It would be nice to confirm that the files we are producing give the correct values when read with other tools, but after taking a look at both the Impala docs and the parquet-mr codebase it doesn't look like anyone else has implemented it. It does look like we're writing it in the correct format as described in the spec. As we will be writing data with it, external confirmation of accuracy would be nice to have. We should probably open a JIRA for testing it once it is added to parquet-mr or the c++ reader based on impala.