[Python][C++] Data corruption when initializing from float16 NumPy array or pandas Series

[Wainberg]

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

>>> import pyarrow as pa
>>> import numpy as np
>>> pa.array(np.array([1], dtype='float16'))
<pyarrow.lib.HalfFloatArray object at 0x7f7bc9ae7dc0>
[
  15360
]
>>> pa.array([np.float16(1)])
<pyarrow.lib.HalfFloatArray object at 0x7fd6bd266800>
[
  15360
]
>>> pa.array([np.float16(1)], type=pa.float16())
<pyarrow.lib.HalfFloatArray object at 0x7f7bc9ae7ee0>
[
  15360
]
>>> pa.array(pd.Series([1], dtype='float16'))
<pyarrow.lib.HalfFloatArray object at 0x7fd6bd266da0>
[
  15360
]

This is on pyarrow 14.0.2, NumPy 1.26.4, pandas 2.2.0, Python 3.12.

Component(s)

Python

---

Duplicate of:

• [C++][Python] Support pretty printing of float16 #36753

[mariosasko]

This has already been reported in #36753 (the print is not implemented, but the values are correctly stored)

[jorisvandenbossche]

Closing as a duplicate of #36753 then