Related
I come from functional languages (e.g. Haskell) and I enjoy a lot on typeclasses to achieve polymorphism which is also a structural approach to implement ad-hoc overloading.
However, recently I'm starting to understand OOP's way to model real problems and I'm curious why do we need dynamic polymorphism in OOP languages (such as Java). In my experience, most of function call can be resolved during compile time as many functional languages do not support subtyping.
So my problem is that, In what kind of situation do we have to use dynamic polymorphism instead of compile-time polymorphism? My guesses are:
When we use the subtype system where we have objects we cannot decide its actual type (e.g. we have a container containing many objects of various types. However, in this situation, why not try Algebraic data type or union type to model the container's element type?).
We only have the object and we do not know its methods' real name, so we have to use the vptr table to help us.
In Haskell, we replace "dynamic polymorphism" with higher-order functions.
Consider the following problem: we want to define a type which denotes a predicate. We will eventually use this type Predicate when we implement our Lists so that we can define the filter function. We would like to be able to easily define the equality predicate, the less-than predicate, and be able to combine two predicates by joining them with "and".
A reasonable Java attempt would look like this.
interface Predicate<T> {
public abstract boolean predicate(T x);
}
class EqualsPredicate<T> implements Predicate<T> {
private final T value;
public EqualsPredicate(T value) {
this.value = value;
}
#Override
public boolean predicate(T x) {
return this.value.equals(x);
}
}
class LessPredicate<T implements Comparable<T>> implements Predicate<T>{
private final T value;
public LessPredicate(T value) {
this.value = value;
}
#Override
public boolean predicate(T x) {
return this.value.compareTo(x) < 0;
}
}
class AndPredicate<T> implements Predicate<T> {
private final Predicate<T> p1;
private final Predicate<T> p2;
public AndPredicate(Predicate<T> p1, Predicate<T> p2) {
this.p1 = p1;
this.p2 = p2;
}
#Override
public boolean predicate(T x) {
return p1.predicate(x) && p2.predicate(x);
}
}
In Haskell, the answer to this conundrum is obvious. We just define
type Predicate t = t -> Bool
makeEqualsPredicate :: Eq t => t -> Predicate t
makeEqualsPredicate = (==)
makeLessPredicate :: Ord t => t -> Predicate t
makeLessPredicate = (<)
makeAndPredicate :: Predicate t -> Predicate t -> Predicate t
makeAndPredicate p1 p2 x = p1 x && p2 x
-- or, even more succinctly, makeAndPredicate = liftA2 (&&)
Java allows "dynamic dispatch" of methods through subclassing. Haskell allows "dynamic dispatch" of functions through higher-order functions.
But wait, you say. Predicate was an interface with only one method. What should we do if we want to have two methods?
Well, if an interface with one method corresponds to a function, an interface with two methods must correspond to a pair of functions. This is just the OOP principle known as "composition over inheritance".
So we can always replace Java-style dynamic polymorphism with Haskell-style higher-order functions.
In fact, you actually see this observation in modern Java as well. As of Java 8, you can add the annotation #FunctionalInterface to an interface with one method, which permits you to create instances of that interface using lambdas. So you could write in Java 8
#FunctionalInterface
interface Predicate<T> {
public abstract boolean predicate(T x);
public static Predicate<J> makeEqualsPredicate(J t) {
return (x) -> t.equals(x);
}
public static Predicate<J implements Comparable<J>> makeLessPredicate(J t) {
return (x) -> t.compareTo(x) < 0;
}
public Predicate<T> makeAndPredicate(Predicate<T> other) {
return (x) -> this.predicate(x) && other.predicate(x);
}
}
With many people's help, currently I've got some of answers I want after reflecting lots of designs. Since Rust has both nice support for static and dynamic polymorphism, I shall use Rust in this answer to demonstrate my points.
I now have 2 points for dynamic dispatch: user-friendly scalability and smaller compiled size .
Point 1: user-friendly scalability
Many people argue that dynamic dispatch is suitable for a situation where you have a container to collect various kinds of objects(of course, different types). For example:
trait MyTrait {
fn method(&self, ...) -> ReturnType;
}
type MyContainer = Vec<Box<MyTrait>>;
fn main() {
...
//the_container : MyContainer
the_container.iter().map(... { x.method(...) } ...) //only demo code
}
In code above, on compile time, we only know that the elements are trait objects, which means the program shall use a vptr-like strategy to find which method to use during executing the expression in the main function.
However, there's another way to implement nearly the same thing:
enum MyContainerTypes {
Type0 A,
Type1 B,
...
}
impl MyTrait for MyContainerType {
fn method(&self, ...) -> ReturnType {
match self {
Type0 a => {...},
Type1 b => {...},
...
}
}
}
type MyContainer = Vec<MyContainerType>;
fn main() {
...
//my_container : MyContainer
my_container.iter().map(... { x.method(...) } ...); //demo
}
In this way, no dynamic polymorphism is required, however, consider the following situation: You are a user of a library which has been designed and you have no access to change definitions like enums inside the library. Now you want to make your own type of ContainerType and you want to reuse codes of existed logic. If you are using dynamic dispatch, the work is simple: just make another impl of your custom container type and everything's fine. Unfortunately, if you are using the static version of the library, it may become a little hard to achieve this goal...
Point 2: Smaller compiled size
Languages like Rust may have to compile a generic function many times, once for each type it’s used with. This could make the binary large, a phenomenon called code bloat in C++ circles.
Let's consider a simpler case:
trait MyTrait {
fn method(&self, ...) -> ReturnType;
}
fn f(x: MyTrait) { ... } //MyTrait is unsized, this is unvalid rust code
fn g<T: MyTrait>(x: T) { ... }
If you have lots of functions like function g, the compiled size may become larger and larger. However this should not be a big issue since most of us have the luxury of ignoring code size for plentiful memory nowadays.
Conclusion
In short, although static polymorphism has many advantages over dynamic polymorphism, there're still some corners dynamic dispatch can work better. Personally I really love Haskell-like's way to treat polymorphism(that's also why I like Rust). I don't think this can be the final best and complete answer, discussions are welcome!
Combining strategies
It suddenly occurred to me that why not combine the static and dynamic strategies? To allow users to further extend our model, we can just leave a small hole for users to fill in later, like:
trait Custom {
fn method(&self) -> ReturnType;
}
enum Object {
A(Type0),
B(Type1),
...
Hole(Box<dyn Custom>)
}
However, in this way, some operations like clone may be a little hard to implement, but I think this is still an interesting idea.
Update
Haskell's existential type also has similar function and implementation as dynamic polymorphism in OOP languages:
data Obj = forall a. (Show a) => Obj a
xs :: [Obj]
xs = [Obj 1, Obj "foo", Obj 'c']
doShow :: [Obj] -> String
doShow [] = ""
doShow ((Obj x):xs) = show x ++ doShow xs
I also found that this existential type can be used to hide some details of types and provide cleaner interface for users to use.
