Why does Set.of disallow duplicate elements?

Stuart Marks stuart.marks at oracle.com
Tue Feb 2 00:37:41 UTC 2021


Indeed it's the case that a varargs method can't determine whether it was called 
with several explicit arguments or whether it was called with an array. However, 
that doesn't have any bearing on whether or not Set.of rejects duplicates.

The model for Set.of is to support a collection-literal-like syntax where the 
programmer can write an arbitrary number of elements in the source code for 
inclusion in the set. Here's an example (though it uses Map.ofEntries instead of 
Set.of, the same rationale applies):


Map<String, TokenType> tokens = Map.ofEntries(
     entry("@",     AT),
     entry("|",     VERTICAL_BAR),
     entry("#",     HASH),
     entry("%",     PERCENT),
     entry(":",     COLON),
     entry("^",     CARET),
     entry("&",     AMPERSAND),
     entry("|",     EXCLAM),
     entry("?",     QUESTION),
     entry("$",     DOLLAR),
     entry("::",    PAAMAYIM_NEKUDOTAYIM),
     entry("=",     EQUALS),
     entry(";",     SEMICOLON)
);


This errors out instead of silently dropping one of the entries.

As an optimization, the API provides several fixed-arg overloads of Set.of. With few 
arguments, the fixed-arg methods are called. If more arguments are added, at a 
certain point it transparently switches to the varargs form. "Transparently" means 
that you can't tell (without counting the arguments) whether a fixed-arg or varargs 
form of Set.of will be called. You don't want the duplicate rejection semantics to 
change if you add or remove an argument that happens to cross the fixed/varargs 
threshold. Thus, Set.of rejects duplicates, whether in fixed or varargs form.

Set.copyOf(Arrays.asList(...)) is the best way to deduplicate an explicit list of 
elements into a set.

