是的，使用HashMap…但是以一种专门的方式:我预见到试图使用HashMap作为伪集的陷阱是Map/Set的“实际”元素和“候选”元素之间可能的混淆，即用于测试是否已经存在相等元素的元素。这不是万无一失的方法，但能让你远离陷阱:

class SelfMappingHashMap<V> extends HashMap<V, V>{
    @Override
    public String toString(){
        // otherwise you get lots of "... object1=object1, object2=object2..." stuff
        return keySet().toString();
    }

    @Override
    public V get( Object key ){
        throw new UnsupportedOperationException( "use tryToGetRealFromCandidate()");
    }

    @Override
    public V put( V key, V value ){
       // thorny issue here: if you were indavertently to `put`
       // a "candidate instance" with the element already in the `Map/Set`: 
       // these will obviously be considered equivalent 
       assert key.equals( value );
       return super.put( key, value );
    }

    public V tryToGetRealFromCandidate( V key ){
        return super.get(key);
    }
}

然后这样做:

SelfMappingHashMap<SomeClass> selfMap = new SelfMappingHashMap<SomeClass>();
...
SomeClass candidate = new SomeClass();
if( selfMap.contains( candidate ) ){
    SomeClass realThing = selfMap.tryToGetRealFromCandidate( candidate );
    ...
    realThing.useInSomeWay()...
}

但是…你现在希望候选对象以某种方式自毁，除非程序员立即将其放入Map/Set…你会希望包含“玷污”候选对象，除非它加入Map，否则任何对它的使用都会使它“被诅咒”。也许你可以让SomeClass实现一个新的Taintable接口。

更令人满意的解决方案是GettableSet，如下所示。然而，要做到这一点，你必须负责SomeClass的设计，以使所有构造函数都不可见(或者…能够并且愿意为它设计和使用包装类):

public interface NoVisibleConstructor {
    // again, this is a "nudge" technique, in the sense that there is no known method of 
    // making an interface enforce "no visible constructor" in its implementing classes 
    // - of course when Java finally implements full multiple inheritance some reflection 
    // technique might be used...
    NoVisibleConstructor addOrGetExisting( GettableSet<? extends NoVisibleConstructor> gettableSet );
};

public interface GettableSet<V extends NoVisibleConstructor> extends Set<V> {
    V getGenuineFromImpostor( V impostor ); // see below for naming
}

实现:

public class GettableHashSet<V extends NoVisibleConstructor> implements GettableSet<V> {
    private Map<V, V> map = new HashMap<V, V>();

    @Override
    public V getGenuineFromImpostor(V impostor ) {
        return map.get( impostor );
    }

    @Override
    public int size() {
        return map.size();
    }

    @Override
    public boolean contains(Object o) {
        return map.containsKey( o );
    }

    @Override
    public boolean add(V e) {
        assert e != null;
        V result = map.put( e,  e );
        return result != null;
    }

    @Override
    public boolean remove(Object o) {
        V result = map.remove( o );
        return result != null;
    }

    @Override
    public boolean addAll(Collection<? extends V> c) {
        // for example:
        throw new UnsupportedOperationException();
    }

    @Override
    public void clear() {
        map.clear();
    }

    // implement the other methods from Set ...
}

你的NoVisibleConstructor类看起来是这样的:

class SomeClass implements NoVisibleConstructor {

    private SomeClass( Object param1, Object param2 ){
        // ...
    }

    static SomeClass getOrCreate( GettableSet<SomeClass> gettableSet, Object param1, Object param2 ) {
        SomeClass candidate = new SomeClass( param1, param2 );
        if (gettableSet.contains(candidate)) {
            // obviously this then means that the candidate "fails" (or is revealed
            // to be an "impostor" if you will).  Return the existing element:
            return gettableSet.getGenuineFromImpostor(candidate);
        }
        gettableSet.add( candidate );
        return candidate;
    }

    @Override
    public NoVisibleConstructor addOrGetExisting( GettableSet<? extends NoVisibleConstructor> gettableSet ){
       // more elegant implementation-hiding: see below
    }
}

