From the beginning of the introduction of [[http://jcp.org/aboutJava/communityprocess/maintenance/JLS/innerclasses.pdf|inner classes]] in Java, developers knew that it would be a mess... and it has been proven again and again. Here is a recent example. Take the following (academic) piece of code:

<code>
public class Anonymous {
	public void foo() {
		class B {}
		new B();
	}
}
</code>

Compile it with your standard Java compiler ''javac''. On both MacOS and Windows, there are two files generated... **BUT**, these two files have different names (!), thus throwing on the ground some of the dedicated [[PADL]] tests (in particular in project [[PADL Creator ClassFile|PADL Creator ClassFile Tests]])

  * On MacOS
    * ''Anonymous.class''
    * ''Anonymous$1B.class''
  * On Windows
    * ''Anonymous.class''
    * ''Anonymous$1$B.class''

Did you notice the difference? on MacOS, the file describing the inner class is named ''Anonymous$1B.class'' with only one $ sign; on Windows, its name is ''Anonymous$1$B.class'' with two $ signs. **Sigh...**