Edit
Thanks #MarkSaving. There's a mistake in Point 2's code, the dyn trait object is unsized and therefore should be changed to a reference or boxed dyn:
fn f(x: Box<dyn MyTrait>) { ... }
I was reading this tutorial on Java 8 where the writer showed the code:
interface Formula {
double calculate(int a);
default double sqrt(int a) {
return Math.sqrt(a);
}
}
And then said
Default methods cannot be accessed from within lambda expressions. The
following code does not compile:
Formula formula = (a) -> sqrt( a * 100);
But he did not explain why it is not possible. I ran the code, and it gave an error,
incompatible types: Formula is not a functional interface`
So why is it not possible or what is the meaning of the error? The interface fulfills the requirement of a functional interface having one abstract method.
It's more or less a question of scope. From the JLS
Unlike code appearing in anonymous class declarations, the meaning of
names and the this and super keywords appearing in a lambda body,
along with the accessibility of referenced declarations, are the same
as in the surrounding context (except that lambda parameters introduce
new names).
In your attempted example
Formula formula = (a) -> sqrt( a * 100);
the scope does not contain a declaration for the name sqrt.
This is also hinted at in the JLS
Practically speaking, it is unusual for a lambda expression to need to
talk about itself (either to call itself recursively or to invoke its
other methods), while it is more common to want to use names to refer
to things in the enclosing class that would otherwise be shadowed
(this, toString()). If it is necessary for a lambda expression to
refer to itself (as if via this), a method reference or an anonymous
inner class should be used instead.
I think it could have been implemented. They chose not to allow it.
Lambda expressions work in a completely different way from anonymous classes in that this represents the same thing that it would in the scope surrounding the expression.
For example, this compiles
class Main {
public static void main(String[] args) {
new Main().foo();
}
void foo() {
System.out.println(this);
Runnable r = () -> {
System.out.println(this);
};
r.run();
}
}
and it prints something like
Main#f6f4d33
Main#f6f4d33
In other words this is a Main, rather than the object created by the lambda expression.
So you cannot use sqrt in your lambda expression because the type of the this reference is not Formula, or a subtype, and it does not have a sqrt method.
Formula is a functional interface though, and the code
Formula f = a -> a;
compiles and runs for me without any problem.
Although you cannot use a lambda expression for this, you can do it using an anonymous class, like this:
Formula f = new Formula() {
#Override
public double calculate(int a) {
return sqrt(a * 100);
}
};
That's not exactly true. Default methods can be used in lambda expressions.
interface Value {
int get();
default int getDouble() {
return get() * 2;
}
}
public static void main(String[] args) {
List<Value> list = Arrays.asList(
() -> 1,
() -> 2
);
int maxDoubled = list.stream()
.mapToInt(val -> val.getDouble())
.max()
.orElse(0);
System.out.println(maxDoubled);
}
prints 4 as expected and uses a default method inside a lambda expression (.mapToInt(val -> val.getDouble()))
What the author of your article tries to do here
Formula formula = (a) -> sqrt( a * 100);
is to define a Formula, which works as functional interface, directly via a lambda expression.
That works fine, in above example code, Value value = () -> 5 or with Formula as interface for example
Formula formula = (a) -> 2 * a * a + 1;
But
Formula formula = (a) -> sqrt( a * 100);
fails because it's trying to access the (this.)sqrt method but it can't.
Lambdas as per spec inherit their scope from their surroundings, meaning that this inside a lambda refers to the same thing as directly outside of it. And there is no sqrt method outside.
My personal explanation for this: Inside the lambda expression, it's not really clear to what concrete functional interface the lambda is going to be "converted". Compare
interface NotRunnable {
void notRun();
}
private final Runnable r = () -> {
System.out.println("Hello");
};
private final NotRunnable r2 = r::run;
The very same lambda expression can be "cast" to multiple types. I think of it as if a lambda doesn't have a type. It's a special typeless function that can be used for any Interface with the right parameters. But that restriction means that you can't use methods of the future type because you can't know it.
This adds little to the discussion, but I found it interesting anyways.
Another way to see the problem would be to think about it from the standpoint of a self-referencing lambda.
For example:
Formula formula = (a) -> formula.sqrt(a * 100);
It would seem that this ought to make sense, since by the time the lambda gets to be executed the formula reference must have already being initialized (i.e. there is not way to do formula.apply() until formula has been properly initialized, in whose case, from the body of the lambda, the body of apply, it should be possible to reference the same variable).
However this does not work either. Interestingly, it used to be possible at the beginning. You can see that Maurice Naftalin had it documented in his Lambda FAQ Web Site. But for some reason the support for this feature was ultimately removed.
Some of the suggestions given in other answers to this question have been already mentioned there in the very discussion in the lambda mailing list.
Default methods can be accessed only with object references, if you want to access default method you'd have an object reference of Functional Interface, in lambda expression method body you won't have so can't access it.
You are getting an error incompatible types: Formula is not a functional interface because you have not provided #FunctionalInterface annotation, if you have provided you'll get 'method undefined' error, compiler will force you to create a method in the class.
#FunctionalInterface must have only one abstract method your Interface has that but it is missing the annotation.
But static methods have no such restriction, since we can access it with out object reference like below.
#FunctionalInterface
public interface Formula {
double calculate(int a);
static double sqrt(int a) {
return Math.sqrt(a);
}
}
public class Lambda {
public static void main(String[] args) {
Formula formula = (a) -> Formula.sqrt(a);
System.out.println(formula.calculate(100));
}
}
I had some confusion about inner classes and lambda expression, and I tried to ask a question about that, but then another doubt arose, and It's probable better posting another question than commenting the previous one.
Straight to the point: I know (thank you Jon) that something like this won't compile
public class Main {
public static void main(String[] args) {
One one = new One();
F f = new F(){ //1
public void foo(){one.bar();} //compilation error
};
one = new One();
}
}
class One { void bar() {} }
interface F { void foo(); }
due to how Java manages closures, because one is not [effectively] final and so on.
But then, how come is this allowed?
public class Main {
public static void main(String[] args) {
One one = new One();
F f = one::bar; //2
one = new One();
}
}
class One { void bar() {} }
interface F { void foo(); }
Is not //2 equivalent to //1? Am I not, in the second case, facing the risks of "working with an out-of-date variable"?
I mean, in the latter case, after one = new One(); is executed f still have an out of date copy of one (i.e. references the old object). Isn't this the kind of ambiguity we're trying to avoid?
A method reference is not a lambda expression, although they can be used in the same way. I think that is what is causing the confusion. Below is a simplification of how Java works, it is not how it really works, but it is close enough.
Say we have a lambda expression:
Runnable f = () -> one.bar();
This is the equivalent of an anonymous class that implements Runnable:
Runnable f = new Runnable() {
public void run() {
one.bar();
}
}
Here the same rules apply as for an anonymous class (or method local class). This means that one needs to effectively final for it to work.
On the other hand the method handle:
Runnable f = one::bar;
Is more like:
Runnable f = new MethodHandle(one, one.getClass().getMethod("bar"));
With MethodHandle being:
public class MethodHandle implements Runnable {
private final Object object;
private final Method method;
public MethodHandle(Object object, java.lang.reflect.Method method) {
this.object = Object;
this.method = method;
}
#Override
public void run() {
method.invoke(object);
}
}
In this case, the object assigned to one is assigned as part of the method handle created, so one itself doesn't need to be effectively final for this to work.