PS这样的NoVisibleConstructor类的一个技术问题:它可能会被反对，这样的类本质上是final的，这可能是不可取的。实际上，你总是可以添加一个虚拟的无参数保护构造函数:

protected SomeClass(){
    throw new UnsupportedOperationException();
}

．.． 这样至少可以让一个子类编译。然后，您必须考虑是否需要在子类中包含另一个getOrCreate()工厂方法。

最后一步是为你的集合成员创建一个抽象基类(注意“element”表示列表，“member”表示集合)，就像这样(如果可能的话——同样，使用包装器类时，类不在你的控制之下，或者已经有一个基类，等等)，以最大限度地隐藏实现:

public abstract class AbstractSetMember implements NoVisibleConstructor {
    @Override
    public NoVisibleConstructor
            addOrGetExisting(GettableSet<? extends NoVisibleConstructor> gettableSet) {
        AbstractSetMember member = this;
        @SuppressWarnings("unchecked") // unavoidable!
        GettableSet<AbstractSetMembers> set = (GettableSet<AbstractSetMember>) gettableSet;
        if (gettableSet.contains( member )) {
            member = set.getGenuineFromImpostor( member );
            cleanUpAfterFindingGenuine( set );
        } else {
            addNewToSet( set );
        }
        return member;
    }

    abstract public void addNewToSet(GettableSet<? extends AbstractSetMember> gettableSet );
    abstract public void cleanUpAfterFindingGenuine(GettableSet<? extends AbstractSetMember> gettableSet );
}

．.． 用法是相当明显的(在你的SomeClass的静态工厂方法中):

SomeClass setMember = new SomeClass( param1, param2 ).addOrGetExisting( set );

要准确回答“为什么Set不提供获取等于另一个元素的元素的操作?”这个问题，答案可能是:因为集合框架的设计者不是很有前瞻性。他们没有预料到您非常合理的用例，天真地试图“建模数学集合抽象”(从javadoc)，只是忘记添加有用的get()方法。

现在回到隐含的问题“如何获得元素”:我认为最好的解决方案是使用Map<E,E>而不是Set<E>来将元素映射到它们自己。通过这种方式，您可以有效地从“set”中检索元素，因为Map的get()方法将使用有效的哈希表或树算法查找元素。如果愿意，可以编写自己的Set实现，提供额外的get()方法，封装Map。

以下答案在我看来是错误的:

“您不需要获取元素，因为您已经有了一个相等的对象”:断言是错误的，正如您在问题中已经表明的那样。两个相等的对象仍然可以具有与对象相等无关的不同状态。目标是访问Set中包含的元素的状态，而不是用作“查询”的对象的状态。

"You have no other option but to use the iterator": that is a linear search over a collection which is totally inefficient for large sets (ironically, internally the Set is organized as hash map or tree that could be queried efficiently). Don't do it! I have seen severe performance problems in real-life systems by using that approach. In my opinion what is terrible about the missing get() method is not so much that it is a bit cumbersome to work around it, but that most programmers will use the linear search approach without thinking of the implications.

因为Set的任何特定实现都可能是随机访问，也可能不是。

You can always get an iterator and step through the Set, using the iterators' next() method to return the result you want once you find the equal element. This works regardless of the implementation. If the implementation is NOT random access (picture a linked-list backed Set), a get(E element) method in the interface would be deceptive, since it would have to iterate the collection to find the element to return, and a get(E element) would seem to imply this would be necessary, that the Set could jump directly to the element to get.

当然，Contains()可能需要做同样的事情，也可能不需要，这取决于实现，但名称似乎不会导致同样的误解。

如果你看一下java.util.HashSet实现的前几行，你会看到:

public class HashSet<E>
    ....
    private transient HashMap<E,Object> map;

HashSet在内部使用HashMap，这意味着如果你直接使用HashMap并使用与键相同的值，你会得到你想要的效果，并节省一些内存。

我在那里做过!!如果你正在使用番石榴，一个快速的方法将它转换为地图是:

Map<Integer,Foo> map = Maps.uniqueIndex(fooSet, Foo::getKey);

如果你有一个相等的对象，为什么你需要集合中的一个?如果它仅与键“相等”，则Map将是更好的选择。

不管怎样，下面的方法就可以了:

Foo getEqual(Foo sample, Set<Foo> all) {
  for (Foo one : all) {
    if (one.equals(sample)) {
      return one;
    }
  } 
  return null;
}

在Java 8中，这可以变成一行代码:

return all.stream().filter(sample::equals).findAny().orElse(null);