[Python] Extra metadata gets added after multiple conversions between pd.DataFrame and pa.Table

[asfimport]

We have a unit test that verifies that loading a dataframe from a .parq file and saving it back with no changes produces the same result as the original file. It started failing with pyarrow 0.8.0.

After digging into it, I discovered that after the first conversion from pd.DataFrame to pa.Table, the table contains the following metadata (among other things):

"column_indexes": [{"metadata": null, "field_name": null, "name": null, "numpy_type": "object", "pandas_type": "bytes"}]

However, after converting it to pd.DataFrame and back into a pa.Table for the second time, the metadata gets an encoding field:

"column_indexes": [{"metadata": {"encoding": "UTF-8"}, "field_name": null, "name": null, "numpy_type": "object", "pandas_type": "unicode"}]

See the attached file for a test case.

So specifically, it appears that dataframe->table->dataframe->table conversion produces a different result from just dataframe->table - which I think is unexpected.

Reporter: Dima Ryazanov / @dimaryaz
Assignee: Phillip Cloud / @cpcloud

Original Issue Attachments:

• fail.py

PRs and other links:

• GitHub Pull Request #1728

_{Note: This issue was originally created as ARROW-1940. Please see the migration documentation for further details.}

[asfimport]

Dima Ryazanov / @dimaryaz:
(The attached test case only fails in Python2, though our unit test fails in both 2 and 3. I can come up with a Python3 test case, too, if that helps.)

[asfimport]

Wes McKinney / @wesm:
Marked for 0.9.0 – we can look into it. I will say that where bytes -> unicode promotions are occurring it may be challenging to preserve a perfect round trip on Python 2 in all cases.

[asfimport]

Phillip Cloud / @cpcloud:
Taking a look at this now.

[asfimport]

Wes McKinney / @wesm:
Issue resolved by pull request 1728
#1728