[ARROW-4675] [Python] Error serializing bool ndarray in py2 and deserializing in py3

• Type: Bug
• Status: Resolved
• Priority: Minor
• Resolution: Fixed
• Affects Version/s: 0.12.0
• Fix Version/s: 0.14.0
• Component/s: Python
• Labels: pull-request-available
• Environment: * pyarrow 0.12.0 ; * numpy 1.16.1 ; * Python 3.7.0, 2.7.15 ; * (macOS 10.13.6)
• External issue URL: https://github.com/apache/arrow/issues/3749
• Assignee: Wes McKinney
• Reporter: Gabe Joseph
• Created: 25/Feb/19 20:49
• Updated: 11/Jan/23 07:35
• Resolved: 19/Jun/19 17:37

Description

np.bool is the only dtype I've found that causes this issue. Both empty and non-empty arrays cause it.

The issue only manifests from py2 to py3; staying within the same version succeeds, as does serializing from py3 and deserializing in py2.

This appears to just be due to Python 2 str being deserialized in Python 3 as bytes; it should be unicode on the py2 end to come back as str in py3. I suppose something in the serialization implementation is writing the dtype (just for bool arrays?) using a str, but haven't dug into it yet.

(two)bash-3.2$ python cereal.py
(two)bash-3.2$ cat cereal.py 
# Python 2
import numpy as np
import pyarrow as pa

data = np.array([], dtype=np.dtype('bool'))
buf = pa.serialize(data).to_buffer()

outstream = pa.output_stream("buffer")
outstream.write(buf)
outstream.close()

# ...switch to python 3 venv...
(three)bash-3.2$ cat decereal.py 
# Python 3
import numpy as np
import pyarrow as pa

instream = pa.input_stream("buffer")
buf = instream.read()

data = pa.deserialize(buf)
print(data)
(three)bash-3.2$ python3 decereal.py 
Traceback (most recent call last):
  File "decereal.py", line 10, in <module>
    data = pa.deserialize(buf)
  File "pyarrow/serialization.pxi", line 448, in pyarrow.lib.deserialize
  File "pyarrow/serialization.pxi", line 411, in pyarrow.lib.deserialize_from
  File "pyarrow/serialization.pxi", line 262, in pyarrow.lib.SerializedPyObject.deserialize
  File "pyarrow/serialization.pxi", line 175, in pyarrow.lib.SerializationContext._deserialize_callback
TypeError: can only concatenate str (not "bytes") to str

Issue Links

• links to: GitHub Pull Request #4611