Your second example is simply not a lambda expression. It's a method reference. In this particular case, it chooses a method from a particular object, which is currently referenced by the variable one. But the reference is to the object, not to the variable one.
This is the same as the classical Java case:
One one = new One();
One two = one;
one = new One();
two.bar();
So what if one changed? two references the object that one used to be, and can access its method.
Your first example, on the other hand, is an anonymous class, which is a classical Java structure that can refer to local variables around it. The code refers to the actual variable one, not the object to which it refers. This is restricted for the reasons that Jon mentioned in the answer you referred to. Note that the change in Java 8 is merely that the variable has to be effectively final. That is, it still can't be changed after initialization. The compiler simply became sophisticated enough to determine which cases will not be confusing even when the final modifier is not explicitly used.
The consensus appears to be that this is because when you do it using an anonymous class, one refers to a variable, whereas when you do it using a method reference, the value of one is captured when the method handle is created. In fact, I think that in both cases one is a value rather than a variable. Let's consider anonymous classes, lambda expressions and method references in a bit more detail.
Anonymous classes
Consider the following example:
static Supplier<String> getStringSupplier() {
final Object o = new Object();
return new Supplier<String>() {
#Override
public String get() {
return o.toString();
}
};
}
public static void main(String[] args) {
Supplier<String> supplier = getStringSupplier();
System.out.println(supplier.get()); // Use o after the getStringSupplier method returned.
}
In this example, we are calling toString on o after the method getStringSupplier has returned, so when it appears in the get method, o cannot refer to a local variable of the getStringSupplier method. In fact it is essentially equivalent to this:
static Supplier<String> getStringSupplier() {
final Object o = new Object();
return new StringSupplier(o);
}
private static class StringSupplier implements Supplier<String> {
private final Object o;
StringSupplier(Object o) {
this.o = o;
}
#Override
public String get() {
return o.toString();
}
}
Anonymous classes make it look as if you are using local variables, when in fact the values of these variables are captured.
In contrast to this, if a method of an anonymous class references the fields of the enclosing instance, the values of these fields are not captured, and the instance of the anonymous class does not hold references to them; instead the anonymous class holds a reference to the enclosing instance and can access its fields (either directly or via synthetic accessors, depending on the visibility). One advantage is that an extra reference to just one object, rather than several, is required.
Lambda expressions
Lambda expressions also close over values, not variables. The reason given by Brian Goetz here is that
idioms like this:
int sum = 0;
list.forEach(e -> { sum += e.size(); }); // ERROR
are fundamentally serial; it is quite difficult to write lambda bodies
like this that do not have race conditions. Unless we are willing to
enforce -- preferably at compile time -- that such a function cannot
escape its capturing thread, this feature may well cause more trouble
than it solves.
Method references
The fact that method references capture the value of the variable when the method handle is created is easy to check.
For example, the following code prints "a" twice:
String s = "a";
Supplier<String> supplier = s::toString;
System.out.println(supplier.get());
s = "b";
System.out.println(supplier.get());
Summary
So in summary, lambda expressions and method references close over values, not variables. Anonymous classes also close over values in the case of local variables. In the case of fields, the situation is more complicated, but the behaviour is essentially the same as capturing the values because the fields must be effectively final.
In view of this, the question is, why do the rules that apply to anonymous classes and lambda expressions not apply to method references, i.e. why are you allowed to write o::toString when o is not effectively final? I do not know the answer to that, but it does seem to me to be an inconsistency. I guess it's because you can't do as much harm with a method reference; examples like the one quoted above for lambda expressions do not apply.
No. In your first example you define the implementation of F inline and try to access the instance variable one.
In the second example you basically define your lambda expression to be the call of bar() on the object one.
Now this might be a bit confusing. The benefit of this notation is that you can define a method (most of the time it is a static method or in a static context) once and then reference the same method from various lambda expressions:
msg -> System.out::println(msg);
In English, a homograph pair is two words that have the same spelling but different meanings.
In software engineering, a pair of homographic methods is two methods with the same name but different requirements. Let's see a contrived example to make the question as clear as possible:
interface I1 {
/** return 1 */
int f()
}
interface I2 {
/** return 2*/
int f()
}
interface I12 extends I1, I2 {}
How can I implement I12? C# has a way to do this, but Java doesn't. So the only way around is a hack. How can it be done with reflection/bytecode tricks/etc most reliably (i.e it doesn't have to be a perfect solution, I just want the one that works the best)?
Note that some existing closed source massive piece of legacy code which I cannot legally reverse engineer requires a parameter of type I12 and delegates the I12 both to code that has I1 as a parameter, and code that has I2 as a parameter. So basically I need to make an instance of I12 that knows when it should act as I1 and when it should act as I2, which I believe can be done by looking at the bytecode at runtime of the immediate caller. We can assume that no reflection is used by the callers, because this is straightforward code. The problem is that the author of I12 didn't expect that Java merges f from both interfaces, so now I have to come up with the best hack around the problem. Nothing calls I12.f (obviously if the author wrote some code that actually calls I12.f, he would have noticed the problem before selling it).
Note that I'm actually looking for an answer to this question, not how to restructure the code that I can't change. I'm looking for the best heuristic possible or an exact solution if one exists. See Gray's answer for a valid example (I'm sure there are more robust solutions).
Here is a concrete example of how the problem of homographic methods within two interfaces can happen. And here is another concrete example:
I have the following 6 simple classes/interfaces. It resembles a business around a theater and the artists who perform in it. For simplicity and to be specific, let's assume they are all created by different people.
Set represents a set, as in set theory:
interface Set {
/** Complements this set,
i.e: all elements in the set are removed,
and all other elements in the universe are added. */
public void complement();
/** Remove an arbitrary element from the set */
public void remove();
public boolean empty();
}
HRDepartment uses Set to represent employees. It uses a sophisticated process to decode which employees to hire/fire:
import java.util.Random;
class HRDepartment {
private Random random = new Random();
private Set employees;
public HRDepartment(Set employees) {
this.employees = employees;
}
public void doHiringAndLayingoffProcess() {
if (random.nextBoolean())
employees.complement();
else
employees.remove();
if (employees.empty())
employees.complement();
}
}
The universe of a Set of employees would probably be the employees who have applied to the employer. So when complement is called on that set, all the existing employees are fired, and all the other ones that applied previously are hired.
