为什么在 Java 中将私有内部 class 成员公开？

Question

What is the reason of declaring a member of a private inner class public in Java if it still can't be accessed outside of containing class? Or can it?

public class DataStructure {
    // ...

    private class InnerEvenIterator {
        // ...

        public boolean hasNext() { // Why public?
            // ...
        }
    }
}

Answer 1

If the InnerEvenIterator class does not extend any class or implement any interface, I think it is nonsense because no other class can access any instance of it.

However, if it extends or implements any other non private class or interface, it makes sense. An example:

interface EvenIterator {
    public boolean hasNext();
}


public class DataStructure {
    // ...

    private class InnerEvenIterator implements EvenIterator{
        // ...

        public boolean hasNext() { // Why public?
            // ...
        }
    }

    InnerEvenIterator iterator;

    public EvenIterator getIterator(){
         return iterator;
    }     

}

Answer 2

This method can be made public in order to indicate that it's semantically public, despite the fact that compiler doesn't enforce visibility rules in this particular case.

Imagine that during some refactoring you need to make this inner class top-level. If this method is private , how would you decide whether it should be made public , or some more restrictive modifier should be used? Declaring method as public tells reader the intentions of original author - this method shouldn't be considered an implementation detail.

Answer 3

It is useful when you implement any interface .

class DataStructure implements Iterable<DataStructure> {

    @Override
    public Iterator<DataStructure> iterator() {
        return new InnerEvenIterator();
    }
    // ...        

    private class InnerEvenIterator implements Iterator<DataStructure> {
        // ...    
        public boolean hasNext() { // Why public?
            // ...
            return false;
        }

        @Override
        public DataStructure next() {
            throw new UnsupportedOperationException("Not supported yet.");
        }

        @Override
        public void remove() {
            throw new UnsupportedOperationException("Not supported yet.");
        }
    }

    public static void main(String[] ex) {
        DataStructure ds = new DataStructure();
        Iterator<DataStructure> ids = ds.iterator();
        ids.hasNext(); // accessable            
    }
}

Answer 4

I think you are missing the implementing the Iterator interface part in your sample code. In that case, you can't make the hasNext() method have any other visibility identifier other than public since that would end up reducing its visibility (interface methods have public visibility) and it won't compile.

Answer 5

There are many combinations of access modifiers which are not useful. A public method in a private inner class is only useful if it implements a public method in a public class/interface.

public class DataStructure {
    // ...

    private class InnerEvenIterator implements Iterator {
        // ...

        public boolean hasNext() { // Why public?
            // ...
        }
    }

    public Iterator iterator() {
        return new InnerEvenIterator();
    }
}

BTW: abstract classes often have public constructors when actually they are protected

Answer 6

If the inner class is private it cannot be accessed by name outside of the outer class. Inner and outer classes have access to each other's private methods and private instance variables. As long as you are within the inner or outer class, the modifiers public and private have the same effect. In your code example:

public class DataStructure {
    // ...

    private class InnerEvenIterator {
        // ...

        public boolean hasNext() { // Why public?
            // ...
        }
    }
}

As far as the class DataStructure is concerned, this is completely equivalent to:

public class DataStructure {
    // ...

    private class InnerEvenIterator {
        // ...

        private boolean hasNext() {
            // ...
        }
    }
}

This is because only DataStructure can access it, so it doesn't matter if you set it to public or private. Either way, DataStructure is still the only class that can access it. Use whichever modifier you like, it makes no functional difference. The only time you can't choose at random is when you are implementing or extending, in which case you can't reduce the access, but you can increase it. So if an abstract method has protected access you can change it to public. Granted neither one actually makes any difference.

If you plan on using an inner class in other classes, and therefore making it public, you probably shouldn't make it an inner class in the first place.

Additionally, I don't see any requirement for inner classes extending or implementing other classes. It might be common for them to do so, but it's certainly not required.

Answer 7

There are multiple aspects which have to be considered here. The following will use the term "nested class" because it covers both non- static (also called "inner class") and static classes (source ).

Not related to private nested classes, but JLS §8.2 has an interesting example which shows where public members in package-private or protected classes could be useful.

Source code

Overriding methods

When your nested class implements an interface or extends a class and overrides one of its methods, then per JLS §8.4.8.3 :

The access modifier of an overriding or hiding method must provide at least as much access as the overridden or hidden method

For example:

public class Outer {
  private static class Nested implements Iterator<String> {
    @Override
    public boolean hasNext() {
      ...
    }
    
    @Override
    public String next() {
      ...
    }
  }
}

The methods hasNext() and next() which override the Iterator methods have to be public because the Iterator methods are public.

