Github user cloud-fan commented on a diff in the pull request: https://github.com/apache/spark/pull/20831#discussion_r175992410 --- Diff: sql/core/src/main/scala/org/apache/spark/sql/execution/columnar/InMemoryTableScanExec.scala --- @@ -169,7 +174,10 @@ case class InMemoryTableScanExec( override def outputOrdering: Seq[SortOrder] = relation.child.outputOrdering.map(updateAttribute(_).asInstanceOf[SortOrder]) - private def statsFor(a: Attribute) = relation.partitionStatistics.forAttribute(a) + // When we make canonicalized plan, we can't find a normalized attribute in this map. + // We return a `ColumnStatisticsSchema` for normalized attribute in this case. --- End diff -- I don't get it. Regardless how copy is implemented in scala, ideally we can just mark `buildFilter` and `partitionFilters` as lazy, and in `doCanonicalize`, create a new `InMemoryTableScanExec`, which won't materialize `partitionFilters` in either the current `InMemoryTableScanExec` or the new `InMemoryTableScanExec`. One problem I can think of is to serialize a canonicalized `InMemoryTableScanExec`, but it should never happen.
--- --------------------------------------------------------------------- To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org For additional commands, e-mail: reviews-h...@spark.apache.org