cloud-fan commented on a change in pull request #27066:
URL: https://github.com/apache/spark/pull/27066#discussion_r448719460
##########
File path:
sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/complexTypeCreator.scala
##########
@@ -539,3 +539,76 @@ case class StringToMap(text: Expression, pairDelim:
Expression, keyValueDelim: E
override def prettyName: String = "str_to_map"
}
+
+/**
+ * Adds/replaces field in struct by name.
+ */
+case class WithFields(
+ structExpr: Expression,
+ names: Seq[String],
+ valExprs: Seq[Expression]) extends Unevaluable {
+
+ assert(names.length == valExprs.length)
+
+ override def checkInputDataTypes(): TypeCheckResult = {
+ if (!structExpr.dataType.isInstanceOf[StructType]) {
+ TypeCheckResult.TypeCheckFailure(
+ "struct argument should be struct type, got: " +
structExpr.dataType.catalogString)
+ } else {
+ TypeCheckResult.TypeCheckSuccess
+ }
+ }
+
+ override def children: Seq[Expression] = structExpr +: valExprs
+
+ private lazy val addOrReplaceExprs = names.zip(valExprs)
+
+ override def dataType: StructType = {
+ val existingStructFields: Seq[(String, StructField)] = structExpr.dataType
+ .asInstanceOf[StructType].map(f => (f.name, f))
+ val addOrReplaceStructFields: Seq[(String, StructField)] =
addOrReplaceExprs.map {
+ case (name, expr) => (name, StructField(name, expr.dataType,
expr.nullable))
+ }
+ StructType(addOrReplace(existingStructFields,
addOrReplaceStructFields).map(_._2))
+ }
+
+ override def foldable: Boolean = structExpr.foldable &&
valExprs.forall(_.foldable)
+
+ override def nullable: Boolean = structExpr.nullable
+
+ override def prettyName: String = "with_fields"
+
+ lazy val evalExpr: Expression = {
+ val existingExprs =
structExpr.dataType.asInstanceOf[StructType].fieldNames.zipWithIndex.map {
+ case (name, i) => (name, GetStructField(structExpr, i, Some(name)))
Review comment:
Another idea: We can use `GetStructField(KnownNotNull(structExpr), i)`,
because we only evaluate it if `IsNull(structExpr)` is false.
Then we can solve https://github.com/apache/spark/pull/27066/files#r448598639
And this is also good for performance: the codegen can skip null check since
we know it can't be null.
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
