int32 conversion at index 0

• GitHub Issue: [Python] Overflow check in datetime conversion in pa.array #48457

[github-actions]

⚠️ GitHub issue #48457 has been automatically assigned in GitHub to PR creator.

[EnricoMi]

Hi @HyukjinKwon, welcome to the Apache Arrow project.

This looks generally good, one possible overflow remark, though.

[EnricoMi]

Happy with this. Just a minor suggestion to rename the method to reflect the "limitation" or "assumption".

[HyukjinKwon]

I believe this should be ready for a look too @AlenkaF 🫡