Hello everyone! I was thinking of specializing `(m...

# kontributors

[Daniel Karnaukh]

Hello everyone! I was thinking of specializing (mutable)setOf using reified genericsto account for enum sets, which are faster than your average HashSet for enums. However, (mutable)setOf stipulates that the returned set maintains insertion order, so I quickly wrote up a LinkedEnumSet and benchmarked it:

BenchEnum.addEnum             avgt   25   1.643 ± 0.135  ns/op
BenchEnum.addEnumSet          avgt   25   1.475 ± 0.050  ns/op
BenchEnum.addMutableSet       avgt   25  10.844 ± 0.687  ns/op

BenchEnum.containsEnum        avgt   25   1.354 ± 0.015  ns/op
BenchEnum.containsEnumSet     avgt   25   1.260 ± 0.044  ns/op
BenchEnum.containsMutableSet  avgt   25   2.800 ± 0.348  ns/op

BenchEnum.iterateEnum        avgt   25  27.879 ± 0.350  ns/op
BenchEnum.iterateEnumSet     avgt   25  10.458 ± 0.105  ns/op
BenchEnum.iterateMutableSet  avgt   25   9.721 ± 0.097  ns/op

BenchEnum.removeEnum          avgt   25   1.573 ± 0.140  ns/op
BenchEnum.removeEnumSet       avgt   25   1.442 ± 0.091  ns/op
BenchEnum.removeMutableSet    avgt   25   3.367 ± 0.803  ns/op

In here, *Enum is the LinkedEnumSet, *EnumSet is the java.util.EnumSet, and *MutableSet is the set returned by toMutableSet. As you can see, LinkedEnumSet is significantly faster than MutableSet except in the case of iteration, which I have no idea why. Does my idea to specialize (mutable)setOf (and the (mutable)mapOf possibly too) have any merit? Benchmark source can be found here.