Artist represents an artist, such as a musician or an actor. An artist has an ego. This ego can increase when others compliment him:
interface Artist {
/** Complements the artist. Increases ego. */
public void complement();
public int getEgo();
}
Theater makes an Artist perform, which possibly causes the Artist to be complemented. The theater's audience can judge the artist between performances. The higher the ego of the performer, the more likely the audience will like the Artist, but if the ego goes beyond a certain point, the artist will be viewed negatively by the audience:
import java.util.Random;
public class Theater {
private Artist artist;
private Random random = new Random();
public Theater(Artist artist) {
this.artist = artist;
}
public void perform() {
if (random.nextBoolean())
artist.complement();
}
public boolean judge() {
int ego = artist.getEgo();
if (ego > 10)
return false;
return (ego - random.nextInt(15) > 0);
}
}
ArtistSet is simply an Artist and a Set:
/** A set of associated artists, e.g: a band. */
interface ArtistSet extends Set, Artist {
}
TheaterManager runs the show. If the theater's audience judges the artist negatively, the theater talks to the HR department, which will in turn fire artists, hire new ones, etc:
class TheaterManager {
private Theater theater;
private HRDepartment hr;
public TheaterManager(ArtistSet artists) {
this.theater = new Theater(artists);
this.hr = new HRDepartment(artists);
}
public void runShow() {
theater.perform();
if (!theater.judge()) {
hr.doHiringAndLayingoffProcess();
}
}
}
The problem becomes clear once you try to implement an ArtistSet: both superinterfaces specify that complement should do something else, so you have to implement two complement methods with the same signature within the same class, somehow. Artist.complement is a homograph of Set.complement.
New idea, kinda messy...
public class MyArtistSet implements ArtistSet {
public void complement() {
StackTraceElement[] stackTraceElements = Thread.currentThread().getStackTrace();
// the last element in stackTraceElements is the least recent method invocation
// so we want the one near the top, probably index 1, but you might have to play
// with it to figure it out: could do something like this
boolean callCameFromHR = false;
boolean callCameFromTheatre = false;
for(int i = 0; i < 3; i++) {
if(stackTraceElements[i].getClassName().contains("Theatre")) {
callCameFromTheatre = true;
}
if(stackTraceElements[i].getClassName().contains("HRDepartment")) {
callCameFromHR = true;
}
}
if(callCameFromHR && callCameFromTheatre) {
// problem
}
else if(callCameFromHR) {
// respond one way
}
else if(callCameFromTheatre) {
// respond another way
}
else {
// it didn't come from either
}
}
}
Despite Gray Kemmey's valiant attempt, I would say the problem as you have stated it is not solvable. As a general rule given an ArtistSet you cannot know whether the code calling it was expecting an Artist or a Set.
Furthermore, even if you could, according to your comments on various other answers, you actually have a requirement to pass an ArtistSet to a vendor-supplied function, meaning that function has not given the compiler or humans any clue as to what it is expecting. You are completely out of luck for any sort of technically correct answer.
As practical programming matter for getting the job done, I would do the following (in this order):
File a bug report with whoever created an interface requiring ArtistSet and whoever generated the ArtistSet interface itself.
File a support request with the vendor supplying the function requiring an ArtistSet and ask them what they expect the behavior of complement() to be.
Implement the complement() function to throw an exception.
public class Sybil implements ArtistSet {
public void complement() {
throw new UnsupportedOperationException('What am I supposed to do');
}
...
}
Because seriously, you don't know what to do. What would be the correct thing to do when called like this (and how do you know for sure)?
class TalentAgent {
public void pr(ArtistSet artistsSet) {
artistSet.complement();
}
}
By throwing an exception you have a chance at getting a stack trace that gives you a clue as to which of the two behaviors the caller is expecting. With luck nobody calls that function, which is why the vendor got as far as shipping code with this problem. With less luck but still some, they handle the exception. If not even that, well, at least now you will have a stack trace you can review to decide what the caller was really expecting and possibly implement that (though I shudder to think of perpetuation a bug that way, I've explained how I would do it in this other answer).
BTW, for the rest of the implementation I would delegate everything to actual Artist and Set objects passed in via the constructor so this can be easily pulled apart later.
How to Solve For Your Specific Case
ArtistSet is simply an Artist and a Set:
/** A set of associated artists, e.g: a band. */
interface ArtistSet extends Set, Artist { }
From an OO perspective, that's not a useful declaration. An Artist is a type of noun, a "thing" that has defined properties and actions (methods).
A Set is an aggregate of things - a collection of unique elements. Instead, try:
ArtistSet is simply a Set of Artists.
/** A set of associated artists, e.g: a band. */
interface ArtistSet extends Set<Artist> { };
Then, for your particular case, the homonym methods are on interfaces that are never combined within the one type, so you have no clash and can program away...
Further, you don't need to declare ArtistSet because you aren't actually extending Set with any new declarations. You're just instantiating a type parameter, so you can replace all usage with Set<Artist>.
How to Solve For the More General Case
For this clash the method names don't even need to be homographic in the english language sense - they can be the same word with same english meaning, used in different contexts in java. Clash occurs if you have two interfaces that you wish to apply to a type but they contain the same declaration (e.g. method signature) with conflicting semantic/processing definitions.
Java does not allow you to implement the behaviour you request - you must have an alternative work-around. Java doesn't allow a class to provide multiple implementations for the same method signature from multiple different interfaces (implementing the same method multiple times with some form of qualification/alias/annotation to distinguish). See Java overriding two interfaces, clash of method names,
Java - Method name collision in interface implementation
Avoid use of Inheritence (extends or implements) and instead use Object Composition (see http://en.wikipedia.org/wiki/Composition_over_inheritance)
E.g. If you have the following
interface TV {
void switchOn();
void switchOff();
void changeChannel(int ChannelNumber);
}
interface Video {
void switchOn();
void switchOff();
void eject();
void play();
void stop();
}
Then if you have an object that is both of these things, you can combine the two in a new interface (optional) or type:
interface TVVideo {
TV getTv();
Video getVideo();
}
class TVVideoImpl implements TVVideo {
TV tv;
Video video;
public TVVideoImpl() {
tv = new SomeTVImpl(....);
video = new SomeVideoImpl(....);
}
TV getTv() { return tv };
Video getVideo() { return video };
}
How can I implement a class which has two superinterfaces having homographic methods?
In Java, a class which has two superinterfaces having homographic methods is considered to have only one implementation of this method. (See the Java Language Specification section 8.4.8). This allows classes to conveniently inherit from multiple interfaces that all implement the same other interface and only implement the function once. This also simplifies the language because this eliminates the need for syntax and method dispatching support for distinguishing between homographic methods based on which interface they came from.
So the correct way to implement a class which has two superinterfaces having homographic methods is to provide a single method that satisfies the contracts of both superinterfaces.
C# has a way to do this. How can it be done in Java? Is there no construct for this?
C# defines interfaces differently than Java does and therefore has capabilities that Java does not.
In Java, the language construct is defined to mean that all interfaces get the same single implementation of homographic methods. There is no Java language construct for creating alternate behaviors of multiply-inherited interface functions based on the compile time class of the object. This was a conscious choice made by the Java language designers.
If not, how can it be done with reflection/bytecode tricks/etc most reliably?
"It" cannot be done with reflection/bytecode tricks because the information needed to decide which interface's version of the homographic method to choose is not necessarily present in the Java source code. Given:
interface I1 {
// return ASCII character code of first character of String s
int f(String s); // f("Hello") returns 72
}
interface I2 {
// return number of characters in String s
int f(String s); // f("Hello") returns 5
}
interface I12 extends I1, I2 {}
public class C {
public static int f1(I1 i, String s) { return i.f(s); } // f1( i, "Hi") == 72
public static int f2(I2 i, String s) { return i.f(s); } // f2( i, "Hi") == 2
public static int f12(I12 i, String s) { return i.f(s);} // f12(i, "Hi") == ???
