BUG: Partial file write to disk on calling to_parquet() with engine='pyarrow' with unsupported dtype

[Anirudhsekar96]

Pandas version checks

• I have checked that this issue has not already been reported.

• I have confirmed this bug exists on the latest version of pandas.

• I have confirmed this bug exists on the master branch of pandas.

Reproducible Example

import pandas as pd
import numpy as np

data = np.arange(2, 10, dtype=np.float16)
df = pd.DataFrame(data=data, columns=['fp16'])
df.to_parquet('./fp16.parquet')

Issue Description

Calling to_parquet() on a dataframe with unsupported write on pyarrow results in partial file dump to disk.

Ref:

• BUG: Parquet does not support saving fp16 GH#44846 #44847
• BUG: Parquet format does not support saving float16 columns #44846

Expected Behavior

Raise ValueError before the partial write.

Installed Versions

python : 3.8.12.final.0
python-bits : 64
OS : Linux
OS-release : 5.11.0-41-generic
Version : #45~20.04.1-Ubuntu SMP Wed Nov 10 10:20:10 UTC 2021
machine : x86_64
processor : x86_64
byteorder : little
LC_ALL : None
LANG : en_US.UTF-8
LOCALE : en_US.UTF-8

pandas : 1.4.0.dev0+1356.gb7991da361.dirty
numpy : 1.21.4
pytz : 2021.3
dateutil : 2.8.2
pip : 21.3.1
setuptools : 59.4.0
Cython : 0.29.25
pytest : 6.2.5
hypothesis : 6.31.3
sphinx : 4.3.1
blosc : None
feather : None
xlsxwriter : 3.0.2
lxml.etree : 4.6.4
html5lib : 1.1
pymysql : None
psycopg2 : None
jinja2 : 3.0.3
IPython : 7.30.1
pandas_datareader: None
bs4 : 4.10.0
bottleneck : 1.3.2
fsspec : 2021.11.0
fastparquet : 0.7.2
gcsfs : 2021.11.0
matplotlib : 3.5.0
numexpr : 2.8.0
odfpy : None
openpyxl : 3.0.9
pandas_gbq : None
pyarrow : 6.0.1
pyxlsb : None
s3fs : 2021.11.0
scipy : 1.7.3
sqlalchemy : 1.4.28
tables : 3.6.1
tabulate : 0.8.9
xarray : 0.18.2
xlrd : 2.0.1
xlwt : 1.3.0
numba : 0.53.1

[Anirudhsekar96]

This pull request closes the issue where we see a partial file flush to disk through pandas when passing a dataframe to to_parquet function without partition columns. This issue was not replicable on pyarrow write_table function directly as the path variable passed was of string type, whereases when we pass a string like path through to_parquet it is converted to a io.BufferWriter type object before being passed to pyarrow.parquet.write_table function.

This pull request adds code to convert the path object to string type if it of io.BufferWriter type.

The issue still persists in the case when partition columns are not none leading to creation of an empty/partially populated folder incase of an error. The case of partition columns not being none was replicable on PyArrow and hence I have raised a bug report on PyArrow JIRA:
https://issues.apache.org/jira/browse/ARROW-15375