Describe the enhancement requested
Currently, when fuzzing a Parquet file, proper errors are ignored since the file is most of the time simply invalid: it's ok to get an error.
However, if reading a Parquet row group was successful, the Parquet reader should have ensured that the resulting Table is structurally sound: calling Table::Validate should succeed, and if it fails then it should report a bug.
Table::ValidateFull, however, would still be allowed to fail, as for example a Parquet UTF8 column could contain invalid UTF8 bytes, and we don't detect that when decoding.
Component(s)
C++, Parquet, Continuous Integration