}
According to the Java language specification, a class implementing I12 must do so in such a way that C.f1(), C.f2(), and C.f12() return the exact same result when called with the same arguments. If C.f12(i, "Hello") sometimes returned 72 and sometimes returned 5 based on how C.f12() were called, that would be a serious bug in the program and a violation of the language specification.
Furthermore, if the author of class C expected some kind of consistent behavior out of f12(), there is no bytecode or other information in class C that indicates whether it should be the behavior of I1.f(s) or I2.f(s). If the author of C.f12() had in mind C.f("Hello") should return 5 or 72, there's no way to tell from looking at the code.
Fine, so I cannot in general provide different behaviors for homographic functions using bytecode tricks, but I really have a class like my example class TheaterManager. What should I do to implement ArtistSet.complement()?
The actual answer to the actual question you asked is to create your own substitute implementation of TheaterManager that does not require an ArtistSet. You do not need to change the library's implementation, you need to write your own.
The actual answer to the other example question you cite is basically "delegate I12.f() to I2.f()" because no function that receives an I12 object goes on to pass that object to a function expecting an I1 object.
Stack Overflow is only for questions and answers of general interest
One of the stated reasons to reject a question here is that "it is only relevant to an extraordinarily narrow situation that is not generally applicable to the worldwide audience of the internet." Because we want to be helpful, the preferred way to handle such narrow questions is to revise the question to be more broadly applicable. For this question I have taken the approach of answering the broadly applicable version of the question rather than actually editing the question to remove what makes it unique to your situation.
In the real world of commercial programming any Java library that has a broken interface like I12 would not accumulate even dozens of commercial clients unless it could be used by implementing I12.f() in one of these ways:
delegate to I1.f()
delegate to I2.f()
do nothing
throw an exception
pick one of the above strategies on a per-call basis based on the values of some members of the I12 object
If thousands or even only a handful of companies are using this part of this library in Java then you can be assured they have used one of those solutions. If the library is not in use by even a handful of companies then the question is too narrow for Stack Overflow.
OK, TheaterManager was an oversimplification. In the real case it is too hard for me to replace that class and I don't like any of the practical solutions you've outlined. Can't I just fix this with fancy JVM tricks?
It depends on what you want to fix. If you want to fix your specific library by mapping all the calls to I12.f() and then parsing the the stack to determine the caller and choosing a behavior based on that. You can access the stack via Thread.currentThread().getStackTrace().
If you run across a caller you do not recognize you may have a hard time figuring out which version they want. For example you may be called from a generic (as was the actual case in the other specific example you gave), like:
public class TalentAgent<T extends Artist> {
public static void butterUp(List<T> people) {
for (T a: people) {
a.complement()
}
}
}
In Java, generics are implemented as erasures, meaning all type information is thrown away at compile time. There is no class or method signature difference between a TalentAgent<Artist> and a TalentAgent<Set> and the formal type of the people parameter is just List. There is nothing in the class interface or method signature of the caller to tell you what to do by looking at the stack.
So you would need to implement multiple strategies, one of which would be decompiling the code of the calling method looking for clues that the caller is expecting one class or another. It would have to be very sophisticated to cover all the ways this could happen, because among other things you have no way of knowing in advance what class it actually expecting, only that it is expecting a class that implements one of the interfaces.
There are mature and extremely sophisticated open source bytecode utilities, including one that automatically generates a proxy for a given class at runtime (written long before there was support for that in the Java language), so the fact that there isn't an open source utility for handling this case speaks volumes about the ratio of effort to usefulness in pursuing this approach.
Okay, after much research, I have another idea to fully accommodate the situation. Since you can't directly modify their code... you can force the modifications yourself.
DISCLAIMER: The example code below is very simplified. My intention is to show the general method of how this might be done, not to produce functioning source code to do it (since that's a project in itself).
The issue is that the methods are homographic. So to solve it, we can just rename the methods. Simple, right? We can use the Instrument package to achieve this. As you'll see in the linked documentation, it allows you to make an "agent" which can directly modify classes as they're loaded or re-modify them even if they've already been loaded.
Essentially, this requires you to make two classes:
An agent class which preprocesses and reloads classes; and,
A ClassFileTransformer implementation which specifies the changes you want to make.
The agent class must have either a premain() or agentmain() method defined, based on whether you want it to begin its processing as the JVM starts up or after it is already running. Examples of this are in the package documentation above. These methods give you access to an Instrumenation instance, which will allow you to register your ClassFileTransformer. So it might look something like this:
InterfaceFixAgent.java
public class InterfaceFixAgent {
public static void premain(String agentArgs, Instrumentation inst) {
//Register an ArtistTransformer
inst.addTransformer(new ArtistTransformer());
//In case the Artist interface or its subclasses
//have already been loaded by the JVM
try {
for(Class<?> clazz : inst.getAllLoadedClasses()) {
if(Artist.class.isAssignableFrom(clazz)) {
inst.retransformClasses(clazz);
}
}
}
catch(UnmodifiableClassException e) {
//TODO logging
e.printStackTrace();
}
}
}
ArtistTransformer.java
public class ArtistTransformer implements ClassFileTransformer {
private static final byte[] BYTES_TO_REPLACE = "complement".getBytes();
private static final byte[] BYTES_TO_INSERT = "compliment".getBytes();
#Override
public byte[] transform(ClassLoader loader, String className,
Class<?> classBeingRedefined, ProtectionDomain protectionDomain,
byte[] classfileBuffer) throws IllegalClassFormatException {
if(Artist.class.isAssignableFrom(classBeingRedefined)) {
//Loop through the classfileBuffer, find sequences of bytes
//which match BYTES_TO_REPLACE, replace with BYTES_TO_INSERT
}
else return classfileBuffer;
}
This is, of course, simplified. It will replace the word "complement" with "compliment" in any class which extends or implements Artist, so you will very likely need to further conditionalize it (for example, if Artist.class.isAssignableFrom(classBeingRedefined) && Set.class.isAssignableFrom(classBeingRedefined), you obviously don't want to replace every instance of "complement" with "compliment", as the "complement" for Set is perfectly legitimate).
So, now we've corrected the Artist interface and its implementations. The typo is gone, the methods have two different names, so there is no homography. This allows us to have two different implementations in our CommunityTheatre class now, each of which will properly implement/override the methods from the ArtistSet.
Unfortunately, we've now created another (possibly even bigger) issue. We've just broken all the previously-legitimate references to complement() from classes implementing Artist. To fix this, we will need to create another ClassFileTransformer which replaces these calls with our new method name.
This is somewhat more difficult, but not impossible. Essentially, the new ClassFileTransformer (let's say we call it the OldComplementTransformer) will have to perform the following steps:
Find the same string of bytes as before (the one representing the old method name, "complement");
Get the bytes before this which represent the object reference calling the method;
Convert those bytes into an Object;
Check to see if that Object is an Artist; and,
If so, replace those bytes with the new method name.
