NativeBatchReader calls checkParquetType on all of the columns on every invocation of loadNextBatch. I tried moving it up to init but some Spark SQL tests expect the exceptions that this generates to come from a call stack going into loadNextBatch rather than init. Perhaps we can stash a boolean that we've checked the columns after the first batch, and then elide the column checks on subsequent batches.
NativeBatchReader calls
checkParquetTypeon all of the columns on every invocation ofloadNextBatch. I tried moving it up toinitbut some Spark SQL tests expect the exceptions that this generates to come from a call stack going intoloadNextBatchrather thaninit. Perhaps we can stash a boolean that we've checked the columns after the first batch, and then elide the column checks on subsequent batches.