I think the current code tests the re-exported Parquet functionalities, not touching the DataFusion-related code. Ideally, we should test the end-to-end Parquet reading process.

The process roughly looks like this:

1. Create a in-memory object_store, and put the Parquet data you generated into the object_store.
2. Register the object_store along with the path to the DataFusion.
3. Run a SQL query from the DataFusion side to see if the results can be read back.

A loosely related test can be found here: https://github.com/XiangpengHao/parquet-viewer/blob/main/src/tests.rs#L9

Originally posted by @XiangpengHao in #15325 (comment)