Search code examples
javacollectionssethashset

Set.of(E... elements) - which Set implementation is it using? and what is its relation to new HashSet<>()?


I wanted to ask - what is the difference between

Set<CurrencyType> set1 = new HashSet<>() {{
    add("a");
    add("b");
    add("c");
}}

and

Set<CurrencyType> set2 = Set.of(
    "a",
    "b",
    "c"
)

In debug mode in a @Test named MySillyTest (this name will come back very soon) i can see that set1 is an instance of MySillyTest$1, but I assume it is just a HashSet. set2 on the other hand is an instance of ImmutableCollections$SetN. What is the real difference between those two? What implementation of java.util.Set is Set.of() using? Is there a performance/memory usage/cpu usage difference between those two?


Solution

  • Don’t know, don’t care

    The concrete class used by Set.of (and List.of, Map.of) is not documented. All we know is that the object returned (a) implements the interface, and (b) is unmodifiable.

    That is all we need to know. We should not care about the particular concrete class used under the covers.

    Being of unknown concrete class gives freedom to the implementors of the of methods.

    • Those programmers are free to optimize according to the nature of your arguments. For example, if you are passing enum arguments, the highly optimized EnumSet class might be used behind the scenes.
    • Those programmers are free to change their concrete class between versions of Java. For example, Java 17 implementation might return one concrete class while Java 18 returns another.

    So you should never depend on a particular concrete class being utilized by the of/copyOf methods.

    You asked:

    What is the real difference between those two?

    In your first one, we know the concrete class. And the resulting set is modifiable.

    In your second one, we do not know the concrete class, nor do we care about the concrete class. And the resulting set is unmodifiable.

    Code Concrete Class Modifiable
    new HashSet<>() {{ add("a"); add("b"); add("c"); }} known modifiable
    Set.of( "a", "b", "c" unknown unmodifiable

    Avoid double-brace

    As others said, it’s generally best to avoid double-brace initialization.

    If you want the convenience of compact literals-style initialization of your modifiable collection, combine with the of methods. You can pass an existing collection to the constructor.

    Set< String > set =
        new HashSet<>(
            Set.of( "a", "b", "c" )  // Pass a collection to constructor. 
        )
    ;
    

    “But I do care …”

    If you do care about the specific details of the implementation, then you should not use Set.of. The implementation details may vary as discussed above, so you should not rely on a specific implementation with Set.of.

    When your project has specific needs, you should explicitly choose a particular implementation that meets those needs. You can choose one of the bundled implementations, obtain one from a third-party library such as Eclipse Collections or Google Guava, or write your own.

    Tip: You can leverage the convenient syntax of Set.of by passing its result to the constructor of your chosen implementation. See the code example above:

    new HashSet<>( Set.of( … ) )