[SPARK-27591] A bug in UnivocityParser prevents using UDT

• Type: Bug
• Status: Resolved
• Priority: Minor
• Resolution: Fixed
• Affects Version/s: 2.4.2
• Fix Version/s: 3.0.0
• Component/s: SQL
• Assignee: Artem Kalchenko Assign to me
• Reporter: Artem Kalchenko
• Created: 29/Apr/19 09:30
• Updated: 12/Dec/22 18:10
• Resolved: 30/Apr/19 23:29

Description

I am trying to define a UserDefinedType based on String but different from StringType in Spark 2.4.1 but it looks like there is a bug in Spark or I am doing smth incorrectly.

I define my type as follows:

class MyType extends UserDefinedType[MyValue] {
  override def sqlType: DataType = StringType
  ...
}

@SQLUserDefinedType(udt = classOf[MyType])
case class MyValue

I expect it to be read and stored as String with just a custom SQL type. In fact Spark can't read the string at all:

java.lang.ClassCastException: org.apache.spark.sql.execution.datasources.csv.UnivocityParser$$anonfun$makeConverter$11 cannot be cast to org.apache.spark.unsafe.types.UTF8String
    at org.apache.spark.sql.catalyst.expressions.BaseGenericInternalRow$class.getUTF8String(rows.scala:46)
    at org.apache.spark.sql.catalyst.expressions.GenericInternalRow.getUTF8String(rows.scala:195)
    at org.apache.spark.sql.catalyst.expressions.JoinedRow.getUTF8String(JoinedRow.scala:102)

the problem is with UnivocityParser.makeConverter that doesn't return (String => Any) function but (String => (String => Any)) in the case of UDT, see UnivocityParser:184

case udt: UserDefinedType[_] => (datum: String) =>
  makeConverter(name, udt.sqlType, nullable, options)

Issue Links

• links to: GitHub Pull Request #24496