Currently ParquetHiveSerde uses BinaryWritable to enclose bytes read from Parquet data. However, existing Hadoop class, BytesWritable, already does that, and BinaryWritable offers no advantage. On the other hand, BinaryWritable has a confusing getString() method, which, if misused, can cause unexpected result. The proposal here is to replace it with Hadoop BytesWritable.
The issue was identified in
HIVE-6367, serving as a follow-up JIRA.