[SPARK-36926][SQL] Decimal average mistakenly overflow

This bug was introduced by apache#33177

When checking overflow of the sum value in the average function, we should use the `sumDataType` instead of the input decimal type.

fix a regression

Yes, the result was wrong before this PR.

a new test

Closes apache#34180 from cloud-fan/bug.

Lead-authored-by: Wenchen Fan <wenchen@databricks.com>
Co-authored-by: Wenchen Fan <cloud0fan@gmail.com>
Signed-off-by: Gengliang Wang <gengliang@apache.org>

sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/aggregate/Average.scala

@@ -94,10 +94,10 @@ case class Average(
   // If all input are nulls, count will be 0 and we will get null after the division.
   // We can't directly use `/` as it throws an exception under ansi mode.
   override lazy val evaluateExpression = child.dataType match {
-    case d: DecimalType =>
+    case _: DecimalType =>
       DecimalPrecision.decimalAndDecimal()(
         Divide(
-          CheckOverflowInSum(sum, d, !failOnError),
+          CheckOverflowInSum(sum, sumDataType.asInstanceOf[DecimalType], !failOnError),
           count.cast(DecimalType.LongDecimal), failOnError = false)).cast(resultType)
     case _: YearMonthIntervalType =>
       If(EqualTo(count, Literal(0L)),

sql/core/src/test/scala/org/apache/spark/sql/DataFrameAggregateSuite.scala

@@ -1416,6 +1416,12 @@ class DataFrameAggregateSuite extends QueryTest
     val df2 = Seq(Period.ofYears(1)).toDF("a").groupBy("a").count()
     checkAnswer(df2, Row(Period.ofYears(1), 1))
   }
+
+  test("SPARK-36926: decimal average mistakenly overflow") {
+    val df = (1 to 10).map(_ => "9999999999.99").toDF("d")
+    val res = df.select($"d".cast("decimal(12, 2)").as("d")).agg(avg($"d").cast("string"))
+    checkAnswer(res, Row("9999999999.990000"))
+  }
 }
 
 case class B(c: Option[Double])