As a side note: JLS §13.4.7 describes that it is possible for a class to increase the access level of one of its methods, even if a subclass overrides it with, without causing linkage errors.

Conveying intention

Access restriction is defined in JLS §6.6.1 :

A member [...] of a reference type [...] is accessible only if the type is accessible and the member or constructor is declared to permit access

[...]

Otherwise, the member or constructor is declared private , and access is permitted if and only if it occurs within the body of the top level type ( §7.6 ) that encloses the declaration of the member or constructor.

Therefore members of a private nested class can (from a source code perspective; see also "Reflection" section) only be accessed from the body of the enclosing top level type. Interestingly the "body" also covers other nested classes:

public class TopLevel {
  private static class Nested1 {
    private int i;
  }

  void doSomething(Nested1 n) {
    // Can access private member of nested class
    n.i++;
  }

  private static class Nested2 {
    void doSomething(Nested1 n) {
      // Can access private member of other nested class
      n.i++;
    }
  }
}

So from a compiler-provided access restriction perspective there is indeed no point in using a public member in a private nested class.

However, using different access levels can be useful for conveying intention, especially (as pointed out by others) when the nested class might be refactored to a separate top level class in the future. Consider this example:

public class Cache {
  private static class CacheEntry<T> {
    private final T value;
    private long lastAccessed;

    // Signify that enclosing class may use this constructor
    public CacheEntry(T value) {
      this.value = value;
      updateLastAccessed();
    }

    // Signify that enclosing class must NOT use this method
    private void updateLastAccessed() {
      lastAccessed = System.nanoTime();
    }

    // Signify that enclosing class may use this method
    public T getValue() {
      updateLastAccessed();
      return value;
    }
  }

  ...
}

Compiled class files

It is also interesting to note how the Java compiler treats access to members of nested classes. Prior to JEP 181: Nest-Based Access Control (added in Java 11) the compiler had to create synthetic accessor methods because the class file could not express the access control logic related to nested classes. Consider this example:

class TopLevel {
  private static class Nested {
    private int i;
  }
    
  void doSomething(Nested n) {
    n.i++;
  }
}

When compiled with Java 8 and inspected with javap -p./TopLevel$Nested.class you will see that a synthetic access$008 method has been added:

class TopLevel$Nested {
  private int i;
  private TopLevel$Nested();
  static int access$008(TopLevel$Nested);
}

This slightly increased the size of the class files and might have decreased performance. This is one reason why package-private (ie no access modifier) access has often be chosen for members of nested classes to prevent creation of synthetic access methods.
With JEP 181 this is no longer necessary ( javap -v output when compiled with JDK 11):

class TopLevel$Nested
...
{
  private int i;
  ...

  private TopLevel$Nested();
  ...
}
...
NestHost: class TopLevel
...

Reflection

Another interesting aspect is reflection. The JLS is sadly not verify specific in that regard, but §15.12.4.3 contains an interesting hint:

If T is in a different package than D, and their packages are in the same module, and T is public or protected , then T is accessible.

[...]

If T is protected , it is necessarily a nested type, so at compile time, its accessibility is affected by the accessibility of types enclosing its declaration. However, during linkage, its accessibility is not affected by the accessibility of types enclosing its declaration. Moreover, during linkage, a protected T is as accessible as a public T.

Similarly AccessibleObject.setAccessible(...) does not mention the enclosing type at all. And indeed it is possible to access the members of a public or protected nested type within non- public enclosing type: test1/TopLevel1.java

package test1;

// package-private
class TopLevel1 {
  private static class Nested1_1 {
    protected static class Nested1_2 {
      public static int i;
    }
  }
}

test2/TopLevel2.java

package test2;

import java.lang.reflect.Field;

public class TopLevel2 {
  public static void main(String... args) throws Exception {
    Class<?> nested1_2 = Class.forName("test1.TopLevel1$Nested1_1$Nested1_2");
    Field f = nested1_2.getDeclaredField("i");
    f.set(null, 1);
  }
}

Here reflection is able to modify the field test1.TopLevel1.Nested1_1.Nested1_2.i without having to make it accessible despite it being inside a private nested class inside a package-private class.

When you are writing code for an environment where untrusted code is run you should keep that in mind to prevent malicious code from messing with internal classes.
So when it comes to the access level of nested types you should always choose the least permissive one, ideally private or package-private.