Hello spark-devs, we hit a similar case with SPARK-28098 when we tried to read a parquet format table which is generated by hive union operation, and I made a quick fix for it.
I'm not sure we should reuse the same configuration with hive or add a new one. And this is my first time to contribute codes to Spark, looks like I need a committer to authorize the testing flow. Any feedback is appreciated!
