Spark 向 Scala 类的构造函数添加隐藏参数

Question

我不知道如何解释，但 Spark 似乎向构造函数添加了一个隐藏（隐式？）参数。这是我在spark-shell 中尝试的代码（在常规 Scala shell 参数列表中为空）：

scala> class A {}
defined class A

scala> classOf[A].getConstructors()(0).getAnnotatedParameterTypes
res0: Array[java.lang.reflect.AnnotatedType] = Array(sun.reflect.annotation.AnnotatedTypeFactory$AnnotatedTypeBaseImpl@5ed65e4b)

由于这个参数，我无法将我的自定义 InputFormat 类传递给 Spark 的 hadoopFile 函数。关于这里发生了什么的任何提示，或者至少如何使用无参数构造函数创建类？

Answer 1

行为似乎与普通 Scala REPL 中的行为相同

$ scala
Welcome to Scala 2.13.3 (Java HotSpot(TM) 64-Bit GraalVM EE 19.3.0, Java 1.8.0_231).
Type in expressions for evaluation. Or try :help.

scala> class A {}
class A

scala> classOf[A].getConstructors()(0).getAnnotatedParameterTypes
val res0: Array[java.lang.reflect.AnnotatedType] = Array(sun.reflect.annotation.AnnotatedTypeFactory$AnnotatedTypeBaseImpl@383864d5)

scala> classOf[A].getConstructors()(0).getParameters
val res1: Array[java.lang.reflect.Parameter] = Array(final $iw $outer)

REPL 使类嵌套（REPL 中的每一行都是外部类的实例化）。这会将外部类的实例作为参数添加到构造函数（$outer 是参数的名称，$iw 是外部类）。您可以按如下方式重现此行为

class X {
  class A {}
}

object App {
  def main(args: Array[String]): Unit = {
    val x = new X

    println(classOf[x.A].getConstructors()(0).getAnnotatedParameterTypes.mkString(","))
    // sun.reflect.annotation.AnnotatedTypeFactory$AnnotatedTypeBaseImpl@2f7c7260

    println(classOf[x.A].getConstructors()(0).getParameters.mkString(","))
    // final X $outer
  }
}

---

如果您在打开编译器选项-Xprint:typer（如scala -Xprint:typer 或spark-shell -Xprint:typer）的情况下运行REPL，您会看到

$ scala -Xprint:typer
Welcome to Scala 2.13.3 (Java HotSpot(TM) 64-Bit GraalVM EE 19.3.0, Java 1.8.0_231).
Type in expressions for evaluation. Or try :help.

scala> class A
[[syntax trees at end of                     typer]] // <console>
package $line3 {
  sealed class $read extends AnyRef with Serializable {
    def <init>(): $line3.$read = {
      $read.super.<init>();
      ()
    };
    sealed class $iw extends AnyRef with java.io.Serializable {
      def <init>(): $iw = {
        $iw.super.<init>();
        ()
      };
      class A extends scala.AnyRef {
        def <init>(): A = {
          A.super.<init>();
          ()
        }
      }
    };
    private[this] val $iw: $iw = new $read.this.$iw();
    <stable> <accessor> def $iw: $iw = $read.this.$iw
  };
  object $read extends scala.AnyRef with java.io.Serializable {
    def <init>(): type = {
      $read.super.<init>();
      ()
    };
    private[this] val INSTANCE: $line3.$read = new $read();
    <stable> <accessor> def INSTANCE: $line3.$read = $read.this.INSTANCE;
    <synthetic> private def writeReplace(): Object = new scala.runtime.ModuleSerializationProxy(classOf[$line3.$read$])
  }
}

class A

所以这个额外的构造函数参数$outer可以获取为$line3.$read.INSTANCE.$iw

scala> classOf[A].getConstructors()(0).newInstance($line3.$read.INSTANCE.$iw)

...

val res0: Object = A@282ffbf5

请注意，编码在不同版本的 Scala 中可能会发生变化。例如，来自 Spark 3.0.1（为 Hadoop 3.2 预构建）的 spark-shell 使用 Scala 2.12.10，$lineXXX.$read.INSTANCE.$iw.$iw 应该代替 $lineXXX.$read.INSTANCE.$iw