Once you've made this second transformer, you can modify the InterfaceFixAgent to accommodate it. (I also simplified the retransformClasses() call, since in the example above we perform the needed check within the transformer itself.)
InterfaceFixAgent.java (modified)
public class InterfaceFixAgent {
public static void premain(String agentArgs, Instrumentation inst) {
//Register our transformers
inst.addTransformer(new ArtistTransformer());
inst.addTransformer(new OldComplementTransformer());
//Retransform the classes that have already been loaded
try {
inst.retransformClasses(inst.getAllLoadedClasses());
}
catch(UnmodifiableClassException e) {
//TODO logging
e.printStackTrace();
}
}
}
And now... our program is good to go. It certainly wouldn't be easy to code, and it will be utter hell to QA and test. But it's certainly robust, and it solves the issue. (Technically, I suppose it avoids the issue by removing it, but... I'll take what I can get.)
Other ways we might have solved the problem:
The Unsafe API
A native method written in C
Both of these would allow you to directly manipulate bytes in memory. A solution could certainly be designed around these, but I believe it would be much more difficult and much less safe. So I went with the route above.
I think this solution could even be made more generic into an incredibly useful library for integrating code bases. Specify which interface and which method you need refactored in a variable, a command line argument, or a configuration file, and let her loose. The library that reconciles conflicting interfaces in Java at runtime. (Of course, I think it would still be better for everyone if they just fixed the bug in Java 8.)
Here's what I'd do to remove the ambiguity:
interface Artist {
void complement(); // [SIC] from OP, really "compliment"
int getEgo();
}
interface Set {
void complement(); // as in Set Theory
void remove();
boolean empty(); // [SIC] from OP, I prefer: isEmpty()
}
/**
* This class is to represent a Set of Artists (as a group) -OR-
* act like a single Artist (with some aggregate behavior). I
* choose to implement NEITHER interface so that a caller is
* forced to designate, for any given operation, which type's
* behavior is desired.
*/
class GroupOfArtists { // does NOT implement either
private final Set setBehavior = new Set() {
#Override public void remove() { /*...*/ }
#Override public boolean empty() { return true; /* TODO */ }
#Override public void complement() {
// implement Set-specific behavior
}
};
private final Artist artistBehavior = new Artist() {
#Override public int getEgo() { return Integer.MAX_VALUE; /* TODO */ }
#Override public void complement() {
// implement Artist-specific behavior
}
};
Set asSet() {
return setBehavior;
}
Artist asArtist() {
return artistBehavior;
}
}
If I were passing this object to the HR department, I'd actually give it the value returned from asSet() to hire/fire the entire group.
If I were passing this object to the Theater for a performance, I'd actually give it the value returned from asArtist() to be treated as talent.
This works as long as YOU are in control of talking to the different components directly...
But I realize that your problem is a single third-party vendor has created a component, TheaterManager, that expects one object for both of these functions and it won't know about the asSet and asArtist methods. The problem is not with the vendors that created Set and Artist, it is the vendor that combined them instead of using a Visitor pattern or just specifying an interface that would mirror the asSet and asArtist methods I made above. If you can convince your one vendor "C" to fix that interface, your world will be a lot happier.
Good luck!
Dog, I have a strong feeling you are leaving out some details that are crucial to the solution. This often happens on SO because
people need to leave out a lot of details to get the question to a reasonable size and scope,
people do not fully understand the problem and the solution (which is why they are asking for help) so they cannot be sure which details are important and which are not, and
the reason the person cannot solve the problem on their own is because they do not understand the importance of this detail, which is the same reason they left it out.
I've said in another answer what I would do about ArtistSet. But keeping the above in mind I will give you another solution to a slightly different problem. Lets say I had code from a bad vendor:
package com.bad;
public interface IAlpha {
public String getName();
// Sort Alphabetically by Name
public int compareTo(IAlpha other);
}
This is bad because you should declare a function returning a Comparator<IAlpha> to implement the sorting strategy, but whatever. Now I get code from a worse company:
package com.worse;
import com.bad.IAlpha;
// an Alpha ordered by name length
public interface ISybil extends IAlpha, Comparable<IAlpha> {}
This is worse, because it is totally wrong, in that it overrides behavior incompatibly. An ISybil orders itself by name length, but an IAlpha orders itself alphabetically, except an ISybil is an IAlpha. They were mislead by the anti-pattern of IAlpha when they could and should have done something like:
public interface ISybil extends IAlpha {
public Comparator<IAlpha> getLengthComparator();
}
However, this situation is still much better than ArtistSet because here the expected behavior is documented. There is no confusion about what ISybil.compareTo() should do. So I would create classes as follows. A Sybil class that implements compareTo() as com.worse expects and delegates everything else:
package com.hack;
import com.bad.IAlpha;
import com.worse.ISybil;
public class Sybil implements ISybil {
private final Alpha delegate;
public Sybil(Alpha delegate) { this.delegate = delegate; }
public Alpha getAlpha() { return delegate; }
public String getName() { return delegate.getName(); }
public int compareTo(IAlpha other) {
return delegate.getName().length() - other.getName().length();
}
}
and an Alpha class that works exactly like com.bad said it should:
package com.hack;
import com.bad.IAlpha;
public class Alpha implements IAlpha {
private String name;
private final Sybil sybil;
public Alpha(String name) {
this.name = name;
this.sybil = new Sybil(this);
}
// Sort Alphabetically
public int compareTo(IAlpha other) {
return name.compareTo(other.getName());
}
public String getName() { return name; }
public Sybil getSybil() { return sybil; }
}
Note that I included type conversion methods: Alpha.getSybil() and Sybil.getAlpha(). This is so I could create my own wrappers around any com.worse vendor's methods that take or return Sybils so I can avoid polluting my code or any other vendor's code with com.worse's breakage. So if com.worse had:
public ISybil breakage(ISybil broken);
I could write a function
public Alpha safeDelegateBreakage(Alpha alpha) {
return breakage(alpha.getSybil).getAlpha();
}
and be done with it, except I would still complain vociferously to com.worse and politely to com.bad.
I have a method that's about ten lines of code. I want to create more methods that do exactly the same thing, except for a small calculation that's going to change one line of code. This is a perfect application for passing in a function pointer to replace that one line, but Java doesn't have function pointers. What's my best alternative?
Anonymous inner class
Say you want to have a function passed in with a String param that returns an int.
First you have to define an interface with the function as its only member, if you can't reuse an existing one.
interface StringFunction {
int func(String param);
}
A method that takes the pointer would just accept StringFunction instance like so:
public void takingMethod(StringFunction sf) {
int i = sf.func("my string");
// do whatever ...
}
And would be called like so:
ref.takingMethod(new StringFunction() {
public int func(String param) {
// body
}
});
EDIT: In Java 8, you could call it with a lambda expression:
ref.takingMethod(param -> bodyExpression);
For each "function pointer", I'd create a small functor class that implements your calculation.
