ARROW-8799: [C++][Parquet] NestedListReader needs to handle empty item batches

[bkietz]

No description provided.

[github-actions]

https://issues.apache.org/jira/browse/ARROW-8799

[bkietz]

CI failure is a minio flake https://github.com/apache/arrow/pull/7181/checks?check_run_id=675715164#step:8:3095

[emkornfield]

@bkietz could you add a unit test?

[bkietz]

@emkornfield @wesm In adding a unit test I've become uncertain of the ColumnReader contract and whether my solution upholds it

• ColumnReader::NextBatch's doccomment states that when no data remains null should be yielded (which I read as: a null ChunkedArray).
• Instead the tests assert that the chunked array contain a single null chunk
• When reading into a dictionary LeafReader does neither of these and instead yields an empty ChunkedArray (for which NestedListReader on master is unprepared, causing ARROW-8799).

If modifying NestedListReader as I have here is unsatisfactory, I could change TransferDictionary to ensure LeafReader yields ChunkedArray{nullptr} when its out of data. What do you think?

[wesm]

+1 -- I think this is a band-aid solution until all these code can be deleted in favor of a complete nested data reading implementation. Just to confirm this fixes ARROW-8799 and so this is OK to merge?

[bkietz]

Confirmed, this resolves the issue