Github user mgaido91 commented on a diff in the pull request:
https://github.com/apache/spark/pull/20023#discussion_r161714618
--- Diff:
sql/catalyst/src/main/scala/org/apache/spark/sql/types/DecimalType.scala ---
@@ -136,10 +137,52 @@ object DecimalType extends AbstractDataType {
case DoubleType => DoubleDecimal
}
+ private[sql] def forLiteral(literal: Literal): DecimalType =
literal.value match {
+ case v: Short => fromBigDecimal(BigDecimal(v))
+ case v: Int => fromBigDecimal(BigDecimal(v))
+ case v: Long => fromBigDecimal(BigDecimal(v))
+ case _ => forType(literal.dataType)
+ }
+
+ private[sql] def fromBigDecimal(d: BigDecimal): DecimalType = {
+ DecimalType(Math.max(d.precision, d.scale), d.scale)
+ }
+
private[sql] def bounded(precision: Int, scale: Int): DecimalType = {
DecimalType(min(precision, MAX_PRECISION), min(scale, MAX_SCALE))
}
+ /**
+ * Scale adjustment implementation is based on Hive's one, which is
itself inspired to
+ * SQLServer's one. In particular, when a result precision is greater
than
+ * {@link #MAX_PRECISION}, the corresponding scale is reduced to prevent
the integral part of a
+ * result from being truncated.
+ *
+ * This method is used only when
`spark.sql.decimalOperations.allowPrecisionLoss` is set to true.
+ *
+ * @param precision
+ * @param scale
+ * @return
+ */
+ private[sql] def adjustPrecisionScale(precision: Int, scale: Int):
DecimalType = {
--- End diff --
@cloud-fan yes, but you have to keep in mind that we are doing so only when
precision is > 38. With some simple math (given `intDigits = precision -
scale`), SQL server is `min(scale, scale + 38 - precision)`. Since we perform
this operation only when precision is greater than 38, the second member is
always the minimum. Which means that in such a case, SQL server behaves like
us, ie. it takes always `38 - intDigits`. When precision is < than 38, instead
we return the input precision and scale, as SQL server does. We are just using
the precision instead of the intDigits for the if.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org
For additional commands, e-mail: reviews-h...@spark.apache.org
