It looks like scala compiles methods in companion objects into static methods, which makes invoking them from java code a little easier. For example, you could write CompanionObject.method() instead of CompanionObject$.MODULE$.method(). However, sometimes seemingly irrelevant code change will break this behavior. I came up with this example to illustrate the problem

$ cat TestCompanion.scala 
class TestCompanion

object TestCompanion {
  def init2 {}
}

@SerialVersionUID(1L)
class TestCompanion2

object TestCompanion2 {
  def init2 {}
}


$ scalac -version
Scala compiler version 2.9.0.1 -- Copyright 2002-2011, LAMP/EPFL

$ scalac TestCompanion.scala
$ javap TestCompanion
Compiled from "TestCompanion.scala"
public class TestCompanion extends java.lang.Object implements scala.ScalaObject{
    public static final void init2();
    public TestCompanion();
}

$ javap TestCompanion2
Compiled from "TestCompanion.scala"
public class TestCompanion2 extends java.lang.Object implements scala.ScalaObject{
    public static final long serialVersionUID;
    public static {};
    public TestCompanion2();
}

So the only difference between TestCompanion and TestCompanion2 is that the latter is annotated with @SerialVersionUID, and init2 is compiled into a static method in TestCompanion but not in TestCompanion2.

Can someone explain why scalac treats these two classes differently? I don't see how the @SerialVersionUID annotation should affect the availability of static methods.