$ spark-shell -Xprint:typer
20/11/25 16:32:16 WARN Utils: Your hostname, dmitin-HP-Pavilion-Laptop resolves to a loopback address: 127.0.1.1; using 192.168.0.103 instead (on interface wlo1)
20/11/25 16:32:16 WARN Utils: Set SPARK_LOCAL_IP if you need to bind to another address
20/11/25 16:32:16 WARN NativeCodeLoader: Unable to load native-hadoop library for your platform... using builtin-java classes where applicable
Using Spark's default log4j profile: org/apache/spark/log4j-defaults.properties
Setting default log level to "WARN".
To adjust logging level use sc.setLogLevel(newLevel). For SparkR, use setLogLevel(newLevel).
Spark context Web UI available at http://192.168.0.103:4040
Spark context available as 'sc' (master = local[*], app id = local-1606314741512).
Spark session available as 'spark'.
Welcome to
      ____              __
     / __/__  ___ _____/ /__
    _\ \/ _ \/ _ `/ __/  '_/
   /___/ .__/\_,_/_/ /_/\_\   version 3.0.1
      /_/
         
Using Scala version 2.12.10 (Java HotSpot(TM) 64-Bit GraalVM EE 19.3.0, Java 1.8.0_231)
Type in expressions to have them evaluated.
Type :help for more information.

scala> class A
[[syntax trees at end of                     typer]] // <console>
package $line14 {
  sealed class $read extends AnyRef with java.io.Serializable {
    def <init>(): $line14.$read = {
      $read.super.<init>();
      ()
    };
    sealed class $iw extends AnyRef with java.io.Serializable {
      def <init>(): $read.this.$iw = {
        $iw.super.<init>();
        ()
      };
      sealed class $iw extends AnyRef with java.io.Serializable {
        def <init>(): $iw = {
          $iw.super.<init>();
          ()
        };
        class A extends scala.AnyRef {
          def <init>(): A = {
            A.super.<init>();
            ()
          }
        }
      };
      private[this] val $iw: $iw = new $iw.this.$iw();
      <stable> <accessor> def $iw: $iw = $iw.this.$iw
    };
    private[this] val $iw: $read.this.$iw = new $read.this.$iw();
    <stable> <accessor> def $iw: $read.this.$iw = $read.this.$iw
  };
  object $read extends scala.AnyRef with Serializable {
    def <init>(): $line14.$read.type = {
      $read.super.<init>();
      ()
    };
    private[this] val INSTANCE: $line14.$read = new $read();
    <stable> <accessor> def INSTANCE: $line14.$read = $read.this.INSTANCE;
    <synthetic> private def readResolve(): Object = $line14.$read
  }
}

defined class A

scala> classOf[A].getConstructors()(0).newInstance($line14.$read.INSTANCE.$iw.$iw)

...

res0: Any = A@6621ab0c

---

在 Scala 2.12.6 scala -Xprint:typer 中产生

$ ./scala -Xprint:typer
Welcome to Scala 2.12.6 (Java HotSpot(TM) 64-Bit GraalVM EE 19.3.0, Java 1.8.0_231).
Type in expressions for evaluation. Or try :help.

scala> class A
[[syntax trees at end of                     typer]] // <console>
package $line3 {
  object $read extends scala.AnyRef {
    def <init>(): $line3.$read.type = {
      $read.super.<init>();
      ()
    };
    object $iw extends scala.AnyRef {
      def <init>(): type = {
        $iw.super.<init>();
        ()
      };
      object $iw extends scala.AnyRef {
        def <init>(): type = {
          $iw.super.<init>();
          ()
        };
        class A extends scala.AnyRef {
          def <init>(): A = {
            A.super.<init>();
            ()
          }
        }
      }
    }
  }
}

defined class A

所以现在类 A 嵌套在一个对象 ($line3.$read.$iw.$iw) 中而不是类中，在这种情况下，不会将附加参数添加到 A 的构造函数中

object X {
  class A {}
}

object App {
  def main(args: Array[String]): Unit = {
    val x = X

    println(classOf[x.A].getConstructors()(0).getAnnotatedParameterTypes.toList)
    // List()

    println(classOf[x.A].getConstructors()(0).getParameters.toList)
    // List()
  }
}