Hi,
On 11/02/2017 12:51 AM, Stuart Marks wrote:
On 11/1/17 10:45 AM, Tagir Valeev wrote:
Set.of:
+ if (coll instanceof
ImmutableCollections.AbstractImmutableSet) {
+ return (Set<E>)coll;
+ } else {
+ return (Set<E>)Set.of(coll.stream().distinct().toArray());
I think that good old Set.of(new HashSet<>(coll).toArray()) would
produce less garbage. distinct() also maintains HashSet internally,
but it removes the SIZED characteristic, so instead of preallocated
array you will have a SpinedBuffer which is less efficient than
AbstractCollection.toArray() implementation which just allocates the
array of exact size. What do you think?
Oh yes, good point. I had initially used stream().distinct() because I
wanted to use distinct()'s semantics of preserving the first equal
element among duplicates. But since I removed that requirement from
the spec, using a HashSet as you suggest is much simpler.
Why not using:
coll.stream().collect(Collectors.toImmutableSet())
As Collectors.toImmutableSet() is currently implemented, with serial
Stream it will create a single HashSet, add all the elements to it and
call Set.of(HashSet.toArray()) with it. Pretty much the same as what
Tagir proposes, but the Collector could be made more efficient in the
future and with it, the optimization would automatically extend to
Set.copyOf()...
Regards, Peter
