How to get the binary name of a java class, if one

Question

The reflection classes and methods as well as class loaders etc. need the so called "binary" names of classes to work with.

The question is, how does one get the binary name if one only has the fully qualified name, i.e. the name that one would use in source code.

For example:

package frege;
public static class RT {
    ....
    public static class X { .... }
}

The fully qualified name of the class would be frege.RT.X. Yet, to get the class object, one needs to write:

Class.forName("frege.RT$X")

and not

Class.forName("frege.RT.X")    // fails with ClassNotFoundException

because X happens to be an inner class of frege.RT.

A possible, but clumsy, solution would be to replace . with $ from backwards, one by one, until Class.forName() doesn't throw ClassNotFoundException anymore or there are no more . to replace.

Is there any better/well known/standard solution? I looked in the API docs for Class, CLassLoader and java.lang.reflect but did not find anything usable.

Answer 1

I think its a safe bet that canonical names specify unique classes. As mentioned above javac will not let you create two classes with the same canonical name from withen a single compilation unit. If you have 2 compilations then you can get into trouble regarding which class you load, but at that point I'd be more worried about a library's package name colliding with your package names, which is avoided by all but the malicious.

For this reason, I think its a safe bet to assume you wont run into that scenario. Along those lines, for those who are interested, I implemented the OP's suggestion (flipping $s to .s), and simply throwing a ClassNotFoundException in the event that it doesn't find any classes with that canonical name, or if it finds two or more that have that name.

   /**
 * Returns the single class at the specified canonical name, or throws a {@link java.lang.ClassNotFoundException}.
 *
 * <p>Read about the issues of fully-qualified class paths vs the canonical name string
 * <a href="http://stackoverflow.com/questions/13331902/how-to-get-the-binary-name-of-a-java-class-if-one-has-only-the-fully-qualified">discussed here</a>.
 */
public static <TStaticallyNeeded> Class<TStaticallyNeeded> classForCanonicalName(String canonicalName)
        throws ClassNotFoundException {

    if (canonicalName == null) { throw new IllegalArgumentException("canonicalName"); }

    int lastDotIndex = canonicalName.length();
    boolean hasMoreDots = true;

    String attemptedClassName = canonicalName;

    Set<Class> resolvedClasses = new HashSet<>();

    while (hasMoreDots) try {
        Class resolvedClass = Class.forName(attemptedClassName);
        resolvedClasses.add(resolvedClass);
    }
    catch (ClassNotFoundException e) {
        continue;
    }
    finally {
        if(hasMoreDots){
            lastDotIndex = attemptedClassName.lastIndexOf('.');
            attemptedClassName = new StringBuilder(attemptedClassName)
                    .replace(lastDotIndex, lastDotIndex + 1, "$")
                    .toString();
            hasMoreDots = attemptedClassName.contains(".");
        }
    }

    if (resolvedClasses.isEmpty()) {
        throw new ClassNotFoundException(canonicalName);
    }

    if (resolvedClasses.size() >= 2) {
        StringBuilder builder = new StringBuilder();
        for (Class clazz : resolvedClasses) {
            builder.append("'").append(clazz.getName()).append("'");
            builder.append(" in ");
            builder.append("'").append(
                    clazz.getProtectionDomain().getCodeSource() != null
                            ? clazz.getProtectionDomain().getCodeSource().getLocation()
                            : "<unknown code source>"
            ).append("'");
            builder.append(System.lineSeparator());
        }

        builder.replace(builder.length() - System.lineSeparator().length(), builder.length(), "");

        throw new ClassNotFoundException(
                "found multiple classes with the same canonical names:" + System.lineSeparator() +
                        builder.toString()
        );
    }

    return resolvedClasses.iterator().next();
}

it still annoys me greatly that the "expected" flow is to hit that catch(NoClass) continue code, but if you've ever told eclipse or intelliJ to auto-break on any exceptions thrown, you'll know this kind of behavior is par for the course.

Answer 2

It now sounds like you want to get the fully qualified name (FQN) from the canonical name. As that is different from working from a simple name I'll add a second answer.

The Sun javac command will not compile a classes if a canonical name conflict would result. However by compiling separately you can still get two different classes with the same canonical name.

An example:

File src1\com\stack\Test.java

package com.stack;

public class Test {
    public static class Example {
        public static class Cow {
            public static class Hoof {
            }
        }
    }

    public static void main(String[] args) throws Exception {
        Class<?> cl1 = Class.forName("com.stack.Test$Example$Cow$Hoof");
        Class<?> cl2 = Class.forName("com.stack.Test.Example.Cow.Hoof");
        System.out.println(cl1.getName());
        System.out.println(cl1.getSimpleName());
        System.out.println(cl1.getCanonicalName());
        System.out.println();
        System.out.println(cl2.getName());
        System.out.println(cl2.getSimpleName());
        System.out.println(cl2.getCanonicalName());
    }
}

File src2\com\stack\Test\Example\Cow\Hoof.java

package com.stack.Test.Example.Cow;

public class Hoof { }

Then to compile and execute:

set CLASSPATH=
mkdir bin1 bin2
javac -d bin1 -sourcepath src1 src1\com\stack\Test.java
javac -d bin2 -sourcepath src2 src2\com\stack\Test\Example\Cow\Hoof.java

set CLASSPATH=bin1;bin2
java com.stack.Test

Producing the output:

com.stack.Test$Example$Cow$Hoof
Hoof
com.stack.Test.Example.Cow.Hoof

com.stack.Test.Example.Cow.Hoof
Hoof
com.stack.Test.Example.Cow.Hoof

Thus two classes have the same canonical name but different FQNs. Even if two classes have the same FQN and same canonical name, they can still be different if they are loaded via different class loaders.

To resolve your issue I see several ways forward you could take.

First you can specify that you match the class with the least amount of nesting and hence the least number of '$'s in the FQN. UPDATE It turns out Sun javac does the exact opposite of this and matches the class with the most nesting.

Second you can test all possible FQNs and throw an exception if there is more than one.

Third, accept that the only unique mapping is with the FQN then only within a specified class loader and re-work you application appropriately. I find it convenient to use the thread context class loader as a default class loader.

Answer 3

A simple name omits a lot of information and it is possible to have many classes with the same simple name. That may make this impossible. For example:

package stack;

/**
 * 
 * @author Simon Greatrix
 */
public class TestLocal {

    public Object getObject1() {
        class Thing {
            public String toString() { 
                return "I am a Thing";
            }
        }
        return new Thing();
    }

    public Object getObject2() {
        class Thing {
            public String toString() { 
                return "I am another Thing";
            }
        }
        return new Thing();
    }

    public Object getObject3() {
        class Thing {
            public String toString() { 
                return "I am a rather different Thing";
            }
        }
        return new Thing();
    }

    /**
     * @param args
     */
    public static void main(String[] args) {
        TestLocal test = new TestLocal();
        Object[] objects = new Object[] {
                test.getObject1(),                
                test.getObject2(),                
                test.getObject3()                
        };

        for(Object o : objects) {
            System.out.println("Object      : "+o);
            System.out.println("Simple Name : "+o.getClass().getSimpleName());
            System.out.println("Name        : "+o.getClass().getName());
        }
    }
}

This produces the output:

Object      : I am a Thing
Simple Name : Thing
Name        : stack.TestLocal$1Thing
Object      : I am another Thing
Simple Name : Thing
Name        : stack.TestLocal$2Thing
Object      : I am a rather different Thing
Simple Name : Thing
Name        : stack.TestLocal$3Thing

As you can see, all three local classes have the same simple name.