s'marks




On 2/1/21 3:01 PM, Aaron Scott-Boddendijk wrote:
>   Dave,
> 
> || Dave said...
> || Okay, I understand this reasoning, but when you want to construct a Set
> from an array, you might be tempted to use Set.of(...) because it looks
> like it supports an
> || array and indeed, you can do Set.of(new int[] {1, 2 }) I believe?
> ||
> || Maybe this is just a quirk because of how varargs work.
> 
> | Rémi said...
> | Set.of(int[]) will call Set.of(E) with E being an int[].
> | but
> | Set.of(new Integer[] { ... }) calls Set.of(...).
> |
> | Yes, exactly, it's a known issue with varargs, you have no way to say, i
> don't want this varargs to be called with an array.
> 
> I think the confusion is the interaction of boxing and varargs.
> 
>> List<Integer> list = List.of(1, 2);
> 
> is actually, once auto-boxing is applied by the compiler, executed as...
> 
>> List<Integer> list = List.of(Integer.valueOf(1), Integer.valueOf(2));
> 
> So the equivalent explicit array form should use `Integer[]` not `int[]`...
> 
>> Integer[] numbers = new Integer[] {1, 2};
>> List<Integer> list = List.of(numbers);
> 
> Interestingly, if you actually wanted a `List<Integer[]>` you would then
> need to say
> 
>> Integer[] numbers = new Integer[] {1, 2};
>> List<Integer> list = List.<Integer[]>of(numbers);
> 
> Which is explicitly telling the compiler what the type arguments are for
> this invocation of the generic method 'of'' (rather than allowing it to use
> type-inference)
> 
> Regarding the use of `Set.copyOf(Arrays.asList(...))`. I do wonder about
> improving the ceremony (because I agree that we want an obvious way of
> getting immutable Sets from non-unique inputs) by following the pattern
> presented in Optional (`Optional.of` and `Optional.ofNullable`) and
> providing `Set.of` and `Set.ofMaybeUnique` (better name needed -
> 'ofOptionallyUnique'?) - to which the implementation could just be
> `Set.copyOf(Arrays.asList(args))` (unless a more efficient path proves
> valuable).
> 
> `Arrays.asList(...array...)` is not all that expensive. It is _not_ an
> ArrayList but a much simpler type with rather trivial implementations for
> most methods (and 'always throws' implementations for methods that are
> unsupported). So not only does it mean that there's no copying occuring to
> make the list but it's even possible that JIT manages enough specialisation
> and inlining to elide the allocation entirely (though in practice this
> doesn't occur as often as we might like).
> 
> --
> Aaron Scott-Boddendijk
> 
> On Mon, Feb 1, 2021 at 10:35 AM <forax at univ-mlv.fr> wrote:
> 
>>
>>
>>
>>
>>
>> De: "dfranken jdk" <dfranken.jdk at gmail.com>
>> À: "Remi Forax" <forax at univ-mlv.fr>
>> Cc: "core-libs-dev" <core-libs-dev at openjdk.java.net>
>> Envoyé: Dimanche 31 Janvier 2021 13:54:44
>> Objet: Re: Why does Set.of disallow duplicate elements?
>>
>>
>>
>>
>> BQ_BEGIN
>>
>> Okay, I understand this reasoning, but when you want to construct a Set
>> from an array, you might be tempted to use Set.of(...) because it looks
>> like it supports an array and indeed, you can do Set.of(new int[] {1, 2 })
>> I believe?
>>
>> BQ_END
>>
>> Set.of(int[]) will call Set.of(E) with E being an int[].
>> but
>> Set.of(new Integer[] { ... }) calls Set.of(...).
>>
>>
>> BQ_BEGIN
>>
>>
>> Maybe this is just a quirk because of how varargs work.
>>
>> BQ_END
>>
>> Yes, exactly, it's a known issue with varargs, you have no way to say, i
>> don't want this varargs to be called with an array.
>>
>>
>> BQ_BEGIN
>>
>>
>> I wondered if there was a canonical way to create a Set from an array, but
>> couldn't find it, maybe I am missing something?
>> I did notice Arrays.asList exists (which makes sense because it creates an
>> ArrayList backed by the array), but not Arrays.asSet.
>>
>> BQ_END
>>
>> asList() reuse the same backing array, you can not do that with asSet() or
>> contains() will be in O(n) in the worst case.
>>
>>
>> BQ_BEGIN
>>
>>
>> So the way I would create a Set from an array would be either
>> Arrays.stream(myArr).collect(Collectors.toUnmodifiableSet()) or new
>> HashSet<>(Arrays.asList(myArray)) or Set.copyOf(Arrays.asList(myArray)).
>>
>> BQ_END
>>
>> yes, the last one is the easy way to create an unmodifiable set from an
>> array.
>>
>>
>> BQ_BEGIN
>>
>>
>> I'm not saying the way it is currently implemented is wrong, it's just
>> something which can suprise developers as it surprised me. :)
>>
>> BQ_END
>>
>> Arrays are currently second class citizen in Java, because they are always
>> modifiable and always covariant (String[] can be seen as a Object[]).
>> We have talked several times to introduce new variants of arrays,
>> non-modifiable one, non-covariant one, etc under the name Array 2.0, but
>> Valhalla generics is a blocker for that project.
>> Once Valhalla is done, it may be a follow up.
>>
>>
>> BQ_BEGIN
>>
>>
>> Kind regards,
>>
>> Dave
>>
>> BQ_END
>>
>>
>> regards,
>> Rémi
>>
>>
>> BQ_BEGIN
>>
>>
>>
>> Op za 30 jan. 2021 om 21:30 schreef Remi Forax < [ mailto:
>> forax at univ-mlv.fr | forax at univ-mlv.fr ] >:
>>
>> BQ_BEGIN
>> Set.of() is the closest way we've got to a literal Set without having
>> introduced a special syntax for that in the language.
>>
>> The idea is that if you conceptually want to write
>> Set<String> set = { "hello", "world" };
>> instead, you write
>> Set<String> set = Set.of("hello", "world");
>>
>> In that context, it makes sense to reject Set constructed with the same
>> element twice because this is usually a programming error.
>> So
>> Set.of("hello", "hello")
>> throws an IAE.
>>
>> If you want a Set from a collection of elements, you can use
>> Set.copyOf(List.of("hello", "hello"))
>>
>> regards,
>> Rémi
>>
>> ----- Mail original -----
>>> De: "dfranken jdk" < [ mailto:dfranken.jdk at gmail.com |
>> dfranken.jdk at gmail.com ] >
>>> À: "core-libs-dev" < [ mailto:core-libs-dev at openjdk.java.net |
>> core-libs-dev at openjdk.java.net ] >
>>> Envoyé: Samedi 30 Janvier 2021 19:30:06
>>> Objet: Why does Set.of disallow duplicate elements?
>>
>>> Dear users,
>>>
>>> While looking at the implementation of Set.of(...) I noticed that
>>> duplicate elements are not allowed, e.g. Set.of(1, 1) will throw an
>>> IllegalArgumentException. Why has it been decided to do this?
>>>
>>> My expectation was that duplicates would simply be removed.
>>>
>>> If I do for instance new HashSet<>(<collection containing duplicates>)
>>> it works and duplicates are removed. To me, it looks a bit inconsistent
>>> to have duplicates removed for a collection passed in the constructor,
>>> but not for a collection (even though it is a vararg array) passed to a
>>> static factory method.
>>>
>>> Kind regards,
>>>
>>> Dave Franken
>>
>> BQ_END
>>
>>
>> BQ_END
>>
>>
>>


More information about the core-libs-dev mailing list