Do not use case class as public API (UserDefinedFunction)

project/MimaExcludes.scala

@@ -227,7 +227,11 @@ object MimaExcludes {
       case ReversedMissingMethodProblem(meth) =>
         !meth.owner.fullName.startsWith("org.apache.spark.sql.sources.v2")
       case _ => true
-    }
+    },
+
+    // [SPARK-26216][SQL] Do not use case class as public API (UserDefinedFunction)
+    ProblemFilters.exclude[MissingClassProblem]("org.apache.spark.sql.expressions.UserDefinedFunction$"),
+    ProblemFilters.exclude[IncompatibleTemplateDefProblem]("org.apache.spark.sql.expressions.UserDefinedFunction")
   )
 
   // Exclude rules for 2.4.x

sql/core/src/main/scala/org/apache/spark/sql/expressions/UserDefinedFunction.scala

@@ -38,114 +38,108 @@ import org.apache.spark.sql.types.DataType
  * @since 1.3.0
  */
 @Stable
-case class UserDefinedFunction protected[sql] (
-    f: AnyRef,
-    dataType: DataType,
-    inputTypes: Option[Seq[DataType]]) {
-
-  private var _nameOption: Option[String] = None
-  private var _nullable: Boolean = true
-  private var _deterministic: Boolean = true
-
-  // This is a `var` instead of in the constructor for backward compatibility of this case class.
-  // TODO: revisit this case class in Spark 3.0, and narrow down the public surface.
-  private[sql] var nullableTypes: Option[Seq[Boolean]] = None
+trait UserDefinedFunction {
 
   /**
    * Returns true when the UDF can return a nullable value.
    *
    * @since 2.3.0
    */
-  def nullable: Boolean = _nullable
+  def nullable: Boolean
 
   /**
    * Returns true iff the UDF is deterministic, i.e. the UDF produces the same output given the same
    * input.
    *
    * @since 2.3.0
    */
-  def deterministic: Boolean = _deterministic
+  def deterministic: Boolean
 
   /**
    * Returns an expression that invokes the UDF, using the given arguments.
    *
    * @since 1.3.0
    */
   @scala.annotation.varargs
-  def apply(exprs: Column*): Column = {
+  def apply(exprs: Column*): Column
+
+  /**
+   * Updates UserDefinedFunction with a given name.
+   *
+   * @since 2.3.0
+   */
+  def withName(name: String): UserDefinedFunction
+
+  /**
+   * Updates UserDefinedFunction to non-nullable.
+   *
+   * @since 2.3.0
+   */
+  def asNonNullable(): UserDefinedFunction
+
+  /**
+   * Updates UserDefinedFunction to nondeterministic.
+   *
+   * @since 2.3.0
+   */
+  def asNondeterministic(): UserDefinedFunction
+}
+
+private[sql] case class SparkUserDefinedFunction(
+    f: AnyRef,
+    dataType: DataType,
+    inputTypes: Option[Seq[DataType]],
+    nullableTypes: Option[Seq[Boolean]],
+    name: Option[String] = None,
+    nullable: Boolean = true,
+    deterministic: Boolean = true) extends UserDefinedFunction {
+
+  @scala.annotation.varargs
+  override def apply(exprs: Column*): Column = {
     // TODO: make sure this class is only instantiated through `SparkUserDefinedFunction.create()`
     // and `nullableTypes` is always set.
-    if (nullableTypes.isEmpty) {
-      nullableTypes = Some(ScalaReflection.getParameterTypeNullability(f))
-    }
     if (inputTypes.isDefined) {
       assert(inputTypes.get.length == nullableTypes.get.length)
     }
 
+    val inputsNullSafe = if (nullableTypes.isEmpty) {
+      ScalaReflection.getParameterTypeNullability(f)
+    } else {
+      nullableTypes.get
+    }
+
     Column(ScalaUDF(
       f,
       dataType,
       exprs.map(_.expr),
-      nullableTypes.get,
+      inputsNullSafe,
       inputTypes.getOrElse(Nil),
-      udfName = _nameOption,
-      nullable = _nullable,
-      udfDeterministic = _deterministic))
-  }
-
-  private def copyAll(): UserDefinedFunction = {
-    val udf = copy()
-    udf._nameOption = _nameOption
-    udf._nullable = _nullable
-    udf._deterministic = _deterministic
-    udf.nullableTypes = nullableTypes
-    udf
+      udfName = name,
+      nullable = nullable,
+      udfDeterministic = deterministic))
   }
 
-  /**
-   * Updates UserDefinedFunction with a given name.
-   *
-   * @since 2.3.0
-   */
-  def withName(name: String): UserDefinedFunction = {
-    val udf = copyAll()
-    udf._nameOption = Option(name)
-    udf
+  override def withName(name: String): UserDefinedFunction = {
+    copy(name = Option(name))
   }
 
-  /**
-   * Updates UserDefinedFunction to non-nullable.
-   *
-   * @since 2.3.0
-   */
-  def asNonNullable(): UserDefinedFunction = {
+  override def asNonNullable(): UserDefinedFunction = {
     if (!nullable) {
       this
     } else {
-      val udf = copyAll()
-      udf._nullable = false
-      udf
+      copy(nullable = false)
     }
   }
 
-  /**
-   * Updates UserDefinedFunction to nondeterministic.
-   *
-   * @since 2.3.0
-   */
-  def asNondeterministic(): UserDefinedFunction = {
-    if (!_deterministic) {
+  override def asNondeterministic(): UserDefinedFunction = {
+    if (!deterministic) {
       this
     } else {
-      val udf = copyAll()
-      udf._deterministic = false
-      udf
+      copy(deterministic = false)
     }
   }
 }
 
-// We have to use a name different than `UserDefinedFunction` here, to avoid breaking the binary
-// compatibility of the auto-generate UserDefinedFunction object.
 private[sql] object SparkUserDefinedFunction {
 
   def create(
@@ -157,8 +151,7 @@ private[sql] object SparkUserDefinedFunction {
     } else {
       Some(inputSchemas.map(_.get.dataType))
     }
-    val udf = new UserDefinedFunction(f, dataType, inputTypes)
-    udf.nullableTypes = Some(inputSchemas.map(_.map(_.nullable).getOrElse(true)))
-    udf
+    val nullableTypes = Some(inputSchemas.map(_.map(_.nullable).getOrElse(true)))
+    SparkUserDefinedFunction(f, dataType, inputTypes, nullableTypes)
   }
 }

sql/core/src/main/scala/org/apache/spark/sql/functions.scala

@@ -4259,7 +4259,7 @@ object functions {
   def udf(f: AnyRef, dataType: DataType): UserDefinedFunction = {
     // TODO: should call SparkUserDefinedFunction.create() instead but inputSchemas is currently
     // unavailable. We may need to create type-safe overloaded versions of udf() methods.
-    new UserDefinedFunction(f, dataType, inputTypes = None)
+    SparkUserDefinedFunction(f, dataType, inputTypes = None, nullableTypes = None)
   }
 
   /**