Beeline reads garbage on "select *" from Parquet files created with Pig or MapReduce

Create a table with Hive or Impala, then populate it with Pig or MapReduce.

With beeline (JDBC client to HiveServer2), "select *" from the table returns empty rows. Selecting all columns explicitly works as expected, and any aggregations as well. This means it is only failing when reading directly from HDFS without launching a MapReduce job.

The Hive CLI, on the other hand, works correctly for all cases.