[FEA] Support First() in windowing context with Integer type

[viadea]

Is your feature request related to a problem? Please describe.
This is a feature request to support first() function in windowing context with double type.

For example:

import org.apache.spark.sql.Row
import org.apache.spark.sql.types._
import org.apache.spark.sql.functions._

val data = Seq(
    Row(Row("Adam ","","Green"),"1","M",1000, "2019-01-01",List("Java","Scala")),
    Row(Row("Bob ","Middle","Green"),"2","M",2000, "2019-01-02",List("Java","Python")),
    Row(Row("Cathy ","","Green"),"3","F",3000, "2019-01-03",List())
)

val schema = (new StructType()
  .add("name",new StructType()
    .add("firstname",StringType)
    .add("middlename",StringType)
    .add("lastname",StringType)) 
  .add("id",StringType)
  .add("gender",StringType)
  .add("salary",IntegerType)
  .add("birthdayStr",StringType)
  .add("language",ArrayType(StringType))
             )

val df = spark.createDataFrame(spark.sparkContext.parallelize(data),schema)
df.withColumn("birthday", to_date(col("birthdayStr"))).write.format("parquet").mode("overwrite").save("/tmp/testparquet")
val df2 = spark.read.parquet("/tmp/testparquet")
df2.createOrReplaceTempView("df2")
df2.printSchema

sql("""SELECT gender, first(salary,true) OVER (PARTITION BY gender ORDER BY salary) FROM df2""").collect

[viadea]

Driver log message:

      !Expression <WindowExpression> first(salary#174, true) windowspecdefinition(gender#173, salary#174 ASC NULLS FIRST, specifiedwindowframe(RangeFrame, unboundedpreceding$(), currentrow$())) cannot run on GPU because the type of orderBy column is not supported in a window range function, found DoubleType

[res-life]

Actually not support first function in window context.
CUDF does not support this currently, please see rapidsai/cudf#9643

[res-life]

@viadea Order by in window only supports int and timestamp, does not supports double float, it's a constraint of CUDF.
Like Count does not support double, you can test:
SELECT gender, count(salary) OVER (PARTITION BY gender ORDER BY salary) FROM df2
So this issue will be: Support First() in windowing context with int and timestamp.

[revans2]

All of these are going to require some CUDF changes to work.

nth_element_aggregation is not supported as a rolling_aggregation in cudf so we cannot implement it in a way that will be able to support null checks like the query above wants.

Cudf does not support range queries on floats or doubles yet. We can add this in though.

[viadea]

@res-life I understand there are limitations on cuDF side. That is fine.
But the issue should not be changed because that is based on a customer's log analysis.
In that, we need to support double, that is the feature request.

[res-life]

Note:
Cudf does not support range queries on floats or doubles yet.
Still depends on cuDF: #6000 (comment)

[res-life]

Split this issue into 2, one for Integer, one for Double #6192

[res-life]

#6000 already supported Integer type.