Define an interface that all the classes will implement, and pass instances of those objects into your larger function. This is a combination of the "command pattern", and "strategy pattern".
#sblundy's example is good.
When there is a predefined number of different calculations you can do in that one line, using an enum is a quick, yet clear way to implement a strategy pattern.
public enum Operation {
PLUS {
public double calc(double a, double b) {
return a + b;
}
},
TIMES {
public double calc(double a, double b) {
return a * b;
}
}
...
public abstract double calc(double a, double b);
}
Obviously, the strategy method declaration, as well as exactly one instance of each implementation are all defined in a single class/file.
You need to create an interface that provides the function(s) that you want to pass around. eg:
/**
* A simple interface to wrap up a function of one argument.
*
* #author rcreswick
*
*/
public interface Function1<S, T> {
/**
* Evaluates this function on it's arguments.
*
* #param a The first argument.
* #return The result.
*/
public S eval(T a);
}
Then, when you need to pass a function, you can implement that interface:
List<Integer> result = CollectionUtilities.map(list,
new Function1<Integer, Integer>() {
#Override
public Integer eval(Integer a) {
return a * a;
}
});
Finally, the map function uses the passed in Function1 as follows:
public static <K,R,S,T> Map<K, R> zipWith(Function2<R,S,T> fn,
Map<K, S> m1, Map<K, T> m2, Map<K, R> results){
Set<K> keySet = new HashSet<K>();
keySet.addAll(m1.keySet());
keySet.addAll(m2.keySet());
results.clear();
for (K key : keySet) {
results.put(key, fn.eval(m1.get(key), m2.get(key)));
}
return results;
}
You can often use Runnable instead of your own interface if you don't need to pass in parameters, or you can use various other techniques to make the param count less "fixed" but it's usually a trade-off with type safety. (Or you can override the constructor for your function object to pass in the params that way.. there are lots of approaches, and some work better in certain circumstances.)
Method references using the :: operator
You can use method references in method arguments where the method accepts a functional interface. A functional interface is any interface that contains only one abstract method. (A functional interface may contain one or more default methods or static methods.)
IntBinaryOperator is a functional interface. Its abstract method, applyAsInt, accepts two ints as its parameters and returns an int. Math.max also accepts two ints and returns an int. In this example, A.method(Math::max); makes parameter.applyAsInt send its two input values to Math.max and return the result of that Math.max.
import java.util.function.IntBinaryOperator;
class A {
static void method(IntBinaryOperator parameter) {
int i = parameter.applyAsInt(7315, 89163);
System.out.println(i);
}
}
import java.lang.Math;
class B {
public static void main(String[] args) {
A.method(Math::max);
}
}
In general, you can use:
method1(Class1::method2);
instead of:
method1((arg1, arg2) -> Class1.method2(arg1, arg2));
which is short for:
method1(new Interface1() {
int method1(int arg1, int arg2) {
return Class1.method2(arg1, agr2);
}
});
For more information, see :: (double colon) operator in Java 8 and Java Language Specification §15.13.
You can also do this (which in some RARE occasions makes sense). The issue (and it is a big issue) is that you lose all the typesafety of using a class/interface and you have to deal with the case where the method does not exist.
It does have the "benefit" that you can ignore access restrictions and call private methods (not shown in the example, but you can call methods that the compiler would normally not let you call).
Again, it is a rare case that this makes sense, but on those occasions it is a nice tool to have.
import java.lang.reflect.InvocationTargetException;
import java.lang.reflect.Method;
class Main
{
public static void main(final String[] argv)
throws NoSuchMethodException,
IllegalAccessException,
IllegalArgumentException,
InvocationTargetException
{
final String methodName;
final Method method;
final Main main;
main = new Main();
if(argv.length == 0)
{
methodName = "foo";
}
else
{
methodName = "bar";
}
method = Main.class.getDeclaredMethod(methodName, int.class);
main.car(method, 42);
}
private void foo(final int x)
{
System.out.println("foo: " + x);
}
private void bar(final int x)
{
System.out.println("bar: " + x);
}
private void car(final Method method,
final int val)
throws IllegalAccessException,
IllegalArgumentException,
InvocationTargetException
{
method.invoke(this, val);
}
}
If you have just one line which is different you could add a parameter such as a flag and a if(flag) statement which calls one line or the other.
You may also be interested to hear about work going on for Java 7 involving closures:
What’s the current state of closures in Java?
http://gafter.blogspot.com/2006/08/closures-for-java.html
http://tech.puredanger.com/java7/#closures
New Java 8 Functional Interfaces and Method References using the :: operator.
Java 8 is able to maintain method references ( MyClass::new ) with "# Functional Interface" pointers. There are no need for same method name, only same method signature required.
Example:
#FunctionalInterface
interface CallbackHandler{
public void onClick();
}
public class MyClass{
public void doClick1(){System.out.println("doClick1");;}
public void doClick2(){System.out.println("doClick2");}
public CallbackHandler mClickListener = this::doClick;
public static void main(String[] args) {
MyClass myObjectInstance = new MyClass();
CallbackHandler pointer = myObjectInstance::doClick1;
Runnable pointer2 = myObjectInstance::doClick2;
pointer.onClick();
pointer2.run();
}
}
So, what we have here?
Functional Interface - this is interface, annotated or not with #FunctionalInterface, which contains only one method declaration.
Method References - this is just special syntax, looks like this, objectInstance::methodName, nothing more nothing less.
Usage example - just an assignment operator and then interface method call.
YOU SHOULD USE FUNCTIONAL INTERFACES FOR LISTENERS ONLY AND ONLY FOR THAT!
Because all other such function pointers are really bad for code readability and for ability to understand. However, direct method references sometimes come handy, with foreach for example.
There are several predefined Functional Interfaces:
Runnable -> void run( );
Supplier<T> -> T get( );
Consumer<T> -> void accept(T);
Predicate<T> -> boolean test(T);
UnaryOperator<T> -> T apply(T);
BinaryOperator<T,U,R> -> R apply(T, U);
Function<T,R> -> R apply(T);
BiFunction<T,U,R> -> R apply(T, U);
//... and some more of it ...
Callable<V> -> V call() throws Exception;
Readable -> int read(CharBuffer) throws IOException;
AutoCloseable -> void close() throws Exception;
Iterable<T> -> Iterator<T> iterator();
Comparable<T> -> int compareTo(T);
Comparator<T> -> int compare(T,T);
For earlier Java versions you should try Guava Libraries, which has similar functionality, and syntax, as Adrian Petrescu has mentioned above.
For additional research look at Java 8 Cheatsheet
and thanks to The Guy with The Hat for the Java Language Specification §15.13 link.
#sblundy's answer is great, but anonymous inner classes have two small flaws, the primary being that they tend not to be reusable and the secondary is a bulky syntax.
The nice thing is that his pattern expands into full classes without any change in the main class (the one performing the calculations).
