Github user rxin commented on a diff in the pull request:
https://github.com/apache/spark/pull/6938#discussion_r34226418
--- Diff:
sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/math.scala
---
@@ -600,3 +604,146 @@ case class Logarithm(left: Expression, right:
Expression)
"""
}
}
+
+case class Round(child: Expression, scale: Expression) extends Expression
with ExpectsInputTypes {
+
+ def this(child: Expression) = {
+ this(child, Literal(0))
+ }
+
+ override def children: Seq[Expression] = Seq(child, scale)
+
+ override def nullable: Boolean = true
+
+ override def foldable: Boolean = child.foldable
+
+ override lazy val dataType: DataType = child.dataType match {
+ case DecimalType.Fixed(p, s) => DecimalType(p, _scale)
+ case t => t
+ }
+
+ override def inputTypes: Seq[AbstractDataType] = Seq(
+ // rely on precedence to implicit cast String into Double
+ TypeCollection(DecimalType, DoubleType, FloatType, LongType,
IntegerType, ShortType, ByteType),
+ TypeCollection(LongType, IntegerType, ShortType, ByteType))
+
+ override def checkInputDataTypes(): TypeCheckResult = {
+ child.dataType match {
+ case _: NumericType => // satisfy requirement
+ case dt =>
+ return TypeCheckFailure(s"Only numeric type is allowed for ROUND
function, got $dt")
+ }
+ scale match {
+ case Literal(value, LongType) =>
+ if (value.asInstanceOf[Long] < Int.MinValue ||
value.asInstanceOf[Long] > Int.MaxValue) {
+ return TypeCheckFailure("ROUND scale argument out of allowed
range")
+ }
+ case _ =>
+ if (scale.dataType.isInstanceOf[IntegralType] && scale.foldable) {
+ // TODO: How to check out of range for foldable LongType
Expression
+ // satisfy requirement
+ } else {
+ return TypeCheckFailure("Only foldable Integral Expression " +
+ s"is allowed for ROUND scale arguments, got ${child.dataType}")
+ }
+ }
+ TypeCheckSuccess
+ }
+
+ private lazy val scaleV = scale.eval(EmptyRow)
+ private lazy val _scale = if (scaleV != null) scaleV.asInstanceOf[Int]
else 0
+
+ override def eval(input: InternalRow): Any = {
+ val evalE = child.eval(input)
+ if (evalE == null || scaleV == null) return null
+ round(evalE)
+ }
+
+ private lazy val round: (Any) => (Any) = typedRound(child.dataType)
+
+ // Using dataType info to find an appropriate round method
+ private def typedRound(dt: DataType)(x: Any): Any = {
+ dt match {
+ case _: DecimalType =>
+ val decimal = x.asInstanceOf[Decimal]
+ if (decimal.changePrecision(decimal.precision, _scale)) decimal
else null
+ case ByteType =>
+ numericRound(x.asInstanceOf[Byte], _scale)
+ case ShortType =>
+ numericRound(x.asInstanceOf[Short], _scale)
+ case IntegerType =>
+ numericRound(x.asInstanceOf[Int], _scale)
+ case LongType =>
+ numericRound(x.asInstanceOf[Long], _scale)
+ case FloatType =>
+ numericRound(x.asInstanceOf[Float], _scale)
+ case DoubleType =>
+ numericRound(x.asInstanceOf[Double], _scale)
+ }
+ }
+
+ private def numericRound[T](input: T, scale: Int)(implicit bdc:
BigDecimalConverter[T]): T = {
+ input match {
+ case f: Float if (f.isNaN || f.isInfinite) => return input
+ case d: Double if (d.isNaN || d.isInfinite) => return input
+ case _ =>
+ }
+ bdc.fromBigDecimal(bdc.toBigDecimal(input).setScale(scale,
BigDecimal.RoundingMode.HALF_UP))
+ }
+
+ override def genCode(ctx: CodeGenContext, ev: GeneratedExpressionCode):
String = {
+ val ce = child.gen(ctx)
+
+ def round(primitive: String, integral: Boolean): String = {
+ val (p1, p2) = if (integral) ("new", "") else ("", ".valueOf")
+ s"""
+ ${ev.primitive} = $p1 java.math.BigDecimal$p2(${primitive}).
+ setScale(${_scale}, java.math.BigDecimal.ROUND_HALF_UP)"""
+ }
+
+ def fractionalCheck(primitive: String, function: String): String = {
+ s"""
+ if (Double.isNaN(${primitive}) || Double.isInfinite(${primitive})){
+ ${ev.primitive} = ${primitive};
+ } else {
+ ${round(primitive, false)}.${function};
+ }"""
+ }
+
+ def decimalRound(): String = {
+ s"""
+ if (${ce.primitive}.changePrecision(${ce.primitive}.precision(),
${_scale})) {
+ ${ev.primitive} = ${ce.primitive};
+ } else {
+ ${ev.isNull} = true;
+ }
+ """
+ }
+
+ val roundCode = child.dataType match {
--- End diff --
I'd do something like this to make it simpler
```scala
if (child.dataType.isInstanceOf[IntegralType]) {
// put the byte/short/.../long handling inline here
} else if (child.dataType.isInstanceOf[DecimalType]) {
// put the decimal type handling inline here
} else if (child.dataType.isInstanceOf[FractionalType]) {
// put the float/double handling inline here
}
```
The other thing I'd consider is to special case for scale == 0, since it is
very common and much faster to run.
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at infrastruct...@apache.org or file a JIRA ticket
with INFRA.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org
For additional commands, e-mail: reviews-h...@spark.apache.org
