StackOverflow2013

Note that there are some explanatory texts on larger screens.

plurals

PO
text
Body
copied!<p>Consider this trivial problem:</p> <pre><code>class A { static void foo(){ } } class B extends A { static void foo(){ } } void test() { A.foo(); B.foo(); } </code></pre> <p>Suppose we remove the <code>foo</code> method from <code>B</code>, and we only recompile <code>B</code> itself, what could happen when we run <code>test()</code>? Should it throw linkage error because <code>B.foo()</code> is no found? </p> <p>According to JLS3 #13.4.12, removing <code>B.foo</code> doesn't break binary compatibility, because <code>A.foo</code> is still defined. This means, when <code>B.foo()</code> is executed, <code>A.foo()</code> is invoked. Remember, there's no recompilation of <code>test()</code>, so this forwarding must be handled by JVM.</p> <p>Conversely, let's remove <code>foo</code> method from <code>B</code>, and recompile all. Even though compiler knows statically that <code>B.foo()</code> actually means <code>A.foo()</code>, it still generate <code>B.foo()</code> in the bytecode. For now, JVM will forward <code>B.foo()</code> to <code>A.foo()</code>. But if in future <code>B</code> gains a new <code>foo</code> method, the new method will be invoked at runtime, even if <code>test()</code> isn't recompiled.</p> <p>In this sense, there is a overriding relation among static methods. When compile sees <code>B.foo()</code>, it must compile it to <code>B.foo()</code> in bytecode, regardless whether <code>B</code> has a <code>foo()</code> today.</p> <p>In your example, when compiler sees <code>BigCage.printList(animalCage)</code>, it correctly infer that it's actually calling <code>Cage.printList(List<?>)</code>. So it needs to compile the call into bytecode as <code>BigCage.printList(List<?>)</code> - the target class must be <code>BigCage</code> here instead of <code>Cage</code>.</p> <p>Oops! Bytecode format hasn't been upgrade to handle method signature like that. Generics information are preserved in bytecode as auxilary information, but for method invocation, it's the old way. </p> <p>Erasure happens. The call is actually compiled into <code>BigCage.printList(List)</code>. Too bad <code>BigCage</code> also has a <code>printList(List)</code> after erasure. At runtime, that method is invoked!</p> <p>This problem is due to the mismatch between Java spec and JVM spec. </p> <p>Java 7 tightens up a little; realizing bytecode and JVM can't handle such situations, it no longer compiles your code:</p> <blockquote> <p>error: name clash: printList(List) in BigCage and printList(List) in Cage have the same erasure, yet neither hides the other</p> </blockquote> <p>Another fun fact: if the two methods have different return types, your program will work correctly. This is because in byte code, method signature includes return type. So there is no confusion between <code>Dog printList(List)</code> and <code>Object printList(List)</code>. See also <a href="https://stackoverflow.com/questions/5527235/type-erasure-and-overloading-in-java-why-does-this-work/5528802#5528802">Type Erasure and Overloading in Java: Why does this work?</a> This trick is only allowed in Java 6. Java 7 forbids it, probably for reasons other than technical ones.</p>

Querying!

Guidance

An individual column

Larger individual text columns get their own page to allow for proper reading.

SQuiL has stopped working due to an internal error.

If you are curious you may find further information in the browser console, which is accessible through the devtools (F12).

Reload