When you instantiate a new class you can pass parameters into that class which can act as constants in your equation--so if one of your inner classes look like this:
f(x,y)=x*y
but sometimes you need one that is:
f(x,y)=x*y*2
and maybe a third that is:
f(x,y)=x*y/2
rather than making two anonymous inner classes or adding a "passthrough" parameter, you can make a single ACTUAL class that you instantiate as:
InnerFunc f=new InnerFunc(1.0);// for the first
calculateUsing(f);
f=new InnerFunc(2.0);// for the second
calculateUsing(f);
f=new InnerFunc(0.5);// for the third
calculateUsing(f);
It would simply store the constant in the class and use it in the method specified in the interface.
In fact, if KNOW that your function won't be stored/reused, you could do this:
InnerFunc f=new InnerFunc(1.0);// for the first
calculateUsing(f);
f.setConstant(2.0);
calculateUsing(f);
f.setConstant(0.5);
calculateUsing(f);
But immutable classes are safer--I can't come up with a justification to make a class like this mutable.
I really only post this because I cringe whenever I hear anonymous inner class--I've seen a lot of redundant code that was "Required" because the first thing the programmer did was go anonymous when he should have used an actual class and never rethought his decision.
The Google Guava libraries, which are becoming very popular, have a generic Function and Predicate object that they have worked into many parts of their API.
One of the things I really miss when programming in Java is function callbacks. One situation where the need for these kept presenting itself was in recursively processing hierarchies where you want to perform some specific action for each item. Like walking a directory tree, or processing a data structure. The minimalist inside me hates having to define an interface and then an implementation for each specific case.
One day I found myself wondering why not? We have method pointers - the Method object. With optimizing JIT compilers, reflective invocation really doesn't carry a huge performance penalty anymore. And besides next to, say, copying a file from one location to another, the cost of the reflected method invocation pales into insignificance.
As I thought more about it, I realized that a callback in the OOP paradigm requires binding an object and a method together - enter the Callback object.
Check out my reflection based solution for Callbacks in Java. Free for any use.
Sounds like a strategy pattern to me. Check out fluffycat.com Java patterns.
oK, this thread is already old enough, so very probably my answer is not helpful for the question. But since this thread helped me to find my solution, I'll put it out here anyway.
I needed to use a variable static method with known input and known output (both double). So then, knowing the method package and name, I could work as follows:
java.lang.reflect.Method Function = Class.forName(String classPath).getMethod(String method, Class[] params);
for a function that accepts one double as a parameter.
So, in my concrete situation I initialized it with
java.lang.reflect.Method Function = Class.forName("be.qan.NN.ActivationFunctions").getMethod("sigmoid", double.class);
and invoked it later in a more complex situation with
return (java.lang.Double)this.Function.invoke(null, args);
java.lang.Object[] args = new java.lang.Object[] {activity};
someOtherFunction() + 234 + (java.lang.Double)Function.invoke(null, args);
where activity is an arbitrary double value. I am thinking of maybe doing this a bit more abstract and generalizing it, as SoftwareMonkey has done, but currently I am happy enough with the way it is. Three lines of code, no classes and interfaces necessary, that's not too bad.
To do the same thing without interfaces for an array of functions:
class NameFuncPair
{
public String name; // name each func
void f(String x) {} // stub gets overridden
public NameFuncPair(String myName) { this.name = myName; }
}
public class ArrayOfFunctions
{
public static void main(String[] args)
{
final A a = new A();
final B b = new B();
NameFuncPair[] fArray = new NameFuncPair[]
{
new NameFuncPair("A") { #Override void f(String x) { a.g(x); } },
new NameFuncPair("B") { #Override void f(String x) { b.h(x); } },
};
// Go through the whole func list and run the func named "B"
for (NameFuncPair fInstance : fArray)
{
if (fInstance.name.equals("B"))
{
fInstance.f(fInstance.name + "(some args)");
}
}
}
}
class A { void g(String args) { System.out.println(args); } }
class B { void h(String args) { System.out.println(args); } }
Check out lambdaj
http://code.google.com/p/lambdaj/
and in particular its new closure feature
http://code.google.com/p/lambdaj/wiki/Closures
and you will find a very readable way to define closure or function pointer without creating meaningless interface or use ugly inner classes
Wow, why not just create a Delegate class which is not all that hard given that I already did for java and use it to pass in parameter where T is return type. I am sorry but as a C++/C# programmer in general just learning java, I need function pointers because they are very handy. If you are familiar with any class which deals with Method Information you can do it. In java libraries that would be java.lang.reflect.method.
If you always use an interface, you always have to implement it. In eventhandling there really isn't a better way around registering/unregistering from the list of handlers but for delegates where you need to pass in functions and not the value type, making a delegate class to handle it for outclasses an interface.
None of the Java 8 answers have given a full, cohesive example, so here it comes.
Declare the method that accepts the "function pointer" as follows:
void doCalculation(Function<Integer, String> calculation, int parameter) {
final String result = calculation.apply(parameter);
}
Call it by providing the function with a lambda expression:
doCalculation((i) -> i.toString(), 2);
If anyone is struggling to pass a function that takes one set of parameters to define its behavior but another set of parameters on which to execute, like Scheme's:
(define (function scalar1 scalar2)
(lambda (x) (* x scalar1 scalar2)))
see Pass Function with Parameter-Defined Behavior in Java
Since Java8, you can use lambdas, which also have libraries in the official SE 8 API.
Usage:
You need to use a interface with only one abstract method.
Make an instance of it (you may want to use the one java SE 8 already provided) like this:
Function<InputType, OutputType> functionname = (inputvariablename) {
...
return outputinstance;
}
For more information checkout the documentation: https://docs.oracle.com/javase/tutorial/java/javaOO/lambdaexpressions.html
Prior to Java 8, nearest substitute for function-pointer-like functionality was an anonymous class. For example:
Collections.sort(list, new Comparator<CustomClass>(){
public int compare(CustomClass a, CustomClass b)
{
// Logic to compare objects of class CustomClass which returns int as per contract.
}
});
But now in Java 8 we have a very neat alternative known as lambda expression, which can be used as:
list.sort((a, b) -> { a.isBiggerThan(b) } );
where isBiggerThan is a method in CustomClass. We can also use method references here:
list.sort(MyClass::isBiggerThan);
The open source safety-mirror project generalizes some of the above mentioned solutions into a library that adds functions, delegates and events to Java.
See the README, or this stackoverflow answer, for a cheat sheet of features.
As for functions, the library introduces a Fun interface, and some sub-interfaces that (together with generics) make up a fluent API for using methods as types.
Fun.With0Params<String> myFunctionField = " hello world "::trim;`
Fun.With2Params<Boolean, Object, Object> equals = Objects::equals;`
public void foo(Fun.With1ParamAndVoid<String> printer) throws Exception {
printer.invoke("hello world);
}
public void test(){
foo(System.out::println);
}
Notice:
that you must choose the sub-interface that matches the number of parameters in the signature you are targeting. Fx, if it has one parameter, choose Fun.With1Param.
that Generics are used to define A) the return type and B) the parameters of the signature.
Also, notice that the signature of the Method Reference passed to the call to the foo() method must match the the Fun defined by method Foo. If it do not, the compiler will emit an error.